Data-driven models for streamflow regionalization in Krishna River Basin, India

Utilization of the Long Short-Term Memory network for predicting streamflow in ungauged basins in Korea

Abstract. Prediction of streamflow at ungauged catchments requires transfer of hydrologic information (e.g., model parameters, hydrologic indices, streamflow values) from gauged (donor) to ungauged (receiver) catchments. A common metric used for the selection of ideal donor catchments is the spatial proximity between donor and receiver catchments. However, it is not clear whether information transfer among nearby catchments is suitable across a wide range of climatic and geographic regions. We examine this issue using the data from 756 catchments within the continental United States. Each catchment is considered ungauged in turn and daily streamflow is simulated through distance-based interpolation of streamflows from neighboring catchments. Results show that distinct geographic regions exist in US where transfer of streamflow values from nearby catchments is useful for retrospective prediction of daily streamflow at ungauged catchments. Specifically, the high predictability catchments (Nash-Sutcliffe efficiency NS > 0.7) are confined to the Appalachian Mountains in eastern US, the Rocky Mountains, and the Cascade Mountains in the Pacific Northwest. Low predictability catchments (NS < 0.3) are located mostly in the drier regions west of Mississippi river, which demonstrates the limited utility of gauged catchments in those regions for predicting at ungauged basins. The results suggest that high streamflow similarity among nearby catchments (and therefore, good predictability at ungauged catchments) is more likely in humid runoff-dominated regions than in dry evapotranspiration-dominated regions. We further find that higher density and/or closer distance of gauged catchments near an ungauged catchment does not necessarily guarantee good predictability at an ungauged catchment.

The objective of this study was to investigate the changes in cropland areas as a result of water availability using Moderate Resolution Imaging Spectroradiometer (MODIS) 250 m time-series data and spectral matching techniques (SMTs). The study was conducted in the Krishna River basin in India, a very large river basin with an area of 265 752 km2 (26 575 200 ha), comparing a water-surplus year (2000–2001) and a water-deficit year (2002–2003). The MODIS 250 m time-series data and SMTs were found ideal for agricultural cropland change detection over large areas and provided fuzzy classification accuracies of 61–100% for various land‐use classes and 61–81% for the rain-fed and irrigated classes. The most mixing change occurred between rain-fed cropland areas and informally irrigated (e.g. groundwater and small reservoir) areas. Hence separation of these two classes was the most difficult. The MODIS 250 m-derived irrigated cropland areas for the districts were highly correlated with the Indian Bureau of Statistics data, with R 2-values between 0.82 and 0.86. The change in the net area irrigated was modest, with an irrigated area of 8 669 881 ha during the water-surplus year, as compared with 7 718 900 ha during the water-deficit year. However, this is quite misleading as most of the major changes occurred in cropping intensity, such as changing from higher intensity to lower intensity (e.g. from double crop to single crop). The changes in cropping intensity of the agricultural cropland areas that took place in the water-deficit year (2002–2003) when compared with the water-surplus year (2000–2001) in the Krishna basin were: (a) 1 078 564 ha changed from double crop to single crop, (b) 1 461 177 ha changed from continuous crop to single crop, (c) 704 172 ha changed from irrigated single crop to fallow and (d) 1 314 522 ha changed from minor irrigation (e.g. tanks, small reservoirs) to rain-fed. These are highly significant changes that will have strong impact on food security. Such changes may be expected all over the world in a changing climate.

Making useful Predictions in Ungauged Basins is an incredibly difficult task given the limitations of hydrologic models to represent physical processes appropriately across the heterogeneity within and among different catchments. Here, we introduce a new method for this challenge, Bayes empirical Bayes, that allows for the statistical pooling of information from multiple donor catchments and provides the ability to transfer parametric distributions rather than single parameter sets to the ungauged catchment. Further, the methodology provides an efficient framework with which to formally assess predictive uncertainty at the ungauged catchment. We investigated the utility of the methodology under both synthetic and real data conditions, and with respect to its sensitivity to the number and quality of the donor catchments used. This study highlighted the ability of the hierarchical Bayes empirical Bayes approach to produce expected outcomes in both the synthetic and real data applications. The method was found to be sensitive to the quality (hydrologic similarity) of the donor catchments used. Results were less sensitive to the number of donor catchments, but indicated that predictive uncertainty was best constrained with larger numbers of donor catchments (but still adequate with fewer donors).

This paper attempts to identify a possible common statistical distribution, to model the annual maximum floods observed at various rivers and streams of Botswana using goodness of fit indices based on K-S Statistics and L-Moment Ratios. Results from the two approaches, suggest that Log-Normal distribution adequately describes the annual maximum flood series obtained from various gauged sites across the country and can possibly be used as a common distribution for estimation of flood quantiles at specified risks which can be used in design of various water resourcesand hydraulic structures across streams and rivers of Botswana. Based on the homogeneity criteria, two standardised curves have been developed for two catchment systems which can be used for the above purpose while dealing with ungauged basins in these systems. Keywords: flood estimation, K-S Statistics, L-Moment Ratios, hydraulic structures, ungauged basins Botswana Journal of Technology Vol. 14(2) 2005: 1-8

Simulation of runoff in ungauged catchments has always been a challenging issue, receiving significant attention more importantly in practical applications. This study aims at calibration of an Artificial Neural Network (ANN) model which is capable to apply in an ungauged basin. The methodology is applied to two sub-catchments located in the Northern East of Iran. To examine the effect of physical characteristics of the catchment on the capability of the model generalization, it is attempted to synthesize effective parameters using empirical methods of runoff estimation. Firstly, the model was designed for a pilot sub-catchment and the statistical comparison between simulated runoff, and target depicted the capability of ANN to accurately estimate runoff over a catchment. Then, the calibrated model was generalized to another sub-catchment assumed as an ungauged basin while there are runoff data to compare the result. The result showed that the designed model is relatively capable to estimate monthly runoff for a homogenous ungauged catchment. The method presented in this study in addition to adding effective spatial parameters in simulation runoff and calibration of model by using empirical methods and the integration of any useful accessible data, examines the adaptability of model to an ungauged catchment.

Modelling ungauged catchments using the catchment runoff response similarity

Streamflow prediction in ungauged basins (PUB) is challenging, and Long Short‐Term Memory (LSTM) is widely used to for such predictions, owing to its excellent migration performance. Traditional LSTM forced by meteorological data and catchment attribute data barely highlight the optimum data integration strategy for LSTM and its migration from data‐rich basins to ungauged ones. In this study, we experimented with 1,897 global catchments and found that LSTM‐corrected Global Hydrological Models (GHMs) outperformed uncorrected GHMs, improving the median Nash‐Sutcliff efficiency (NSE) from 0.03 to 0.66. Notably, there was a large gap between traditional LSTM modeling in ungauged basins and autoregressive modeling in data‐rich basins, and GHM‐forced LSTM were an effective way to close this gap in ungauged basins. The spatial heterogeneity of the performance of GHM‐forced LSTM was mainly influenced by three metrics (dryness, the leaf area index and latitude), which described the hydrological similarity among catchments. Weaker hydrological similarity among continental catchments results in larger variability in GHM‐forced LSTM, with the best performance in Siberia (NSE, 0.54) and the worst in North America (NSE, 0.10). However, the migration performance of GHM‐forced LSTM was significantly improved (NSE, 0.63) in ungauged basins when hydrological similarity was considered. This study stressed the advantages of GHM‐forced LSTM and due significance should be attached to hydrological similarities among catchments to improve hydrological prediction in ungauged catchments.

Forty-five Coupled Model Intercomparison Project 5 (CMIP5)-based global climate models are assessed for two major river basins in India, namely, Krishna and Godavari on precipitation rate. About 44 and 59 grids of 1° × 1° size are studied for this purpose, respectively. Metrics employed for evaluation of GCMs are average absolute relative anomaly (AARA), normalised root mean square anomaly (NRMSA), absolute normalised mean bias anomaly (ANMBA) and correlation coefficient (CoC). Entropy method is applied to obtain weights of AARA, NRMSA, ANMBA and CoC. Ranks of 45 GCMs are obtained through a Euclidean distance-based technique, the compromise programming. Additive ranking is employed to aggregate the grid-wise ranking pattern.

Prediction in Ungauged Basins (PUB) is an important task for water resources planning and management and remains a fundamental challenge for the hydrological community. In recent years, geostatistical methods have proven valuable for estimating hydrological variables in ungauged catchments. However, four major problems restrict the development of geostatistical methods. We established a new information diffusion model based on genetic algorithm (GIDM) for spatial interpolating of runoff in the ungauged basins. Genetic algorithms (GA) are used to generate high-quality solutions to optimization and search problems. So, using GA, the parameter of optimal window width can be obtained. To test our new method, seven experiments for the annual runoff interpolation based on GIDM at 17 stations on the mainstream and tributaries of the Yellow River are carried out and compared with the inverse distance weighting (IDW) method, Cokriging (COK) method, and conventional IDMs using the same sparse observed data. The seven experiments all show that the GIDM method can solve four problems of the previous geostatistical methods to some extent and obtains best accuracy among four different models. The key problems of the PUB research are the lack of observation data and the difficulties in information extraction. So the GIDM is a new and useful tool to solve the Prediction in Ungauged Basins (PUB) problem and to improve the water management.

In a mountainous country like Nepal, most of the gauge stations are located at low altitudes, and getting reliable hydrological data at intake sites, most of which are located at high mountains, is difficult. The main objective of the study is to apply the ENKI model system in a regional calibration of the Saptakoshi Basin and to test the reliability of the model in this area and extract the runoff at ungauged sites. ENKI is a modular framework to implement hydrological models, which operates on distributed data within a spatial region. Models can be created by a set of user-defined subroutines to custom design the functionality, and the model can be calibrated in a region making it suitable for prediction in ungauged catchments. Climatic data from 1999 to 2008 are used for the model calibration in the Saptakoshi Basin. While processing data, some errors and inconsistencies in the flow data were found, and this reduced the potential calibration sites in the catchment. The results showed that the modeled discharge of independent and upstream catchments were well in agreement with observed data, while less agreement was found with downstream basins where observed data were inconsistent. Good quality observation data and the availabilityof enough data govern the quality of the simulation with the model, so the importance of data quality cannot bedisregarded. Finally, the obtained regional parameters were applied to extract the runoff data at the intake site of the Tamor hydropower plant. Further improvement of simulation results can be achieved with good quality of data and thus uncertainties in the parameters can be reduced. Even though there are some uncertainties in the results of model, the conclusion of this study is that the regional model, ENKI, could be a very helpful tool to predict the hydrological variables at the ungauged basins in a mountainous country like Nepal.DOI: http://dx.doi.org/10.3126/hn.v14i0.11269HYDRO Nepal JournalJournal of Water, Energy and EnvironmentVolume: 14, 2014 JanuaryPage: 61-72

<p>Rainfall-runoff models are widely used tools in catchment hydrology. Their evaluation is mostly based on comparing observed and simulated discharge values and various statistical objective functions have been proposed to evaluate the agreement between these time series. However, model evaluations that are based on statistical objective functions often does not provide the modeller with much insight on why the model fails to represent the hydrology of the real-world system.  Other, hydrologically meaningful indices or signatures have been proposed instead that quantify the hydrologic response characteristics of the catchment. They can also be regionalized and thus provide a potential opportunity for model evaluation in ungauged basins.</p><p>Our study investigates how to best integrate hydrological signatures in an objective function for model evaluation to shift the focus of objective functions to evaluate basic hydrological functions of catchments. We propose a signature-based hydrologic efficiency metric that can be derived from locally observed or regionalised hydrologic signatures.  The metric improves upon the Kling-Gupta Efficiency (KGE) metric by replacing its three components with hydrologic signatures characterising the water balance (or bias), the damping (or variance) and the timing of flows (or correlation). Additionally, we use these hydrologic signatures with the physical characteristics (i.e. catchment attributes) in some regionalization approaches such as linear, nonlinear regression and random forests for streamflow predictions in ungauged catchments. We test our ideas on a large and diverse sample of 582 UK catchments using the CAMELS-GB dataset and show that the performance of the proposed metric works well.</p>

Flooding is a recurrent hazard occurring worldwide, resulting in severe losses. The preparation of a flood susceptibility map is a non-structural approach to flood management before its occurrence. With recent advances in artificial intelligence, achieving a high-accuracy model for flood susceptibility mapping (FSM) is challenging. Therefore, in this study, various artificial intelligence approaches have been utilized to achieve optimal accuracy in flood susceptibility modeling to address this challenge. By incorporating the grey wolf optimizer (GWO) metaheuristic algorithm into various models—including recurrent neural networks (RNNs), support vector regression (SVR), and extreme gradient boosting (XGBoost)—the objective of this modeling is to generate flood susceptibility maps and evaluate the variation in model performance. The tropical Manimala River Basin in India, severely battered by flooding in the past, has been selected as the test site. This modeling utilized 15 conditioning factors such as aspect, enhanced built-up and bareness index (EBBI), slope, elevation, geomorphology, normalized difference water index (NDWI), plan curvature, profile curvature, soil adjusted vegetation index (SAVI), stream density, soil texture, stream power index (SPI), terrain ruggedness index (TRI), land use/land cover (LULC) and topographic wetness index (TWI). Thus, six susceptibility maps are produced by applying the RNN, SVR, XGBoost, RNN-GWO, SVR-GWO, and XGBoost-GWO models. All six models exhibited outstanding (AUC above 0.90) performance, and the performance ranks in the following order: RNN-GWO (AUC: 0.968) > XGBoost-GWO (AUC: 0.961) > SVR-GWO (AUC: 0.960) > RNN (AUC: 0.956) > XGBoost (AUC: 0.953) > SVR (AUC: 0.948). It was discovered that the hybrid GWO optimization algorithm improved the performance of three models. The RNN-GWO-based flood susceptibility map shows that 8.05% of the MRB is very susceptible to floods. The modeling found that the SPI, geomorphology, LULC, stream density, and TWI are the top five influential conditioning factors.

Study RegionThe study area consists of 44 catchments across Ireland. Study FocusWe regionalize two hydrological models (GR4J and GR6J) to produce continuous discharge simulations and compare performance in simulating high, median and low flow conditions with other established approaches to prediction in ungauged basins and a simple benchmark of using the median parameter set across all catchments. These include K-nearest neighbor (KNN) and statistical methods for predicting flow quantiles using catchment characteristics. Different objective functions were selected for different parts of flow regime and the success of different methods for regionalizing hydrological model parameters; including multiple linear regression (MLR), non-linear regression (NL) and random forests (RF) were evaluated. New Hydrological insights for the RegionAll regionalization approaches perform well for average flow conditions. The GR4J model regionalized using RF performs best for simulating high flows, though all regionalized models underestimate the median annual flood. GR6J regionalized using RF performs best for low flows. While KNN and statistical approaches that directly leverage physical catchment descriptors provide comparable median performances across catchments, the spread in relative error across our sample is reduced using regionalized hydrological models. Our results highlight that the choice of hydrological model, objective functions for optimization and approach to linking model parameters and physical catchment descriptors significantly influence the success of regionalization for low and high flows.

Abstract. In this paper the potential of discharge-based indirect calibration of the probability-distributed model (PDM), a lumped rainfall-runoff (RR) model, is examined for six selected catchments in Flanders. The concept of indirect calibration indicates that one has to estimate the calibration data because the catchment is ungauged or scarcely gauged. A first case in which indirect calibration is applied is that of spatial gauging divergence: because no observed discharge records are available at the outlet of the ungauged catchment, the calibration is carried out based on a rescaled discharge time series of a very similar donor catchment. Both a calibration in the time domain and the frequency domain (also known as spectral domain) are carried out. Furthermore, the case of temporal gauging divergence is considered: limited (e.g. historical or very recent) discharge records are available at the outlet of the scarcely gauged catchment. Additionally, no time overlap exists between the forcing and discharge records. Therefore, only an indirect spectral calibration can be performed in this case. To conclude also the combination case of spatio-temporal gauging divergence is considered. In this last case only limited discharge records are available at the outlet of a donor catchment. Again the forcing and discharge records are not concomitant, which only makes feasible an indirect spectral calibration. For most catchments the modelled discharge time series is found to be acceptable in the considered cases. In the case of spatial gauging divergence, indirect temporal calibration results in a better model performance than indirect spectral calibration. Furthermore, indirect spectral calibration in the case of temporal gauging divergence leads to a better model performance than indirect spectral calibration in the case of spatial gauging divergence. Finally, the combination of spatial and temporal gauging divergence does not lead to a notably worse model performance compared to the case of spatial gauging divergence.