3 views

Uploaded by SudharsananPRS

- SR744_Flood Risk User Guide_Presentacion Alto Impacto
- bttr5t
- erotion.pdf
- terrestrial biomes study guide
- double ring.pdf
- Comparison of Run-Off Computed by Strange’s Table and ‘Dry Damp Wet’ Method
- GROUP 9 - GYREXIR M. TOMAS.docx
- Vision IAS CSP 2019 Test 3 Questions
- lec14
- Labortary simulation studies on
- 4.1 Raised Beds and Waru Waru Cultivation
- Mark Scheme for Holiday Student Block 3
- Chapter 1: Introduction to the Manual
- Dynamics of storm-driven suspended sediments in a headwater catchment described by multivariable modeling
- Hydrology
- The SuDS Manual C697
- Commercial Dispatch eEdition 2-5-19
- Precipitation
- Lakshayjit Report
- Odonata

You are on page 1of 26

Chapter 11 A simple exponential rainfall depth-intensity distribution for application in erosion modelling

Abstract Most runoff and erosion models require rainfall intensity data of considerable to high temporal resolution, which restricts their current application and predictive potential. The present paper proposes a simple exponential rainfall depth-intensity distribution involving only two parameters to characterise storms: rainfall depth (P) and depth-averaged rainfall intensity (). On the basis of this distribution, analytical expressions were derived for relevant (R)USLE and GUEST model parameters. The approach was investigated using tipping bucket rainfall intensity measurements during 30 storms of 33-81 mm. Parameters of both models could be recovered with high accuracy using only P and . A comparison of results obtained with the original (high temporal resolution) measurements and data resampled into five-minute intervals showed that resampling affected model outcomes in all cases. The proposed rainfall depth-intensity distribution combines physically meaningful, clearly defined variables with theoretical simplicity and high descriptive accuracy and provides an excellent scope for use in runoff and erosion modelling. Parts of this chapter are submitted to Water Resources Research, with A.G.C.A. Meesters, J. Schellekens and L.A. Bruijnzeel as co-authors

11.1. Introduction 11.1.1. The role of rainfall intensity in runoff and erosion modelling Water erosion is closely linked to the magnitude of rainfall intensity in that (i) rain is able to detach more particles through rain splash when falling at higher intensities (Chapter 7), and (ii) higher rainfall intensities lead to higher infiltration excess runoff rates that are capable of keeping more sediment in transport as well as actively entraining soil particles under the right conditions (Rose, 1993). Various attempts have been made to incorporate the characteristics of rainfall intensity distribution during storms into erosion models. In the original and Revised USLE approaches (Wischmeier and Smith, 1958, 1978; Renard et al., 1997) the influence of rainfall intensity is encapsulated in a rainfall erosivity or R factor, which expresses the empirical correlation found between soil loss and the product of storm kinetic energy and maximum 30-minute rainfall intensity (EI30, Wischmeier and Smith, 1958). On the basis of increased experimental

199

CHAPTER 11 - A SIMPLE RAINFALL DEPTH-INTENSITY DISTRIBUTION understanding of erosion processes, more physically based models have been developed in recent years. Examples include WEPP (Flanagan and Nearing, 1995; Laflen et al., 1997), EUROSEM (Morgan et al., 1992), LISEM (De Roo et al., 1996) and GUEST (Rose, 1993; Yu et al., 1997b). In all cases, however, model application requires a time series of rainfall intensity with moderate to high temporal resolution. This imposes an important restriction to their use in terms of simulation, because rainfall intensity data of sufficient temporal resolution are not always available, but even more so for predictive purposes because storm patterns cannot be predicted. This problem has been addressed in two ways, viz.: (i) by using synthesised storms, usually assuming an exponential rainfall intensity distribution over time (Nicks et al., 1995; Brown and Foster, 1987) or (ii) by empirical correlations between certain rainfall characteristics and more readily available data. Examples of the last approach include the empirical relationships between the rainfall erosivity factor within (R)USLE over longer periods and monthly or annual rainfall (Renard and Freimund, 1994) and, within the GUEST approach, relationships between hydrological model parameters and gross runoff coefficient and peak rainfall intensity for short time intervals (e.g. five minutes; Yu et al., 1997b; Yu and Rose, 1999a, b). Both approaches either explicitly or implicitly assume that the distribution of rainfall intensity during individual storms follows certain statistical rules. 11.1.2. Errors associated with tipping bucket technology A second issue with regard to the use of rainfall intensity data for runoff and erosion modelling relates to (re-)sampling. Rainfall intensity is commonly measured using tipping bucket rainfall recorders that register each time a nominal rainfall depth (normally 0.1-0.25 mm) has fallen. Although not strictly necessary, rainfall measurements are usually sampled (by the recording device) or resampled (by the user) into time intervals of fixed length, thereby averaging the intensity range of rainfall during the interval. The choice of interval length is dictated by data availability or based on a trade-off between the required accuracy and computational time in subsequent applications, but is usually between 1 and 15 minutes (cf. Yu et al., 1997b). However, resampling tipping bucket measurements into fixed time intervals leads to a number of errors, which can be classified as errors relating to (i) time averaging, and (ii) resolution. Time averaging refers to the fact that the variation in rainfall intensity within the interval is lost through resampling. One of the consequences is that rainfall intensity peaks will be under-estimated. Another effect of time averaging occurs if the storm starts and/or ends during a time interval. In that case, resampled data will suggest rain to have fallen during the entire interval, thus stretching the storm over a period longer than in reality. A similar over-estimation of storm duration occurs if the storm is interrupted by one or more dry spells. The effects of time averaging on erosion model calculations are potentially significant. Problems caused by time-averaging can be minimised by choosing a small time interval, but this will increase computational time as well as result in resolution problems, which are particularly noticeable if actual rainfall intensity is lower than the minimum nominal rainfall intensity (R* in mm h-1), given by the ratio of nominal rainfall depth increment (h in mm) over resampled time interval length (t in h). In that case, rain appears to fall at intensity R* during the interval of tipping only, while the other intervals appear dry (Yu et al., 1997a). The errors are greatest with short resampling intervals (t) and can result in an under-estimation of storm duration, whereas resampling errors also

200

CHAPTER 11 - A SIMPLE RAINFALL DEPTH-INTENSITY DISTRIBUTION occur at intensities higher than R* but without affecting storm duration. Resolution problems may also affect model output, although their effect is likely to be less pronounced than those associated with time averaging, given the marginal importance of low rainfall intensities reflected in erosion models. 11.1.3. Aim of this study The current study has three main aims. Firstly, it investigates whether a simple exponential distribution of rainfall intensity involving only two parameters accurately describes measurements of rainfall intensity made with a tipping bucket device. Next, it is examined whether the two distribution parameters can be used to derive analytical expressions for the hydrological parameters used in the (R)USLE and GUEST models. The predictive performance of these expressions is then tested: model parameter values predicted by the distribution parameters are compared with those calculated from measurements made with a tipping bucket rainfall recorder during 30 individual storms of 33 to 81 mm in West Java, Indonesia. Furthermore, the effect of resampling rainfall intensity data on erosion model outputs is evaluated by comparing values calculated from the original measurements and data resampled in five-minute intervals. Finally, the paper investigates to what extent an approach involving variable time steps reduces sampling problems. 11.2. Rainfall depth-intensity distribution theory and application 11.2.1. The rainfall depth-intensity distribution and storm duration When describing temporal variations in rainfall intensity often a stochastic approach is adopted, in which the frequency or recurrence interval of various combinations of rainfall depth and duration is plotted and described by a log-normal function (e.g. Chow, 1996). A somewhat similar approach, which assumes an exponential relationship between rainfall intensity and the associated frequency, was taken to describe the variation of rainfall intensity within storms by Nicks et al. (1995) and Brown and Foster (1987). A drawback of both methods is that they involve total storm duration, which cannot be calculated from tipping bucket data without introducing some measure of uncertainty. More importantly, it can be questioned whether this type of distribution does right to the observation that prolonged periods of low rainfall intensity, although contributing significantly to overall storm duration, are generally insignificant for surface runoff and erosion. An alternative approach is not to consider the time during which rain falls at a particular intensity, but the depth of rain falling at this intensity. A simple distribution that expresses this relationship is an exponential equation of the form: R P(> R ) = P exp R [11.1]

where P(>R) (in mm) denotes the depth of rainfall that falls at an intensity higher than R (in mm h-1), P (in mm) total storm depth, and (in mm h-1) the depth-averaged rainfall

201

CHAPTER 11 - A SIMPLE RAINFALL DEPTH-INTENSITY DISTRIBUTION intensity of the storm. The value of P is measured while is readily calculated from (original or resampled) tipping bucket measurements as shown further on. An advantage of this approach is that low-intensity rainfall has less influence on the calculated distribution parameter than in the earlier mentioned time-based approaches, because of the limited rainfall depth represented by this type of rainfall. By contrast, rain falling at high intensities will influence much more, which is an advantage from a runoff and erosion modelling perspective. A preliminary survey of rainfall data presented below suggests that Eq. [11.1] in many cases described the rainfall intensity distribution well; examples illustrating the best, worst and average fit obtained are shown further on in Figs. 11.1 a, b, and c, respectively. The probability distribution function (PDF), expressing the probability [D(R)] that rain falls at an intensity R, can be derived from the exponential function described by Eq. [11.1]: D ( R) = R 1 R d P exp = exp R R R dR [11.2]

It follows that the total time (dt in h) during which rainfall intensity is between R and R+dR, is given by:

dt (R ) = PD( R)dR R

[11.3]

The duration (TR in h) of rainfall falling at intensities between R1 and R2 can now be found by integration of Eq. [11.3] for rainfall intensities between 0 and infinity. Substitution of u=R/ and dR= du and subsequent solution yields: TR PD( R)dR P = = R R R1

R2 R2

P R R = E1 1 E1 2 R R R

[11.4]

where E1(x) is the exponential integral function (better known as the Theis well function in groundwater hydrology) for which numerical approximations are available (e.g. Fetter, 1994; see also Appendix A). However, total storm duration (TP in h) cannot be calculated with Eq. [11.4], because E1(x) has an infinitely high value for x=0. This reflects that a depth of rain falling at an infinitely small intensity will take an infinitely long time to precipitate. In some ways this problem is comparable to the sampling errors associated with tipping bucket measurements: as stated earlier the amount of rain that falls at an intensity less than minimum nominal intensity R* will be recorded as rain falling at this very intensity. The amount of rain P(<R*), falling at an intensity less than R*, can be found by integration of Eq. [11.5] between 0 and R*:

202

R*

*

[11.5]

Given that this rain will be recorded by the tipping bucket device as rain falling at constant intensity R* the corresponding rainfall duration (T P(<R*) in h) can be approximated by: TP (< R* ) =

* P 1 exp R * R R

[11.6a]

This simplifies matters, since the total storm duration can now be written as:

TP = TP (< R* ) + TP ( R* )

The time that rainfall intensity exceeds R* is given by (cf. Eq. [11.4]):

* P R* P R TP (R R* ) = E1 E1 = E1 R R R R R

[11.6b]

[11.7]

The (R)USLE approach to estimate soil loss involves calculating the sum of EI30 values for individual storms occurring over a long period of time, although use of a threshold storm depth has also been suggested (Renard et al., 1997). For reasons of consistency, this rainfall erosivity factor, or R factor, will be denoted here by EKR30, i.e. the product of total storm kinetic energy EK (in J m-2) and peak 30-minute rainfall intensity R30 (in mm h-1). The kinetic energy flux EK,dens(R) (in J m-2 h-1) associated with rain falling at an intensity R equals the product of rainfall intensity and rainfall kinetic energy load (eK in J m-2 mm-1), and can be estimated from (cf. Renard et al., 1997; Chapter 7):

E K ,dens (R ) = R e(R ) = R emax [1 a exp( bR )]

[11.8]

where emax (in J m-2 mm-1), a, and b (in h mm-1) are empirical parameters varying somewhat between locations (Chapter 7). The total amount of storm kinetic energy (EK in J m-2) is found by combining Eqs. [11.2] and [11.8]. The resulting equation describes the product of rain falling at a particular intensity and the corresponding kinetic energy load, followed by integration and rearranging:

203

[11.9a]

Pemax EK = R

a aR R [0 0] = Pemax 1 bR + 1 bR + 1

[11.9b]

The other rainfall parameter needed in the (R)USLE is the maximum average rainfall intensity for 30 consecutive minutes during a storm, R30 (in mm h-1). In principle, the maximum amount of rain falling in a time interval T can be calculated from the exponential depth-intensity distribution using Eq. [11.7] inversely. The maximum intensity RT (in mm h-1) occurring during T is related to the corresponding rainfall depth by (cf. Eq. [11.1]):

RT =

P(> R) P R = exp T T R

[11.10]

T R R = R invE1 P

[11.11b]

To calculate R using Eq. [11.11b] requires the inverse of the exponential integral function, invE1(u). Numerical approximations can be obtained by Newton-Raphson iteration; examples for two ranges of u values are given in Appendix 11.A. Combining Eqs. [11.10] and [11.11b] yields: RT

T R P = exp invE1 P T

[11.12]

Maximum average rainfall intensity during a period T=30 minutes can now be estimated using Eq. [11.12]. However, Eq. [11.12] does not take the chronological distribution of rainfall into account: a storm made up of separate cloudbursts covering more than 30 minutes will result in an R30 value that is lower than predicted by Eq.

204

CHAPTER 11 - A SIMPLE RAINFALL DEPTH-INTENSITY DISTRIBUTION [11.12] even if the overall distribution conforms to Eq. [11.2]. It may be questioned which value leads to better estimates (see below). 11.2.3. Hydrological GUEST model parameters In the GUEST (Griffith University Erosion System Template) model a distinction is made between the situation in which sediment concentration is controlled by rainfall detachment, and that in which runoff itself actively entrains soil and sediment particles (Rose, 1993). The model uses a spatially variable infiltration model (SVIM; Yu et al., 1997c, 1998; Yu, 1999), based on the assumption that on a particular area (e.g. a runoff plot or a field) an exponential distribution of maximum infiltration rates occurs. As a consequence, at increasing rainfall intensities an increasingly large area starts to contribute to runoff, as expressed by an exponential relationship between rainfall intensity (R) and the infiltration rate at that intensity [I(R) in mm h-1]: R I (R ) = I m 1 exp I m [11.13]

where Im (in mm h-1) denotes the spatially averaged maximum infiltration rate reached when the entire area generates runoff. Runoff rate is calculated using Eq. [11.13] as the difference between instantaneous rainfall intensity and infiltration rates. Yu (1999) included an initial amount of infiltration (F0 in mm) that needs to take place before runoff can be generated according to Eq. [11.13], and described the attenuation of runoff peaks over larger distances by a kinematic wave approximation. Soil loss is subsequently calculated as the product of runoff and sediment concentration and both are estimated using rainfall intensity data over small time intervals. When erosion is dominated by rainfall detachment, a linear relationship is assumed between rainfall intensity and sediment concentration. This assumption is not entirely consistent with the literature on the subject (e.g. Kinnell, 1990) and, more importantly, the theory has not yet been tested too rigorously. The theory for runoffdriven erosion is much better developed and validated, both through laboratory studies (e.g. Rose, 1993) and field experiments over a range of conditions (e.g. Rose et al., 1997a; Yu and Rose, 1999b). Runoff-driven erosion is thought to dominate in many cases, in particular on steeper slopes (Rose et al., 1997a). The GUEST approach is based on the recognition of a maximum sediment concentration sustainable for a particular combination of runoff rate, topography and soil characteristics, referred to as the transport-limited concentration (ct in kg m-3). The flow-weighted average transportlimited concentration for the event ( c t in kg m-3) is defined by:

ct = kQe

m

[11.14]

where k is a coefficient calculated from, inter alia, slope gradient and length, surface roughness and sediment properties (it is noted that the unit of k depends on m). The value of m is assumed to be 0.4 for turbulent flow based on Mannings equation (Rose, 1993). The effective runoff rate (Qe in mm h-1) represents the flow-weighted average runoff rate, calculated as:

205

Q dt Qe = Qtot

m +1

1 m

[11.15]

where Qm+1dt is the sum of instantaneous runoff rates raised to the power 1+m. The total amount of soil loss for the event (M in g m-2) is eventually calculated from c t and total runoff depth (Qtot in mm):

M = ct Qtot = kQe

)Q

tot

[11.16]

where is a type of soil erodibility coefficient (01) included to account for the fact that part of the runoffs erosive energy is used to erode cohesive soil and, therefore, cannot be used to maintain sediment concentration (Rose, 1993) (it is noted here that strictly speaking, (16) is only dimensionally correct for =1, which can be avoided by expressing c t as a mixing ratio and expressing Qtot in g m-2) Below, analytical expressions relating runoff depth (Qtot) and effective runoff rate (Qe) to the rainfall distribution parameters P and are derived. To calculate the amount of runoff contributed by rain falling at an intensity R, Eqs. [11.2] and [11.13] may be combined to describe the amount of runoff produced during the time dt(R) that rainfall intensity is between R and R+dR:

R R 1 exp exp dR I R m

[11.17]

The total runoff amount (Qtot) during the rainfall event is found by integration and substitution with u=R/Im:

P Im 1 Qtot = R 0 R

R I m I m R 1 exp exp dR = P 1 F I R R R m

[11.18a]

where f(x) is defined as (cf. Eq. [5.1.32], Abramowitz and Stegun, 1965):

F (x ) =

[11.18b]

Im R Qtot = P 1 ln1 + R I m

[11.19]

206

Apparently, the runoff coefficient (Qtot/P) depends only on the ratio between Im and . To calculate Qe from the rainfall intensity distribution, it can be defined as (cf. Eq. [11.15]):

m+1 m Q dt Qe = 0 Qtot

1

[11.20]

Whereas Qtot is already obtained from Eq. [11.19], the following integral still needs to be solved (note that Im is temporarily substituted by I* to avoid confusion):

Q

0

m +1

dt = P Q

0

m +1

P m+1 = I * F m, I * , R R

m +1

dR [11.21a]

where the function F(m,I*,) depends on the values of m and I*/ and, after introducing the integration variable u=R/I* is given by:

1 I* m +1 F m, I * , R = exp u {u [1 exp( u )]} du u R 0

[11.21b]

This function cannot be solved analytically, but is easily evaluated numerically with the aid of a computer. Eqs. [11.19] and [11.21b] may be combined to yield Qe (cf. Eq. [11.20]):

1 m R R Qe = I m F m, I m , R ln1 + I m Im

1

[11.22]

Apparently Qe depends on the individual values of m, Im and , but is independent of storm depth (P). Using the values of Qtot and Qe predicted by Eqs. [11.19] and [11.22], respectively, soil loss (M) can be calculated using Eq. [11.16] once k and are known. Below, the effect of using predicted values for Qtot and Qe on soil loss estimates will be investigated by introducing the normalised soil loss M0. If it is assumed that =1 (representing non-cohesive substrate; Rose, 1993), M0 can be defined as (cf. Eq. [11.16]):

M0 = M m m = Qe Qtot Qe Qtot k

[11.23]

207

11.3. Materials and methods

Rainfall intensity measurements were made about 40 km east of Bandung in West Java, Indonesia (7-03S, 108-04W) at an altitude of ca. 575 m a.s.l. The area experiences a humid tropical climate with a drier season (average monthly rainfall less than 60 mm) generally extending from July until September whereas mean annual rainfall is about 2650 mm. Rainfall intensity was measured using single tipping bucketlogger systems that recorded the time of tipping to the nearest second. Because individual instruments occasionally malfunctioned, data collected by three different tipping bucketlogger systems were used. The systems were calibrated and the data subsequently corrected in a non-linear manner. The resulting nominal rainfall depth increment (h) was 0.072-0.102 mm. The instruments were calibrated following Marsalek (1981); details may be found in Appendix 11.B. During a period of almost fifteen months (17 November 1998 to 7 February 2000) a total of 3827 mm of rainfall was recorded, distributed over 247 storms. If there were more than four hours between two consecutive tips, these were assumed to represent two different storm events. The first tip of each storm was assumed to represent rain falling at an intensity equal to that corresponding to the second tip. The rainfall represented by the first tip may actually have been less than h if it included water from the preceding storm remaining in the bucket, while the reverse reasoning applies to the last tip of each storm. The resulting difference in storm depth will always be less than h, but the effect on inferred storm duration can be more significant. The 30 largest storms were selected from the full data set, accounting for 1478 mm or 39% of total recorded rainfall and varying in depth between 33.3 and 81.0 mm. To investigate the effect of resampling on calculated and predicted storm and erosion model parameters, the original, variable time step rainfall intensity data were used as well as data resampled into five-minute intervals (starting on the hour), representative for the range of interval lengths commonly used in modelling. The data calibration and resampling procedure was carried out using custom-written software (LOGC; J. Schellekens, unpublished).

To calculate the depth-averaged rainfall intensity () the rainfall intensity data series was considered to be made up of n intervals, each being assigned a rainfall intensity Ri (in mm h-1) and a duration ti (in h). can then be calculated as:

R=

(R t )

n 2 i =1

[11.24]

Because the time intervals do not have to be of equal length, calculations could be made using both the original and resampled data. The coefficient of determination (r2) between the observed and predicted probability of occurrence of each tip falling at a particular intensity, was calculated for both data types. Using the original and resampled data sets and the corresponding P and values, the expressions for the various hydrological parameters given in Section 11.2 were tested.

208

Storm duration (TP) was calculated from the five-minute data using Eqs. [11.6ab] and [11.7]. Separate values for the duration of rainfall at intensities higher and lower than minimum nominal rainfall intensity (R*) were also derived. For comparative purposes, the duration of rainfall in excess of 1 mm h-1 was also calculated from the original data. This value was chosen to remain in the same order of magnitude as R* values for the resampled five-minute interval data (0.86-1.22 mm h-1 and 0.97 mm h-1 on average for the respective gauges). Time-averaged rainfall intensity ( R in mm h-1) was calculated as the ratio of rainfall depth over storm duration. Total kinetic energy (EK) and maximum 30-minute rainfall intensity (R30) for use in the (R)USLE were calculated using the theory outlined in Section 11.2.2. Values of the coefficients in Eq. [11.8] were based on a review of the available literature (Chapter 7: emax=28.3 J m-2 mm-1, a=0.52 and b=0.042 h mm-1, respectively). For the resampled data, R30 was calculated as the highest average rainfall intensity during six consecutive fiveminute intervals per storm (Renard et al., 1995). A slightly different approach was followed for the variable time-step data to reduce the effect of starting time: the data were resampled into one-minute intervals and these were used to determine the highest amount of rainfall during 30 consecutive minutes. The GUEST model parameters storm runoff depth (Qtot), effective runoff rate (Qe) and normalised soil loss (M0) were calculated from the original and resampled rainfall data and predicted from the corresponding values using the theory explained in Section 11.2.3. This was done for average maximum infiltration rates (Im) of 5, 50 and 500 mm h-1, respectively. No initial amount of infiltration (F0) or kinematic flow approximation were applied, and values of m=0.4 and =1 were used where applicable. Most calculations were made in a spreadsheet environment, with numerical approximations for the normal and inverse exponential integral functions given in Appendix 11.A. The only exception was the calculation of Qe from Im and , which invoked a MATLAB programme to evaluate the integral given in Eq. [11.21b]. Depending on the parameter, cumulative and/or average values were calculated for the 30 storms. The difference between cumulative (or average) values was computed, as well as the average error of estimate associated with predictions by the distribution theory, both in absolute (in the corresponding unit) and relative terms (as a percentage of the calculated value). Finally, model efficiency (ME) sensu Nash and Sutcliffe (1970) was determined to provide a measure of distribution theory performance.

11.4. Results

Basic characteristics of the 30 selected storms are listed in Table 11.1, along with average values per storm size class and an indication of the agreement between measured and predicted probabilities of rainfall intensity exceedence, for both the original and the resampled data. The 30 storms had an average depth of 47.1 mm (range 33.0-81.0 mm) and average values of 46.125.9 mm h-1 (last number indicates standard deviation) and 39.921.2 mm h-1 for the original and resampled data, respectively. The average coefficient of determination (r2) between observed and predicted probabilities of rainfall intensity exceedence for the 30 storms was 0.97 and 0.96 (st. dev. 0.03 in both cases) for the respective data types. Distributions for three storms illustrating the worst, best and closest-to-average agreement are shown in Figs. 11.1a, b and c, respectively. By and large, the exponential rainfall intensity distribution

209

Original data P mm 33-35 35-40 40-45 45-60 55-65 65-81 All storms N 5 6 6 4 5 4 30 Total P mm 165 220 250 197 288 294 1413 mm h-1 36.0 30.6 43.5 35.4 70.5 66.1 46.1 r

2

Resampled data mm h-1 31.7 26.7 37.4 31.5 59.3 58.0 39.9 r2 0.96 0.98 0.97 0.98 0.93 0.95 0.96

Table 11.1. Average values of depth-averaged rainfall intensity () calculated from tipping bucket measurements in their original form and resampled into 5-minute intervals, respectively, for 30 storms grouped into six storm depth (P) classes. The average correlation coefficient (r2) between observed and predicted frequencies of rainfall intensity exceedence for individual storms (cf. Fig. 11.1) are given as a measure of agreement.

described the observations rather well, particularly the original measurements. In all cases, the highest rainfall intensities were lost upon resampling and hence the depthaveraged values derived from resampled data were somewhat lower compared with the original data. This is also shown in Fig. 11.2a, which suggests a rather well-defined relationship between the respective series. Both data sets also indicate a weak, positive relationship between storm depth (P) and (Fig. 11.2b). The durations of periods with rainfall intensities exceeding the nominal minimum value R* (assumed to be 1 mm h-1 for the original data, cf. Section 3) and those with intensities less than R* (resampled data only) are listed in Table 11.2. The predicted total duration with rain falling at intensities >1 mm h-1 (original data) was very similar to that observed (111 versus 109 h). For individual storms the standard error was on average 0.8 h or 30% of the duration observed. By contrast, predicted and observed durations for the resampled data show poorer agreement: the duration of high intensity rainfall was generally over-estimated, while that of low intensities was under-estimated. Together, this resulted in a non-systematic difference between predicted and observed total storm duration (standard error of estimate of 1.5 h or 36%) and a reasonable agreement between cumulative duration of the 30 storms (166 versus 149 h; Table 11.2). Depending on storm duration, the predicted and observed time-averaged rainfall intensities ( R ) for rainfall at an intensity exceeding R* also differed. The standard error of estimate was 25% (4.3 mm h-1) of the observed intensity for the original data and 28% (3.4 mm h-1) for the resampled data (Table 11.2).

Calculated and predicted cumulative and average values of storm kinetic energy (EK), maximum 30-minute rainfall intensity (R30) and their product (EKR30) for the 30 sample storms are listed in Table 11.3, together with indicators of the agreement between these, both for the original and the resampled data. Values for individual storms are compared in Fig. 11.3.

210

60

(a)

50 P(>R) (mm) 40 30 20 10 0 0 100 200 300 R (mm h -1 ) 400 500

50

(b)

40 P(>R) (mm) 30 20 10 0 0 20 40

R (mm h )

-1

60

80

40

(c)

30

P(>R) (mm)

20

10

0 0 20 40

-1

60

80

100

R (mm h )

Fig. 11.1. Examples of the exponential rainfall depth-intensity distribution (curved line) fitted to the frequency distribution original tipping bucket data (stepwise line), plotted as the amount of rainfall P(>R) (in mm) falling at an intensity exceeding R (in mm h-1). The worst, average and best fits among the 30 tested storms are shown, as well as the frequency distribution for resampled data (dashed line) (note that scales vary). (a) Worst fit (r2=0.89), for a 56.3 mm storm on 28 March 1999 (=133 and 108 mm h-1, for the original and resampled data, respectively; (b) most average fit (r2=0.97), for a 42.2 mm storm on 15 December 1998 ( values of 29.2 and 29.1 mm h-1, respectively); (c) best fit (r2=1.00), for a 38.3 mm storm on 15 March 1999 ( values of 20.1 and 15.2 mm h-1, respectively).

211

140

-1

120 100 80 60 40 20 0 0

(a)

1:1

50

100

-1

150

140 (b) 120

original -orig. = 0.52 P 1.14 r2 = 0.30

100

-1

resampled

)

(mm h

80 60 40 20 0 0 20 40 60 P (mm) 80 100

-resamp. = 0.49 P 1.12 r = 0.30

2

Fig. 11.2. Relationships between (a) depth-averaged intensity () values derived from the original ( -orig.) and the resampled tipping bucket data (-resampl.); and (b) the respective storm rainfall depth (P) and values.

Parameters

-1

Predicted Total Avg. 110.7 123.8 42.4 166.2 3.7 4.1 1.4 5.5 13.8 9.4

Indicators of performance S.E. Total St. Rel. ME Est. diff. Err.Est. 0.8 1.0 0.9 1.5 4.3 3.4 2% 24% -15% 11% -9% -14% 30% 39% 72% 36% 25% 28% 0.62 0.41 0.23 0.25 0.76 0.74

Original Resamp.

TP(>1)

TP(>R*) h TP(R*) h TP h

R P(>1) mm h -1 R (>R*) mm h

Original Resamp.

Table 11.2. Average and cumulative or overall values of storm duration (TP in h) and timeaveraged rainfall intensity ( R in mm h-1) calculated from original and resampled tipping bucket measurements, respectively, and calculated using the event-based equations, with various indicators of the agreement between calculated and predicted values added (ME denotes model efficiency sensu Nash and Sutcliffe, 1971). 212

Parameters a

Calculated Total Avg. 33129 1766 32823 1726 1104 47.7 59 1094 47.0 58

Predicted Total Avg. 32447 1854 31831 1703 1082 51.5 62 1061 48.2 57

Indicators of performance S.E. Total St. Rel. ME Est. diff. Err.Est. 32 6.6 7.3 38 5.6 6.5 -2% 7% 5% -3% 3% -1% 3% 18% 17% 3% 14% 13% 0.99 0.91 0.97 0.99 0.93 0.98

Resampled data J m-2 EK (E) mm h-1 R30 (I30) EKR30 (EI30) 01000 b

symbols used in original (R)USLE terminology are given in brackets. has unit J mm m-2 h-1 (1000 J mm m-2 h-1/58.75 ft ton inch acre-1 h-1).

Table 11.3. Average and cumulative values of the RUSLE parameters storm kinetic energy (EK in J m-2) and 30-minute maximum rainfall intensity (R30 in mm h-1) and their product (EKR30, or EI30 in (R)USLE terminology) as calculated from original and resampled measurements, with various indicators of the agreement between calculated and predicted values.

Predicted storm kinetic energy was very close to calculated amounts (Fig. 11.3a); the error of estimate was only 3% for both the original and the resampled data, while the total predicted amounts were only 2% and 3% less than those calculated for the respective data types. Furthermore, cumulative EK calculated from the resampled data was 1% less than that calculated from the original data. The agreement between predicted and calculated values of 30-minute maximum intensity was somewhat less good. For the data resampled into one-minute intervals (representing the original data, see Section 11.3.2) a standard error of 18% (6.6 mm h-1) was found, while the predicted average R30 was 7% higher than that calculated (51.5 versus 47.7 mm h-1). Agreement was slightly better for the data resampled in five-minute intervals, with an error of estimate of 14% (5.6 mm h-1) and a predicted average value 3% higher than calculated (Table 11.3, Fig. 11.3b). R30 values representing six consecutive five-minute intervals were on average 1.51.7% lower than those representing 30 consecutive one-minute intervals. For comparison, the maximum amounts of rainfall for 30 not necessarily consecutive minutes of original (variable time step) data was also calculated. This resulted in an average R30 value of 54.4 mm h-1, i.e. 15.7% higher than that calculated for six consecutive five-minute intervals (47.0 mm h-1), while for individual storms the difference varied from 0 to 94%. Predicted and calculated values of the product EKR30 also showed good agreement (Table 11.3, Fig. 11.3c). For the original data the average difference was 17% (7.3 J m-1 h-1) while the cumulative difference over the 30 storms was 5%. Resampled data even produced a somewhat better agreement, resulting in an average difference of 13% (or 6.5 J m-1 h-1) and a cumulative difference of only 1% (Table 11.3).

213

2,500

(a)

Predicted E K (J m )

-2

2,000

1,500

1,000

original resam pled

-2

2,500

Calculated E K (J m

120

(b)

Predicted R 30 (mm h ) 100 80 60 40 20 0 0 20 40 60 80

-1 -1

100 )

120

Calculated R 30 (mm h

100,000

50,000

Calculated E K R 30

Fig. 11.3. Agreement between calculated values of RUSLE parameters and values predicted using the exponential rainfall depth-intensity distribution: (a) storm kinetic energy (EK), (b) maximum 30-minute rainfall intensity (R30), and (c) their product (EKR30) for both the original () and resampled data (+).

214

Parameters Im (mm h-1) Runoff depth (Qtot, mm) Original 5 50 500 5 50 500

Indicators of performance Total St. Rel. diff. Err.Est. -2% -6% -2% -4% -7% -2% -9% 9% 43% -10% 20% 69% -5% -2% 14% -7% 0% 22% 5% 5% 1% 6% 6% 1% 8% 20% 54% 8% 31% 74% 6% 4% 16% 7% 6% 22% ME

Pred. S.E. Est. 34.4 13.6 2.2 33.3 12.5 1.9 37.8 30.8 8.2 32.7 25.5 6.3 150 57 5.6 138 49 4.4 1.8 0.9 0.04 2.0 1.0 0.04 3.9 4.3 2.5 3.6 5.1 2.6 10 2.4 0.7 11 2.8 0.8

1053 433 66 1043 404 58 n.a. n.a. n.a. n.a. n.a. n.a. 4762 1765 148 4465 1463 109

1031 408 65 1000 374 57 n.a. n.a. n.a. n.a. n.a. n.a. 4503 1723 169 4138 1467 133

35.1 14.4 2.2 34.8 13.5 1.9 41.7 28.2 5.7 36.2 21.2 3.7 159 59 4.9 149 49 3.6

0.98 0.99 1.00 0.98 0.98 1.00 0.97 0.96 0.83 0.97 0.91 0.49 0.99 1.00 0.98 0.98 1.00 0.96

Resampled

Table 11.4. Cumulative and average values of the GUEST model parameters storm runoff depth (Qtot in mm), effective runoff rate (Qe in mm h-1) and normalised soil loss (M0 - see text for explanation), as calculated from original and resampled tipping bucket rainfall intensity data, and predicted using event-based equations. Various indicators of the agreement between calculated and predicted values are also listed.

Predicted and calculated values of the GUEST model parameters runoff depth (Qtot), effective runoff rate (Qe) and normalised soil loss (M0) are listed in Table 11.4, along with various indicators of the agreement between calculated and predicted values. Stormbased values for the three parameters are plotted in Fig. 11.4 and 11.5. Predicted and calculated runoff depths agreed remarkably well for both original and resampled data, with model efficiencies of 0.98-1.00 (Fig. 11.4a-c, Table 11.4). The average difference was only 1-6% (0.04-2.0 mm) and the accuracy actually increased with increasing average maximum infiltration rate (Im). The predicted total runoff amount for the 30 storms was 2-7% less than that calculated.

215

Predicted Q e (mm h )

80 (d) I m=5 mm h

-1 -1

60

40

20

0 0 20 40 60

-1

80

Calculated Q e (mm h )

-1

-1

80

20

40

10

0 0 40 80

-1

120

Calculated Q e (mm h )

Predicted Q e (mm h )

-1

50

(f) I m=500 mm h

-1

40 30 20 10 0

10

20

30

-1

40

50

Calculated Q e (mm h )

Fig. 11.4. Agreement between calculated values of (a-c) storm runoff depth (Qtot in mm) and (d-f) effective runoff rate (Qe in mm h-1) as defined in GUEST theory on one hand, and values predicted by the distribution parameters P and using both the original () and resampled data (+). Calculations were made for three contrasting values of average maximum infiltration rate Im (5, 50 mm h-1 and 500 mm h-1).

216

200

100

Predicted M 0

80

40

0 0 40 80 Calculated M 0 120

Predicted M 0

20

10

0 0 10 20 30 Calculated M 0

Fig. 11.5. Agreement between calculated values of normalized soil loss (M0) calculated as explained in the text on one hand, and values predicted by the distribution parameters P and using both the original () and resampled data (+). Calculations were made for three contrasting values of average maximum infiltration rate Im: (a) 5, (b) 50 mm h-1 and (c) 500 mm h-1.

217

A slight under-estimation of runoff depth seemed to occur, particularly for the intermediate Im value, but in most cases this effect was less than that introduced by data resampling (Table 11.4). By comparison, effective runoff rate (Qe) was predicted somewhat less well by the use of the exponential distribution theory. Agreement was still good for low and intermediate values of Im, with associated standard errors of estimate of 8% and 20-31% for Im values of 5 and 50 mm h-1, respectively. However, predictions for Im=500 mm h-1 resulted in an over-estimation of Qe by 43% and 69% on average for the original and resampled data, respectively. The over-estimation of Qe appeared to be systematic, in particular for the resampled data (Fig. 11.4) As observed earlier for runoff depth, this over-estimation compensated the effect of resampling somewhat: the average Qe value predicted from resampled values was only 10% higher than that based on the original data (Table 11.4). A similar compensating effect was found for Im=50 mm h-1, but not for the lowest Im value where a tendency of under-estimation was observed (Fig. 11.4a). Normalised soil loss (M0) was again predicted quite well by the distribution parameters, as illustrated in Fig. 11.5. The results indicate a slight under-estimation (by 5-7%) for Im=5 mm h-1 and an over-estimation (by 14-22%) for the highest maximum infiltration value. Predicted and calculated values agreed best for Im=50 mm h-1, with a difference of only -2 to 0% for total amounts (4-6% for individual storms; Table 11.4). Again, the effect of resampling was compensated somewhat by the use of the exponential distribution theory: the difference between total M0 as calculated for the resampled and original data was reduced from 26% to 10% in the case of Im=500 mm h-1, while it remained within 17% for the other two infiltration rates.

11.5. Discussion

Resampling of rainfall intensity data into five-minute intervals led to reduced values (Table 11.1). This illustrates the loss of information associated with the time-averaging that occurs upon resampling. The exponential depth-intensity distribution tested here appears to be least successful in predicting duration-related parameters such as total storm duration and time-averaged rainfall intensity (Table 11.2) and this was expected. Storm duration is significantly influenced by the duration of rain falling at low intensities but the depth-intensity distribution tends to give too little weight to this intensity range. Likewise, because of its direct dependence on storm duration the same holds for timeaveraged rainfall intensity ( R ). On a more general note, a fundamental difference emerges from this study between erosion parameters based on rainfall depth and those based on rainfall duration. Given the fact that high intensities are of greatest importance in soil erosion, the usefulness of time-based parameters such as storm duration is open to discussion. For example, the weather generator developed by Nicks et al. (1995) for use in WEPP modelling emphasises on storm duration. Other rainfall-related parameters in the WEPP model are also sensitive to low rainfall intensities, in particular if the soil effective soil hydraulic conductivity (Foster et al., 1995) is low. This reflects the discrete nature of the GreenAmpt infiltration approach used in WEPP. As the infiltration model is one of the most important components of WEPP (Brazier et al., 2000), the model could well benefit from a more realistic, gradual transition from no runoff to runoff conditions (cf. Yu, 1999).

218

Storm kinetic energy (EK) was predicted quite well using only the distribution parameters and P; on average, predicted values differed by no more than 3% from those calculated (Table 11.3). Such differences are modest compared with variations in the rainfall intensity-kinetic energy relationship that occur not only between locations (about 15%), but also between and during individual storms at a single location (Chapter 7). For example, simultaneous measurements of rainfall intensity and kinetic energy at a site in South-eastern Australia (Rosewell, 1986) were used in Chapter 7 to demonstrate that storm kinetic energy can only be predicted with an accuracy of about 20% at best because of inter- and intra-storm variations, although estimates were generally better for storms exceeding ca. 25 mm. In view of these uncertainties, the added error introduced by the use of exponential distribution theory is marginal. In fact, the exponential distribution theory may be used to compare predictions of rainfall kinetic energy (EK) according to different parameterisations of Eq. [11.8]. For example, Renard et al. (1997) proposed a parameterisation of Eq. [11.8] for use within the context of the (R)USLE, based on measurements of rainfall intensity and drop size in the Southern U.S.A. (Brown and Foster, 1987: emax =29 J mm-1, a=0.72 and b=0.05 h mm-1). For a given value of , predictions of EK with the latter parameterisation can be compared with the results obtained with the presently used version (Chapter 7). Testing with values derived for the 30 sample storms used in the present study (13-133 mm h-1) suggested that the parameterisation of Brown and Foster (1987) results in values that were between 13% lower and 1% higher, for the lowest and highest value, respectively. The predicted cumulative kinetic energy for the 30 storms was 3% lower. Such a difference is similar to the -8% found in a comparative exercise in Chapter 7 using 24 storms of overall somewhat smaller intensity. Predictions of 30-minute maximum average intensity (R30) using the exponential distribution theory had an average standard error of 18% when calculated from data that had been resampled into one-minute intervals, with an overall over-estimation of 7% (Table 11.3). The reason for this is primarily that the chronology of rainfall during events is not taken into account when using : the predicted 30-minutes of high rainfall intensity may have occurred as separate clusters instead of during a continuous period. Indeed, data inspection indicated that eleven out of the 30 resampled storm hyetographs showed peak rainfall intensities occurring during two or more bursts covering a period of more than 30 minutes on aggregate. Disregarding this chronology and instead ordering intervals by rainfall intensity led to R30 values that were 16% higher on average, and even as much as 94% for one particular storm. Similar differences of 10-30% were observed for rainfall records from 200 sites in Australia and the U.S.A. (B. Yu, pers. comm.). By comparison, the timing of the five-minute intervals was much less important: R30 values for six consecutive five-minute intervals were on average 1.5% lower than values for 30 one-minute intervals (Table 11.3). Because of the good estimates of storm EK and R30 the prediction of their product EKR30 was of comparable quality (Table 11.3). To reduce the error of estimate in EKR30, which is primarily associated with R30, the actual storm pattern would have to be taken into account. Even if possible, its simulation for predictive applications would require computational efforts that do not seem to be justified by the physical basis of the (R)USLE. In fact, the authors are not aware of any evidence that two consecutive cloud

219

bursts should result in more erosion than the same two cloud bursts separated by a few more minutes. Arguably, the dependency of R30 values on the adopted time resolution should be of more concern. Increasing the interval of resampling from one to five minutes led to R30 values that were about 2% lower and this error can be expected to increase for larger time intervals. (R)USLE rainfall factors, calculated as the average annual sums of storm EKR30 values for a certain location, are often based on rainfall intensity data resampled into time intervals larger than five minutes. For example, iso-erodent maps showing the spatial distribution of R-factors for the USA in Renard et al. (1997) are based on 15-minute data at best and, in most cases, on regressions between these data and 60-minute average intensities. For these empirical relationships Renard et al. (1997) suggested coefficients of determination that were in excess of 0.8. Clearly, the errors introduced by such approximations are much greater than those associated with the theory outlined in this study. An alternative approach to estimate R factors may therefore involve investigation of the frequency of occurrence for different combinations of P and , which could also serve a much wider range of hydrological applications.

Parameters used in the GUEST model (Rose et al., 1997a) were predicted well by the exponential distribution theory (Table 11.4). In particular, storm runoff depth (Qtot) could be predicted within rather narrow margins (1-6%). Although differences were somewhat larger for effective runoff rate (Qe), these too could be predicted within acceptable limits. Moreover, the impact of the estimation error for Qe in practical applications will be reduced for three reasons, viz.: (i) Qe is raised by the power m=0.4 and (ii) subsequently by (having a value between 0 and 1), and (iii) it is finally multiplied by runoff depth (Qtot) to obtain a soil loss estimate (cf. Eq. [11.16]). Estimation errors in Qe and Qtot balanced each other to some extent and in some cases (Table 11.4). As a result, predictions of normalised soil loss (M0) were much closer to calculated values suggested by Qe predictions, resulting in model efficiency values in excess of 0.96 and considerably better in most cases (Table 11.4). As such, it appears that the exponential rainfall depthintensity distribution proposed here may be combined profitably with SVIM/GUEST theory. The sound physical basis, relatively transparent formulation and low parameter demand for both GUEST theory and the proposed depth-intensity distribution hold considerable promise for future modelling. Unlike more empirical approaches to estimate Qe in the absence of high resolution rainfall intensity data (Yu et al., 1997b; Yu and Rose, 1999a), the theory outlined in the present study retains a clear physical basis. At the same time, the fact that the peak rainfall intensity and runoff coefficient used in these empirical approaches are clearly defined functions of the relevant variables (Im, and P, cf. Eqs. [11.12] and [11.19]) provides a physical explanation for these relationships. The initial infiltration (F0) and routing of excess rainfall over larger areas by a kinematic wave approximation as used in the GUEST model (Yu, 1999) could not be included in the present distribution model, because this requires knowledge of actual storm patterns. However, if initial infiltration is assumed not to take place before steady state infiltration, but in addition to this at the start of the storm, then it can be easily included in storm-based runoff modelling. Neglecting runoff routing may be justified when small and/or steep runoff plots with correspondingly small time lags are involved, but may present a problem on long hillslope segments. A simple approach to this problem may include a reduction factor based on relevant parameters, such as surface

220

400

I m=5 mm h

-1

300 Predicted M 0

2 -1

200

100

I m=500 mm h r =0.61

2

-1

Predicted E K R 30

Fig. 11.6. Relationships between the rainfall erosivity factor within the (R)USLE (EKR30) and normalised soil loss (M0) following the GUEST approach (see text for explanation), as calculated for three contrasting values of average maximum infiltration (Im) and the 30 tested storms, using the proposed exponential rainfall depth-intensity distribution theory. The lines and coefficients of determination relate to linear regression equations through the origin.

roughness, slope gradient and slope length. An application of the GUEST model expressions, including the suggested approach to avoid the need for actual storm patterns, to predict runoff and erosion from backsloping bench-terraces in a humid tropical upland environment of West Java, Indonesia is described in Chapters 12, 13 and 14.

The hydrological parameters calculated following the (R)USLE and GUEST approaches may be compared. The rainfall factor (EKR30) for the (R)USLE and normalised soil loss (M0) for GUEST were calculated using P and values derived from the original data for the 30 selected storms. In the case of GUEST, this was done with an assumed value of =1 and average maximum infiltration rates (Im) of 5, 50 and 500 mm h-1, respectively. The results are plotted in Fig. 11.6, indicating that an almost proportional relationship exists between soil loss predictions by the two models for individual values of Im. This finding is by no means trivial when considering the contrasting mathematical formulations of the two parameters. It suggests that the GUEST theory may in fact provide a theoretical basis for the R factor in the (R)USLE, while at the same time the voluminous (R)USLE database appears to support this part of the GUEST theory. The factor of proportionality between the two model components depends on Im, reflecting that in the (R)USLE approach the influence of soil infiltration capacity is incorporated in the soil K factor (Renard et al., 1997). Kinnell (1997) found that the performance of the (R)USLE approach could be enhanced by using a storm runoff coefficient instead of the R-factor, but also stated that this meant that the soil K-factor would have to be re-evaluated. These remarks are supported by the current study, which suggests that runoff coefficient can be described as a function of depth221

averaged rainfall intensity () and infiltration rate (Im) (Eq. [11.19]), whereas the rainfall factor can be described as a function of P and alone.

11.6. Concluding remarks

In spite of its simplicity, the tested rainfall depth-intensity distribution theory involving only two parameters (P and ) that are easily derived from original tipping bucket data, proved capable of predicting various runoff and erosion modelling parameters with encouraging accuracy. The data for individual storms (for example as depicted in Fig. 11.1a, b) suggest that in some cases this distribution can be expressed even more precisely by introducing additional parameters. This would be detrimental to model simplicity and transparency, however, while it may also be questioned whether runoff and soil erosion can be expected to be predictable with a correspondingly high accuracy. Soil erosion processes are notoriously variable in space and time and may behave in a non-predictable fashion to some extent. Arguably, therefore, the model presented here provides sufficient accuracy to give useful storm-based runoff and erosion estimates (cf. Table 11.4). Calculating the storm parameters from original (as opposed to resampled) rainfall intensity data avoids some errors associated with resampling and reduces the amount of computational effort. The simplicity of the proposed distribution model also provides good scope for its predictive use in runoff and erosion modelling. A preliminary investigation involving the 30 storms analysed here suggested a reasonably well-defined statistical relationship between storm depth (P) and depth-averaged rainfall intensity (): the ratios of calculated over average values predicted by the regression equation in Fig. 11.2b seemed to be log-normally distributed (not shown). Similarly, (log-normal) rainfall depth-duration curves (e.g. Chow, 1996) are widely used in flood prediction and available for many locations. These may also be used to find statistical relationships between P and using equations similar to Eq. [11.12], provided assumptions can be made about the time-scale associated with separate storm events. This again illustrates that the currently found combination of clear, physically-defined variables, theoretical simplicity and high accuracy holds considerable potential for runoff and erosion modelling.

Appendix 11.A. Numerical approximations of the normal and inverse exponential integral functions

The following infinite series is suggested in many text books on groundwater hydrology (e.g. Fetter, 1994) as a numerical approximation of the exponential integral E1(x):

x2 x3 x4 E1 (x ) = 0.5772 ln x + x + + ... 2 2! 3 3! 4 4!

[11.A.1]

When only the terms shown in Eq. [11.A.1] are used the error involved is generally well below 2% for 0<x<0.25. However, the use of Eq. [11.A.1] leads to significant errors if it is used for higher values of x, even if more terms are used. It was found that the following equation yields a better approximation for the range 0.25<x<2.0:

222

[11.A.2]

Tests showed that the error associated with Eq. [11.A.2] is usually much less than 1%. A similar phenomenon occurs for the inverse exponential integral function, invE1(u). The following numerical approximation was found for the range 1<u<4:

[11.A.3]

The error was found to be (typically much) less than 1%. The following numerical approximation was found for the range 0.05<u<1:

[11.A.4]

The error associated with Eq. [11.A.4] is less than 3% for values near both ends of the range and much less for intermediate values.

Appendix 11.B. Measurement and calibration of the tipping bucket rain gauge The tipping bucket-logger systems used in the present study was developed at the Vrije Universiteit Amsterdam and consisted of a 25.5 cm diameter sharp-rimmed funnel discharging into a single plastic spoon-like bucket. The bucket has a maximum nominal volume of 5.1 cm3 and, after tipping, is returned to the horizontal position by a magnet. The equivalent rainfall depth increment (h) corresponding to one tip was 0.072-0.102 mm and was adjustable. The tipping bucket operated a magnetic switch sending a pulse signal to the data logger which recorded the time of tipping to the nearest second. Theoretically, double recordings would occur if the tipping frequency would exceed 1 Hz (corresponding to an actual intensity of more than ca. 430 mm h-1). At intensities above 325 mm h-1, however, the funnel did not drain fast enough to accommodate all rainfall. This must have led to an under-estimation of very high intensities, but had the advantage that the force of the incoming water did not exceed the magnet retracting force. The logger therefore recorded all rainfall eventually and double values did not occur in the recorded data set. Furthermore, a non-linear relation exists between rainfall intensity and the pulse frequency of a tipping bucket recorder, because the bucket needs a certain amount of time to tip over (Bruce and Clark, 1966). In a twinned bucket system water comes in unrecorded during the short period lapsing between the start of bucket movement and the moment incoming water enters the other bucket. Marsalek (1981) tested three twinned tipping bucket systems and found that this missed time led to a relative underestimation that increased with rainfall intensity. His instruments had a bucket volumefunnel area ratio corresponding to 0.25 mm rainfall depth increment and the underestimation increased to some 3-5% at an intensity of 100 mm h-1. For the single tipping bucket instrument used in the current study some rainfall also went unrecorded if water entered the bucket before it reached its horizontal position. Laboratory calibration in which flow rate and tipping interval were measured using balance and stopwatch (cf. Marsalek, 1981) indicated a response similar to that of the systems tested by Marsalek, with a tipping time of 0.36 s versus 0.32-0.43 s, respectively. Because the present system

223

recorded smaller rainfall increments, however, the frequency of tipping and therefore also the effective under-estimation were larger at high rainfall intensities (ca. 10% at 100 mm h-1). The data were corrected using the theory of Marsalek (1981), assuming rainfall intensity during tipping equalled the average rainfall intensity between the time of the tip (tx+1) and the preceding tip (tx). The corrected rainfall intensity Ri, between tx and tx+1 is given by (cf. Marsalek, 1981):

Ri =

(t x +1 t x ) t *

[11.B.1]

where t* is the missed time (0.36 s). Eq. [11.B.1] was used to compute rainfall intensity for time intervals of variable length. The value of h varied somewhat between the different instruments and, in addition, changed slightly sometimes during the measurement period, probably because of some algal growth in the tipping bucket. Therefore, its value was adjusted using rainfall depths measured in manual rain gauges placed next to the tipping bucket systems.

224

- SR744_Flood Risk User Guide_Presentacion Alto ImpactoUploaded byKarla Zárate
- bttr5tUploaded byPeriko DelosPalotes
- erotion.pdfUploaded byRatih
- terrestrial biomes study guideUploaded byapi-238476134
- double ring.pdfUploaded byMing Chan
- Comparison of Run-Off Computed by Strange’s Table and ‘Dry Damp Wet’ MethodUploaded byIJSTE
- GROUP 9 - GYREXIR M. TOMAS.docxUploaded bygyrexir
- Vision IAS CSP 2019 Test 3 QuestionsUploaded byvikram
- lec14Uploaded byNeatha Chim
- Labortary simulation studies onUploaded byGurpreet Bedi
- 4.1 Raised Beds and Waru Waru CultivationUploaded bya_mohid17
- Mark Scheme for Holiday Student Block 3Uploaded bycgmalia
- Chapter 1: Introduction to the ManualUploaded byjalham
- Dynamics of storm-driven suspended sediments in a headwater catchment described by multivariable modelingUploaded byenrico66
- HydrologyUploaded byCostinel Cristescu
- The SuDS Manual C697Uploaded byGANGA RS
- Commercial Dispatch eEdition 2-5-19Uploaded byThe Dispatch
- PrecipitationUploaded byPori
- Lakshayjit ReportUploaded byAmit Chaudhary
- OdonataUploaded byantaeus71
- br no 767-1Uploaded byAnonymous ToMbL1jgTg
- GCSE Geography Revision Guide Paper 2Uploaded byKatie
- NIOEC-SP-00-11Uploaded bydonya
- Landslide Detection System using AVR microcontrollerUploaded byIRJET Journal
- Topic 2Uploaded bygajeel1991
- 1000Series_ReclamationRulesUploaded bypolobook3782
- Journal.pmed.1001308Uploaded by2010MTsang
- Minesite Water Management HandbookUploaded byedatgka
- Swimming Against The Tide in Search of Solutions By Carla J. ZambelliUploaded bythereadingshelf
- the restless atmosphereUploaded byapi-263822026

- Experimental Setup for Fibre Reinforced CompositeUploaded bySudharsananPRS
- GW Quality Assessment - IranUploaded bySudharsananPRS
- Assessment of Water Quality Using GISUploaded bySudharsananPRS
- Low-Flow Estimation and PredictionUploaded byDawit Abraham
- Water Quality AssessmentUploaded bySudharsananPRS
- Design of Concrete Gravity DamUploaded bySudharsananPRS
- Numerical Modelling and HydraulicsUploaded byrajesh005
- Water Resources YieldUploaded bySudharsananPRS
- Storm Drainage Design and Technical Criteria Manual 012006Uploaded bywesprit
- WEES Community Watersheds SPW 2009Uploaded bySudharsananPRS
- IS 10500 - 2012Uploaded byMehul Jain
- Groundwater Studies UNESCOUploaded bySudharsananPRS
- tK7P8EjrybIeUploaded byissairaq
- Open Channel FlowUploaded byRabar Muhamad
- GW Quality Using WQIUploaded bySudharsananPRS
- Civil Engineering_1 2012Uploaded byankitchhabra5727
- Ias - Civil_i - 2009Uploaded bySudharsananPRS
- Ias - Civil_engg_ii - 2011Uploaded bySudharsananPRS
- IAS - Civil Engineering Paper II - 2014Uploaded bySudharsananPRS
- CE-S5-2016-PUploaded byAjay Goel
- IAS - Civil Engineering Paper II - 2013Uploaded bySudharsananPRS
- CE-S7-2016-PUploaded byAjay Goel
- Ias - Civil_ii - 2009Uploaded bySudharsananPRS
- Ias - Civil-Engineering_2 - 2012Uploaded bySudharsananPRS
- IAS - Civil Engineering Paper I - 2013Uploaded bySudharsananPRS
- Ias - Civil_engg_i - 2011Uploaded bySudharsananPRS
- Enginnering Hydrology bookUploaded byRajat Garg
- IAS - Civil Engineering Paper I - 2014Uploaded bySudharsananPRS
- Fuzzy MaterialUploaded byahmed s. Nour
- Control of Water Pollution From Agriculture - Idp55e-PDFUploaded bySudharsananPRS

- EcosystemwUploaded byAllen Andicoy
- Facts for Swinburne DebateUploaded byMani Jack
- 420r10006 (RFS2 Regulatory Impact Analysis)Uploaded bycastilloehx
- Understanding Soil NutrientsUploaded bySaibal Chakraborty
- Fao EucaliptoUploaded bysilvia_a_
- Dennis Martinez (2011)Uploaded byJane Marina Apgar
- science in ohio blizzard bag 2Uploaded byapi-239855791
- Green OlympiadUploaded byPankaj Yadav
- November 21, 2013Uploaded byThe Delphos Herald
- Tropical Rain ForestsUploaded byapi-3821061
- Disaster Risk Reduction (DRR) for Food and Nutrition SecurityUploaded byJeffrey Marzilli
- Carrying Capacity of Mines in Bellary District EMPRI 2007 09Uploaded byAnonymous ufxnMzUhK
- Erosion and Reservoir Sedimentation Report 1Uploaded byIssa Eliasdrdr
- 9535Uploaded byChris Nash
- EV20001_L8_SoilPolnUploaded bySiddhant Karhadkar
- C Footprint of Pineapple Production and Transport WAFFUploaded byRomy Serilo
- Soil Erosion.pdfUploaded bycatarac19
- Crop RotationUploaded byRaj Kumar
- 00000102Uploaded byScary Creatures
- TenCate Mirafi SP Soil Protector_tcm32-31018Uploaded bypriodeep chowdhury
- 12 English Core Notes ReadingUploaded byZara Hasan
- 47105 PDFUploaded bylcoaguilap
- GULLY EROSION THREAT AND NIGERIA ROAD TRANSPORTATION:A CASE STUDY OF SOUTHEASTERN NIGERIA.Uploaded byopata cletus
- SoilUploaded byAnsari Fahad Anis
- ARE 1110 Final Exam GuideUploaded byASUPREMEA
- Identification of Critical Soil Erosion Prone Areas and AnnuaUploaded byDennyLumbanRaja
- Energy and Environment BooksUploaded byFatima Mir
- Characterization of Soil Erosion under different Agricultural Land Use Types in a Semi-Arid RegionUploaded byAZOJETE UNIMAID
- Population Dynamics in Developing CountriesUploaded byAzmat
- Issues and Scopes in Environmental BiotechnologyUploaded bybparker_3