Professional Documents
Culture Documents
where < R > and < Q > are long-term (generally greater than a year [see, e.g., Gerrits et al., 2009]) mean rainfall
and mean streamflow, respectively. 𝜙 is the dryness index of the basin, which is the ratio of the mean potential
evapotranspiration (solar energy) to the mean rainfall. Equation (1) suggests that partitioning of rainfall into
streamflow and evapotranspiration is determined by dryness index alone. In other words, it is a ‘universal’
zero-parameter model. Many forms of the function f (𝜙) can be found in the hydrologic literature; among them
( ( ) ( ))0.5
the most popular one is the Budyko function: f (𝜙) = 1 − 𝜙 ⋅ tanh 𝜙−1 ⋅ 1 − e−𝜙 [Budyko, 1948]. The
strength of equation (1) is that it requires only easily obtainable climate data for prediction of streamflow,
crucial for solving various water resources and ecological management problems, particularly in ungagued
regions [e.g., Berghuijs et al., 2014; Doulatyari et al., 2015; Greve et al., 2015; Singh and Kumar, 2015].
The limitation of equation (1), as its definition states, is that it is not suitable for a small (say, daily) timescale. At
a small timescale, hydrologic partitioning is far more complex due to involvement of factors that equation (1)
ignores. Hydrologists generally conceptualize different catchment processes by introducing multiple model
parameters whose values must be determined through calibration. Once a model’s parameter values are
determined, it essentially operates like a zero-parameter dynamic model for predicting future streamflow.
The calibrated parameter values are often mathematically related to measurable catchment characteristics
in order to predict flow in ungauged basins [e.g., Oudin et al., 2008]. The main issue with this approach,
however, is that a model’s parameters can interact with each other, making it difficult to find their physical
meanings [Oudin et al., 2008; Beven, 2011]. In other words, for no hydrological model there is a unique set of
parameter values.
©2016. American Geophysical Union. In this study, I propose that zero-parameter dynamic hydrologic modeling can be achieved by considering
All Rights Reserved. “dryness index” as a dynamic or time-varying property. In particular, this study introduces a “decay function”
that determines the dryness index of a basin at an instant of time from antecedent rainfall and energy inputs,
which thereby allows for dynamic hydrologic modeling. The proposed model is then tested considering daily
data from 63 U.S. Geological Survey (USGS) basins.
Figure 1. (a) A graphical illustration of how a basin delays in transforming rainfall (R) into streamflow (Q) and
evapotranspiration (ET). The blue blocks represent rainfall events. Purple and green lines, respectively, are hypothetical
Q and ET response curves. If we consider a short time span as shown in the figure (solid red rectangle), inflow (R) will not
be equal to outflow (Q + ET); i.e., we have to consider the effect of storage in the mass balance equation. However, over
a long time span, as shown in the figure (dashed red rectangle), R will be mostly transformed into Q and ET: R = Q + ET.
(b) Schematic diagram explaining continuous hydrologic partitioning as envisaged in this study. A basin is assumed to
be consisting of two conceptual zones, I and II. Rainfall (R, solid blue arrow) or solar energy (PET, solid green arrow)
entering into the basin at an instant of time needs to first pass through zone I which has ET demand equals to the PET.
The remaining rainwater ( ) and energy ( ) interact with each other in zone II to produce effective rainfall (ER) and
rainfall loss (RL) depending on the dryness state of the basin.
the total functional solar energy (F ) determining Θ at a time due to past rainfall and solar energy inputs can
be computed as
t
F(t) = (𝜏) ⋅ (1 + (𝜓 − 1)𝜖 ⋅ (t − 𝜏))1∕(1−𝜓) ⋅ d𝜏 (2a)
∫t−N
t
F(t) = (𝜏) ⋅ (1 + (𝜓 − 1)𝜖 ⋅ (t − 𝜏))1∕(1−𝜓) ⋅ d𝜏 (2b)
∫t−N
where 𝜏 is a random variable. Thus, Θ, which controls hydrologic partitioning at a time, is a function of F
and F . N here is the time period for which antecedent input of rainfall or solar energy affects Θ (its relevance
is explained in section 5).
Similar to the definition of dryness index, I define instantaneous dryness index (𝜑) as the dryness index at any
point of time as 𝜑(t) = F(t)∕F(t). Again, similar to equation (1), hydrologic partitioning at an instant of
time can be expressed as
Figure 2. Results from a sample basin (USGS ID: 7058000) for a period of 2 years starting from 1 October 1967).
(a) The daily rainfall and PET data; (b) the “functional rainfall,” “functional solar energy,” and instantaneous aridity index
(secondary axis) time series for the same time period. (c) Modeled discharge time series obtained using equation (4) at
daily timescale, and observed discharge for the same time period is displayed for visual comparison. For reference,
the effective rainfall time series obtained by using equation (3) is plotted in the same panel (secondary axis).
Equation (3) can be implemented using the Budyko function. The same decay function is used to obtain Q
time series from ER time series (ER decays to produce Q). Assuming that the principle of superposition is valid
here, discharge can be expressed as
[ t ]
d
Q(t) = − ER(𝜏) ⋅ (1 + (𝜓 − 1)𝜖 ⋅ (t − 𝜏))1∕(1−𝜓) ⋅ d𝜏 (4)
dt ∫0
The negative sign indicates that effective rainfall decays to produce discharge. The next section discusses
application of the proposed model to real basins.
Figure 3. (a) NSE and (b) R2 maps for the study basins. The computation was carried out at daily time step using the
proposed model. Although no clear patterns emerge, NSE and R2 maps seem to indicate regional influences on the
model’s performance. In particular, the model performs poorly in western catchments (generally dry) and well in
eastern catchments (generally wet).
ith day is computed as 𝜑i = Fi ∕Fi , after which ERi is obtained using the Budyko function as ERi = i ⋅
[ ( ) ]
1 − (𝜑i ⋅ tanh 𝜑−1
i[
⋅ (1 − e−𝜑i ))0.5 (see equation (3)). Discharge time series is then obtained in discrete form
∑i ]
as Qi = j=0 ERj ⋅ (1 + (𝜓 − 1)𝜖 ⋅ (i − j))1∕(1−𝜓) − (1 + (𝜓 − 1)𝜖 ⋅ (i + 1 − j))1∕(1−𝜓) .
To determine the values of 𝜓 , 𝜖 , and N, a trial and error method is followed and 15 basins are randomly selected
from the data set (see Table S1). For this, two criteria are considered: (i) the average modeled discharge from
all the basins is approximately equal to the average observed discharge from them (mass balance) and (ii)
the median of the recession flow power law exponents (𝛼m ) of the modeled −dQ∕dt = kQ𝛼 curves [Biswal and
Marani, 2010] being nearly equal to that of the observed curves. A recession curve is defined as a continuously
decreasing discharge time series of time span at least 5 days, and computation of 𝛼 is done by following the
least square regression method. The reasoning behind analysis of −dQ∕dt versus Q curves individually is that
although 𝛼 remains fairly constant, the coefficient k varies significantly across events [Biswal and Marani, 2010;
Shaw and Riha, 2012]. Recession curves are considered here in the analysis as they typically follow a decay
function with universal features [Biswal and Marani, 2014].
4.3. Results
Note that the choice of N mainly depends on 𝜓 . Theoretically, for 𝜓 ≥ 2 (𝜓 = 2 in our case), N cannot be ∞ as
the expressions for F and F (equation (2)) would diverge. Otherwise also N cannot be infinite due to data
limitations. Thus, in this study, N was fixed at 365, which can be considered to be sufficiently large. The first
criterion (but not the second criterion) was observed to be significantly sensitive to N, particularly when its
value was small (see Figure S1). Following the two criteria mentioned in subsection 4.2, the values of 𝜓 and 𝜖
were found to be 2 and 0.4 day−1 for the 15 basins. That means the decay function is
x(0)
x(t) = (5)
1 + 0.4 ⋅ t
where t is in days. This was then used for computation for every study basin. Figure 2 shows simulation results
for a sample basin for 2 years of time period. The average modeled Q for the basins is 1.12 mm/d, in comparison
to average observed Q of 1.06 mm/d. For validation, the analysis was repeated considering the rest of the study
Figure 4. (a) Annual average discharge predicted by the proposed model against observed annual average discharge. Modeled annual discharge was
obtained by averaging daily discharge values. (b) For comparison, annual discharge obtained by the Budyko model is plotted against observed annual
discharge. (c) Annual average discharge by the proposed model versus annual average discharge by the Budyko model. A good agreement between the
two models suggests that the proposed model preserves the essential features of the Budyko function.
basins, which showed 1.00 mm/d of average modeled Q against 1.04 mm/d of average observed Q. Similarly,
the second criterion was fairly satisfied for both the data sets (see Figure S2 in the supporting information).
Next, comparison of daily observed discharge time series and modeled discharge time series was done for
each of the 63 study basins by calculating the Nash-Sutcliffe efficiency (NSE) and coefficient determina-
tion (R2 ) (see Figure 3 and Table S1). The 25th, 50th, and 75th percentiles of NSE and R2 are, respectively,
(0.18, 0.41), (0.41, 0.53), and (0.50, 0.60). Note that to incorporate travel time delays in channel networks,
discharge time series were shifted (by 0–2 days) so as to achieve maximum NSE (Table S1). To compare the
proposed model with the Budyko model in depicting long-term water balance (equation (1)), modeled daily
discharge time series were averaged to produce annual average discharge time series. R2 between observed
and modeled values is found to be 0.87 for the proposed model considering discharge time series for all the
basins together, whereas for the Budyko model it is 0.91 (Figures 4a and 4b). Interestingly, the comparison
between the proposed model and the Budyko model showed R2 of 0.94 (Figure 4c). Further, the long-term
average 𝜑 (< 𝜑 >) showed a power law relationship with 𝜙 (Figure S3 in the supporting information).
(median NSE and R2 , respectively, are 0.41 and 0.53), particularly considering the fact that similar perfor-
mances are reported by some of the complex hydrological models. For example, Duan et al. [2006] considered
12 basins from the MOPEX database and used regionalized parameter values for 8 established hydrologic
models, which gave median NSEs ranging from 0.2 to 0.6 . The proposed model gives poor performance espe-
cially in arid basins (see the Table S1). This is understandable as hydrological models in general perform poorly
in arid regions.
The proposed model showed slightly poor performance in terms of predicting annual average discharge
compared to the Budyko model (Figures 4a and 4b), which might be due the fact that it is more likely to be
affected by observational errors as the computation was done at daily time step. However, the correlation
between annual average discharge predicted by the proposed model and that by the Budyko model is rela-
tively stronger (Figure 4c). This might be implying that the concept of instantaneous dryness index is robust
and that a better f (𝜙) (equation (1)) would mean a better f (𝜑) (equation (3)). In other words, this study quan-
titatively links short-term hydrologic partitioning with long-term hydrologic partitioning with the help of the
concept of dryness index. Thus, a fundamental question that arises here is what is the physical meaning of
the dryness index (< PET >∕ < R >) of a basin in the first place? Future studies need to answer this question,
as it is observed that dryness index is not equal to long-term average instantaneous dryness index. However,
it is clear that the dryness state of a basin is a dynamic phenomenon, not a static phenomenon as assumed
by equation (1).
The model’s performance varies significantly across the study basins (see Table S1 and Figure 3), which might
be suggesting that there are important local factors that are ignored by the model. That means there are
scopes for improvement. Nevertheless, the proposed model in its present form appears to reproduce daily
rainfall-runoff processes quite well. Finally, the proposed model should not be interpreted as a replacement for
traditional hydrological models with multiple calibration parameters for operational hydrologic forecasting
purposes; this was neither the purpose of this study. The main argument here is that the proposed model can
be used as an alternative platform for flow prediction in ungauged regions. The model’s performance may be
improved in future by incorporating meaningful physical characteristics, just like physical characteristics are
added to equation (1) to improve long-term hydrological partitioning [Shao et al., 2012; Gentine et al., 2012;
Wang and Tang, 2014].
Brutsaert, W., and J. L. Nieber (1977), Regionalized drought flow hydrographs from a mature glaciated plateau, Water Resour. Res., 13(3),
637–643, doi:10.1029/WR013i003p00637.
Budyko, M. (1948), Evaporation Under Natural Conditions, Gidrometeorizdat, Leningrad, English translation by IPST, Jerusalem.
Doulatyari, B., A. Betterle, S. Basso, B. Biswal, M. Schirmer, and G. Botter (2015), Predicting streamflow distributions and flow duration curves
from landscape and climate, Adv. Water Resour., 83, 285–298, doi:10.1016/j.advwatres.2015.06.013.
Duan, Q., et al. (2006), Model parameter estimation experiment (MOPEX): An overview of science strategy and major results from the
second and third workshops, J. Hydrol., 320(1), 3–17, doi:10.1016/j.jhydrol.2005.07.031.
Falcone, J. A., D. M. Carlisle, D. M. Wolock, and M. R. Meador (2010), GAGES: A stream gage database for evaluating natural and altered flow
conditions in the conterminous United States: Ecological archives, Ecology, 91(2), 621–621, doi:10.1890/09-0889.1.
Gentine, P., P. D’Odorico, B. R. Lintner, G. Sivandran, and G. Salvucci (2012), Interdependence of climate, soil, and vegetation as constrained
by the Budyko curve, Geophys. Res. Lett., 39(19), L19404, doi:10.1029/2012GL053492.
Gerrits, A., H. Savenije, E. Veling, and L. Pfister (2009), Analytical derivation of the Budyko curve based on rainfall characteristics and a simple
evaporation model, Water Resour. Res., 45(4), W04403, doi:10.1029/2008WR007308.
Greve, P., L. Gudmundsson, B. Orlowsky, and S. I. Seneviratne (2015), Introducing a probabilistic Budyko framework, Geophys. Res. Lett., 42(7),
2261–2269, doi:10.1002/2015GL063449.
Hargreaves, G. H., and Z. A. Samani (1985), Reference crop evapotranspiration from temperature, Appl. Eng. Agric., 1(2), 96–99,
doi:10.13031/2013.26773.
Jakeman, A., and G. Hornberger (1993), How much complexity is warranted in a rainfall-runoff model?, Water Resour. Res., 29(8), 2637–2649,
doi:10.1029/93WR00877.
Oudin, L., V. AndréAssian, C. Perrin, C. Michel, and N. Le Moine (2008), Spatial proximity, physical similarity, regression and ungaged
catchments: A comparison of regionalization approaches based on 913 French catchments, Water Res. Res., 44(3413), W03413,
doi:10.1029/2007WR006240.
Sankarasubramanian, A., and R. M. Vogel (2003), Hydroclimatology of the continental United States, Geophys. Res. Lett., 30(7), 1363,
doi:10.1029/2002GL015937.
Shao, Q., A. Traylen, and L. Zhang (2012), Nonparametric method for estimating the effects of climatic and catchment characteristics on
mean annual evapotranspiration, Water Resour. Res., 48(3), W03517, doi:10.1029/2010WR009610.
Shaw, S. B., and S. J. Riha (2012), Examining individual recession events instead of a data cloud: Using a modified interpretation
of dQ/dt-Q streamflow recession in glaciated watersheds to better inform models of low flow, J. Hydrol., 434, 46–54,
doi:10.1016/j.jhydrol.2012.02.034.
Singh, R., and R. Kumar (2015), Vulnerability of water availability in India due to climate change: A bottom-up probabilistic Budyko analysis,
Geophys. Res. Lett., 42(22), 9799–9807, doi:10.1002/2015GL066363.
Sivapalan, M., M. A. Yaeger, C. J. Harman, X. Xu, and P. A. Troch (2011), Functional model of water balance variability at the catchment scale:
1. Evidence of hydrologic similarity and space-time symmetry, Water Resour. Res., 47, W02522, doi:10.1029/2010WR009568.
United States Department of Agriculture (1972), National Engineering Handbook, Section 4: Hydrology, Washington, D. C.
Wang, D., and Y. Tang (2014), A one-parameter Budyko model for water balance captures emergent behavior in Darwinian hydrologic
models, Geophys. Res. Lett., 41(13), 4569–4577, doi:10.1002/2014GL060509.