Simulation of streamflow with statistically downscaled daily rainfall using a hybrid of wavelet and GAMLSS models

ABSTRACT

A rainfall–streamflow model is proposed, in which a downscaled rainfall series and its wavelet-based decomposed sub-series at optimum lags were used as covariates in GAMLSS (Generalized Additive Model in Location, Scale and Shape). GAMLSS is applied in climate change impact assessment using CMIP5 general climate model to simulate daily streamflow in three sub-catchments of the Onkaparinga catchment, South Australia. The Spearman correlation and Nash-Sutcliffe efficiency between the observed and median simulated streamflow values were high and comparable for both the calibration and validation periods for each sub-catchment. We show that the GAMLSS has the capability to capture non-stationarity in the rainfall–streamflow process. It was also observed that the use of wavelet-based decomposed rainfall sub-series with optimum lags as covariates in the GAMLSS model captures the underlying physics of the rainfall–streamflow process. The development and application of an empirical rainfall–streamflow model that can be used to assess the impact of catchment-scale climate change on streamflow is demonstrated.     

Simulated response of an intermittent stream to rainfall frequency patterns

The response of intermittent catchments to rainfall is complex and difficult to model. This study uses the spatially distributed CATchment HYdrology (CATHY) model to explore how the frequency of daily rainfall (λ) can affect the hydrologic regime of intermittent catchments. After a multi-objective calibration and validation of CATHY against experimental measurements of streamflow and groundwater levels in a catchment used as a pasture, the role of λ in affecting streamflow characteristics was explored using different scenarios. With different values of λ for the dry and wet periods of the year, CATHY showed that a series of frequent rainfall events was often associated with incipient streamflow, independent of the season. Activation of streamflow during the wet season was related to multiple factors and was not often associated with the shallow groundwater levels near the outlet of the catchment. The interplay between rainfall depth and intensity acted as the most important factor for the generation of streamflow. Using the difference between accumulated rainfall and evapotranspiration as a measure of wetness, saturated subsurface flow mechanism generated streamflow in simulations with wetness at least three times larger than mean wetness of other simulations. Although groundwater uprise near the outlet did not effectively contribute to streamflow in the initial days of flow, it strongly correlated with the magnitude of the runoff coefficient. Values of λ close or equal to the maximum value in the wet season can sustain the connectivity between groundwater and streamflow in the riparian zone. This connectivity increases the catchment wetness, which consequently results in an increase of the generated streamflow. Our study showed that rainfall regimes characterized by different λ were able to identify distinct flow regimes typical of either intermittent, ephemeral, or nonflowing catchments. Decrease of λ in the wet season is likely associated with a reduction of streamflow, with a shift of flow regime from intermittent to ephemeral or no-flow.     

Estimation of rainfall elasticity of streamflow in Australia

Abstract

Estimates of rainfall elasticity of streamflow in 219 catchments across Australia are presented. The rainfall elasticity of streamflow is defined here as the proportional change in mean annual streamflow divided by the proportional change in mean annual rainfall. The elasticity is therefore a simple estimate of the sensitivity of long-term streamflow to changes in long-term rainfall, and is particularly useful as an initial estimate of climate change impact in land and water resources projects. The rainfall elasticity of streamflow is estimated here using a hydrological modelling approach and a nonparametric estimator. The results indicate that the rainfall elasticity of streamflow (? _P ) in Australia is about 2.0–3.5 (observed in about 70% of the catchments), that is, a 1% change in mean annual rainfall results in a 2.0–3.5% change in mean annual streamflow. The rainfall elasticity of streamflow is strongly correlated to runoff coefficient and mean annual rainfall and streamflow, where streamflow is more sensitive to rainfall in drier catchments, and those with low runoff coefficients. There is a clear relation-ship between the ? _P values estimated using the hydrological modelling approach and those estimated using the nonparametric estimator for the 219 catchments, although the values estimated by the hydrological modelling approach are, on average, slightly higher. The modelling approach is useful where a detailed study is required and where there are sufficient data to reliably develop and calibrate a hydrological model. The nonparametric estimator is useful where consistent estimates of the sensitivity of long-term streamflow to climate are required, because it is simple to use and estimates the elasticity directly from the historical data. The nonparametric method, being model independent, can also be easily applied in comparative studies to data sets from many catchments across large regions.     

Improved estimators of correlation and R2 for skewed hydrologic data

ABSTRACT

The coefficient of determination R² and Pearson correlation coefficient ρ = R are standard metrics in hydrology for the evaluation of the goodness of fit between model simulations and observations, and as measures of the degree of dependence of one variable upon another. We show that the standard product moment estimator of ρ, termed r, while well-behaved for bivariate normal data, is upward biased and highly variable for bivariate non-normal data. We introduce three alternative estimators of ρ which are nearly unbiased and exhibit much less variability than r for non-normal data. We also document remarkable upward bias and tremendous increases in variability associated with r using both synthetic data and daily streamflow simulations from 905 calibrated rainfall–runoff models. We show that estimators of ρ = R accounting for skewness are needed for daily streamflow series because they exhibit high variability and skewness compared to, for example, monthly/annual series, where r should perform well.     

Conceptually-based shot noise modeling of streamflows at short time interval

A conceptual-stochastic approach to short time runoff data modelling is proposed, according to the aim of reproducing the hydrological aspects of the streamflow process and of preserving as much as possible the dynamics of the process itself. This latter task implies preservation of streamflow characteristics at higher scales of aggregation and, within a conceptual framework, involves compatibility with models proposed for the runoff process at those scales. At a daily time scale the watershed response to the effective rainfall is considered as deriving from the response of three linear reservoirs, respectively representing contributions to streamflows of large deep aquifers, with over-year response lag, of aquifers which run dry by the end of the dry season and of subsurface runoff. The surface runoff component is regarded as an uncorrelated point process. Considering the occurrences of effective rainfall events as generated by an independent Poisson process, the output of the linear system represents a conceptually-based multiple shot noise process. Model identification and parameter estimation are supported by information related to the aggregated runoff process, in agreement to the conceptual framework proposed, and this allows parameter parsimony, efficient estimation and effectiveness of the streamflow reproduction. Good performances emerged from the model application and testing made with reference to some daily runoff series from Italian basins.     

Modelling streamflow trends for a watershed with limited data: case of the Litani basin,Lebanon

Abstract

Streamflow variability in the Upper and Lower Litani basin, Lebanon was modelled as there is a lack of long-term measured runoff data. To simulate runoff and streamflow, daily rainfall was derived using a stochastic rainfall generation model and monthly rainfall data. Two distinct synthetic rainfall models were developed based on a two-part probabilistic distribution approach. The rainfall occurrence was described by a Markov chain process, while the rainfall distribution on wet days was represented by two different distributions (i.e. gamma and mixed exponential distributions). Both distributions yielded similar results. The rainfall data were then processed using water balance and routing models to generate daily and monthly streamflow. Compared with measured data, the model results were generally reasonable (mean errors ranging from 0.1 to 0.8?m³/s at select locations). Finally, the simulated monthly streamflow data were used to investigate discharge trends in the Litani basin during the 20th century using the Mann-Kendall and Sen slope nonparametric trend detection methods. A significant drying trend of the basin was detected, reaching a streamflow reduction of 0.8 and 0.7 m³/s per decade in January for the Upper and Lower basin, respectively.

Editor D. Koutsoyiannis; Associate editor Sheng Yue

Citation Ramadan, H.H., Beighley, R.E., and Ramamurthy, A.S., 2012. Modelling streamflow trends for a watershed with limited data: case of the Litani basin, Lebanon. Hydrological Sciences Journal, 57 (8), 1516–1529.     

Conceptually-based shot noise modeling of streamflows at short time interval

A conceptual-stochastic approach to short time runoff data modelling is proposed, according to the aim of reproducing the hydrological aspects of the streamflow process and of preserving as much as possible the dynamics of the process itself. This latter task implies preservation of streamflow characteristics at higher scales of aggregation and, within a conceptual framework, involves compatibility with models proposed for the runoff process at those scales. At a daily time scale the watershed response to the effective rainfall is considered as deriving from the response of three linear reservoirs, respectively representing contributions to streamflows of large deep aquifers, with over-year response lag, of aquifers which run dry by the end of the dry season and of subsurface runoff. The surface runoff component is regarded as an uncorrelated point process. Considering the occurrences of effective rainfall events as generated by an independent Poisson process, the output of the linear system represents a conceptually-based multiple shot noise process. Model identification and parameter estimation are supported by information related to the aggregated runoff process, in agreement to the conceptual framework proposed, and this allows parameter parsimony, efficient estimation and effectiveness of the streamflow reproduction. Good performances emerged from the model application and testing made with reference to some daily runoff series from Italian basins.     

Daily streamflow trends in Western versus Eastern Norway and their attribution to hydro-meteorological drivers

Regional warming and modifications in precipitation regimes has large impacts on streamflow in Norway, where both rainfall and snowmelt are important runoff generating processes. Hydrological impacts of recent changes in climate are usually investigated by trend analyses applied on annual, seasonal, or monthly time series. None of these detect sub-seasonal changes and their underlying causes. This study investigated sub-seasonal changes in streamflow, rainfall, and snowmelt in 61 and 51 catchments respectively in Western (Vestlandet) and Eastern (Østlandet) Norway by applying the Mann–Kendall test and Theil–Sen estimator on 10-day moving averaged daily time series over a 30-year period (1983–2012). The relative contribution of rainfall versus snowmelt to daily streamflow and the changes therein have also been estimated to identify the changing relevance of these driving processes over the same period. Detected changes in 10-day moving averaged daily streamflow were finally attributed to changes in the most important hydro-meteorological drivers using multiple-regression models with increasing complexity. Earlier spring flow timing in both regions occur due to earlier snowmelt. Østlandet shows increased summer streamflow in catchments up to 1100 m a.s.l. and slightly increased winter streamflow in about 50% of the catchments. Trend patterns in Vestlandet are less coherent. The importance of rainfall has increased in both regions. Attribution of trends reveals that changes in rainfall and snowmelt can explain some streamflow changes where they are dominant processes (e.g., spring snowmelt in Østlandet and autumn rainfall in Vestlandet). Overall, the detected streamflow changes can be best explained by adding temperature trends as an additional predictor, indicating the relevance of additional driving processes such as increased glacier melt and evapotranspiration.     

A stochastic model of low flows

A model is developed for annual low flow hydrographs. Its two primary components reflect the fact that hydrologic processes during streamflow rise (function of water input) and recession (function of basin storage) are different. Durations of periods of rise (wet intervals) and recession (dry intervals) are modelled by discrete probability distributions — negative binomial for dry intervals and negative binomial or modified logarithmic series for wet intervals depending on goodness of fit. During wet intervals, the total inflow is modelled by the lognormal distribution and daily amounts are allocated according to a pattern-averaged model. During dry intervals, the flow recedes according to a deterministic-stochastic recession model. The model was applied to three Canadian basins with drainage area ranging from 2210 to 22000 km² to generate 50 realizations of low flow hydrographs. The resulting two standard-error confidence band for the simulated probability distribution of annual minimum 7-day flows enclosed the probability distribution estimated from the observed record. A sensitivity analysis for the three basins revealed that in addition to the recession submodel, the most important submodel is that describing seasonality. The state of the basin at the beginning of the low flow period is of marginal importance and the daily distribution of input is unimportant.     

A stochastic model of low flows

A model is developed for annual low flow hydrographs. Its two primary components reflect the fact that hydrologic processes during streamflow rise (function of water input) and recession (function of basin storage) are different. Durations of periods of rise (wet intervals) and recession (dry intervals) are modelled by discrete probability distributions — negative binomial for dry intervals and negative binomial or modified logarithmic series for wet intervals depending on goodness of fit. During wet intervals, the total inflow is modelled by the lognormal distribution and daily amounts are allocated according to a pattern-averaged model. During dry intervals, the flow recedes according to a deterministic-stochastic recession model. The model was applied to three Canadian basins with drainage area ranging from 2210 to 22000 km² to generate 50 realizations of low flow hydrographs. The resulting two standard-error confidence band for the simulated probability distribution of annual minimum 7-day flows enclosed the probability distribution estimated from the observed record. A sensitivity analysis for the three basins revealed that in addition to the recession submodel, the most important submodel is that describing seasonality. The state of the basin at the beginning of the low flow period is of marginal importance and the daily distribution of input is unimportant.     

Factors influencing long range dependence in streamflow of European rivers

Investigating long range dependence of river flows, especially in connection with various climate and storage related factors, is important in order to improve stochastic models for long range dependence and in order to understand deterministic and stochastic variability in long‐term behaviour of streamflow. Long range dependence expressed by the Hurst coefficient H is estimated for 39 (deseasonalized) mean daily runoff time series in Europe of at least 59 years using five estimators (rescaled range, regression on periodogram, Whittle, aggregated variances, and least squares based on variance). All methods yield estimates of H > 0.5 for all data sets. The results from the different estimators are significantly positively correlated for all pairs of methods indicating consistency of the methods used. Correlations between H and various catchment attributes are also analysed. H is strongly positively correlated with catchment area. Apparently, increasing storage with catchment area translates into increasing long range dependence. H is also positively correlated with mean discharge and air temperature and negatively correlated with the mean specific discharge and the seasonality index (maximum Pardé coefficient). No significant correlation is found between the Hurst coefficient and the length of the analyzed time series. The correlations are interpreted in terms of snow processes and catchment wetness. Copyright © 2012 John Wiley & Sons, Ltd.     

Relationship between point rainfall, average sampled rainfall and ground truth at the event scale in the Sahel

Geostatistical techniques are used to quantify the reference mean areal rainfall (ground truth) from sparse raingaugenetworks. Based on the EPSAT-Niger event cumulative rainfall, a linear relationship between the ground truth considered as the mean area rainfall estimated from the densely available raingauge network and the area rainfall estimated from sparse network are derived. Also, a linear relationship between the ground truth and point rainfall is established. As it was reported experimentally by some authors, the slope of these relationships is less than one. Based on the geostatistical framework, the slope and the ordinate at the origin can be estimated as a function of the spatial structure of the rainfall process. It is shown that the slope is smaller than one. For the special case of one gauge inside a fixed area or a Field Of View (FOV), an areal reduction factor is derived. It has a limit value which depends only on the size of the area and the spatial structure of the rainfall process. The relative variance error of estimating the FOV cumulative rainfall from point rainfall is also given.     

Relationship between point rainfall, average sampled rainfall and ground truth at the event scale in the Sahel

Geostatistical techniques are used to quantify the reference mean areal rainfall (ground truth) from sparse raingaugenetworks. Based on the EPSAT-Niger event cumulative rainfall, a linear relationship between the ground truth considered as the mean area rainfall estimated from the densely available raingauge network and the area rainfall estimated from sparse network are derived. Also, a linear relationship between the ground truth and point rainfall is established. As it was reported experimentally by some authors, the slope of these relationships is less than one. Based on the geostatistical framework, the slope and the ordinate at the origin can be estimated as a function of the spatial structure of the rainfall process. It is shown that the slope is smaller than one. For the special case of one gauge inside a fixed area or a Field Of View (FOV), an areal reduction factor is derived. It has a limit value which depends only on the size of the area and the spatial structure of the rainfall process. The relative variance error of estimating the FOV cumulative rainfall from point rainfall is also given.     

Multifractal detrended fluctuation analysis of streamflow series of the Yangtze River basin,China

Scaling and multifractal properties of the hydrological processes of the Yangtze River basin were explored by using a multifractal detrended fluctuation analysis (MF‐DFA) technique. Long daily mean streamflow series from Cuntan, Yichang, Hankou and Datong stations were analyzed. Using shuffled streamflow series, the types of multifractality of streamflow series was also studied. The results indicate that the discharge series of the Yangtze River basin are non‐stationary. Different correlation properties were identified within streamflow series of the upper, the middle and the lower Yangtze River basin. The discharge series of the upper Yangtze River basin are characterized by short memory or anti‐persistence; while the streamflow series of the lower Yangtze River basin is characterized by long memory or persistence. h(q) vs q curves indicate multifractality of the hydrological processes of the Yangtze River basin. h(q) curves of shuffled streamflow series suggest that the multifractality of the streamflow series is mainly due to the correlation properties within the hydrological series. This study may be of practical and scientific importance in regional flood frequency analysis and water resource management in different parts of the Yangtze River basin. Copyright © 2008 John Wiley & Sons, Ltd.     

Streamflow estimation using optimal regional dependency function

The determination of spatial dependency of regionalized variable (ReV) is important in engineering studies. Regional dependency function that leads to calculation of weighting coefficients is required in order to make regional or point‐wise estimations. After obtaining this dependency function, it is possible to complete missing records in the time series and locate new measurement station. Also determination of regional dependency function is also useful to understand the regional variation of ReV. Point Cumulative Semi‐Variogram (PCSV) is another methodology to understand the regional dependency of ReV related to the magnitude and the location. However, this methodology is not useful to determine the weighting coefficient, which is required to make regional and point‐wise estimations. However, in Point Semi‐Variogram (PSV) proposed here, weighting coefficient depends on both magnitude and location. Although the regional dependency function has a fluctuating structure in PSV approach, this function gradually increases with distance in PCSV. The study area is selected in Mississippi river basin with 38 streamflow stations used for PCSV application before. It is aimed to compare two different geostatistical models for the same data set. PSV method has an ability to determine the value of variable along with optimum number of neighbour stations and influence radius. PSV and slope PSV approaches are compared with the PCSV. It was shown that slope slope point semi‐variogram (SPSV) approaches had relative error below 5%, and PSV and PCSV methods revealed relative errors below 10%. Copyright © 2009 John Wiley & Sons, Ltd.     

Bivariate stochastic modelling of ephemeral streamflow

Streamflow time series in arid and semi‐arid regions can be characterized as a sequence of single discrete flow episodes or clusters of hydrographs separated by periods of zero discharge. Here, two point process models are presented for the joint occurrence of flow events at neighbouring river sites. The first allows for excess clustering by adding autocorrelated errors to an empirically derived seasonally varying probability of an event and is extended to the case of the joint occurrence of flow events in two catchments. The second approach is to explicitly model the occurrences of clusters of events and the bivariate point process of event occurrences within them at both sites. For the two models, the magnitude of event peaks are assumed to be drawn from continuous distributions with seasonally varying parameters. Rises and recessions in discharge are interpolated between the peaks using regression estimates of hydrographs. The models are fitted to mean daily flows at two sites in Namibia and demonstrated to provide realistic simulations of the hydrology. Copyright © 2002 John Wiley & Sons, Ltd.     

Stochastic simulation of droughts. Part 1: point droughts

The number of drought events derived from the historic streamflow or rainfall series will be limited and produce results that are not very reliable. This study proposes a drought simulation methodology that uses a long sequence of synthetically generated monthly streamflow/rainfall series, from which it is possible to drive a large sample of drought events and the prediction of drought characteristics will be reliable. The modified Herbst method has been used to identify droughts in the generated streamflow and rainfall series. The drought simulation procedure is illustrated with a case study of the Bhadra reservoir catchment in Karnataka State, India. Monthly droughts were derived from both historic and generated monthly streamflow and rainfall series. The important drought characteristics were determined and the suitable probability distribution for each parameter was arrived at after studying seven different probability models. The use of the probability curves thus derived has been illustrated with examples (referred to in Part 1 as ‘point droughts’). Similarly, the development and application of stochastic models for the prediction of regional drought parameters have been illustrated with examples in the accompanying paper (Part 2: regional droughts). Copyright © 2007 John Wiley & Sons, Ltd.     

A cascade approach to continuous rainfall data generation at point locations

High-resolution temporal rainfall data sequences serve as inputs for a range of applications in planning, design and management of small (especially urban) water resources systems, including continuous flow simulation and evaluation of alternate policies for environmental impact assessment. However, such data are often not available, since their measurements are costly and time-consuming. One alternative to obtain high-resolution data is to try to derive them from available low-resolution information through a disaggregation procedure. This study evaluates a random cascade approach for generation of high-resolution rainfall data at a point location. The approach is based on the concept of scaling in rainfall, or, relating the properties associated with the rainfall process at one temporal scale to a finer-resolution scale. The procedure involves two steps: (1) identification of the presence of scaling behavior in the rainfall process; and (2) generation of synthetic data possessing same/similar scaling properties of the observed rainfall data. The scaling identification is made using a statistical moment scaling function, and the log–Poisson distribution is assumed to generate the synthetic rainfall data. The effectiveness of the approach is tested on the rainfall data observed at the Sydney Observatory Hill, Sydney, Australia. Rainfall data corresponding to four different successively doubled resolutions (daily, 12, 6, and 3 h) are studied, and disaggregation of data is attempted only between these successively doubled resolutions. The results indicate the presence of multi-scaling behavior in the rainfall data. The synthetic data generated using the log–Poisson distribution are found to exhibit scaling behaviors that match very well with that for the observed data. However, the results also indicate that fitting the scaling function alone does not necessarily mean reproducing the broader attributes that characterize the data. This observation clearly points out the extreme caution needed in the application of the existing methods for identification of scaling in rainfall, especially since such methods are also prevalent in studies of the emerging satellite observations and thus in the broader spectrum of hydrologic modeling.     

Comparison of daily and sub-daily SWAT models for daily streamflow simulation in the Upper Huai River Basin of China

Despite the significant role of precipitation in the hydrological cycle, few studies have been conducted to evaluate the impacts of the temporal resolution of rainfall inputs on the performance of SWAT (soil and water assessment tool) models in large-sized river basins. In this study, both daily and hourly rainfall observations at 28 rainfall stations were used as inputs to SWAT for daily streamflow simulation in the Upper Huai River Basin. Study results have demonstrated that the SWAT model with hourly rainfall inputs performed better than the model with daily rainfall inputs in daily streamflow simulation, primarily due to its better capability of simulating peak flows during the flood season. The sub-daily SWAT model estimated that 58 % of streamflow was contributed by baseflow compared to 34 % estimated by the daily model. Using the future daily and 3-h precipitation projections under the RCP (Representative Concentration Pathways) 4.5 scenario as inputs, the sub-daily SWAT model predicted a larger amount of monthly maximum daily flow during the wet years than the daily model. The differences between the daily and sub-daily SWAT model simulation results indicated that temporal rainfall resolution could have much impact on the simulation of hydrological process, streamflow, and consequently pollutant transport by SWAT models. There is an imperative need for more studies to examine the effects of temporal rainfall resolution on the simulation of hydrological and water pollutant transport processes by SWAT in river basins of different environmental conditions.     

Estimating daily inflows of large lakes using a water‐balance‐based runoff coefficient scaling approach

Daily river inflow time series are highly valuable for water resources and water environment management of large lakes. However, the availability of continuous inflow data for large lakes is still relatively limited, especially for large lakes situated within humid plain regions with tens or even hundreds of tributaries. In this study, we choose the fifth largest freshwater Lake Chaohu in China as our study area to introduce a new approach to reconstruct historical daily inflows at ungauged subcatchments of large lakes. This approach makes use of water level, lake surface rainfall, evaporation from the lake, and catchment rainfall observations. Rainfall–runoff relationship at a reference catchment was analysed to select rainfall input and estimate run‐off coefficient firstly, and the run‐off coefficient was then transferred to ungauged subcatchments to initially estimate daily inflows. Run‐off coefficient was scaled to adjust daily inflows at ungauged subcatchments according to water balance of the lake. This approach was evaluated using sparsely measured inflows at eight subcatchments of Lake Chaohu and compared with the commonly used drainage area ratio method. Results suggest that the inflow time series reconstructed from this approach consistent well to corresponding observations, with mean R² and Nash–Sutcliffe efficiency values of 0.69 and 0.6, respectively. This approach outperforms drainage area ratio method in terms of mean R² and Nash–Sutcliffe efficiency values. Accuracy of this approach holds well when the number of water‐level station being used decreased from four to one.