Long short-term memory networks enhance rainfall-runoff modelling at the national scale of Denmark

This study explores the application of long short-term memory (LSTM) networks to simulate runoff at the national scale of Denmark using data from 301 catchments. This is the first LSTM application on Danish data. The results were benchmarked against the Danish national water resources model (DK-model), a physically based hydrological model. The median Kling-Gupta Efficiency (KGE), a common metric to assess performance of runoff predictions (optimum of 1), increased from 0.7 (DK-model) to 0.8 (LSTM) when trained against all catchments. Overall, the LSTM outperformed the DK-model in 80% of catchments. Despite the compelling KGE evaluation, the water balance closure was modelled less accurately by the LSTM. The applicability of LSTM networks for modelling ungauged catchments was assessed via a spatial split-sample experiment. A 20% spatial hold-out showed poorer performance of the LSTM with respect to the DK model. However, after pre-training, that is, weight initialisation obtained from training against simulated data from the DK-model, the performance of the LSTM was effectively improved. This formed a convincing argument supporting the knowledge-guided machine learning (ML) paradigm to integrate physically based models and ML to train robust models that generalise well.

Prediction of flood quantiles at ungauged catchments for the contiguous USA using Artificial Neural Networks

In this study, we utilise Artificial Neural Network (ANN) models to estimate the 100- and 1500-year return levels for around 900,000 ungauged catchments in the contiguous USA. The models were trained and validated using 4,079 gauges and several selected catchment descriptors out of a total of 25 available. The study area was split into 15 regions, which represent major watersheds. ANN models were developed for each region and evaluated by calculating several performance metrics such as root-mean-squared error (RMSE), coefficient of determination (R2) and absolute percent error. The availability of a large dataset of gauges made it possible to test different model architectures and assess the regional performance of the models. The results indicate that ANN models with only one hidden layer are sufficient to describe the relationship between flood quantiles and catchment descriptors. The regional performance depends on climate type as models perform worse in arid and humid continental climates. Overall, the study suggests that ANN models are particularly applicable for predicting ungauged flood quantiles across a large geographic area. The paper presents recommendations about future application of ANN in regional flood frequency analysis.

On the selection of precipitation products for the regionalisation of hydrological model parameters

Over the past decades, novel parameter regionalisation techniques have been developed to predict streamflow in data-scarce regions. In this paper, we examined how the choice of gridded daily precipitation (P) products affects the relative performance of three well-known parameter regionalisation techniques (spatial proximity, feature similarity, and parameter regression) over 100 near-natural catchments with diverse hydrological regimes across Chile. We set up and calibrated a conceptual semi-distributed HBV-like hydrological model (TUWmodel) for each catchment, using four P products (CR2MET, RF-MEP, ERA5, and MSWEPv2.8). We assessed the ability of these regionalisation techniques to transfer the parameters of a rainfall-runoff model, implementing a leave-one-out cross-validation procedure for each P product. Despite differences in the spatio-temporal distribution of P, all products provided good performance during calibration (median Kling–Gupta efficiencies (KGE′s) > 0.77), two independent verification periods (median KGE′s >0.70 and 0.61, for near-normal and dry conditions, respectively), and regionalisation (median KGE′s for the best method ranging from 0.56 to 0.63). We show how model calibration is able to compensate, to some extent, differences between P forcings by adjusting model parameters and thus the water balance components. Overall, feature similarity provided the best results, followed by spatial proximity, while parameter regression resulted in the worst performance, reinforcing the importance of transferring complete model parameter sets to ungauged catchments. Our results suggest that (i) merging P products and ground-based measurements does not necessarily translate into an improved hydrologic model performance; (ii) the spatial resolution of P products does not substantially affect the regionalisation performance; (iii) a P product that provides the best individual model performance during calibration and verification does not necessarily yield the best performance in terms of parameter regionalisation; and (iv) the model parameters and the performance of regionalisation methods are affected by the hydrological regime, with the best results for spatial proximity and feature similarity obtained for rain-dominated catchments with a minor snowmelt component.

Evaluation of Catchments’ Similarity by Penalization in the Context of Engineering Tasks—A Case Study of Four Slovakian Catchments

The present study deals with the similarity of catchments, which is a preliminary investigation before performing various water resource analyses and computations regarding other catchments, e.g., catchments’ similarity may be utilized in the context of analogous calculations of river flows in catchments without measured flows. In this paper, the penalization method of evaluating similarity is proposed; this method is appropriate for tasks in which fewer catchments are analyzed for engineering purposes. In addition to the various physical characteristics of the catchment, the “catchment’s calibrability” property is also formulated and evaluated. A methodology that used specific flows from catchments in a case study from Slovakia successfully verified the proposed penalization method. This verification confirmed that physical similarity, as evaluated using the proposed penalization methodology, also helps to identify hydrological similarity, i.e., finding the most similar catchment to a given catchment in terms of the rainfall-runoff process. Such a finding can be helpful, e.g., in the computation of the mentioned flows in ungauged catchments. Determining unmeasured flows can help to solve many engineering tasks, such as various technical calculations during the design of small reservoirs, defining the potential of a given stream for supplying irrigation, flood protection, etc.

Implementing a Proxy-Basin Strategy to Assess the Transposability of a Hydrological Model in Geographically Similar Catchments

Hydrological modelling is the most common way to investigate the spatial and temporal distribution of regional water resources. The reliability and uncertainty of a model depend on the efficient calibration of hydrological parameters. However, in complex regions where several subcatchments are defined, calibration of parameters is often difficult due to a lack of observed data. The transposability of hydrological models is of critical importance for assessing hydrological effects of land use and climatic changes in ungauged watersheds. Our study implemented a Proxy-Catchment Differential Split-Sample (PBDSS) strategy to assess the transposability of the conceptual hydrological model J2000 in three different subcatchments with similar physiographic conditions in Western Turkey. For dry and wet scenarios, the model was calibrated and validated for five years (2013–2017) in two selected catchments (Kayirli and Ulubey). Afterwards, it was validated by predicting the streamflow in the Amasya catchment, which has similar physical and climatic characteristics. The approach comprises transferring J2000 model parameters between different catchments, adjusting parameters to reflect the prevailing catchment characteristics, and validating without calibration. The objective functions showed a reliable model performance with Nash–Sutcliffe Efficiency (E) ranging from 0.72 to 0.82 when predicting streamflow in the study subcatchments for wet and dry conditions. An uncertainty analysis showed good agreement between the ensemble mean and measured runoff, indicating that the sensitive parameters can be used to estimate discharge in ungauged catchments. Therefore, the J2000 model can be considered adequate in its transposability to physically similar subcatchments for simulating daily streamflow.

Meteorological spatiotemporal drought impact assessment on water storage of Kolar reservoir, India

Climate change remains one of the main problems of the century which causes abrupt change in metrological phenomena. Hydrological drought assessment is very critical for ungauged catchments and important for the water storage and management of the reservoir. The research analyses the temporal variation of the drought events in Kolar River basin located in Madhya Pradesh state, India. Also, critical watershed is also identified which suffers most of the drought events. To accomplish this, four drought indices namely Standardized Precipitation Index, Reconnaissance Drought Index, Rainfall Deciles and Modified China Z-Index is used for the analysis. Rainfall from three rain gauge stations namely Birpur, Brijesh Nagar and Ichhawar from the sites were collected from the site ranging from 1988 to 2018 for thirty years. The contemporary research can be treasured source of scientific basis of policy formation and decision of stakeholders involved. Also, the analysis indicates the changing trends in rainfall patterns of the basin which affects the storage and preventive measures to be incorporated in the basin for the future availability of water for agricultural purposes and ever-growing domestic use water demand. It shows based on analysis that middle region of Kolar river basin is more vulnerable to drought in recent years.

Rainfall spatial variability in the application of catchment morphing for ungauged catchments

Catchment Morphing (CM) is a newly proposed approach to apply fully distributed models for ungauged catchments and has been experimented in several catchments in the UK. As one of the most important input datasets for hydrological models, rainfall spatial variability is influential to the stream variabilities and simulation performance. A homogenous rainfall was utilized in the previous experiments with Catchment Morphing. This study applied a spatially distributed rainfall from CEH-GEAR rainfall dataset in the morphed catchment for ungauged catchments as the follow-on study. Three catchments in the UK were used for rainfall spatial analysis and CEH-GEAR rainfall data were adopted for additional spatial analysis. The results demonstrate the influence of rainfall spatial information to the model performance with CM and illustrate the ability of morphed catchment to tackle with spatially varied information. More spatially distributed information is expected to be introduced for a wider application of CM.

Geomorphology-Wavelet based approach to rainfall runoff modeling for data scarce semi-arid regions, Kolar River catchment, India

The conceptual and physical mathematical model of rainfall-runoff modeling uses various parameters such as land use land cover, soil type classification, rainfall, atmospheric data such as temperature, evapotranspiration, solar radiation and wind speed, etc. But these data may not be available for developing countries and data scares semi-arid watershed. Also, the problem is even more critical for ungauged catchments and where manual record is maintained of water level and rainfall data. To address this issue, trend analysis is performed using Mann-Kendall test and Sen’s slope test which shows significant trend change stressing the need for new method for runoff prediction for better water resource management. In this study, a total of four models namely nonlinear autoregressive model with exogenous inputs lumped (LNARX), nonlinear autoregressive model with exogenous geomorphometrically processed inputs (GNARX), wavelet nonlinear autoregressive model with exogenous inputs (WLNARX) and nonlinear autoregressive model with exogenous geomorphometrically processed inputs (WGNARX). Ten models with different input combinations were selected based on their performance are analyzed for all the four networks. The best performing model for these networks is model no. 6 with WGNARX network with NSE 0.97 and RMSE 0.97 and with least value of RMSE. This method can be applied to data scarce region where data available are available for shorter duration and helpful for ungauged catchments also.