Hydrological Investigation of Swat River Basin Using GIS, Remote Sensing and Snowmelt Runoff Modeling

Zakir Hussain Dahri

Thesis Submitted for the Degree of Master of Applied Science (Geographic Information Systems)

Department of Geomatics Faculty of Engineering

The University of Melbourne

April, 2008

DECLARATION

This is to certify that (i) The thesis comprises only my original work, (ii) Due acknowledgement has been made in the text to all other material used, and (iii) The thesis is approximately 20,300 words in length exclusive of tables, maps, and references.

April, 2008

Zakir Hussain Dahri

DEDICATION

To my beloved mother to whom I am most inspired in my life and my personality is largely her reflection

and

To my beloved late son who I lost at the age of less than two and half years during the course of this study

ii

ABSTRACT
The snowcover and glaciers of HKH region of Pakistan are one of the largest repositories of inland cryosphere outside Polar Regions and obviously the lifeline of Pakistani people. However, reliable estimates of the snow area extent and snowmelt runoff have been lacking in this largely inaccessible and data sparse region. The study utilized GIS, RS and hydrological modeling techniques to evaluate the distribution of snowcover, estimated snowmelt runoff and statistically related both these variables. A very high variability of snowcover and associated snowmelt runoff during the entire calendar year is observed. Snowfall usually starts abruptly in September and October months but the following four main winter months (Nov Feb) generally bring in most of the snowfall and snowcover is increased from less than 2 % in August to about 64 % by the end of January or in early February. Snowmelt generally continues throughout the year but contribution of winter snowmelt runoff is often very low. Unlike snowfall, snowmelt runoff usually progresses gradually and smoothly and is more easily predictable. The summer snowmelt normally gets momentum in March and increases linearly from around 30 60 m3/sec to 400 760 m3/sec in late June or early July. It declines gradually thereafter reducing to 30 - 50 m3/sec in December. The Dec Feb snowmelt runoff normally tends to remain same. The results reveal Swat river basin of Pakistan as predominantly a snow-fed as the annual snowmelt runoff contribution to the total runoff may ranges from 65 75 %. The study observes a definite response of observed river discharges and simulated snowmelt runoff to seasonal snowcover changes, i.e. an association of low stream flows with high snow area extent during the winter season (Sep Feb), an increase in discharge associated with a decrease of snow area extent during the early summer (Mar Jun), and decrease in discharge with decreasing snowcover in the late summer, monsoon season (Jul mid Sep). It employs the daily records of snowcover and relates them with the daily river discharges and snowmelt runoff and also develops prediction model for the total runoff volume of the four main summer months (May Aug).

iii

ACKNOWLEDGEMENTS

I have tremendous appreciation and gratitude to my research advisor Dr. Joseph H. Leach, Department of Geomatics, for his invaluable guidance, constructive ideas, positive criticism and constant encouragement during the course of this research study. I am highly indebted to thank AusAID for sponsoring me and also to my parent department PARC for allowing me to avail this opportunity. I am extremely grateful to my friend Faisal Masood Qureshi, PhD scholar at the Department of Geomatics, for his valuable and constant help especially in GIS and RS related issues and also for a wonderful company throughout my stay here in Melbourne. Sincere thanks are also due to my friend and colleague Dr Bashir Ahmad for arranging met and flow data and also for his invaluable guidance and support. Due acknowledgement is extended to NASAs NSIDC, WAPDA-Pakistan, and PMD for free distribution of valuable data The support offered by the Department of Geomatics at the University of Melbourne is also duly acknowledged. The competence of faculty and friendliness of staff has truly complemented this academic experience. Finally, I owe my deepest gratitude to my parents, brothers, sisters and other close relatives for their constant support and sacrifice. The role and courage of my wife is unforgettable especially when we lost our beloved son during my stay here. She has been splendidly brave, always ready for sacrifice and proved to be a devoted life partner.

iv

CONTENTS
DECLARATION DEDICATION ABSTRACT ACKNOWLEDGEMENTS CONTENTS LIST OF ABBREVIATIONS LIST OF TABLES LIST OF FIGURES 1. INTRODUCTION 1.1 1.2 1.2 2 Background Problem Statement and Justification Study Objectives I II III IV V VIII IX X 1 1 3 5 6 6 6 7 9 10 13 17 17 19 20 22

LITERATURE REVIEW 2.1 2.2 2.3 2.4 2.5 2.6 General Properties of Snow Remote Sensing of Snow Process of Snowmelt Snowmelt Runoff Modeling Related Research

3.

DESCRIPTION OF THE STUDY AREA 3.1 3.2 3.3 3.4 Physiography Landuse Pattern Climatic Conditions Hydrological Characteristics and Water Resources

4.

METHODOLOGY 4.1 4.2 Outline The MODIS Instrument 4.2.1 4.2.2 4.3 MODIS Snow Mapping Algorithm MODIS Snowcover Products

25 25 25 27 30 33 34 35 36 39 40 41 41 42 44 45 46 46 48 48 49 49 51 51 51 57 69 69 71

The Snowmelt Runoff Model 4.3.1 Governing Equation 4.3.2 Model Accuracy Assessment

4.4 4.5 4.6 4.7

Data Acquiring and Database Development River Network and Watershed Delineation Image Processing & Classification Derivation of Model Input Parameters 4.7.1 Basin Boundary and Zone Areas 4.7.2 Temperature Lapse Rate and Degree Days

4.7.3 Precipitation 4.7.4 Snow Area Extent

4.7.5 Runoff Coefficients 4.7.6 4.7.7 4.8 4.9 4.10 5. Recession Coefficient Rainfall Contribution Area and Time Lag

Model Calibration and Verification Model Simulations Model Development

RESULTS AND DISCUSSION 5.1 5.2 5.3 5.4 Outline Parameter Estimation Snowcover Estimation Snowmelt Runoff Modeling 5.4.1 5.4.2 Calibration and Verification Results Simulation Results

vi

5.5

Relationship of Snow Area Extent with River Discharge and Snowmelt Runoff 79 87 87 88 89 90

6.

CONCLUSIONS AND RECOMMENDATIONS 6.1 6.2 6.3 Conclusions Limitations Recommendations

REFERENCES

vii

LIST OF ABBREVIATIONS
AMSR-E a.s.l AVCS AVHRR BCM CMG DAAC DEM DHVSM EB EOS ESSA GIS GPS HKH ICIMOD LIDAR MODIS MCM MAF NASA NDSI NDVI NOAA NSIDC NWFP PARC PMD SAE SHE SMMR SR SRM SRTM SSM/I TI UBC USGS VHRR WAPDA Advanced Microwave Scanning Radiometer Above Mean Sea Level Advanced Vidicon Camera System Advanced Very High Resolution Radiometer Billion Cubic Meters Climate Modeling Grid Distributed Active Archive Center Digital Elevation Model Distributed HydrologyVegetationSoil Model Energy Balance Earth Observation System Environmental Science Service Administration Geographic Information System Global Positioning System Hindukush-Karakoram-Himalaya International Centre for Mountain Development Light Detection and Ranging Moderate Resolution Imaging Spectroradiometer Million Cubic Meters Million Acre Feet National Normalized Difference Snow Index Normalized Difference Vegetation Index National Oceanographic and Atmospheric Administration National Snow and Ice Data Centre North West Frontier Province Pakistan Agricultural Research Council Pakistan Meteorological Department Snow Area Extent European Hydrological System Scanning Multi-channel Microwave Radiometer Scanning Radiometer Snowmelt Runoff Model Shuttle Radar Topography Mission Special Sensor Microwave/Imager Temperature Index University of British Columbia United States Geological Survey Very High Resolution Radiometer Water and Power Development Authority

viii

LIST OF TABLES

Table 4.1 Table 4.2 Table 4.3 Table 4.4 Table 5.1

MODIS spectral bands and their primary uses.. 27 MODIS data product inputs to the MODIS snowmap algorithm.. 30 Summary of the MODIS collection 5 snow data products 31 Classes of the processed MOD10A2 dataset. 33 Area under permanent and temporary snow cover for three study years. 68

Table 5.2

Year round simulation statistics for different study years. 71

ix

LIST OF FIGURES

Figure 1.1 Figure 3.1 Figure 3.2

Spatial distribution of average annual rainfall in Pakistan... Location map of the study area in Pakistan.. 3-D view of the true color Lanndsat-7 image draped over DEM of the study area

2 18

19 21 21

Figure 3.3 Figure 3.4 Figure 3.5

Variability of average monthly temperature at the three places... Variability of average monthly precipitation at the three places. Average and at 60 % probability river discharges measured at Chakdara...........

22

Figure 3.6

Variability of average monthly discharge at Chakdara for the 1 half of a year 23 Variability of average monthly discharge at Chakdara for the 2nd half of a year. 24

st

Figure 3.7

Figure 4.1

Conceptual model (flow chart) of the adopted methodological approach... 26 47

Figure 4.2 Figure 5.1

Recession flow plot Qn vs Qn+1 for Swat river basin Delineated river network and watershed area of the whole Swat basin and study area.

52 53 53 54

Figure 5.2 Figure 5.3 Figure 5.4 Figure 5.5

Elevation zones, their areas & mean hypsometric elevation Area-elevation (hypsometric) curve of the upper Swat river basin. Average daily minimum and maximum temperature at Kalam... Meteorological stations used for computation of temperature lapse rate

55

Figure 5.6 (a)

Relationship between temperature and elevation for Jan and Feb months... 55

Figure 5.6 (b)

Relationship between temperature and elevation for Mar and Apr months... 56

Figure 5.6 (c)

Relationship between temperature and elevation for May and June months... 56

Figure 5.6 (d)

Relationship between temperature and elevation for Jul and Aug months... 56

Figure 5.6 (e)

Relationship between temperature and elevation for Sep and Oct months... 57

Figure 5.6 (f) Relationship between temperature and elevation for Nov and Dec months...
Figure 5.7 (a) Figure 5.7 (b)

57 60

Temporal variation of snowcover in the upper Swat basin (Jan - Apr) Temporal variation of snowcover in the upper Swat basin (MayAug) ...

61 62 63

Figure 5.7 (c) Figure 5.8 (a) Figure 5.8 (b)

Temporal variation of snowcover in the upper Swat basin (Sep-Dec). Temporal variation of snowcover in the Zone-A (6861500 m a.s.l).. Temporal variation of snowcover in the Zone-B (1501 2500 m a.s.l)...

63

Figure 5.8 (c)

Temporal variation of snowcover in the Zone-C (2501 3500 m a.s.l)... 63

Figure 5.8 (d)

Temporal variation of snowcover in the Zone-D (3501 4500 m a.s.l)... 64

Figure 5.8 (e)

Temporal variation of snowcover in the Zone-E (4501 5808 m a.s.l)... 64 64 65 67 68 70 70 71

Figure 5.8 (f) Temporal variation of snowcover in the whole basin... Figure 5.9 Figure 5.10 Figure 5.11 Figure 5.12 Figure 5.13 Figure 5.14 Figure 5.15 Comparison of snowcover variation in different years. Permanent and temporary/seasonal snow cover... Glacier location and extent as identified by PARC & ICIMOD 2005. Simulated and observed river flows for calibration year of 2003 Simulated and observed river flows for verification year of 2004... Simulated and observed river flows for verification year of 2002... Cumulative runoff components in various zones for the simulation year 2004 (Red is initial snow, green is new snow, and blue is contribution of rain).. Figure 5.16 Computed snowmelt and rainfall runoff components for the Year

73

2002... 74

xi

Figure 5.17

Computed snowmelt and rainfall runoff components for the Year 2003... 74

Figure 5.18

Computed snowmelt and rainfall runoff components for the Year 2004... 75

Figure 5.19

Average contributions of the two runoff components to the total runoff generated from the basin 77 77

Figure 5.20 Figure 5.21

Contribution of two runoff components to the total monthly runoff Average monthly distribution of snowmelt runoff in Jan Jun months...

78

Figure 5.22

Average monthly distribution of snowmelt runoff in Jul Dec months... 78

Figure 5.23

Temporal distribution of average daily snow area extent, observed river discharge and simulated snowmelt runoff 82

Figure 5.24

Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for March June months 83

Figure 5.25

Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for July August months... 84

Figure 5.26

Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for September February months.. 85

Figure 5.27

Prediction model for estimating May Aug runoff volume from the snowcover estimated on May 1-8. 86

xii

CHAPTER

ONE

INTRODUCTION
1.1 Background

The research community over the last several years has been putting momentous efforts into studying the potential impacts of changing climate on water resources as water has become a major limiting factor in most of the worlds agricultural development. As the Earths population has been growing rapidly and more stress is put on the land to fulfill the livelihood requirements of an ever-increasing population, one question remains ambiguous that how hydrologic resources will be affected. The global climate change models predict greatest changes at higher latitudes (Rees 2006) and higher altitudes of northern hemisphere, which mostly accommodate earths cryosphere. It is therefore imperative to monitor these regions to look for manifestation of global climate change. Pakistan is predominantly a dry country of the warm temperate zone. Its climate is transitional between that of Central Asia and the monsoonal lands of South Asia, and varies considerably with latitude, altitude, aspect and localized relief. Temperatures may reach as low as 26C over the northern high mountains, and as high as 52C over the south-eastern lowland arid plains. The mountainous and sub-mountainous areas of the northeast can receive over 1700 mm of precipitation annually, in contrast to only 30 mm in the arid plains of southwest Balochistan. In general, Pakistan is one of the worlds most arid countries with an average annual rainfall of only 292 mm. More than threefourth of the country receives less than 250 mm of annual rainfall (Figure 1.1). About 70 % of the total rainfall occurs in the monsoon season (July September), and is hardly used by crops directly as at that time the crops are near to their harvesting stage. Moreover, rainfall intensity of monsoon rains is generally higher resulting in greater surface runoff and lower absorption by the soil. Consequently, the agriculture and in turn population and economy of the country are heavily dependent on an average annual influx of about 180 BCM (billion cubic meters) of river water mostly derived from snowmelt in the Hindu Kush-Karakoram-Himalayan (HKH) region into the Indus river

system. The snow and glaciers of HKH region act as frozen reservoirs, capturing snow and rain, holding the water and releasing it into the rivers which feed the lower Indus plains. However, river inflows in summer are almost four times that of winter flows, necessitating enormous resources and efforts to control flooding and store water. The current per capita water storage capacity is only 150 m3 compared to over 5000 m3 in USA and Australia and 2200 m3 in China (WB 2005).

Figure 1.1 Spatial distribution of average annual rainfall in Pakistan

The snowcover and glaciers of HKH region are the largest repository of inland cryosphere outside Polar Regions. Significant portion of this snow and glacier cover is temporary and seasonal in nature. Seasonal snow cover is formed by consecutive

snowfall during the snow-accumulation season and gradually disappears during the subsequent snowmelt season. Temporary snow cover, which can also be specified as short-lived snow, is formed by a snowstorm during the snowmelt season, and exists only 2

for a few hours or a few days. While permanent snow cover is retained for many years, seasonal and temporary snow cover has major impact on regions renewable fresh water resources. However, water availability in this regionin terms of temporal as well as spatial distributionis expected to be highly vulnerable to anticipated climate changes (Sing et al 1997). The Indus river flows are estimated to be worst affected as it may loose about 27 % its total flows by 2050 (Arnel 1999). Hence, one of the most important and recent thrusts in hydrological research in Pakistan is the monitoring of glacier and snowcover changes and impact assessment of that variation on regions water resources. These changes are the result of natural processes as well as anthropogenic influences. 1.2 Problem Statement and Justification

Snow and glaciers are the frozen reservoirs of fresh water and cover a significant part of many mountain chains on the globe. In Pakistan about 5218 glaciers covering an area of 15,040 sq. km were identified in the ten sub-basins of Indus River System namely Swat, Chitral, Gilgit, Hunza, Shigar, Shyok, Upper Indus, Shingo, Astor and Jhelum covering HKH region of Pakistan (PARC and ICIMOD 2005). These glaciers constitute 11.7 % of the total area of these basins and are an important source of fresh water in Pakistan as 50 85 % of the countrys total flows come from melting snows and glaciers of the this region (Tarar 1982; Hewitt 1985; PARC and ICIMOD 2005). The major tributaries of the Indus River originate from the HKH region and have their upper catchments in the high mountain snow covered areas and flow through steep mountainous slopes. This factor and the perennial nature of these rivers provide excellent conditions for the development of hydropower resources. Rainfall during the monsoon season further adds to this potential. Snowmelt season in Pakistan generally coincides with the monsoon rainfall thereby augmenting the surface runoff often bringing heavy floods in the lower southern plain areas of the county resulting in substantial loses in terms of property and lives. As such the Government of Pakistan is undertaking massive program of hydropower projects under the Water Vision-25 program to have greater control over the available water resources and store water for next season and possibly for dry years as well as provide cheap source of renewable energy. The planning of such 3

new multi-purpose projects on HKH rivers in Pakistan emphasizes the need for reliable estimates of the snow extent and snow and glacier runoff because it provides a more dependable and perennial flow. Despite their well recognized importance and potential, little attempts have been made to assess in detail the contributions of snowmelt runoff in these rivers, although a few studies, e.g. Tarar 1982, Hewitt 1985, DeScally 1994, PARC and ICIMOD 2005, etc provide some insight in to the important aspects. Reliable predictions of snowmelt runoff generally require comprehensive snow surveys. Such surveys on a large scale are almost impossible for highly rugged and mostly inaccessible mountain topography of HKH region and for a resource poor country like Pakistan. Also, the ground data collection methods cannot provide either the desired areal coverage (due to large areal extent or access problem) or observational frequency. There are also procedural errors in point measurements and their extrapolation to large basins (Tarar 1982). Moreover, ground methods of snow surveys are often expensive, time consuming and difficult. Hence, due to lack of field data, unreliable and, often, late prediction of water availability usually results in ill planning and management of precious fresh water resources. Early prediction of snowcover extent and expected snowmelt runoff allows efficient planning and management of water resources for hydropower generation, regulation of discharge through reservoirs for flood control, and for irrigation, industrial and domestic water-supply. In the HKH river basins, which for the most parts are inaccessible due to extreme climate and highly rugged terrain and where snow cover data from conventional methods are either nonexistent or are very limited, satellite remote sensed observations provide the attractive and perhaps the only viable alternative for acquiring snow cover data necessary for hydrologic forecasting of snowmelt runoff. Recent advances in GIS, remote sensing and hydrological modeling techniques allow their powerful integration. In the field of snowmelt runoff modeling, such integration provides valuable basis for better understanding of snow accumulation and snowmelt runoff processes within the catchments, as well as for incorporating the spatial variability of hydrological and geographical variables and their impacts on catchment responses (Ahmad 2005). The research hypothesis of this study builds on such an integration and

utilizes remotely sensed satellite imagery of MODIS instrument aboard the Terra spacecraft for snow cover mapping. The WinSRM (Snowmelt Runoff Model for Windows) is used for snowmelt runoff modeling while all the analysis and map overlays are supported by GIS technique using ArcGIS 9.2. 1.3 Study Objectives

Climate change is likely to affect basins water resources so there is a need to monitor and estimate the fresh water resource base (snowcover) and assess the impacts of its variation on net water availability. Moreover, WAPDA plans to construct two dams each at Kalam and 5 km upstream of Munda Headworks with live storage capacities of 0.32 BCM (0.26 MAF) and 0.826 BCM (0.67 MAF) respectively under the Water Vision-25 program. The analysis carried out in this study will evaluate temporal availability of surface water resources and help optimally design and operate these projects. The specific objectives of this research study are; 1. Estimation of spatial and temporal distribution of snowcover through satellite remote sensing, 2. Estimation and quantification of snowmelt and rainfall runoff components through hydrological modeling, and 3. Development of snowmelt runoff prediction models.

CHAPTER LITERATURE
2.1 General

TWO REVIEW

Remote sensing and hydrology of snow are indeed wide and quite mature disciplines in themselves and in fact a large amount of related theories and contemporary literature is available. This chapter is fairly specific emphasizing only the basic concepts and the most relevant aspects to this study. For a more detailed and comprehensive review, interested readers are encouraged to refer Rees 2006; Sing & Sing 2001; and US Army of Corps Engineers 1956. 2.2 Properties of Snow

Snow is a mixture of ice crystals, liquid water and air; and forms from the crystallization of ice particles in the atmosphere during precipitation. The newly formed snow generally crystallizes in hexagonal shapes with grain size varying from 0.01 0.5 mm but they alter greatly over time due to metamorphosis and can form different shapes and sizes. Snow pack below 0oC temperature is dry snow and it hardly contains any liquid water but in wet snow at or above 0oC significant quantities of liquid water may be present. Wetness by volume typically ranges up to about 10 %. The total amount of water contained in a snow pack is specified by the snow water equivalent (SWE), which is depth of liquid water layer produced by the melting of all snow pack. If the density of snow pack is uniform, the typical value of SWE is around one-third of its depth (Rees 2006). A typical density of freshly fallen snow is about 0.1 gm/cc. However, as the snow ages, its density increases as a result of compaction by wind and gravity, and through thermal metamorphism (Sing & Sing 2001). Thermal properties of snow such as specific heat, latent heat of fusion, thermal quality, thermal conductivity, thermal diffusivity, and cold content are vital for computation of snow ablation, snowmelt, and energy balance of sow pack.

Reflective properties of snow are determined by its albedo and dielectric constant. Albedo of snow is the ratio of the reflected to the incoming solar radiation. Higher albedo values indicate greater reflection of incoming radiation. Spectral reflectivity of snow depends on grain size and shape, impurity content, liquid water content, depth, surface roughness, and solar elevation angle (Hall and Martinec 1985). Depending on the condition of the snowcover surface and the height of sun, the value of its albedo may vary from 0.29 for very porous, dirty, saturated with water snow to 0.86 for clean, compact and dry snow (Sing and Sing 2001). Moreover, the reflective properties of snow significantly differ in the various regions of the electromagnetic spectrum. Since ice constituting snow is in highly divided form, usually 109 particles per cubic meter, the fresh dry snow looks white and is highly reflective in the visible range (0.4-0.65 m). In the short-wave infrared region, however, it has strong absorbing characteristics. In the thermal infrared region its reflection is very low and does not exceed 1% for grain sizes above 100 m (Rees 2006). In the microwave region reflective properties are mainly controlled by dielectric constant, which is the measure of the response of a material to an applied electric field, such as electromagnetic wave and is the function of radiation and frequency. The greater the difference between the dielectric constant of snow and that of external medium, the greater the reflection coefficient, hence propagation of microwave radiation through dry snow is generally dominated by scattering. Because real part of dielectric constant of ice is practically constant throughout the microwave region and snow is a low loss dielectric medium, the real part of dielectric constant of snow depends only on the snow density and is given by
s = 1 + 1.9s

For a snow density of

0.3 gm/cc, the above equation yields dielectric constant of 1.57. 2.3 Remote Sensing of Snow

Satellite-Based Remote Sensing Technology has revolutionized the monitoring of spatial and temporal distribution of snow area extent (SAE) and snow depth in the complex natural conditions at regional and global scales. Satellite remote sensing involves making inferences about the nature of particular objects at the earth from the characteristics of the electromagnetic radiation received at the sensor and establishes relationship between objects physical properties and the received radiation. Because of higher albedo and

highly reflective nature, snow offers a good contrast with most other natural surfaces, except cloud, in the visible region. Hence, it is well suited to satellite remote sensing. Due to this effect, snow was detected from space in the first ever satellite image obtained through TIROS-1 weather satellite in its April 1960 launch (Singer and Popham 1963). Later on snow was mapped from space on a weekly basis following the launch of the Environmental Science Service Administration (ESSA-3) satellite which carried the Advanced Vidicon Camera System (AVCS) that operated in the spectral range of 0.50.75 m with a spatial resolution at nadir of 3.7 km. However large scale purposeful snowcover mapping in the northern hemisphere intensified after the National Oceanographic and Atmospheric Administration (NOAA) launched a variety of sensors, including the Scanning Radiometer (SR), Very High Resolution Radiometer (VHRR) and Advanced Very High Resolution Radiometer (AVHRR). Microwave remote sensing products, like SMMR, SSM/I, and AMSR-E, are generally used for global scale studies because of their coarse resolution (25 km), daily observational frequency and no influence of cloud cover. Products derived from optical instruments using reflected solar radiation, such as AVHRR, MODIS, and Landsat, etc., have higher spatial resolution and are better for regional studies, but heavily depend on suitable weather conditions, especially clear sky (no clouds). The high cost and low temporal resolution (16 days) of Landsat data are an obstacle to its wide application in monitoring snow, even though it has much higher spatial resolution (30 m) than MODIS and AVHRR. The NOAA-AVHRR frequency is twice every 24 hours (one daytime pass and one nighttime pass) but very high resolution (1 km) may be insufficient for snow mapping outside the polar regions particularly on small basins. The Moderate Resolution Imaging Spectroradiometer (MODIS) is one of the most sophisticated and recent instrument carried over Terra/Aqua spacecrafts, which offers good alternative. Due to a wide range of spectral bands (36), daily observational frequency and relatively higher spatial resolution (500 m) its use for snowcover detection is preferable.

2.4

Process of Snowmelt

Watersheds store water in its various forms including snow, which may range from a newly fallen crystalline snow to glacial ice. The release of water from various forms of snow and ice results from the net heat exchange between snow pack and its surrounding environment, but the rate of melting is different for each form due to their varying thermal properties. Light fresh snow melts faster than the old snow that has been altered to ice. The energy balance or heat budget of a snow pack, which governs the production of melt water, accounts the incoming energy, outgoing energy and the change in energy storage of the snow pack for a given period of time. If all the heat fluxes toward the snow pack are considered positive and those away considered negative, the sum of these fluxes is equal to the energy available for melting of the snow pack for a given time period. Hm = Hrs + Hrt + Hs + Hl + Hg + Hp (6)

where Hm is the energy available for melting of snow pack; Hrs is the net solar radiation; Hrt is the net thermal radiation; Hs is the sensible or convective heat transfer from air; Hl is the latent heat of evaporation, condensation or sublimation; Hg is the heat transfer through conduction from underlying ground; and Hp is the heat content of precipitation. The solar radiation (Hrs) is the net of incoming minus reflected solar radiation while thermal radiation (Hrt) is primarily the net of incoming radiation from the atmosphere, clouds, and surrounding vegetation minus the outgoing blackbody radiation from the snow pack itself. Sensible heat transfer occurs when the air temperature is different from the snow pack temperature. If the air is colder, Hs is negative conversely it will be positive. Latent heat is the energy released during a phase change of water from vapor to liquid to solid when condensation onto the snow pack occurs, or conversely, it is the energy extracted from the snow pack when evaporation or sublimation from the snow pack occurs. Condensation, evaporation or sublimation depends on the humidity of the air and the water vapor pressure gradient between the air and the snow surface. If the humidity is high, such that the vapor pressure of the air is greater than that at the snow surface, the vapor pressure gradient is towards the snow resulting in condensation and, in this case Hl is positive. If the air is dry, evaporation and/or sublimation will occur, and Hl 9

will be negative. The Hg will be positive if the snow is colder than the underlying soil and negative if the snow is warmer, whereas Hp will be positive if the temperature of the precipitation is warmer than the snow and negative if it is colder. Only the positive value of Hm will result in melting of snow. The relative importance of the above described energy balance terms involved in melting of snow pack depends on time and local conditions. For example radiation melting dominates when wind is calm, whereas melting due to sensible heat flux dominates in warm and windy conditions (Sing and Sing 2001). When all the components of energy balance equation are known and Hm is positive, the melting of snow pack is given by:

M =

Hm w L

(7)

Where M is the depth of melt water (m/day), L is the latent heat of fusion (333.5 kJ/kg),

w is the density of water (1000 kg/m3), and is the thermal quality of snow. The
thermal quality of snow pack is the ratio of the heat input required to produce a given amount of water from snow relative to that required to melt the same quantity of water from pure ice at 0oC. It is usually found in the range of 0.80 1.1. 2.5 Snowmelt Runoff Modeling

Snowmelt runoff is a major component of the hydrologic cycle in many regions. Its modeling needs knowledge of site specific climatic conditions, comprehension of basin characteristics and an understanding of various processes associated with snow accumulation, snowcover properties, snowcover distribution, surface energy exchange, water retention and movement through snow pack, snow soil interaction, and routing of generated snowmelt runoff (Sing and Sing 2001). Computation of snowmelt from a snow pack can precisely be accomplished using the energy balance approach described above. The energy balance models, also known as physically based models, use fundamental physical principles and equations that describe the physics of processes in each component of the energy balance (Dingman 1994). Generally, different models simulate the surface energy balance in similar ways, with

10

more or less complex treatments of albedo, and often ignoring some of the less important energy terms. However, there is considerable variation between models in the ways in which the internal distribution of heat and mass are represented within the snow profile. Many models treat the snow pack as a single, lumped layer. This is true, for example, for the snow hydrology component of the SHE model (Morris 1982; Abbot et al. 1986), the DHVSM (Wigmosta et al. 1994), and the Hadley Centre land surface scheme (Essery 1997). In these models, internal state variables such as temperature or density are treated as average values for the whole snow pack. The most complex layered models utilizes vertically distributed implementations of coupled partial differential equations to represent heat and mass transfer (Anderson 1976; Brun et al. 1989; Jordan 1991; Morris et al. 1993). These models simulate details of snow pack stratigraphy, temperature gradients and melt water movement. These models are perhaps most suitable for examining processes occurring on short, hourly times scales, such as nocturnal refreezing of the surface and melt water outflow from the base of snow. Theoretically, the physically based models are most accurate and have applicability in a wider range of conditions and environments. However, the major disadvantage of such models is their large and complex data requirements. In most cases some of the variables are not observed at all and are often estimated inducing some degree of errors. Alternatively, conceptually index models use one or more variables in an empirical expression to estimate snowcover energy exchange. Air temperature is the best and most commonly used index, but other variables such as net radiation, wind speed, vapor pressure and solar radiation may also be used. The temperature index, also known as degree-day method, is more popular and widely used because air temperature reasonably represents the energy flux and at the same time it is relatively an easy parameter to measure, extrapolate and even forecast. The temperature index models physically lump all the components of the surface energy balance into a degree-day melt factor, which is a proportionality coefficient that calculates melt rates on the basis of air temperature (normally in excess of some threshold value) alone. Several operational models, including the Snowmelt Runoff Model-SRM (Martinec 1975 & 2007; HBV (Bergstrom 1975), used to forecast runoff from mountainous areas use temperature index approach. 11

The main advantage of temperature index models is the data requirements may be limited to as little as average daily air temperatures, the most easily measured and widely available meteorological variable. However, this is also potentially their biggest drawback as factors other than air temperature control melt rates. In particular, radiation is often the most important factor controlling melt rates in mid-latitude mountainous areas; and although air temperature and net radiation may be correlated over the course of several weeks (Ferguson 1999), simple temperature index models cannot incorporate variation in radiation receipt directly. Moreover, even though air temperature is obviously an important control over turbulent fluxes, wind speed and surface roughness also play a role and are not included in a degree-day melt factor. Hence, snowmelt prediction through the conceptually index models can be significantly improved by incorporating vapor pressure, net radiation and wind speed rather than the temperature alone. Given the advantages and disadvantages of both conceptual temperature index and physically based energy balance models, a number of attempts have been made to generate hybrid approaches, which tend to keep the simplicity of the degree day approach and accuracy of the energy balance approach by explicitly incorporating other important components of the surface energy balance, principally the radiation. These extended formulation models include that of Anderson 1973, whose combined approach used degree-day formulation during dry periods and a simplified empirical energy balance formulation during rainy periods. The UBC runoff model (Quick and Pipes 1977) and the HYMET runoff model (Tangborn 1984) both add the use of daily temperature range as a measure of cloud cover, and thus radiation. The most common addition though to temperature index-type models has been the simple incorporation of measured shortwave radiation (Martinec 1989) or net radiation (Martinec and de Quervain 1975; Kustas and Rango 1994; Brubaker et al. 1996). Pipes and Quick (1987) found that partial energy based (EB) depicted better results than the temperature index (TI) models in both small and large basins in British Columbia, and far better results in a heavily glacierized Karakoram basin where temperature index (TI) drastically underestimated radiation melt at higher elevations.

12

Irrespective of the choice of method, modeling of the spatial distribution of snowcover and melt usually is accomplished by dividing a watershed into a number of smaller land units based on topographic facets such as elevation bands and hill slopes or by geometrical subdivision into grid squares. Within individual land units all hydrological processes are parameterized or described by physical and/or empirical formulas. 2.6 Related Research

Areal extent of snowcover has been an important variable for a number of uses including snowmelt runoff prediction in snow-fed basins (Martinec, 1975 & 1985; Hall & Martinec 1985), for accurate specification of the boundary conditions in surface-atmospheric modeling (Dai et al 2003; Zeng et al 2001), and for modeling atmospheric, hydrological, and ecological processes (Simic et al 2004). Snowmelt runoff modeling in high mountain areas based on periodical snowcover mapping derived from earth observation satellites has been regularly reported in the literature particularly after 1970s when a breakthrough was achieved in satellite based snowcover mapping (Martinec1973; Odegaard & Ostrem 1977; Rango et al 1977; Gupta et al 1982; Baumgartner et al 1985; Kumar et al 1991; Martinec et al 1991; Seidel & Martinec 1992; Rango & Martinec 1999). Since, runoff regimes in most of the northern basins are mainly controlled by the melting snowcover; snowmelt runoff modeling in these basins has been important aspect of hydrology. However, due to a wide variety of input data needs, it is also a cumbersome issue as most of the data are not available. Consequently attempts have been made to simplify the effort. Regression and other models of runoff based on satellite derived estimates of snow covered areas have underscored the importance of seasonal snowmelt in the HKH region (Rango et al 1977; Qureshi & Umar 1978; Tarar 1982; Dey et al 1983 & 1988; Ramamoorthi 1983 & 1987; Makhdoom & Solomon 1986; Kumar 1991) and elsewhere (Odegaard and Ostrem 1977; Yang et al 2003; Zhou et al 2005). The snowmelt runoff volume in the snowmelt period in a given basin was statistically related to the area covered by snow at the start of snowmelt season. More precise attempts generated regression models on monthly or even weekly basis.

13

Odegaard & Ostrem (1977); using Landsat satellite data empirically related the snowcover area with runoff in Norwegian catchment and observed that the snowcover area can be used to forecast the expected snowmelt runoff with a reasonably good accuracy. Rango et al (1977) developed linear regression models for estimating seasonal runoff volume in the Kabul and Indus River basin from the single time ESSA and NOAA satellite- observed snowcover data in the western Himalayan basin. Dey et al (1983) employing similar approach developed similar models for the same area using NOAAVHRR snowcover datasets of the following six years. They also extended the earlier work of Rango et al (1977) and developed linear regression model by combining both the datasets. Tatar (1982) also found a significant correlation between the variations in March or April snowcover and the summer-season runoff for several basins of the Indus system in the Himalayas. He concluded that the Landsat snow-coverage data for remote areas were susceptible to yield seasonal stream flow predictions by applying a liner regression equation and the relationships were at best preliminary and would need to be improved and refined by supplementation through field data collection and application of improved analytical tools and procedures. In an attempt to relate snowcover area with snowmelt runoff volume, Gupta et al (1982) found that for a particular sub-catchment the relationship between snow area extent and snowmelt runoff was independent of geographic factors like solar illumination, catchment orientation and relative location. Instead, geomorphical factors such as size of sub-catchment, permanent snowcover, average altitude, lithology, and stream have major impact. Consequently he suggested different relations for each sub-catchment. Makhdoom and Solomon (1986) examined the usefulness of the snowmelt forecasting models from the snowcover for the Indus basin of Pakistan and found their limited practicability due to variability of snow water equivalent (SWE) for the same snowcover. They emphasized the need for improved estimation of snow area extent and information on depth and density of snowcover. 14

Dey and Sharma (1989), using NOAA-4 satellite imagery, tested the SRM (Martinec 1983) for a large Kabul river basin to assess the accuracy of model simulation in a subtropical environment and its performance on a daily basis for the snowmelt season. They observed very poor simulation due to unrepresentative lapse rate, extremely marginal climatic data, and larger difference between the mean elevation of the total basin and that of snowmelt contributing area. DeScally (1994) found strong correlations between point field measurements of the annual maximum of snow pack water equivalent and of total winter precipitation in the Kunhar sub-basin of Chinab River and total annual discharge. The total winter snowfall also showed significant correlation with annual discharge. Monsoon rainfall appeared to be a very poor indicator of annual discharge. Mashayekhi and Mahjoub (1991), using field data, developed multiple linear regression models for snowmelt forecasting from Karadj river basin, Iran and found their regression model more accurate for seasonal forecast than the monthly forecasts. Yang et al (2003) using long term NOAA snowcover data examined and compared the weekly mean stream flow with the weekly basin snowcover extent in large Siberian watersheds and developed statistically significant weekly runoff-snowcover logarithmic relations for weeks of strong snowmelt. This approach is good when there is very high variability of snowcover and associated stream flow and they could not be related accurately for longer durations. But when a definite trend is prominent and best-fit regression line can be drawn for longer durations, then it is better to develop regression models for monthly or even seasonal discharges. A regression analysis of stream flow and snow area extent conducted by Zhou, et al 2005 for the Upper Rio Grande River Basin, USA depicted statistically significant logarithmic decay function of the stream flow with snow area extent for the daily as well eight-day MODIS snowcover products, but the correlation coefficient from the 8-day product was larger than that from the daily product.

15

PARC & ICIMOD (2005) developed inventory of glaciers, glacial lakes and dangerous glacial lake in the HKH region of Pakistan using one time Landsat-7 imagery and calculated the glacier area and ice reserve in each of the 10 sub basins of Pakistani Indus Basin. However, they did not estimate or quantify the seasonal variation of snowcover. The major problem with most of the above referenced studies was that they used satellite imagery having either very low temporal resolution (mostly single time or in some cases few images) or very high spatial resolution of over 1 km. The meteorological data used was not much representative of the study area due to its unavailability. Moreover, they do incorporate the contribution of rainfall which may have significant impacts particularly in rainfed areas and the observed flows may not be the true representative of the associated snowcover. The summer flows are completely the function of winter snowcover only when no any snowfall is observed in the summer months. Hence, the basins receiving considerable amount of snowfall in summer months may not be modeled correctly with that approach. This study on the other hand takes care of these factors and utilizes continuous time series satellite data having relatively higher spatial resolution. It also employs a different approach of relating daily values of snowcover and discharges. Moreover, the more representative meteorological data and recent improvements in the SRM further add to the applicability, suitability and recognition of this research study.

16

CHAPTER DESCRIPTION
3.1 Physiography

THREE STUDY AREA

OF

THE

The Swat valley, stretched mainly over the Swat River Basin of North West Frontier Province (NWFP) of Pakistan, is one of the most beautiful valleys of the country. Because most parts of the valley are geographically located within the monsoon belt, the valley is largely greener and more fertile than the valleys further north. The Swat valley is famous as the land of waterfalls, lakes, lush green hills and other beautiful natural gifts bestowed upon it by the nature. The valley also offers some of the best walking trails in Pakistan, as well as excellent opportunities for fishing and climbing. The excavated archaeological sites here range from prehistoric caves through Aryan graveyards to Buddhist monasteries. The Swat River originates in upper Swat between Shandur and Kalam towns tumbling through pine forests hemmed in by snowcapped mountains through two of its originating tributaries, the Ushu (north-eastern) and Utrot (north-western) rivers, which together form the Swat River near Kalam. The river drains parts of the Hindu Kush, Dir, Swat, and Kohistan ranges in the western territory of Pakistan. The Panjkora River is its major tributary joining it on the right side downstream of Chakdara town and upper Swat canal. After passing through eastern parts of Bajour and Mohmand agencies close to Munda Headworks, the river unites the Kabul River near Nowshera in the NWFP, which ultimately joins the Indus River downstream of Tarbela dam at Atock. The study is undertaken in the catchment area of upper Swat River upstream of Chakdara gauge station. The upper Swat River Basin is located between the latitude and longitude range of 34.57 to 35.896 and 71.928 to 72.834 decimal degrees respectively covering an area of 5713.38 km2 (Figure 3.1), which is about 39 % of the total area of the Swat river basin. Its northern part has high mountainous rugged terrain with elevation range of 2000 5808 m a.s.l., whereas the southern part is relatively flat with elevation range of 686 2000 m a.s.l. having some crop fields on either side of the river.

17

Figure 3.1 Location map of the study area in Pakistan

18

3.2

Landuse Pattern

The most part of the basin lies in the active monsoon belt and possesses coniferous forest dominant landuse. The other major landuse types in the basin are scrub, and alpine forests, agriculture, and grassland. Figure 3.2 presents a Three-Dimensional view of the rough idea of the landuse pattern obtained through Landsat-7 TM image draped over Digital Elevation Model (DEM) of the study area. Socio-economic conditions in the area are generally poor. Snowcover at higher elevations at northern part is quite prominent in this image.

Figure 3.2 3-D view of the true color Lanndsat-7 image draped over DEM of study area

19

3.3

Climatic Conditions

Based on the historic as well as prevailing climatic conditions, the study area can be divided into two parts. The upper north-eastern part Kalam and surrounding areas comprises very rugged mountain topography and may receive a maximum temperature of 37 oC in June at Kalam to as low as 18.2 oC in January at Shandur. The lower southeastern part near Saidu and Chakdara is relatively flat, receiving considerably higher temperatures ranging from -2 oC in January to as high as 45 oC in June. Similarly, the precipitation pattern in the lower south-western part is influenced by the summer monsoon rainfall, which originates in the Bay of Bengal and after crossing India reach Pakistan in early July and continue till late September. The upper north-eastern part on the other hand is dominated by the winter rainfall mainly received from the Western Disturbances, which come from the Mediterranean and after passing through Iran and Afghanistan enter Pakistan in December and continue till early April. The northern highlands receive most of winter precipitation in the form of snow. Figures 3.3 and 3.4 present average monthly temperature and precipitation received at Saidu (1990 2005 record), Kalam (2003 2006 record) and Shandur (2003 2006 record) met stations respectively. The mean annual precipitation at Saidu is 1086 mm, of which 56% falls in summer (Apr Sep). Kalam receives little more annual precipitation but summer precipitation is only 33% of the total of 1376 mm. Due to higher elevation most of the winter precipitation at Kalam falls in the form of snow, whereas relatively higher summer monsoon rain at lower elevations augmented by snowmelt runoff, sometimes causes heavy floods in the area. At Shandur, however, the weather becomes significantly dry with only 208 mm of annual precipitation. Generally weather becomes gradually drier if some one goes further north from Kalam. This diverse climate coupled with very high variability of altitude ranging from 686 5808 m a.s.l. provide conducive environment for significant snow accumulation in the winter months and subsequent snow ablation in the following summer months.

20

Saidu 30 25 20 Temperature (oC) 15 10 5 0 -5 -10 -15 Jan Feb Mar Apr May Jun

Kalam

Shandur

Jul

Aug

Sep

Oct

Nov

Dec

Month

Figure 3.3 Variability of average monthly temperature at the three places

Saidu 225 200 Precipitation (mm) 175 150 125 100 75 50 25 0 Jan Feb Mar Apr May Jun

Kalam

Shandur

Jul

Aug

Sep

Oct

Nov

Dec

Month

Figure 3.4 Variability of average monthly precipitation at the three places

21

3.4

Hydrological Characteristics and Water Resources

The upper Swat is predominantly a snow-fed river basin as under optimum conditions about 80% of its area can receive snowfall and over 74% of the total flows can come from snowmelt runoff. Most of that snowcover is concentrated in the northern part at an elevation exceeding 2500 m a.s.l.. In winter the areas even at 1500 m elevation can be blocked by snow, which however melts in the summer and one can drive up beyond Kalam and from there trek north to either Chitral or Gilgit valleys. The upper reaches of the Kohistan-Swat ranges are mostly covered with snow and glaciers. PARC and ICIMOD (2005) identified six types of glaciers present in the basin. These are Mountain, Cirque, Ice cap, Niche, Ice apron, and Valley. Mountain glaciers are the dominant type followed by Valley glaciers. Pakistans Water and Development Authority (WAPDA) has established two river gauge stations each at Kalam and Chakdara towns on the upper Swat River. The flows at Kalam gauge come predominantly from snowmelt runoff, whereas at Chakdara considerable contribution of summer monsoon rainfall runoff is also received. Figure 3.5 presents average and at 60 % probability (3 out of 5) daily river discharges measured at the Chakdara gauge station.
Average 600 Daily Discharge (Cumec) 500 400 300 200 100 0 60% Probability

Figure 3.5 Average and at 60 % probability river discharges measured at Chakdara 22

M onth

The average monthly flows observed at the Chakdara gauge station can be estimated by the second order polynomial function for the two halves of a calendar year as shown in Figures 3.6 and 3.7. The lowest average monthly flows of about 41.93 m3/sec (0.112 BCMbillion cubic meters) are observed in January, whereas the highest flows observed in June are over ten times (425.81 m3/sec or 1.14 BCM) that of January flows. Similarly, on an average, about 80 % flows are received in Kharif (summer) cropping season (Apr Sep) leaving only 20 % for Rabi (winter) cropping season (Oct Mar). The minimum annual flows of 3924.954 MCM were observed in 2001-02 because of severe drought, while maximum annual flows of 6803.035 MCM were observed in 2004-05 when drought was over, depicting the variability of 1.73 times between minimum and maximum annual flows. This very high monthly, seasonal as well as annual variability of upper Swat river flows necessitates comprehensive study of the available hydrological resources and development of appropriate models for predictions of water resources to ensure their better planning and management.

450 Average Monthly Obseved Discharge (Cumec) 400 350 300 250 200 150 100 50 0 1 2 3 4 5 6 y = 10.503x + 4.2595x + 24.424 R = 0.9995
2 2

Calender M onth Number

Figure 3.6 Variability of average monthly discharge at Chakdara for the 1st half of a year

23

Average Monthly Obseved Discharge (Cumec)

450 400 350 300 250 200 150 100 50 0 7 8

y = 19.203x - 205.87x + 605.35 R = 0.9938

2

10

11

12

Calender M onth Number

Figure 3.7 Variability of average monthly discharge at Chakdara for the 2nd half of a year

24

CHAPTER

FOUR

METHODOLOGY
4.1 Outline

This chapter describes the MODIS instrument, its snowmap algorithms and snowcover products and discusses the accuracy of snowcover products. The Snowmelt Runoff Model is also described followed by the details of the important methodological steps employed to achieve the specified research objectives. Finally the statistical approach used to develop the prediction models of snowmelt runoff is explained. However, the overall conceptual model or flow chart of the basic methodological approach adopted to accomplish the study is summarized in the Figure 4.1. 4.2 The MODIS Instrument

The Moderate Resolution Imaging Spectroradiometer (MODIS) is a key instrument aboard the Terra spacecraft launched on December 18, 1999 and the Aqua spacecraft, launched on May 4, 2002. Terra's orbit around the Earth is timed so that it passes from north to south across the equator in the morning, while Aqua passes south to north over the equator in the afternoon. MODIS instrument acquires images in 36 spectral bands between 0.405 and 14.385 m for different uses (Table 3.1). A 55 degree scanning pattern via a two-side scan mirror at the EOS orbit of 705 km achieves a swath of 2,330 km cross track by 10 km along track (at nadir) each scan and views the Earths entire surface ranging from every day at high latitudes to every other day at low latitudes (Justice et al. 1998). Its spatial resolution varies with spectral band, and ranges from 250 m to 1 km at nadir. The 1st two bands are imaged at a nominal resolution of 250 m at nadir, next five bands at 500 m, and the remaining 29 bands at 1 km. This study uses snow products of the TMODIS-Terra, because band 6 on Aqua spacecraft is only partly functional and most of algorithm development and testing work is done on MODIS-Terra products. Hence the algorithms and data products described here in primarily refer to MODIS-Terra sensor. These algorithms and products for MODIS-Aqua sensor however, only slightly change. 25

Digital Elevation Model

MODIS Snow Cover Imagery

Daily Climatic Data

Daily Observed Discharge

Stream Network & Watershed Delineation Elevation Zones Snowcover Distribution

Model Input Parameters

Snowmelt Runoff Model

Model Calibration
No

OK
Yes

Model Simulations

Snowmelt Runoff

Rainfall Runoff

Regression Models

Regression Models

Figure 4.1 Conceptual models (flow chart) of the adopted methodological approach

26

Table 4.1 MODIS spectral bands and their primary uses. Primary Use Land/Cloud/ Aerosols Boundaries Land/Cloud/ Aerosols Properties Band Bandwidth (m) Primary Use 1 0.620 0.670 Surface/ 2 0.841 0.876 Cloud Temperature 3 0.459 0.479 4 0.545 0.565 5 1.230 1.250 Atmospheric Temperature 6 1.628 1.652 7 2.105 2.155 Cirrus Clouds Water Vapor Ocean Color/ 8 0.405 0.420 Phytoplankton/ 9 0.438 0.448 Biogeochemistry Cloud 10 0.483 0.493 Properties 11 0.526 0.536 Ozone 12 0.546 0.556 Surface/ Cloud 13 0.662 0.672 Temperature 14 0.673 0.683 Cloud Top 15 0.743 0.753 Altitude 16 0.862 0.877 Atmospheric 17 0.890 0.920 Water Vapor 18 0.931 0.941 19 0.915 0.965 Band Bandwidth (m) 20 3.660 - 3.840 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 3.929 - 3.989 3.929 - 3.989 4.020 - 4.080 4.433 - 4.498 4.482 - 4.549 1.360 - 1.390 6.535 - 6.895 7.175 - 7.475 8.400 - 8.700 9.580 - 9.880 10.780 - 11.280 11.770 - 12.270 13.185 - 13.485 13.485 - 13.785 13.785 - 14.085 14.085 - 14.385

4.2.1

MODIS Snow Mapping Algorithm

The development of the MODIS snow mapping algorithm (snowmap) is chronicled in detail elsewhere (Hall et al 1995; Klein et al 1998; Hall et al 2001 and 2002; Hall and Riggs 2007; Riggs et al 2006), hence only a cursory overview is presented here. The snowmap (Hall et al 1995) is the basis for all MODIS snow cover products. However, the algorithm has been continuously evolving as limitations become apparent in early versions of data. The basic techniques used in the snowmap algorithm are groupedcriteria incorporating the normalized difference between bands, threshold-based criteria tests, and decision rules (Hall et al 2001).

27

The first test of snow detection uses the Normalized Difference Snow Index (NDSI) approach, which is an effective way to distinguish snow from many other surface features taking advantage of strong visible reflectance and strong short-wave IR absorbing characteristics of the snow pack. The NDSI is defined as the difference of reflectances observed in a visible band such as MODIS band 4 (0.555 m) and a short-wave infrared band such as MODIS band 6 (1.640 m) divided by the sum of the two reflectances.

NDSI =

Band 4 Band 6 Band 4 +Band 6

(1)

Generally, snow is characterized by higher NDSI values than other surface types and pixels. A pixel is mapped as snow if the NDSI value is 0.4 and the reflectance in MODIS band 2 is greater than 0.11. However, if the reflectance in MODIS band 4 is less than 0.10 then the pixel will not be mapped as snow even if the other criteria are met (Hall et al 2001 and 2002). This minimum reflectance test screens low reflectance surfaces, e.g. water that may have a high NDSI value from being erroneously detected as snow. However, in forest areas snow-covered pixels may have considerably lower NDSI values and to correctly classify these pixels as snow-covered, NDSI and NDVI are used together to the pixels that have an NDSI value in the range of 0.1 to 0.4. MODIS bands 2 and 1 are used to calculate NDVI.

NDVI =

Band 2 Band1 Band 2 +Band1

(2)

Snow cover tends to lower the NDVI therefore pixels with NDVI value of 0.1 may be mapped as snow even if the NDSI < 0.4 (Klein et al 1998). Moreover, pixels with an absolute reflectance of greater than 0.11 in MODIS band 2 and greater than 0.10 in MODIS band 1 are determined as snow. Because of higher reflectance of clouds in near-infrared wavelengths the NDSI generally separates snow from most obscuring cumulus clouds, but it cannot always discriminate optically-thin cirrus clouds from snow. Instead, cloud discrimination is accomplished by using the MODIS cloud mask product, MOD35L2, (Ackerman et al. 1998; Plat nick et al.

28

2003), which employs a series of visible and infrared threshold and consistency tests to specify confidence that an unobstructed view of the Earths surface is observed. An indication of shadows affecting the scene is also provided. Land and inland waters are masked with the 1 km resolution land/water mask, contained in the MODIS geolocation product (MOD03). In Collection 5 the land/water mask made by the Boston University (BU) team based on EOS data is used. The 1 km data of the land/water mask is applied to the four corresponding 500 m pixels in the snow algorithm to analyze inland waters. Thermal mask is used to improve the snow mapping accuracy and to eliminate the spurious snow especially in warm climates. Using MODIS infrared bands 31 (10.78 11.28 m) and 32 (11.7712.27 m), a split window technique (Key et al., 1997) is used to estimate ground temperature (Hall et al., 2002). If the temperature of a pixel is >283 K then the pixel will not be mapped as snow (Riggs et al., 2006). The collection 5 snowmap algorithm also includes computation of fractional snow cover for all land and inland water body pixels in a swath. Fractional snow cover is calculated using the regression equation of Salomonson and Appel 2004, which is based on a statistical-linear relationship developed between the NDSI from MODIS and the true subpixel fraction of snow cover as determined using Landsat scenes from Alaska, Canada and Russia. Table 4.2 summarizes the data inputs to the MODIS snowmap algorithm. The accuracy of snowmap has been tested over a variety of surface covers relative to other derived snow cover maps; errors were estimated for seven different land covers using Landsat Thematic Mapper and MODIS Airborne Simulator data prior to the MODIS launch. In addition, it is fully automated thus reducing or eliminating biases due to human subjectivity which are problematic in long-term climatology studies (Hall et al 2001). Under ideal conditions of illumination, clear skies and several centimeters of snow on a smooth surface the snow algorithm is about 93-100% accurate at mapping snow (Hall and Riggs 2007).

29

Table 4.2 MODIS data product inputs to the MODIS snowmap algorithm. Earth Science Data Long Name Type (ESDT) MOD02HKM Data Used

MODIS Level 1B Reflectance for MODIS Calibrated and Geolocated bands: Radiances 1 (0.645 m) 2 (0.865 m) 4 (0.555 m) 6 (1.640 m)

MOD021KM MODIS

Level 1B Calibrated and 31 (11.28 m) Geolocated Radiances 32 (12.27 m) MODIS Geolocation Land/Water Mask Solar Zenith Angles Sensor Zenith Angles Latitude Longitude

MOD03

MOD35L2

MODIS Cloud Mask

Cloud Mask Flag Unobstructed Field of View Flag Day/Night Flag After Riggs et al 2006

4.2.2

MODIS Snowcover Products

MODIS snow products produced through the snowmap algorithm described above are archived at and distributed by the National Snow and Ice Data Center (NSIDC), which is one of NASAs eight Distributed Active Archive Centers (DAACs). The collection 5 MODIS snow data products are currently produced as a sequence of seven products (Table 4.3) beginning with a 5 min swath segment (granule) at a nominal pixel spatial resolution of 500 m and a nominal swath coverage of 2330 km (cross track) by 2030 km (along track) and progressing, through spatial and temporal transformations, to a monthly global gridded product (Hall et al 2002; Riggs et al 2006; Hall and Riggs 2007). 30

Table 4.3 Summary of the MODIS collection 5 snow data products. Earth Product Science Data Level Type (ESDT) MOD10L2 L2 Nominal Spatial Data Array Resolution Dimensions 1354 km by 500 m 2000 km 1200 km by 500 m 1200 km Temporal Resolution Swath (scene) Map Projection None (lat, lon referenced)

MOD10L2G

L2G

day of Sinusoidal multiple coincident swaths Day Sinusoidal

MOD10A1

L3

1200 km by 500 m 1200 km 1200 km by 500 m 1200 km 360 by 180 0.05 (global) 0.05 360 by 180 0.05 (global) 0.05 360 by 180 0.05 (global) 0.05

MOD10A2 MOD10C1 MOD10C2 MOD10CM

L3

Eight days

Sinusoidal

L3 L3 L3

by Day by Eight days by Month

Geographic Geographic Geographic

After Riggs et al 2006 The swath product (MOD10L2) takes input of the MODIS calibrated data products presented in Table 2, and other criteria specified in the snowmap algorithm and has two snow cover fields (snow extent and fractional snow cover) at 500 m spatial resolution for each swath (Riggs et al 2006; Hall and Riggs 2007). The snow cover field classifies each cloud-free land or inland water body pixel as snow-covered or snow-free, while fractional snow cover field provides the percent of snow cover within each pixel for land and inland water bodies. The resultant snow cover maps are the consequence of the snowmap algorithm, which identifies snow covered land, snow covered ice on inland water and computes fractional snow cover. After the swath product, each product in the sequence assimilates accuracy and error from the preceding product (Riggs et al 2006).

31

The second product, MOD10L2G, is a multidimensional data product created by mapping the pixels from the MOD10L2 granules for a day to the appropriate Earth locations on the sinusoidal map projection. The third product, MOD10A1, is a tile of daily snow cover maps at 500 m spatial resolution. The daily observation that is selected from multiple observations in a MOD10L2G cell is selected using a scoring algorithm to select the observation nearest local noon and closest to nadir. The fourth product, MOD10A2, is an eight-day composite of MOD10A1 to show maximum snow extent. The MOD10C1 is daily global snow cover map in a geographic map projection created by assembling MOD10A1 daily tiles and binning the 500 m cell observations to the 0.05 spatial resolution of the Climate Modeling Grid (CMG) cells. Similarly, the global eightday snow cover product, MOD10C2, is created by assembling MOD10A2 daily tiles. There are several different data-product levels starting from level 1B (L1B), which is a swath (scene) of MODIS data geolocated to latitude and longitude centers of 1 km resolution pixels. A level 2 (L2) product is a geophysical product that remains in latitude and longitude orientation of L1B; it has not been temporally or spatially manipulated. A level 2 gridded (L2G) product is in a gridded format of a map projection. The L2G algorithm creates a gridded product necessary for the level 3 products. A level 3 (L3) product is a geophysical product that has been temporally and or spatially manipulated, and is in a gridded map projection format and comes as a tile of the global grid. A full description of the products and levels is provided in the MODIS Snow Products User Guide (Riggs et al., 2006) and product documentation available at the MODIS website. The study utilizes the MODIS/Terra Snow Cover 8-Day L3 Global 500m Grid (MOD10A2) data set, which composites eight-days of input from MOD10A1 to generate maximum snow extent for the period and tracks the chronology of snow observations for each day. The product gives classified (Table 4.4) image of eight day period showing a maximum eleven classes presented in the Table 3.4. The eight day periods begins on the first day of the year and extends into the next year. The product can be produced with two to eight days of input, as there may not always be eight days of input, because of various reasons. If snow cover is found for any day, then the cell in the Maximum Snow Extent field is labeled as snow. If no snow is found, but there is one value that occurs more than

32

once, that value is placed in the cell. Similarly if a cell is observed as other than cloud on any of the eight days the algorithm assumes a cloud free period and labels the pixel with the observed value. This logic minimizes cloud-cover extent, such that a cell needs to be cloud-obscured for all days in order to be labeled cloud. Table 4.4 Classes of the processed MOD10A2 dataset. Maximum Snow Extent Coded Integer Values Sample Value Explanation 0 data missing 1 no decision 11 night 25 snow free land / forest 37 inland water 39 ocean 50 cloud 100 lake ice 200 snow 254 detector saturated 255 fill

4.3

The Snowmelt Runoff Model

The snowmelt runoff model (SRM), also known as Martinec Model or MartinecRungo Model is a semi-distributed, deterministic and degree-day hydrological model especially designed to simulate and forecast daily stream flow in mountain basins where snowmelt is major runoff factor (Martinec et al 2007). The model utilizes ambient air temperature values combined with a degree-day coefficient in order to estimate the ablation factor of the snow cover (Martinec et al 1998) and takes input of snow covered area and its variation along meteorological data (Martinec et al 1983). The model can also be used to evaluate the effect of climate change on seasonal snow cover and snowmelt runoff. The SRM was originally developed for small European basins but with the breakthrough achieved in estimating snow cover through satellite remote sensing and model improvements, it can now be applied in mountain basins of any size and any elevation range throughout the world (Martinec et al 2007).

33

The study utilizes the Windows Version 1.11 of the Snowmelt Runoff Model (WinSRM), which is the most recent version. The WinSRM provides an excellent environment for snowmelt runoff modeling in mountain basins. The basin area is divided in to a suitable number of elevation zones (not exceeding 16) and various input parameters including basin characteristics, climatic variables, snow covered area, runoff coefficients, recession coefficients, etc are specified for each elevation zone. The model manages a physical database of both input and output for a given basin. Each simulation in the model is a unique entity operating on a 2 366 days. Different simulations can be sequenced for greater time periods. Unlike most of the non-deterministic hydrological models, the input parameters for the SRM are not calibrated or optimized from the historical records. Instead those are either derived from field measurements or estimated through physical laws, theoretical principles and empirical or regression relations (Martinec et al 2007) as unsatisfactory results have been improved not by adjusting input parameters but by correcting the errors in datasets and input of variables. For this reason the model does not necessarily require calibration and can be used for ungauged basins as well. However occasional adjustments, never exceeding the range of physically and hydrologically acceptable values, are often done. In the rugged and high mountain regions and in a country like Pakistan where adequate field data is hardly gathered and meteorological data are only available at a limited density and also in lower valleys, model calibration and allowable adjustment of certain input parameters is unavoidable. 4.3.1 Governing Equation

Daily water produced from snowmelt and rainfall is computed, superimposed on the calculated recession flow and transformed into daily discharge from the basin according to the following equation.

Qn+1 = [cSnan (Tn + Tn )Sn + cRnPn ]

A.10000 (1 kn+1 ) + Qnkn+1 86400

(3)

34

where: Q c a T T S P = average daily discharge [m3 s-1] = runoff coefficient expressing the losses as a ratio (runoff/precipitation), with cS referring to snowmelt and cR to rain = degree-day factor [cm oC-1 d-1] indicating the snowmelt depth resulting from 1 degree-day = number of degree-days [oC d] = the temperature lapse rate correction factor [oC d] = ratio of the snow covered area to the total area = precipitation contributing to runoff [cm]. A pre-selected threshold temperature, TCRIT, determines whether this contribution is rainfall (immediate) or snow (delayed). A k = area of the basin or zone [km 2] = recession coefficient indicating the decline of discharge in a period without snowmelt or rainfall: k n = Qm+1/Qm (m, m + 1 are the sequence of days during a true recession flow period). = sequence of days during the discharge computation period. Equation (1) is written for a time lag between the daily temperature cycle and the resulting discharge cycle of 18 hours. In this case, the number of degreedays measured on the nth day corresponds to the discharge on the n + 1 day.
Various lag times can be introduced by a subroutine.

10000/86400 = conversion from cmkm2 d-1 to m3 s-1 If the study area is divided into certain number of zones then the above equation is repeated for each zone and the sum of all gives total discharge from the basin. 4.3.2 Model Accuracy Assessment

Apart from the first glance visual inspection of the actually measured and simulated daily flows in its graphics section, the SRM uses two well established accuracy criteria, the

35

coefficient of determination (R2) and the volume difference (Dv), which are computed as follows (Martinec et al 2007):

R = 1
2

(Q
n i =1 n i =1

Qi'

(Q Q )
i

(4)

where:

Qi Qi

is the measured daily discharge is the computed (simulated) discharge is the average measured discharge of the given season year is the number of daily discharge values

Q
n

The deviation of the runoff volume, Dv (%) is computed as follows:

VR VR' Dv = 100 VR
where:
VR

(5)

is the measured seasonal or annual runoff volume is the computed seasonal or annual runoff volume

V R' and Dv values closer to zero. 4.4

The perfect matching and highly accurate simulation will result in R2 values closer to one

Data Acquiring and Database Development

The primary data required to accomplish this research study are the digital elevation model (DEM) of the basin; remotely sensed satellite imagery of MODIS instrument; and daily records on temperature, precipitation and outflows from the basin. The other data required for model input are mainly derived from these records or estimated through physical laws and adjusted further during model calibration and verification. A DEM is the most versatile and widely used representation of a terrain for the continuous variation of the relief over space. It is a raster representation in which each grid cell records elevation of the earths surface and reflects a view of terrain as a field of elevation values. The recorded elevation is often the elevation of the cells central point

36

but in some cases it may be mean elevation of the entire grid cell. The digital elevation data are usually organized into three data structures regular grids, triangulated irregular networks, and contours depending on the source and/or preferred method of analysis. The square-grid digital elevation models have emerged as the most widely used data structure during the past decade because of their simplicity (i.e. simple elevation matrices that record topological relations between data points implicitly) and relative ease of computer implementation. The elevation data are vital for some applications including prediction of the effects of global warming and rising sea levels in coastal areas. However, for many other important applications the value of DEM lies in its ability to produce important derivative measures and fields such as aspect, slope, flow direction, flow accumulation, stream network and watershed delineation, etc. through calculation and transformation. The most important applications of DEM can be seen in physical geography, geomorphology, hydrology, ecology, soil conservation, forest and watershed management, etc. There has been an increasing trend during the recent past in the use of DEMs in terrain analysis of the earths surface due to advancements in computer based GIS technologies and easy availability of digital data. Most of the currently available digital elevation datasets are the product of photogrammetric data capture data are collected by decoding stereo air photos and by manually or automatically extracting satellite pictures using stereograph plotters. Additional elevation datasets can be acquired by contour digitalization, field surveys, global positioning system (GPS), and light detection and ranging (LIDAR). In 2003, the National Aeronautics and Space Administration (NASA) of the United States released the Shuttle Radar Topography Mission (SRTM) dataset for some regions, with 3 arc-second resolution for the globe, and 1 arc-second for the US. This dataset superseded the previous global dataset of topography, the GTOPO30, produced by the United States Geological Survey (USGS). The SRTM DEM data was produced using radar images gathered from NASAs shuttle. Two antennae received the reflected radar pulses at the same time, one antenna located in the shuttles cargo bay, the other at the tip

37

of a 60-m-long mast. This configuration allowed single-pass radar interferometry, and consequently the generation of a highly accurate global elevation model with a vertical accuracy of 6 m and a horizontal pixel spacing of 30 m (Jarvis et al 2004). The study utilizes digital elevation data acquired from the NASAs SRTM, which freely distributes such data through the CGIAR-CSI GeoPortal (http://srtm.csi.cgiar.org). The latest version (V3) of the SRTM DEM data is available in the shape of tiles and is projected in WGS-84 projection. This version of SRTM DEM data does not contain any data holes where water or heavy shadow prevents the quantification of elevation. These are generally small holes, which nevertheless render the data less useful, especially in fields of hydrological modeling. The remotely sensed satellite imagery from the MODIS instrument is processed at the NASAs Goddard Space Flight Center and a number of products are developed using the best available techniques and theories. The MODIS snowmap products are archived at and distributed freely by the National Snow and Ice Data Center (NSIDC), of NASA, USA. The study utilizes the collection 5, level 3, MODIS / Terra eight daily maximum snow extent (MOD10A2) data set, which composites eight-days of input from MOD10A1 daily snow cover product to generate maximum snow extent for the period and also tracks the chronology of snow observations for each day. The MOD10A2 snowmap products are available as a 500 m grid (at the equator) projected in Sinusoidal World projection in the shape of tiles. The study area is covered by the h23v05 tile. In all 141 MOD10A2 snowmap products for three years 2002 2004 were downloaded and processed further in a GIS environment to estimate altitudinal, spatial and temporal distribution of snowcover in the study area. The daily meteorological data for a number of met stations is collected from Pakistan Meteorological Department (PMD). In 2002-03 PMD has established a meteorological station at Kalam at a height of 2103 m amsl. This met station is located at the centre of the northern part of the study area. Saidu Sharif and Shandur met stations are located just outside the study area at an elevation of 961 and 3719 m amsl respectively. There are few other met stations located outside of the study basin and their temperature data is used to

38

estimate the temperature lapse rates due to elevation difference. The study however utilizes temperature data of Kalam met station and precipitation data of Kalam, Shandur and Saidu met stations for different elevation zones. Daily river discharge data for the study area is available from Surface Water Hydrology Project (SWHP) of Water and Power Development Authority (WAPDA), Pakistan for two gauge station namely Chakdara and Kalam over the upper Swat River. Chakdara gauge station is located at the lowest end and is the exit point of all the runoff generated in the basin. Therefore flow data of this gauge station is used for calibrating and verification of the SRM and also developing relationship with the snowcover. 4.5 River Network and Watershed Delineation

The two tiles of SRTM DEM data (srtm_51_05 and srtm_51_06) were mosaic and subset for the study area using the ERDAS Imagine software. The SRTM DEM data has spatial resolution of 0.0008333 degrees, which becomes approximately 92.6 m by 75.6 m for the study area. Since most of the analysis in ArcGIS environment is automatically performed using square cells rather than rectangular, unless the user specifies differently. Therefore the DEM data was re-sampled to a cell size of 77.2 m2. This cell size when multiplied with six gives the cell size of the MOD10A2 snowmap data product of MODIS. This matching of the cell sizes of both the data sets is necessary as it will help perform further GIS analysis and simplify map overlays. The re-sampled DEM was then re-projected into Pak-1 projection, which is the modified form of Lambert Conformal Conic projection and is the standard projection for Pakistan used by Survey of Pakistan and most of the other organizations. ArcHydro extension of the ArcGIS 9.2 was used to delineate the river network and their drainage areas. Before generating the flow direction grid, the sinks present in the original DEM were filled in and a depression less DEM was generated. A sink is a cell or set of spatially connected cells whose flow direction cannot be assigned one of the eight neighboring cell values in a flow direction grid. This can occur when all neighboring cells are higher than the processing cell, or when two cells flow into each other creating a two-cell loop. Sinks are considered to have undefined flow directions and are assigned a 39

value that is the sum of their possible directions. The flow direction function in Arc Hydro Tools assigns to each cell a number corresponding to which of the 8 neighboring cells lies on the path of steepest descent. The direction of flow is determined by finding the direction of steepest descent or gradient from each cell. This is calculated as drop = change in z value / distance * 100. The distance is determined between cell centers. The flow accumulation grid is created from the flow direction grid by accumulating the weight for all cells that flow into each down slope cell. Cells of undefined flow direction will only receive flow; they will not contribute to any downstream flow. A cell is considered to have an undefined flow direction if its value in the flow direction grid is anything other than 1, 2, 4, 8, 16, 32, 64, or 128. The accumulated flow is based upon the number of cells flowing into each cell in the output grid. The results of flow accumulation are used to create a stream network by applying a threshold value to subset cells with a high accumulated flow. Higher threshold values will delineate major streams and lower threshold values will define minor streams. The resultant raster of stream definition is used for calculating stream segmentation or stream links, which are the sections of a stream channel connecting two successive junctions, a junction and the outlet, or a junction and the drainage divide. Finally, the catchment areas for each stream are delineated from the stream link raster using the catchment grid delineation option. 4.6 Image Processing & Classification

The MODIS snowmap data products are produced through intensive processing and analysis using the best available techniques and algorithms and often do not require further processing work generally required in remote sensing techniques. However, some specific processing is necessary for achieving particular study objectives. The MOD10A2 dataset was acquired for three calendar years (2002 2004) and was converted from HDF format to imagine using IRDAS Imagine software. The MODIS snowcover data products have spatial resolution of 463.3127165 m2 for the study area. The images were resampled into 77.21878608 m2 cell size to match the cells size of the DEM, while conserving all the properties of the original dataset. This re-sampling generated 36 cells from a single cell of the original dataset. The images were then re-projected and subset

40

for the study area using ArcGIS and ERDAS Imagine softwares. In all, 140 images for the selected three years study period were processed and analyzed in GIS to determine the altitudinal, spatial and temporal distribution of the snow cover in the different elevation zones of the study area. 4.7 Derivation of Model Input Parameters

The SRM has modest input data requirements. Besides the DEM, daily records of temperature, precipitation, and snowcover are the basic input variables. The other input parameters are mainly derived from these records and outflows from the basin. The following paragraphs highlight the general procedure adopted to derivation of these input parameters. 4.7.1 Basin Boundary and Zone Areas

The basin boundary is usually defined by the location of stream gauge (or some arbitrary point on the course of stream, while watershed divide is identified from the digital elevation model using suitable GIS software such as ArcHydro in this case. Due to higher elevation range of 686 5808 m amsl the basin is divided into five elevation zones (Zone-A to Zone-E). The area occupied by each elevation zone and mean hypsometric elevation of each zone is determined. The mean hypsometric elevation can be determined either from the area elevation curve or manually by weighted average technique. The area-elevation (hypsometric) curve is the plot of cumulative area versus elevation. The zonal mean hypsometric elevation ( h ) can be determined from this curve by balancing the areas above and below the mean elevation. The manual weighted average technique calculates the percent area under each individual elevation and multiplying that area with its corresponding elevation and summing up this elevation. Alternatively, it can also be determined by calculating cumulative elevation of each zone, and then the elevation and number of counts for each elevation grid are multiplied and cumulated. The cumulative of the product of the number of counts and elevation is then divided by the cumulative elevation of each zone. This study adopted the weighted average technique to determine the mean hypsometric elevation of each elevation zone. The mean hypsometric elevation

41

of each zone is used as an elevation to which the base or reference station temperatures are extrapolated for the calculation of degree days. 4.7.2 Temperature Lapse Rate and Degree Days

The model accepts either daily average temperature or both minimum and maximum daily temperatures. These values can be input as basin wide or different values for each zone. Although air temperature is a continuous field, usually point measurements are recorded at each but distant meteorological station. In a mountain terrain air temperature is significantly dependent on the elevation rather than the horizontal location. The environmental temperature lapse rate is about 6.5 oC/km in the troposphere, which may be used in the absence of any actual local data (Singh and Singh 2001). The model can take input of one or several met stations. With the input of single station, temperature values are extrapolated from the reference elevation of area (usually elevation of the met station) to the mean hypsometric elevation of each zone using the temperature lapse rate, which is change in temperature per unit of elevation. If the user wants to use separate met station for each zone, the temperature values must have already been lapsed with respect to the entered reference elevation as the program does not accepts separate reference elevation for each zone. In this case it is better to use either input from a good, reliable and true representative met station or prepare a single synthetic station from data of multiple stations. If the elevation of the selected met station is equal to the mean elevation of the study area then the possible errors in the lapse rate are to some extent cancelled because of both upward and downward extrapolation (Martinec 2007). Significant errors however may occur with too much difference in the elevation of met station and mean elevation of the study area and in such cases correct estimation of the temperature lapse rate becomes important. Although the elevation of the only met station of the basin at Kalam is within the active hydrological zone, it is well below the mean hypsometric elevation of the study area. Due to higher elevation range of the study there is need to calculate the lapse rate of temperature due to difference of elevation. For this purpose temperature records of few other stations located outside the study area at various elevations are used to determine 42

the temperature lapse rates for different months. The altitudinal adjustment (T) in the models governing equation is computed through the following formula.

T = . hst h .
where: hst
h

1 ) 100

(6)

is the temperature lapse rate (oC per 100 m) is the altitude of selected met station (m) is the mean hypsometric elevation of each elevation zone (m)

Because the average temperatures always refer to a 24 hour period starting at 6.00 hrs, they become degree-days, T (oC.d). Degree-day factor ( a ) can be determined by comparing degree-day values (temperature values above a certain base temperature) with the daily decrease of snow water equivalent. However, the data on variation of SWE is rarely available. In the absence of any detailed data, the degree day factor can be calculated from the following empirical relation (Martinec 1960):

a = 1 .1 .
where:

s w

(7)

is the degree day factor (cm/oC/d), and s & w are densities of snow

and water respectively. Density of snow usually varies from 0.3 to 0.55 gm/cc resulting in value of degree-day factor in the range of 0.35 0.61, with lower value recommended for fresh snow and snow under forest canopy. However, slightly higher values have also been reported in the snow melt runoff modeling studies (Martinec 2007). The degree-days factor is used to convert the number of degree-days, T (oC.d) in to the daily snowmelt depth, M (cm) by:

M = a.T

(7)

The degree-day factor does not account for the other components of the energy balance 43

notably the solar radiation, wind speed, and latent heat of condensation and its values are extremely variable over the time because changing properties of snow significantly influence the snow melting process. 4.7.3 Precipitation

The correct evaluation of true representative precipitation in mountain basins is a real challenge as it usually has high variability depending on geographical location, elevation, direction of air currents, height of mountain barriers, vegetation cover, etc. Unlike temperature, which tends to change gradually, precipitation may not be continuous and it may have abrupt and very high spatial variability. The estimates of spatial precipitation are also highly uncertain unless a good network precipitation gauges exist. In this area of HKH region, the precipitation is caused by different weather systems during different seasons of a year and varies from place to place because of highly rugged topography of the HKH mountains. Arora et al 2006 studied the spatial, altitudinal and seasonal variability of rainfall in the Chenab basin of the Himalayan region and found elevation, distance and direction of wind currents to be equally important in explaining the variability in annual rainfall distribution. The study area possesses only one whether station (Kalam) inside its boundaries. The out side met stations, except Saidu and Shandur, are far away from its boundaries. Moreover, the lower southern part of the study area is located in the monsoon belt, whereas the outside stations are located in relatively dry zone. The example is the north-western part of Northern Areas where despite higher elevations the area is relatively dry. The Shandur, Kalam and Saidu met stations are not only located almost at the three ends (head, centre and tail) of the basin but also at varying elevations of 3719, 2103 and 961 m amsl respectively. The close examination of the data of these three stations revealed only 14 % variation in the mean annual precipitations of Saidu and Kalam and most of the rainstorms occur at almost similar times. The Shandur area however is considerably dry because of its location outside the monsoon belt, but the rainstorms here also tend to occur at the same times. Hence, it is assessed better to use the precipitation data of these three stations separately for different zones rather than conduct analysis to determine the 44

precipitation lapse rate, as in case of temperature. Moreover, instead of synthesizing the data of these three stations, the precipitation data of Saidu station is used for Zone-A area as it is very close to this zone and also their elevation is quite closer. Similarly, Zones-B, -C and -D areas very closely match the characteristics of the Kalam met station. The characteristics of the last zone (Zone-E) are best matched by the Shandur met station, which is located further north of the basin. Hence, precipitation data of this station is used for the last zone area. Critical temperature determines whether the precipitation is in the form of rain or snowfall. Usual values range from 0 3 oC with higher values in snow accumulation periods, but it can never be less than 0 oC (Martinec 2007). This parameter is more important for year round simulations which model both snow accumulation and snow ablation periods. For precipitation identified to be snow, model accounts its delayed effect on runoff generation differently for snow covered and snow free areas. The new snow that falls over the previously snow covered area is assumed to become part of the seasonal snow pack and its effect is included in the normal depletion curve of the snow coverage. The new snow falling over the snow free area is considered as precipitation to be added to snowmelt, with this effect delayed until the next day warm enough to produce melting. However, it is difficult to differentiate exactly between rain and snow because the temperature used is the daily average while precipitation may occur at any time during the day and that particular moment may be warmer or colder than the assigned temperature value. 4.7.4 Snow Area Extent

Information on the temporal, spatial and altitudinal distribution of snow cover in the area of interest is the heart of snowmelt runoff modeling with the SRM. To estimate snow cover in high mountain rugged terrain, satellite remote sensing is more suitable alternative than the field measurements. The eight daily, maximum snow extent (MOD10A2) snow cover product of the MODIS instrument was processed to determine the altitudinal, spatial and temporal distribution of snow cover in the study area using GIS and remote sensing techniques. Since each MOD10A2 snow cover product gives

45

snow cover for the eight days, an abrupt change in snow cover is usually observed from the map of one time period to another. This effect was smoothed by taking snow cover of these products for only two middle days and estimating it for the rest six days (three days before and three days after) using linear interpolation from the previous and next image. Few pixels of inland water and lake ice in some of the images were simply neglected and any image with considerable cloud cover was considered as an outlier and excluded from the Snowcover analysis. The Snowcover for that time period was also estimated by liner interpolation between the two time images. 4.7.5 Runoff coefficients

The runoff coefficient takes care of the losses from the basins available water resources (rain + snow) during its journey to the outlet. The average value of runoff coefficient for a particular basin is given by the ratio of annual runoff to annual precipitation. The comparison of historical precipitation and runoff ratios provide starting point for estimation of runoff coefficient. However, more often it varies throughout the year as a result of changing temperature, vegetation and soil moisture conditions. Moreover, very high uncertainty involved in the measurement of true representative precipitation poses serious difficulties in its correct estimation. For this reason, among SRM parameters, the runoff coefficient is the primary candidate for adjustment during model calibration. Runoff coefficient is usually higher for snow melt than for rainfall due to effect of cold water soil hydraulic conductivity. 4.7.6 Recession Coefficient

Stream flow recession represents withdrawal of water from the storage with no or little inflow. Analysis of historical discharge data is usually a good way to determine recession coefficient (k). The discharge on a given day (Qn) is plotted on the logarithmic scale against the value of discharge on the following day (Qn+1) as shown in the Figure 4.2. An envelop is drawn to enclose most of the points and the lower envelop line represents the extreme discharge decline, i.e. the recession without any partial delay by possible precipitation or snowmelt.

46

Figure 4.2 Recession flow plot Qn vs Qn+1 for Swat river basin For a snow fed basin, the value of recession coefficient changes with time due to changes in the characteristics of the drainage basin. For example, the changes in the snow covered area and depth of snow pack with time influence the recession trend of the basin. The recession coefficient will always be less than unity (normally greater than 0.9), and also not constant but may increases with decreasing discharges and is given by:
k n +1 = xQn y

(9)

where:

the constants x and y are determined by solving the above equation for two Qs and Ks from the Figure 4.2.

Theoretically, k can exceed the value of one in some cases of very small discharges in large basins but practically large basins usually have large discharges. However, the model avoids such cases by preventing k values from exceeding 0.99. The estimated x and y values must fulfill this condition, Qmin > x1/y. Recession coefficient can also be adjusted by comparing the measured and simulated flows during calibration.

47

4.7.7

Rainfall Contribution Area and Time Lag

Snow pack is usually dry before and during early snowmelt season and most of the rain falling on snow pack is normally retained by it. Only snow free area contributes to rainfall runoff during that period. However, at some later stage the snow pack becomes wet and the rain falling afterwards can flow as runoff. The user has to decide which time periods snow pack in a particular area and height will be dry and assign that input to the model accordingly. For large basins with multiple elevation zones, the time lag changes during the snowmelt season as a result of changing spatial distribution of snow cover with respect to the basin outlet. Generally the time lag in a basin increases as the snow line retreats. If there is uncertainty, the time lag can be adjusted in order to improve the synchronization of the measured and simulated peaks of average daily flows. 4.8 Model Calibration and Verification

The mountain hydrology is mainly the function of topography and meteorology (Ahmad and Joya 2003). The knowledge about interaction of these components of mountain hydrology is generally limited and qualitative in nature. Therefore there is more reliance on river flow data of the mountain areas which largely represent the hydrological responses of all the existing topographical factors and meteorological events taking place in the mountain regions (Sing & Kumar 1997; Siddiqui et al 2003). The SRM normally does not require calibration as its input parameters are generally derived from the field data and historical records through physical laws and empirical relationships. However, gathering of all the required data is only a dream for a highly rugged mountain terrain in a country like Pakistan, where inaccessibility and lack of resources generally limit collection of such data. Hence, calibration of the model and some adjustment of few input parameters is quite necessary and in fact the user gains more confident over the simulation results. Therefore, the SRM was calibrated against the daily river inflows of year 2003 and was validated by backward as well as forward verification for the daily flows of years 2002 and 2004.

48

The accuracy of the model calibration is judged from the two well established accuracy criteria, the coefficient of determination (R2) and the deviation of runoff volumes (Dv), which are described earlier in the equations (4) and (5) respectively. 4.9 Model Simulations

Once the model is adequately calibrated, it can be run for a number of scenarios as per requirements. One of the major objectives of this study is to quantify the snowmelt and rainfall runoff, which can be achieved by the following two ways; i.e. either by setting the runoff coefficient for rainfall as zero or setting the SRM to run with zero input of precipitation. Practically it is not possible as this precipitation is the only source of snowfall. But, since the SRM does not take input of precipitation to be converted as snow, instead it takes the input of daily snowfall from outside source such as remote sensing, the results are unaffected. The other major objective is to relate the snowmelt runoff with the observed snowcover for different times. When the calibrated model is run differently for three years, it takes temperature input of that particular year and in this way the effect of temperature is also incorporated in snowmelt runoff generation. Since temperature is the only source of energy to melt the available snow, it can not be set to zero as in case of precipitation. Instead the effect of temperature was normalized (equalized) by assigning the average values of daily minimum and maximum temperatures for all the three runs (2002-04), rather than their own temperature data. This way whatever the effect of temperature is, it remains the same for each run and only the effect of varying snowcover is simulated. The simulated runoff achieved this way is then related with the average observed snowcover for different months. 4.10 Model development

As described in the preceding paragraph, the simulations with the input of normalized temperature are run to derive the variation of snowmelt runoff only as the function of snowcover change. These daily runoffs are then averaged to compute the average

49

monthly runoff, which are plotted against the average monthly observed snowcover obtained similar way and the best fit regression model is developed for forecasting the one variable from the other. Similarly, such regression models are also developed by relating the snowcover with the actually observed river runoff. The river runoff however also contains the rainfall runoff component, which has very high variability as compared to snowcover and snowmelt runoff. Even then the prediction models developed this way can be a good tool for a rough estimate of runoff from the snowcover.

50

CHAPTER RESULTS AND

FIVE DISCUSSION

5.1

Outline

The results and discussion chapter has been divided into four main sections. The 1st part discusses the results of derived input parameters while the 2nd component covers altitudinal, spatial, and temporal distribution of snowcover estimated through remote sensing in the Swat River Basin. The 3rd part presents snowmelt runoff modeling results including SRM calibration, verification and simulations. The last segment is dedicated to the development of relationship between the observed river flows as well as simulated snowmelt runoff with the computed snowcover for different time intervals. 5.2 Parameter Estimation

The analysis started with delineation of river network and watershed boundaries of the study area from the SRTM DEM data using ArcHydro GIS software. Figure 5.1 presents the delineated river network and basin boundary of the whole Swat Basin and study area (upper Swat basin). The total area of the basin is 5713.38 sq. km with a mean hypsometric elevation of 2727.2 m. Since the SRM represents a semi-distributed approach, considering each catchment section with similar hydrological characteristics as a single unit (hydrological response unit, HRU), the basin is divided into five elevation zones (Zone-A to Zone-E) keeping in view the available elevation range of 686 m 5808 m, as described in the Figure 5.2 The area occupied by each elevation zone is 23.16, 23.10, 19.47, 26.99, and 7.28 % of the total basin area respectively. The plot of cumulative area versus elevation (area-elevation curve) is presented in the Figure 5.3. The mean hypsometric elevation for each elevation zone is 1133.42, 1956.63, 3014.76, 4007.57, 4726.55 m respectively. The mean hypsometric elevation of each zone is used as an elevation to which the base or reference station temperatures are extrapolated for the calculation of degree days. The elevation distribution depicts northern part of the basin with high mountainous terrain having elevation range of 1500 5808 m, while the southern part is relatively flat with elevation ranging from 686 2500 m a.s.l.

51

Figure 5.1 Delineated river network and watershed area of the Swat River Basin

52

Figure 5.2 Elevation zones, their areas & mean hypsometric elevation.

Figure 5.3 Area-elevation (hypsometric) curve of the upper Swat river basin 53

Due to the sensitivity of temperature to elevation and higher elevation range of the study basin, the temperature data of its sole met station at Kalam has been extrapolated. Figure 5.4 presents the average daily minimum and maximum temperature at Kalam. The temperature lapse rates for different months are calculated using the temperature records of few other stations located outside the study area at various elevations (Figure 5.5). The average monthly temperature of all these met stations is plotted against their elevations and the best fit regression line is drawn. A very high linear correlation can be found between the temperature and elevation for all the met stations throughout the year. Figure 5.6 presents these regression models which are used to compute the values of the temperature lapse rate for different months. The computed lapse rates are 0.68, 0.69, 0.69, 0.67, 0.70, 0.73, 0.62, 0.61, 0.64, 0.68, 0.66, and 0.65 oC / 100 m for January to December months respectively. These lapse rates were input to the SRM for all its scenarios. However, for precipitation there is no need of such a practice. Instead as mentioned earlier, precipitation data of Saidu is used for Zone-A area, Kalam for ZonesB, -C, and D areas and Shandur for Zone-E area. The recession coefficient was calculated from the discharge data of 2001-2005, whereas the other parameters were adjusted during model calibration and verification.

M aximum
35 Average Daily Temperature (oC) 30 25 20 15 10 5 0 -5 -10 -15
M J A J F M J

M inimum

Month

Figure 5.4 Average daily minimum and maximum temperature at Kalam

54

Figure 5.5 Meteorological stations used for computation of temperature lapse rate.

15 10 Temperature (oC) 5 0 -5 -10 -15 -20 0 1000 2000 3000 y = -0.0068x + 13.999 R2 = 0.985

Jan y = -0.0069x + 16.205 R2 = 0.9763

Feb

4000

5000

Elevation (m)

Figure 5.6 (a) Relationship between temperature and elevation for Jan and Feb months

55

25 20 Temperature (oC) 15 10 5 0 -5 -10 -15 0 1000 2000 3000 y = -0.0069x + 21.074 R2 = 0.9747

Mar

Apr

y = -0.0067x + 26.234 R2 = 0.9752

4000

5000

Elevation (m)

Figure 5.6 (b) Relationship between temperature and elevation for Mar and Apr months

35 30 Temperature (oC) 25 20 15 10 5 0 -5 0 1000 2000 3000 y = -0.007x + 31.11 R2 = 0.9755

May y = -0.0073x + 36.314 R2 = 0.9539

Jun

4000

5000

Elevation (m)

Figure 5.6 (c) Relationship between temperature and elevation for May and June months

35 30 Temperature (oC) 25 20 15 10 5 0 0 1000 2000 3000 y = -0.0061x + 36.113 R2 = 0.8926

Jul

Aug

y = -0.0062x + 36.721 R2 = 0.8866

4000

5000

Elevation (m)

Figure 5.6 (d) Relationship between temperature and elevation for Jul and Aug months 56

30 25 Temperature (oC) 20 15 10 5 0 -5 -10 0 1000 2000 3000 y = -0.0068x + 27.148 R2 = 0.9558

Sep

Oct

y = -0.0064x + 32.896 R2 = 0.937

4000

5000

Elevation (m)

Figure 5.6 (e) Relationship between temperature and elevation for Sep and Oct months

20 15 Temperature (oC) 10 5 0 -5 -10 -15 -20 0 1000 2000 3000 y = -0.0065x + 15.784 R2 = 0.989

Nov

Dec

y = -0.0066x + 21.005 R2 = 0.982

4000

5000

Elevation (m)

Figure 5.6 (f) Relationship between temperature and elevation for Nov and Dec months 5.3 Snowcover Estimation

Snowcover estimation is an integral part of hydrological modeling as it provides basic information for calculating snowmelt runoff from any snow-fed basin. The areal extent of snowcover is two-dimensional information and is an important variable for snowmelt runoff computations, because each daily melt water volume stored in a basin is obtained as a product of this area and the associated snowmelt depth. Determining contribution of snowmelt runoff to total river runoff has great practical significance as snowmelt runoff is more dependable source of fresh water. Unfortunately, in the high mountainous terrain

57

with an extreme and harsh climate, such as HKH region of Pakistan, where highly rugged terrain provide limited accessibility and little ground control, it is very difficult to monitor metrological data and snowcover information accurately on a continuous basis. The ruggedness further complicates the definition of snow line owing to occurrence of snow in patches. In such circumstances satellite remote sensing has great value and seems to be the only viable alternative, as it can provide repetitive data on snow area extent at different, regular time intervals. The study utilizes MODIS snowcover products to estimate snow area extent in the Swat river basin of Pakistan. The MODIS 8-daily (level 3, version 5) maximum snow extent composite snowcover product (MOD10A2) was processed in a GIS environment to determine spatial and temporal variation of snow cover in the basin. The MOD10A2 snow cover product contains information on the presence or absence of a number of classes (Table 3.4). The study area is covered by the h23v05 tile of the MODIS sinusoidal grid. In all 140 MOD10A2 images of that grid, distributed over three years period (Jan 2002 to Dec 2004), were analyzed. Figure 5.7 presents a selected sequence of time series GIS processed snowcover maps of the basin for the three years. The gradual or sometimes abrupt increase in areal extent of snowcover during winter months and its gradual decrease during the subsequent summer season, a typical phenomenon of the mountain snow hydrology, is prominent in all the maps of Figure 5.7. The MOD10A2 snowcover product was further processed using GIS techniques to determine altitudinal distribution of the snowcover. Figures 5.8 present temporal and altitudinal variation of snowcover for the five elevation zones during three year study period, whereas Figure 5.9 shows three years average conditions. The analysis and visual observation of the generated snowcover maps and developed graphs reveal that snowfall and subsequent snowmelt in the Swat river basin is highly variable in terms of altitude, space and time. Winter snowfall usually starts by the mid to late September initially at higher elevations and snow area may be increased abruptly from less than 2% in August to about 10 20 % of the total basin area. Occasional and unpredictable rainstorms in September and

58

October months sometimes bring immediate and abrupt but significant increase in snowcover area and snowcover may cover about 45% of the total basin area by the end of October. However, the following few weeks are unable to maintain that tempo and consequently some decline in snowcover is usually observed in many cases due to subsequent and immediate melting of that fresh and temporary snowcover. The main winter months (Nov Feb) generally bring in most of the snowfall and snowcover keeps accumulating reaching its peak area by the end of January or early February covering about 58 64 % of the basin area. Significant snowfall at lower elevations is also witnessed during these main winter months as the snowcover gets extended down to valleys in southern parts and snowline may reach at elevations less than 1500 m. However, this snowcover at lower elevations completely disappears by mid to late March when snowmelt season starts. At higher elevations above 3500 m a.s.l. snow continues to fall even in March and April months as can be observed in Figures 5.8 d and 5.8 e when snow area in 2003 was increased in both these months. Although some occasional snowfall has also been witnessed during the main summer months (May Aug) particularly when continued rainfall significantly brings down the temperature. However, it rarely happens as snow in these months does not last for long but rather melts very soon. So, practically this occasional summer snowfall does not make any difference and can easily be neglected. Hence, this four month (May Aug) period can be termed as purely snowmelt season during which snowcover gradually declines. Based on the three years daily snowcover observed through remote sensing, the average monthly snowcover in the upper Swat river basin can best be described by the fourth order polynomial function with highest peak in February and lowest peak in August months as depicted in the Figure 5.9, which also compares average monthly snowcover among three years. Figures 5.8 (e) and 5.9 depict very high variability of mid September to late October snowcover among the three years period due mainly to uncertainty and variability of precipitation and marginal temperatures during this period.

59

Figure 5.7 (a) Temporal variation of snowcover in the upper Swat basin (Jan - Apr)

60

Figure 5.7 (b) Temporal variation of snowcover in the upper Swat basin (May-Aug)

61

Figure 5.7 (c) Temporal variation of snowcover in the upper Swat basin (Sep Dec)

62

2002 Daily Snowcover (% of Zone-A Area) 30 25 20 15 10 5 0

2003

2004

Average

Figure 5.8 (a) Temporal variation of snowcover in the Zone-A (686 1500 m a.s.l)

J
50 40 30 20 10 0

M onth

2002 Daily Snowcover (% of Zone-B Area)

2003

2004

Ave rage

Figure 5.8 (b) Temporal variation of snowcover in the Zone-B (1501 2500 m a.s.l)

J
100 80 60 40 20 0

M onth

2002 Daily Snowcover (% of Zone-C Area)

2003

2004

Ave rage

Figure 5.8 (c) Temporal variation of snowcover in the Zone-C (2501 3500 m a.s.l)

M onth

63

2002 Daily Snowcover (% of Zone-D Area) 100 80 60 40 20 0

M

2003

2004

Ave rage

Figure 5.8 (d) Temporal variation of snowcover in the Zone-D (3501 4500 m a.s.l)

J
100 80 60 40 20 0

M onth

2002 Daily Snowcover (% of Zone-E Area)

2003

2004

Average

Figure 5.8 (e) Temporal variation of snowcover in the Zone-E (4501 5808 m a.s.l)

J
70 60 50 40 30 20 10 0

M onth

2002 Daily Snowcover (% of Basin Area)

2003

2004

Average

Figure 5.8 (f) Temporal variation of snowcover in the whole basin

M onth

64

2002 70 Average Monthly Snowcover (% of Basin Area) 60 50 40 30 20 10 0 1 2 3 4

4

2003

2004
3

Average
2

y = -0.0368x + 1.1749x - 11.052x + 28.518x + 33.747 R = 0.9895

2

10

11

12

Calender M onth Number

Figure 5.9 Comparison of average monthly snowcover variation in different years

Snowmelt generally continues throughout the year but contribution of winter snowmelt runoff is often insignificant. Flow during the winter season is usually augmented from surface flow due to seasonal rains, sub-surface flow, and ground-water contribution and is termed as the base flow. Unlike snowfall, snowmelt usually progresses gradually and smoothly and is more easily predictable. The summer snowmelt normally gets momentum in the month of March which also brings in some new snows at times of cold waves accompanied with precipitation particularly at higher elevations. The net outcome however is towards snowmelt. At first the snow starts disappearing rapidly from valleys at southern parts of the basin and from elevations less than 2500 m in early March, which gradually widens and the snowline retreats upward as the summer season progresses and temperature gets increased. At elevations greater than 4500 snowmelt starts in late April and continues till mid September. Starting in March, the rate of snowcover retreat reaches at its peak in June and thereafter declines rapidly up to August and consequently snowmelt runoff also reaches at its peak in late June or early July and thereafter declines gradually up to August. During July to mid September temperatures are usually sufficient enough to melt

65

the snow and snowmelt is mainly the function of available snow, which is mostly concentrated at highest elevations and is about to finish. Minimum snowcover is usually observed in the late August until the new snowfall season starts in September. During the monsoon season, the peak snowmelt runoff sometimes is augmented by monsoon rains to produce higher discharges and occasional peak floods sometimes destroying the infrastructure. The three years snowcover monitoring with remote sensing shows that under conducive climatic conditions, the maximum snow area extent may cover about 64 % of the total area of the basin during January-February to as low as 1.7 % in late August during the snowmelt season. However, not always the same area receives snowfall. Spatial analysis of the three years snowcover maps in a GIS environment (Figure 5.10 and Table 5.1) show that about 79.14 % of the area received snowfall at any time during 2002 2004. This area can be termed as area which generally accommodates temporary and seasonal snowfall. A handful of 20.72 % never received snowfall during that period, while only in 0.14 % (8.187 sq. km) area of the basin the snowcover remained in tact and could not be melted during that three years period. This area can be termed as permanent snow. It means that the entire basin predominantly accommodates temporary and seasonal snowcover, which is an important element of the hydrological cycle of the basin and major contributor to the basins fresh water resources. PARC and ICIMOD (2005) identified six types of 200 glaciers present in the basin covering an area of about 195.84 sq. km using the single time Landsat-7 ETM+ Imagery obtained in September-October 2001 as shown in Figure 5.11. The size of these 200 glaciers varied from 0.06084 sq. km to 8.02 sq km. The number of glaciers having size less than 0.25 sq. km were 46 covering an area of 7.43 sq. km. The present study however observed minimum snowcover of only 97.13 sq. km in the 1st week of September 2003 during the three years study period and only 8.187 sq. km permanent snowcover, which could not be melted during that period. The location of this permanent snowcover is also by and large different than the glacier location identified by the PARC and ICIMOD. The difference in glacier area between the two studies may be

66

attributed to the reasons; either they may have overestimated the glacier area as in SepOct months winter snowfall gets started; or these glaciers might have retreated during the meantime. The other reason may be underestimation of this study due to coarse resolution of 500 m (0.25 sq. km), which might have overlooked smaller sized glaciers. However, one thing seems quite clear that significant flow does take place from the glacier melt particularly in July Sep months.

Figure 5.10 Permanent and temporary/seasonal snow cover

67

Table 5.1 Area under permanent and temporary snow cover for three study years Year 2002 2003 2004 2002-2004 Permanent Land (sq. km) 1617.960 (28.32) 1809.644 (31.67) 1675.184 (29.32) 1183.646 (20.72) Temporary Snow (sq. km) 4043.269 (70.77) 3888.715 (68.06) 3981.328 (69.68) 4521.546 (79.14) Permanent Snow (sq. km) 52.150 (0.91) 15.021 (0.26) 56.867 (1.00) 8.187 (0.14)

Figure 5.11 Glacier location and extent as identified by PARC & ICIMOD 2005.

68

5.4 5.4.1

Snowmelt Runoff Modeling Calibration and Verification Results

After derivation of the variables and parameters necessary for model input, the SRM was run and calibrated for the river flows of year 2003. During the process of calibration deficiencies in some of the input parameters were identified which were adjusted accordingly. The WinSRM program includes a good facility of graphical display of the simulated and observed hydrographs of the river runoff. This visual examination at the first glance shows whether the simulation adequately represents the flow conditions or not. Additionally, the SRM uses two well established and statistically valid accuracy criteria, namely, the coefficient of determination (R2) and the deviation of runoff volume (Dv) to evaluate model calibration in quantitative terms. Again the model displays the results of both these criteria terms and there is no need of any manual calculations or graphical representations. After calibrating the model for 2003 river flows, the model was run for 2004 to verify the calibration by inputting the daily records of temperature, precipitation and snowcover for that year. Few deficiencies in some of the input parameters were observed and these parameters were adjusted once again to match the flow regimes of both years. Similarly the model was verified for 2002 year with temperature, precipitation and snowcover inputs of its own and input parameters were further refined. With this forward as well backward verification, the SRM is ready for any simulations as per requirements. Figures 5.12, 5.13 and 5.14 show the plots of the observed and simulated river flows for years 2002, 2003 and 2004 respectively, while Table 5.2 presents the simulation statistics and calibration results of the two accuracy parameters. The coefficient of determination is 79.60, 82.37 and 80.15 for years 2002, 2003 and 2004 respectively and the volume difference for these years is 2.815, - 4.077 %, and 3.202 % respectively. The minus sign indicates overestimation of simulated runoff by the SRM. These calibration and verification results can be termed quite good and well under acceptable limits as SRM have been applied in the past with 60 % and 8 % values of both these criteria

69

respectively (Martinec 1995). Hence this calibrated and verified model can be used for simulation and forecasting.

Figure 5.12 Simulated and observed river flows for calibration year of 2003

Figure 5.13 Simulated and observed river flows for verification year of 2004

70

Figure 5.14 Simulated and observed river flows for verification year of 2002

Table 5.2 Year round simulation statistics for different study years Simulation Measured Year Runoff Volume (106 m3) 2002 2003 2004 4465.183 5742.862 5874.324 Simulated Runoff Volume (106 m3) 4590.864 5977.021 5686.182 Volume Difference (%) - 2.815 - 4.077 3.202 Coefficient of Determination (R2) 0.7960 0.8237 0.8015

5.4.2

Simulation Results

Keeping in view the specific objectives of this study three scenarios have been developed. The first scenario runs the model with each years own data and computes the daily runoff. The second scenario runs the model for each years data but with no rainfall to calculate the respective share of two runoff components i.e. snowmelt and rainfall

71

runoff. The third scenario runs the model for each year with no rainfall and with normalized (putting historical average temperature values rather than each years own temperature data) temperature. This scenario is developed to normalize the effect of temperature. It means whatever the effect of temperature is, it remains the same for each year and only the effect of snowcover change on snowmelt runoff is simulated. The distribution and share of simulated potential runoff for different zones is shown in the Figure 5.15. These components of accumulated runoff are the total (potential) depth of water which could be generated at the source. The contribution of new snow is computed from the input of precipitation and critical temperature, which determines the form of precipitation. The runoff water reached at the gauge station however will be significantly lower due to runoff losses. Figures 5.16, 5.17 and 5.18 present the simulated snowmelt and rainfall runoff components for the three study years at the Chakdara gauge station computed through the SRM. The graphs of Figure 5.15 are just to have general idea of the contribution offered by the three runoff components. Very similar trend, in terms of start and end times, is quite clear in Figures 5.16, 5.17 and 5.18 as well. These figures clearly indicate dominancy of snowmelt runoff as the basin is predominantly a snow-fed. However, there is also significant contribution of rainfall runoff particularly in Mar May and Jul Sep periods, whereas the rainfall contribution of the rest six months is considerably less. Although there occurs higher precipitation in the winter months but it usually falls in the form of snow due to low temperatures. June on the other hand is the driest month of year so rainfall contribution during this month is also very low. However, due to maximum temperatures during June, it receives highest snowmelt runoff. Since, snowmelt runoff in a snow-fed basin is mainly the function of temperature and available snowcover therefore it responses accordingly with the change in temperature and available snowcover.

72

Figure 5.15 Cumulative runoff components in various zones for the simulation year 2004 (Red is initial snow, green is new snow, and blue is contribution of rain).

73

Snowmelt Discharge 800 Average Daily Discharge (Cumec) 700 600 500 400 300 200 100 0

Rainfall Discharge

Figure 5.16 Computed snowmelt and rainfall runoff components for the Year 2002

J
800 Average Daily Discharge (Cumec) 700 600 500 400 300 200 100 0
J

Figure 5.17 Computed snowmelt and rainfall runoff components for the Year 2003

F
F

M onth

Snowmelt Discharge

Rainfall Discharge

M onth

74

Snowmelt Discharge 800 A verage D aily D ischarge (C umec) 700 600 500 400 300 200 100 0
M M

Rainfall Discharge

Figure 5.18 Computed snowmelt and rainfall runoff components for the Year 2004 The snowmelt runoff from Dec Feb mostly remains in between 30 50 m3/sec due to less variability of temperatures in these three months. The summer snowmelt runoff however has very high variability among the months as well as among the years. At elevations less than 1500 m (Zone-A area, which is located in the active monsoon belt) flow is mainly coming from rainfall runoff, which contributes almost 70% of the total zonal runoff. The contribution of snowmelt runoff in this elevation zone is received during main winter months only. In the areas having elevation range of 1500 2500 m, rainfall contribution starts by mid March to mid December, during rest of the period it falls as snow. Snow continues to melt almost throughout the year in this zone, however during December to January; its contribution is very little and is mainly coming from fresh snow. In Zone-C area (elevations range of 2500 3500 m), snowmelt usually starts in mid March and continues till the end of November with almost similar trend of rainfall runoff. At higher elevation range of 3500 4500 m (Zone-D area) snowmelt commences in late

M onth

75

April and continues up to September and at further high elevation range of greater than 4500 m, the snowmelt runoff is only generated by the mid of May to mid September. The rainfall contribution in both these zones is mainly received in monsoon season. Precipitation during rest of the period is in the form of snowfall. On the basis of three years simulation results, the study basin is predominantly a snowfed as the annual snowmelt runoff contribution to the total runoff may ranges from 65 75 %. Figure 5.19 shows the average contribution of the two runoff components (snowmelt and rainfall) for each month to the total runoff from the basin. About 65.5 % of the total runoff (45.9 % snowmelt and 19.6 % rainfall) is generated in the four main summer months (May Aug). The results further suggest that about 30 60% of the total rain fall runoff occurs in monsoon season (Jul Sep) and about 25 50 % in Mar to May period. Figure 5.20 presents the average share (in per cent) of each runoff components to the total monthly runoff generated in different months. The average contribution of snowmelt runoff to the total monthly runoff is 98.5, 91.2, 61.3, 61.6, 70.8, 83.0, 67.6, 53.3, 61.5, 73.1, 82.5, and 86.7 % for Jan Dec months respectively. The average monthly snowmelt discharge from the basin estimated through SRM can be described by the third order polynomial functions for the two halves of a calendar year, which are presented by the Figures 5.21 and 5.22 for January to June and July to December respectively. Snowmelt runoff reaches at its peak by the end of June and declines gradually in both ways i.e. before and thereafter.

76

Snowmelt Runoff
24 Average Monthly Contribution to Total Runoff (%) 20 16 12 8 4 0 Jan Fe b M ar Apr M ay Jun

Rainfall Runoff

Jul

Aug

Se p

Oct

Nov De c

M onth

Figure 5.19 Average contribution of the two runoff components to the total runoff generated from the basin.

Snowmelt Runoff
100 90 Average Monthly Sim ulated Discharge (% ) 80 70 60 50 40 30 20 10 0 Jan Feb Mar Apr May Jun

Rainfall Runoff

Jul

Aug

Sep

Oct

Nov

Dec

Month

Figure 5.20 Contribution of two runoff components to the total monthly runoff

77

400 350 300 250 200 150 100 50 0

Average Monthly Snowmelt Discharge (Cumec)

y = 0.7517x + 10.356x - 37.49x + 56.917 R = 0.9905

2

Calendar M onth Number

Figure 5.21 Average monthly distribution of snowmelt runoff in Jan Jun months

Average Monthly Snowmelt Discharge (Cumec)

300 250 200 150 100 50 0 7 8 9 10 11 12 Calendar M onth Number y = -3.6406x + 50.159x - 235.08x + 425.97 R = 0.9991
2 3 2

Figure 5.22 Average monthly distribution of snowmelt runoff in Jul Dec months

78

5.5

Relationship of Snow Area Extent with River Discharge and Snowmelt Runoff

The snowfall and corresponding snowmelt in a particular river basin are mainly the function of topography and meteorology. The topographical factors may remain unchanged while the meteorological factors may have very high temporal, spatial and altitudinal variability. Fortunately the SRM takes input of daily snowcover determined through remote sensing and models only the snowmelt process by taking input of temperature and precipitation data. As per objectives of the study the SRM was run for different scenarios discussed in preceding paragraphs. This section summarizes the results of these scenarios and relates the simulated snowmelt runoff (without rainfall runoff component achieved through normalized temperature input) and observed river discharges at the Chakdara gauge station with the estimated snow area extent. The study has quantified the monthly and seasonal cycles and variations of river flows and simulated runoff with the snow area extent and identified a clear correspondence of river flows and simulated snowmelt runoff to the change in snow area extent. In practice normally the summer snowmelt runoff is mainly the function of winter snowcover as snowfall hardly occurs during summer. Consequently, some researchers (Rango et al 1977; Dey et al 1983; Tarar 1982) have related winter snowcover (normally at the end of winter when snowfall is stopped) with the total summer season runoff volume. However, winter snowcover can be related with summer snowmelt runoff when there is no snowfall during the forecasting period and snowcover data is not available for that period. Moreover, relating winter snowcover with summer runoff volume seems an impractical approach as farmers as well as water resources planners are more interested in daily, weekly or monthly discharges rather than total seasonal volume. Further, most of the past studies have related snowcover with observed river runoff volume, which incorporates the rainfall runoff component which usually have higher variability and low predictability and may also have significant contribution particularly in rainfed areas and the observed flows may not be the true representative of the associated snow cover.

79

This study on the other hand employs the daily records of snowcover which show that snowfall can take place during eight months (Sep Apr) and even minute amounts can be observed during the four main summer months. Therefore relating winter snowcover with total summer runoff volume will give correct estimates for only the four main summer months (May August) and may not be a wise approach as for as the other seasons and monthly are daily estimates are concerned. Instead this study not only relates the daily river discharges with the daily snow area extent but also develops prediction model for the total runoff volume of the four main summer months. The study also relates the simulated snowmelt runoff (excluding rainfall runoff component) with the snow area extent. Relating snow area extent with the snowmelt runoff (excluding rainfall runoff) rather than the observed river flows, which also contain rainfall runoff component, is absolutely logical concept particularly for a snow-fed river basin. The contribution of rainfall runoff may be added after the estimation to compute total river discharge. But the major problem with this kind of approach lies in the uncertainty involved in the simulated snowmelt runoff estimation. However, selection of the best model, accurate estimation of model input parameters and adequate calibration and verification of the selected model may significantly avoid this problem. To achieve this, the study uses the SRM, which takes daily inputs of snowcover during the whole simulation period and the effect of specified snowcover in the winter months is superseded by the subsequent snowcover inputs on the following days. Hence, the snowcover keeps changing as an outside input rather than modified by the model itself during the different simulation time periods. Therefore the model results dont totally describe summer snowmelt runoff as the function of winter snowcover. Due to unpredictability and high variability of weather, snowcover is subjected to vary each year for different days and months. The same is also true for temperature, which strongly influence the melting of snowcover. For example certain month of a year may receive significantly different snowcover at its start during different years due to variation or shift of weather, and this difference may sometimes exceed the average total snowcover depletion in that month. In such cases there may be very high variability of snowmelt runoff and estimates based on average conditions might contain some degree of

80

error. It means that there is need to relate the snowcover with runoff for a particular month in both ways, i.e. for year to year variation and for during month or year variation. Hence, it is really difficult to exactly relate snowcover with river discharge as snowmelt runoff on each day of the year may be significantly different because of temperature and available snowcover variation particularly in snowmelt season. If we neglect the variation of temperature on a particular day of multiple years, we can easily relate snowcover with runoff for that particular day and such an ideal situation will lead development of 365 regression models (separate model for each day of the year). However, this would be an absolutely impractical approach as no one likes such a large number of models. Obviously, snowmelt runoff on a particular day is directly proportional with the snowcover available on that day. The magnitude of this proportion, however, may be significantly different for each day of a year due to temperature variation. If we average the snowcover and corresponding runoff of each day of multiple years then snowmelt runoff may behave systematically during the course of year and can be related with the available average snowcover through regression models for different time intervals (months or season). However, since this approach uses average conditions, it may result in serious errors in extreme cases when a particular day or month receives significantly different snowcover than the average snowcover due to drastic change or shift in weather conditions. As described earlier, this study utilizes snowcover and corresponding river discharge and snowmelt runoff data of three years (2002 2004), which are averaged and then related with the average snow area extent using the regression analysis. To quantify and analyze the relationship of mean daily snow area extent with the observed river discharges and simulated snowmelt runoff from the basin, their records for the period of 2002-2004 are examined and compared in the Figure 5.23. It clearly indicates a definite response of observed river discharges and simulated snowmelt runoff to seasonal snow cover changes, i.e. an association of low stream flows with high snow area extent during the winter season (Sep Feb), an increase in discharge associated with a decrease of snow area extent during the early summer (Mar Jun), and decrease in discharge with decreasing snowcover in the late summer, monsoon season (Jul mid Sep).

81

Snowcover
70 60 Av Daily Snowcover (% of Basin Area) 50 40 30 20 10 0
J

River Discharge

Snowmelt Runoff
700 600 500 400 300 200 100 0 Av Daily Discharge (Cumecs)

Figure 5.23 Temporal distribution of average daily snow area extent, observed river discharge and simulated snowmelt runoff Based on the three years time series data of MODIS snowcover products for the study area, the regression models for various time periods are developed to estimate the average daily river discharge and snowmelt runoff from the average daily snowcover available at different times of the year. The estimated average daily snowcover is plotted against the average daily observed river discharge and simulated snowmelt runoff computed with zero input of rainfall and normalized temperature in the upper Swat river basin of Pakistan. The results generally confirm a very strong linkage between the river flows and snow area extent in the basin. The first of these relationships is presented in the Figure 5.24 for the early summer snowmelt season starting from March and extending up to June. The relationship of average daily snow area extent with the observed daily river discharge for this period can be described by the negative linear regression model as the river discharge increases with decrease in corresponding snowcover. Its relationship with the daily simulated snowmelt runoff is also negative but slightly different and is best explained by the third order polynomial function. This difference between the two regression models is due to incorporation and variation of rainfall runoff component in the river discharges.

M onth

82

River Discharge 600 500 400 300 200 100 0 0 10 20 30

Snowmelt Runoff

y = -9.2607x + 558.38 R = 0.9507 y = -0.0022x + 0.3374x - 23.303x + 628.54 R = 0.9522

2 3 2 2

A verage D aily D ischarge (C umec)

40

50

60

Average Daily Snowcover (% of Basin Area)

Figure 5.24 Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for March June months. It is worth mentioning that increase in river discharge is not due to decrease in snowcover, rather decrease in snowcover is due to its melting, which ultimately increases river discharge. Moreover, this inverse relationship is only true for the first part of the snowmelt season during which availability of snowcover is generally not a limiting factor and snowmelt runoff is largely the function of available temperature. But as the melting season progresses, the available snowcover gets depleted and it starts limiting the snowmelt runoff more than the temperature. Consequently, the relationship of the snowcover with the snowmelt runoff and river discharge during this second part of the snowmelt season (July August) is completely different from that for the first part. During this late summer monsoon period most of the temporary and seasonal snowcover at lower to medium elevations is melted and snowmelt runoff mainly comes from the permanent snow and glaciers of higher elevations. The relationship of average daily snowcover with the average daily river discharge and snowmelt runoff for this second part of snowmelt season (July August) is shown in the Figure 5.25. Unlike the previous model, this regression model shows positive relationship of average daily snowcover with 83

the two runoffs. Also, there is exchange in type of regression model between the two relationships. The average daily snowcover now relates the simulated snowmelt runoff linearly, whereas its relationship with the average daily observed river discharge can be simplified by the second order polynomial function. The river discharge during the early July month tends to remain constant but slightly and occasional greater river discharges in mid or late July than the early July month are due to greater contribution of rainfall runoff component during that period, otherwise snowmelt runoff decreases linearly during the following period.

River Discharge 600 500 400 300 200 100 0 0 2 4 y = -14.331x + 210.35x - 276.33 R = 0.8785
2 2

Snowmelt Runoff

Average Daily Discharge (Cumec)

y = 65.898x - 99.637 R = 0.9489

2

10

Average Daily Snowcover (% of Basin Area)

Figure 5.25 Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for July August months.

The relationship of the average daily snowcover with the river discharge and snowmelt runoff for the remaining six moths (September February), the winter season, is shown in the Figure 5.26. This relationship shows a completely different condition. The two runoff discharges are now decreasing with increase in snowcover. Apparently, this is an unbelievable trend as snowcover has always positive impact on the snowmelt runoff. It is again worth mentioning that snowmelt runoff still causes the snowcover to deplete but

84

due to onslaught of winter season, the new snowfall of the season has been started and the net effect on the snowcover is positive resulting in increase of snow area extent. The relationship of snowcover with both the runoffs for this time period is described by the third order polynomial function.

River Discharge 200 180 Average Daily Discharge (Cumec) 160 140 120 100 80 60 40 20 0 0 10 20 30

Snowmelt Runoff
3 2

y = -0.0012x + 0.1305x - 6.2054x + 179.88 R = 0.8375 y = 0.0007x - 0.0647x + 0.6011x + 79.582 R = 0.9414
2 3 2 2

40

50

60

70

Average Daily Snowcover (% of Basin Area)

Figure 5.26 Relationship of average daily snowcover with average daily simulated snowmelt runoff and average daily observed runoff for September February months. Apart from the above predictive regression models based on daily data, the total runoff volume of the four main summer months (May August) is related with the snowcover observed at the start of May month for the purpose of seasonal water resources planning. These four months are selected because there occurs hardly any snowfall during the May August period and their contribution to total river discharge is about 64%. To accomplish this objective snowcover data at the start of May (1 8th May) month for few other years was processed to develop prediction model of hydrological significance. The previous models, developed for prediction of daily flows, were based on three years average data, whereas this model is developed for incorporating snowcover and runoff data of five years, i.e. 2001 2005.

85

The results of this seasonal model are summarized in the Figure 5.27. This model also shows a very strong association of the observed river runoff volume of the four main summer months with the late April snowcover. The relationship of snowcover at the start of May with the total runoff volume for the four months is explained by the linear function of regression model. The correlation coefficients for all the models are quite high depicting strong correlations.

10 May-Aug Runoff Volume (BCM)

y = 0.1645x - 1.57 R = 0.8966

2

0 0 10 20 30 40 50 60

Snowcover on May 1-8 (% of Basin Area)

Figure 5.27 Prediction model for estimating May Aug runoff volume from the snowcover estimated on May 1-8.

86

CHAPTER CONCLUSIONS AND

SIX

RECOMMENDATIONS

6.1

Conclusions

The altitudinal, spatial and temporal distribution of snowcover in the Swat river basin of Pakistan was successfully evaluated using remotely sensed satellite imagery of the MODIS instrument, GIS techniques and snowmelt runoff modeling. A very high variability of snowcover during the calendar year was observed. Snowfall usually starts abruptly by the mid to late September increasing snow area extent from less than 2 % in August to about 10 20 % by the end of September. More abrupt increase in snowcover is observed in October month and snow area extent sometimes may cover 45 % of the basin area. The main winter months (Nov Feb) generally bring in most of the snowfall and snowcover keeps accumulating reaching its peak area of about 64 % by the end of January or early February. Significant snowfall at lower elevations is also witnessed during this period as the snowcover gets extended down to valleys in southern parts and snowline may reach at elevations less than 1500 m. Snowfall also continues in March and April months at higher elevations but the net result during this period is towards snowcover depletion due to its greater melting at the lower elevations. The occasional and very little snowfall at the highest elevations during the main summer months (May August) does not have any practical value therefore this period can be termed as purely snowmelt season during which snowcover gradually declines from around 40% at the start to less than 2% by the end of August. Snowmelt generally continues throughout the year but contribution of winter snowmelt runoff is often very low. Unlike snowfall, snowmelt runoff usually progresses gradually and smoothly and is more easily predictable. The summer snowmelt normally gets momentum in the month of March and increases linearly from around 30 60 m3/sec to more than 400 m3/sec to as high as 760 m3/sec in late June or early July. Snowmelt runoff thereafter declines gradually up to December reducing to 30 - 50 m3/sec. The December February snowmelt runoff normally tends to remain same. The July mid

87

September runoff is believed to be coming from the melting of permanent snow and glacier melt at the highest elevations as most of the snowcover at lower to medium elevations is finished. The runoff of the following period is primarily coming from the fresh snowfall precipitating in these months. On the basis of three years simulation results, the study basin is found predominantly a snow-fed as the annual snowmelt runoff contribution to the total runoff may ranges from 65 75 %. About 65.5 % of the total runoff (45.9 % snowmelt and 19.6 % rainfall) is generated in the four main summer months (May Aug). The results further suggest that about 30 60% of the total rain fall runoff occurs in monsoon season (Jul Sep) and about 25 50 % in Mar to May period. The average contribution of snowmelt runoff to the total monthly runoff is 98.5, 91.2, 61.3, 61.6, 70.8, 83.0, 67.6, 53.3, 61.5, 73.1, 82.5, and 86.7 % for Jan Dec months respectively. The study has quantified the monthly and seasonal cycles and variations of river flows and simulated runoff with the snow area extent and identified a clear correspondence of river flows and simulated snowmelt runoff to the change in snow area extent. The study observes a clear and definite response of observed river discharges and simulated snowmelt runoff to seasonal snow cover changes, i.e. an association of low stream flows with high snow area extent during the winter season (Sep Feb), an increase in discharge associated with a decrease of snow area extent during the early summer (Mar Jun), and decrease in discharge with decreasing snowcover in the late summer, monsoon season (Jul mid Sep). It employs the daily records of snowcover and relates the average daily snowcover with the daily river discharges and snowmelt runoff and also develops prediction model for the total runoff volume of the four main summer months (May Aug). 6.2 Limitations The study can give reasonably good estimates for average weather conditions. The developed regression equations better model the flow conditions for a progressing or continuing year. But if the snowcover of a particular day or month

88

is significantly different from the average conditions, then the results might incur some degree of error. The prediction models developed for snowmelt runoffs are based on the results of SRM application therefore the limitations of the model and uncertainties involved in input data are incorporated. However, selection of the best model, accurate estimation of model input parameters and adequate calibration and verification of the selected model may significantly avoid this problem. The study uses MODIS snowcover products directly, without testing and validating their accuracy in study area, therefore inaccuracies, if any, incorporated in this data are also accumulated. 6.3 Recommendations There is need to test and validate the MODIS snowcover products in the HKH region of Pakistan and compare its accuracy with the actually observed field data. The study findings are based on only three years daily records of snow area extent, any expansion in the study time period may improve the developed models. Also, there is need to test the developed regression models against the observed river flows and snow extent areas for different years.

89

References
Abbott, M.B., Bathurst, J.C., Cunge, J.A., OConnell, P.E. and Rasmussen, J. (1986), An introduction to the European Hydrological System/Systme Hydrologique Europen, SHE. 2. Structure of a physically-based, distributed modeling system. Journal of Hydrology 87, 6177. Ackerman, S. A., Strabala, K. I., Menzel, P. W. P., Frey, R. A., Moeller, C. C., & Gumley, L. E. (1998). Discriminating clear sky from clouds with MODIS. Journal of Geophysical Research, 103 (D24), 32141 32157. Ahmad, B. (2005), Development of a Distributed Hydrological Model Coupled with Satellite Data for Snowy Basins, thesis submitted to the University of Tokyo in partial fulfillment of requirements for the degree of Doctor of Engineering. Ahmad, S. and Joya M.F. (2003), Northern Areas Strategy for Sustainable Development Background Paper: Water, IUCN Pakistan, Northern Areas Progamme, Gilgit, Pakistan Anderson, E. A. (1973). National Weather Science River Forecast System Snow Accumulation and Ablation Model. Bethesda, Maryland, US National Oceanic and Atmospheric Administration. Anderson, E. A. (1976). A point energy and mass balance model of a snow cover. Silver Spring, Maryland, US National Oceanic and Atmospheric Administration. Arnell, N.W., (1999), Climate change and global water resources. Global Environmental Change, 9, S51S67. Arora, M., Singh, P., Gole, N.K. and Singh, R.D. (2006), Spatial Distribution and Seasonal Variability of Rainfall in a Mountainous Basin in the Himalayan Region, Water Resources Management (2006) 20: 489508. Springer. Baumgartner, M., Seidel, K., Martinec, J., Haefner, H., and Itten, K. I. (1985). Snow Cover Mapping for Runoff Simulations Based on Landsat-MSS Data in an Alpine

90

Basin. In Remote Sensing Instrumentation: Technology for Science and Applications, Vol. 1, pages 30-38. University of Massachusetts, Amherst, IEEE, No. 85CH2162-6. Bergstrom, S. (1975). The development of a snow routine for the HBV-2 model. Nordic Hydrology 6: 73-92. Barnes, W. L., Xiong, X. and Salomonson, V.V., (2003), Status of Terra MODIS and Aqua MODIS, J. Adv. Space Res., Vol. 32, no. 11, pp. 20992106. Brubaker, K., Rango, A. and Kustas, W. (1996) Incorporating radiation inputs into the Snowmelt Runoff Model. Hydrological Processes 10, 132943. Brun, E., Martin, E., Simon,V., Gendre, C., and Coleou, C. (1989). An energy and mass model of snow cover suitable for operational avalanche forecasting. Journal of Glaciology 35 (121): 333-342. Cline, D.W and Dozier, J. (1998), Estimating the spatial distribution of snow in

mountain basins using remote sensing and energy balance modeling, Water Resources Research, Vol. 34, No. 5, Pages 1275 1285, May 1998. Dai, Y. J., Zeng, X. B., Dickinson, R. E., Baker, I., Bonan, G. B., Bosilovich, M. G., & et al. (2003), The common land model, Bulletin of the American Meteorological Society, 84(8), 1013 1023. DeScally, F.A. (1994), Relative importance of snow accumulation and monsoon rainfall data for estimating annual runoff, Jhelum basin, Pakistan. Hydrological Sciences Journal, 39 (3), June 1994.

http://www.cig.ensmp.fr/~iahs/hsj/390/hysj_39_03_0199.pdf Dey, B., Sharma, V. K., Goswami D.C. and Subba Rao P. (1988), Snow Cover, Snowmelt and Runoff in the Himalayan River Basins, NASA CR 182434. Dingman, S. L., (1994) Physical Hydrology, Prentice-Hall, Inc.

91

Eigdir, A.N. (2003), Investigation of the snowmelt runoff in the Orumiyeh region, using modeling, GIS and remote sensing techniques, Thesis, submitted to the International Institute of Geo-Information Science and Earth Observation in partial fulfillment of the requirements for the degree of Masters of Science in watershed and environmental management, ITC, Netherlands Essery, R. (1997). Seasonal snow cover and climate change in the Hadley Centre GCM. Annals of Glaciology 25 : 362-366. Ferguson, R.I. (1999). Snowmelt runoff models Progress in Physical Geography 23,2 (1999) pp. 205227 Gomez-Landesa, E and Rango, A. (2002), Operational snowmelt runoff forecasting in the Spanish Pyrenees using the snowmelt runoff model, Hydrological Processes: 16, 1583 1591, John Wiley & Sons, Ltd. Guenther, B., Godden, G. D., Xiong, X., Knight, E. J., Qiu, S., Montgomery, H., Hopkins, M. M., Khayat, M. G., & Hao, Z. (1998). Prelaunch algorithm and data format for the level 1 calibrations products for the EOS-AM1 Moderate Resolution Imaging Spectroradiometer (MODIS). IEEE Transactions on Geoscience and Remote Sensing, 36(4), 11421150. Hall, D.K., Riggs, G.A. and Salomonson, V.V. (2006), MODIS snow and sea ice products, Earth Science Satellite Remote Sensing--Volume I: Science and Instruments. Qu J editor. Springer-Verlag Press: Berlin, Heidelberg. Hall, D.K. and Riggs, G.A. (2007), Accuracy assessment of the MODIS snow products, Hydrological Processes, Volume 21, Issue 12 , Pages 1534 - 1547 Hall, et al. (2001), Algorithm theoretical basis document (ATBD) for MODIS snow & sea ice mapping algorithms, http://modis.gsfc.nasa.gov/data/atbd/atbd_mod10.pdf Hall, D.K. Riggs, G.A. and Salomonson, V.V., Nicolo, E.D. and Klaus, J.B. (2002), MODIS snow-cover products, Remote Sensing of Environment, 83:181-194. http://modis-snow-ice.gsfc.nasa.gov/pap_snowcover02.pdf 92

Hall, D.K., Riggs, G.A. and Salomonson, V.V. (2006), updated weekly. MODIS/Terra snow cover 8-day L3 global 500m grid V005, [list the dates of the data used]. Boulder, Colorado USA: National Snow and Ice Data Center. Digital media. Hall, D.K., Riggs, G.A. and Salomonson, V.V. (1995), Development of methods for mapping global snow cover using moderate resolution imaging spectroradiometer data, Remote Sensing of Environment, 54, pp 127-140. Hall, D.K. and Martinec, J. (1985), Remote sensing of ice and snow, London: Chapman & Hall. Hewitt, K. (1985), Snow and ice hydrology in remote, high mountain regions: the Himalayan sources of the river Indus. Snow and Ice Hydrology Project, Working Pap. no. 1, Wilfrid Laurier University, Waterloo, Ontario, Canada. Hu, et al. (2006), Gridded snowmelt and rainfall-runoff CWMS hydrologic modeling of the Red River of the North Basin, Journal of hydrologic Engineering ASCE, March April 2006, Pages 91 100 Jarvis, A., Reuter, H.I., Nelson, A., Guevara, E. (2006), Hole-filled SRTM for the globe Version 3, available from the CGIAR-CSI SRTM 90m Database:

http://srtm.csi.cgiar.org. Jordan, R. E. (1991), A one-dimensional temperature model for a snow cover: Technical documentation for SNTHERM.89. Hanover, New Hampshire, US Army Corps of Engineers. Justice, et al (1998), The Moderate Resolution Imaging Spectroradiometer (MODIS): land remote sensing for global change research. IEEE Transactions on Geoscience and Remote Sensing, 36: 12281249. Key, J. R., Collins, J. B., Fowler, C., & Stone, R. S. (1997), High-latitude surface temperature estimates from thermal satellite data, Remote Sensing of Environment, 61: 302 309.

93

Klein, A. G. and Barnett, A.C. (2003), Validation of daily MODIS snow cover maps of the Upper Rio Grande River Basin for the 20002001 snow year, Remote Sensing of Environment, 86: 162176 Klein, A.G., Hall, D.K. and Riggs, G.A. (1998a), Global snowcover monitoring using MODIS, In 27th International Symposium on Remote Sensing of Environment, June 8-12, 1998, Troms, pp. 363-366. Klein, A.G., Hall, D.K. and Riggs, G.A. (1998b), Improving snow cover mapping in forests through the use of a canopy reflectance model, Hydrological Processes, 12:1723-1744. Kustas, W. P. and Rango, A. (1994), A simple energy budget algorithm for the snowmelt runoff model. Water Resources Research 30 (5): 1515-1527. Lal, M., et al ( ), Climate Change in Asia Leavesley, G.H. (1989), Problems of snowmelt runoff modeling for a variety of physiographic and climatic conditions, Hydrological SciencesJournal, 349 (6), December, 1994. http://www.cig.ensmp.fr/~iahs/hsj/340/hysj_34_06_0617.pdf Lee, S.A., Klein, G. and Over, T.M. (2005), A comparison of MODIS and NOHRSC snow-cover products for simulating stream flow using the Snowmelt Runoff Model Hydrological Processes, 19, 29512972 (2005), John Wiley & Sons, Ltd. Li, et al. (2007), Automatic mapping of snow cover depletion curves using optical remote sensing data under conditions of frequent cloud cover and temporary snow, Hydrological Processes (2007), John Wiley & Sons, Ltd Makhdoom, M. T. A. and Solomon, S. I. (1986) Attempting flow forecasts of Indus river, Pakistan using remotely sensed snow cover data, Nord. Hydrol., Vol.17, pp.171-184. Martinec, J., Rango, A. and Roberts, R. (1995), Snowmelt Runoff Model (SRM) Users Manual, Version 4.0, ftp://hydrolab.arsusda.gov/pub/srm/srm4.pdf

94

Martinet, J. and de Quervain, M.R., (1975), The effect of snow displacement by avalanches on snowmelt and runoff. In: Snow and Ice Syrup., Moscow, International Association of Scientific Hydrology, Gentbrugge, Publ. 104, pp. 364-377. Martinec, J. (1975). Snowmelt runoff model for stream flow forecasts, Nordic Hydrology, 6, 145154. Martinec, J. (1985). Snowmelt runoff models for operational forecasts, Nordic Hydrology, 16, 129136. Mashayekhi, T. and Mahjoub, M. (1991), Snowmelt runoff forecasting: case study of Karadj reservoir, Iran, in Snow, Hydrology and Forests in High Alpine Areas/Proceedings of the Vienna Symposium, August 1991). IAHS Pub. No. 205, 1991. Maurer, E.P., Rhoades, J.D., Dubayah, R.O. and Lettenmaier, D.P. (2003), Evaluation of the snow-covered area data product from MODIS, Hydrological Processes: 17, 59 71, John Wiley & Sons, Ltd. McGuire M., Wood, A.W., Hamlet, A.F. and Lettenmaier, D.P. (2006), Use of satellite data for stream flow and reservoir storage forecasts in the Snake River Basin, Journal of Water Resources Planning and Management, 132: 97110, ASCE. Morris, E. M., Anderson, P.S., Bader, H.P., Weilenmann, P. and Blight, C. (1993). Modeling mass and energy exchange over polar snow using the DAISY model. in: Snow and ice covers: interactions with the atmosphere and ecosystems. Proc. symposia, Yokohama IAHS Publication no. 223: 53-60 Morris, E.M. (1982) Sensitivity of the European hydrological system snow models, International Association of Hydrological Sciences Publication 138, 22231. Nagler, T., Helmut, R., Petra, M., and Florian, M. (2007), Assimilation of

meteorological and remote sensing data for snowmelt runoff forecasting, Remote Sensing of Environment, XXXX, Article in Press, Elsevier Inc.

95

Odegaard, H.A. and Ostrem, G. (1977), Application of Landsat imagery for snow mapping in Norway. Nor. Water Resour. Electr. Board, Publa., pp. 1 61. PARC and ICIMOD (2005), Inventory of glaciers and glacial lakes and identification of potential glacial lake outburst floods (GLOFs) affected by global warming in the mountains of Himalayan Region, International Centre for Mountain Development (ICIMOD) and Pakistan Agricultural Research Council (PARC). Parde, M., Gota, K.and Royer, A. (2006), Inversion of a passive microwave snow emission model for water equivalent estimation using airborne and satellite data, Remote Sensing of Environment 111, pp. 346356, Elsevier Inc. Pipes, A. and Quick, M.C. (1987) Modelling large scale effects of snow cover. International Association of Hydrological Sciences Publication 166, 15160 Platnick S, King MD, Ackerman SA, Menzel WP, Baum BA, Riedi JC, Frei RA. (2003), The MODIS cloud products: algorithms and examples from Terra. IEEE Transactions on Geoscience and Remote Sensing, 41(2): 459473. Quick, M. C. and Pipes, A. (1977), The UBC watershed model. Hydrological Sciences Bulletin 221: 153-161. Rango, A., Solomonson, V.V. and Foster, J.L. (1977), Seasonal stream flow estimations in the Himalayan region employing meteorological satellites, Water Resources Research 13:109 112. Rees, W.G. (2006), Remote Sensing of Snow and Ice, CRC, Taylor and Francis Richard, C. and Gratton, D. J. (2001), The importance of the air temperature variable for the snowmelt runoff modeling using the SRM, Hydrological Processes: 15, 3357 3370, John Wiley & Sons, Ltd. Riggs, G.A., Hall, D.K. and Salomonson, V.V. (2006), MODIS snow products user guide to collection 5. http://modis-snow-ice.gsfc.nasa.gov/sug_c5.pdf

96

Salomonson, V.V. and Appel, I. (2004), Estimating fractional snow cover from MODIS using the normalized difference snow index, Remote Sensing of Environment, 89 (2004) 351360, Elsevier Inc. Seidel, K. and Martinec, J. (1992), Operational snow cover mapping by satellites and real time runoff forecasts. In Proceedings ISSGH 92, Kathmandu, Nepal, pages 123-132. IAHS Publication. No. 218. Siddiqui, K.M., Ahmad, S., Khan, A.R., Bari, A., Sheikh, M.M., and Khan, A.H.(2003), Global change impact assessment for the Himalayan mountain regions of Pakistan, country case studies of Siran and Hunza valleys report-2002, APN Project # 2002-03 Simic, A., Fernandes, R., Brown, R., Romanov, P., & Park, W. (2004), Validation of VEGETATION, MODIS, and GOES plus SSM/I snowcover products over Canada based on surface snow depth observations, Hydrological Processes, 18(6), 1089 1104. Singer, F.S. and Popham, R.W. (1963) Non-meteorological observations from weather satellites, Astronautics and Aerospace Engineering, 1(3), pp 89-92. Singh, P., Jain, S. K. and Kumar, N. (1997), Estimation of snow and glacier-melt contribution to the Chenab River, Western Himalaya, Mountain Research and Development, Vol. 17, No. 1. pp. 49-56. http://links.jstor.org/sici?sici=02764741%28199702%2917%3A1%3C49%3AEOSAGC%3E2.0.CO%3B2-9 Singh, P. and Singh, V.P. (2001), Snow and Glacier Hydrology, Water Science and Technology Library, Volume 37, Kluwer Academic Publishers, The Netherlands Tangborn, W. V. (1984). Prediction of glacier derived runoff for hydro-electric development. Geografiska Annaler, Series A 66 (3): 257-265. Tarar, R.N. (1982), Water resources investigation in Pakistan with the help of Landsat imagery snow surveys 1975-1978, I Hydrological Aspects of Alpine and High Mountain Areas, Proceedings of the Exeter Symposium, July 1982, IAHS Pub. no. 138. 97

Tekelia, et al. (2005), Using MODIS snow cover maps in modeling snowmelt runoff process in the eastern part of Turkey, Remote Sensing of Environment, 97 (2005) 216 230, Elsevier Inc. US Army Corps of Engineers (1956), Snow Hydrology, summary report on the snow investigations, North Pacific Div. Corps of Engineers, Portland, Oregaon. Wang, X., Xie, H. and Liang, T. (2007), Evaluation of MODIS snow cover and cloud mask and its application in Northern Xinjiang, China, Remote Sensing of Environment, XX (2007) XXXX, Article in Press, Elsevier Inc. Wigmosta, M. S., Vail, L. W. and Lettenmaier, D. P. (1994) A Distributed HydrologyVegetation Model for Complex Terrain, Water Resources Research, Vol. 30, No. 6: 1665-1679. Winiger, M., Gumpert, M. and Yamout, H. (2005), KarakorumHindukushwestern Himalaya: assessing high-altitude water resources, Hydrological Processes, 19, 2329 2338 (2005), John Wiley & Sons, Ltd. World Bank (2005), Pakistans water economy: running dry, Pakistan water strategy draft of June 23, 2005. Yang, D., Robinson, D., Zhao, Y., Estilow, T. and Ye, B. (2003), Stream flow response to seasonal snow cover extent changes in large Siberian watersheds, J. Geophys. Res., 108(D18), 4578, doi:10.1029/2002JD003149. Zeng, X., Shaikh, M., Dai, Y. -J., Dickson, R. E., & Myneni, R. (2001), Coupling of the common land model to NCAR community climate model, Journal of Climate, 15, 1832 1854. Zhou, X., Xie, H., Hendrickx, J. M. H., (2005), Statistical evaluation of remotely sensed snow-cover products with constraints from stream flow and SNOTEL measurements, Remote Sensing of Environment, 94: 214231.

98