Understanding snow and ice melt dynamics is vital for flood risk assessment and effective water resource management in populated river basins sourced in inaccessible high-mountains. This study provides an AI-enabled hybrid approach integrating glacio-hydrological model outputs (GSM-SOCONT), with different machine learning and deep learning techniques framed as alternative ‘computational scenarios, leveraging both physical processes and data-driven insights for enhanced predictive capabilities. The standalone deep learning model (CNN-LSTM), relying solely on meteorological data, outperformed its counterpart machine learning and glacio-hydrological model equivalents. Hybrid models (CNN-LSTM1 to CNN-LSTM15) were trained using meteorological data augmented with glacio-hydrological model outputs representing ice and snow-melt contributions to streamflow. The hybrid model (CNN-LSTM14), using only glacier-derived features, performed best with high NSE (0.86), KGE (0.80), and R (0.93) values during calibration, and the highest NSE (0.83), KGE (0.88), R (0.91), and lowest RMSE (892) and MAE (544) during validation. Finally, a multi-scale analysis using different feature permutations was explored using wavelet transformation theory, integrating these into the final hybrid model (CNN-LSTM19), which significantly enhances predictive accuracy, particularly for high-flow events, as evidenced by improved NSE (from 0.83 to 0.97) and reduced RMSE (from 892 to 442) during validation. The comparative analysis illustrates how AI-enhanced hydrological models improve the accuracy of runoff forecasting and provide more reliable and actionable insights for managing water resources and mitigating flood risks - despite the paucity of direct measurements.
1 November 2022 in Journal of Water and Climate Change
Hydrological models play a key role in simulating and assessing climate and land use/cover (LULC) change impacts on hydrology in a watershed. In this study, the impact of climate and LULC change was investigated using the Soil and Water Assessment Tool (SWAT) model. The simulated and observed streamflow showed a good agreement. Both Nash–Sutcliffe Efficiency (NSE) and coefficient of determination (R2) were found to be greater than 0.7 during the calibration (1985–2002) and validation (2003–2012) period. The water balance components were simulated with inputs from downscaled Global Climate Models (GCMs) data (i.e., future scenario (2030–2100) relative to a baseline period (1974–2004)) under RCP4.5 and RCP8.5, and hypothetical generated LULC change scenarios. All GCMs projected an increase in temperature over the Kabul River Basin (KRB), whereas there was a lack of agreement on projected precipitation among GCMs under both emission and future scenarios. Water yield (WYLD) and evapotranspiration (ET) were projected to decrease in the 21st century. Average annual WYLD was projected to increase under the agriculture-dominant scenario, whereas it decreased under forest and grassland-dominant scenarios. These results are valuable for relevant agencies and stakeholders to adopt measures to counter the negative impacts of climate and LULC change on water resources.
20 January 2022 in Journal of Water and Climate Change
Several satellite-based and reanalysis products with a high spatial and temporal resolution have become available in recent decades, making it worthwhile to study the performance of multiple precipitation forcing data on hydrological modeling. This study aims to examine the veracity of five precipitation products employing a semi-distributed hydrological model, i.e., the Soil and Water Assessment Tool (SWAT) to simulate streamflow over the Chenab River Basin (CRB). The performance indices such as coefficient of determination (R2), Nash–Sutcliffe efficiency (NSE) and percentage bias (PBIAS) were used to compare observed and simulated streamflow at daily and monthly scales during calibration (2015–2018) and validation (2019–2020). The hydrologic performance of European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis (ERA) 5-Land (ERA5) was very good at daily (calibration R2=0.83, NSE=0.81, PBIAS=−6%; validation R2=0.75, NSE=0.74, PBIAS=−9.6%) and monthly ( calibration R2=0.94, NSE=0.94, PBIAS=−3.3%; validation R2=0.91, NSE=0.89, PBIAS=−3.2%) scales. This study suggests that the ERA5 precipitation product was the most reliable of the five precipitation products, while the CHIRPS performance was the worst. These findings contribute to highlighting the performance of five precipitation products and reference in the selection of precipitation data as input data to the SWAT model in similar regions.
29 November 2021 in Journal of Water and Climate Change
Climate change has implications for water resources by increasing temperature, shifting precipitation patterns and altering the timing of snowfall and glacier melt, leading to shifts in the seasonality of river flows. Here, the Soil & Water Assessment Tool was run using downscaled precipitation and temperature projections from five global climate models (GCMs) and their multi-model mean to estimate the potential impact of climate change on water balance components in sub-basins of the Upper Indus Basin (UIB) under two emission (RCP4.5 and RCP8.5) and future (2020–2050 and 2070–2100) scenarios. Warming of above 6 °C relative to baseline (1974–2004) is projected for the UIB by the end of the century (2070–2100), but the spread of annual precipitation projections among GCMs is large (+16 to −28%), and even larger for seasonal precipitation (+91 to −48%). Compared to the baseline, an increase in summer precipitation (RCP8.5: +36.7%) and a decrease in winter precipitation were projected (RCP8.5: −16.9%), with an increase in average annual water yield from the nival–glacial regime and river flow peaking 1 month earlier. We conclude that predicted warming during winter and spring could substantially affect the seasonal river flows, with important implications for water supplies.
14 October 2022 in Remote Sensing Applications: Society and Environment
Seasonal snowpacks, characterized by their snow water equivalent (SWE), can play a major role in the hydrological cycle of montane environments with months of snow accretion followed by episodes of melt controlling flood risk and water resource availability downstream. Quantifying the temporal and spatial patterns of snowpack accumulation and its subsequent melt and runoff is an internationally significant challenge, particularly within mountainous regions featuring complex terrain with limited or absent observational data. Here we report a new approach to snowpack characterization using open-source global satellite and modelled data products (precipitation and SWE) greatly enhancing the utility of the widely used Soil and Water Assessment Tool (SWAT). The paper focusses on the c. 23,000 km2 Chenab river basin (CRB) in the headwaters of the Indus Basin, globally important because of its large and growing population and increasing water insecurity due to climate change. We used five area-weighted averaged satellite, gridded and reanalysis precipitation datasets: ERA5-Land, CMORPH, TRMM, APHRODITE and CPC UPP. As well as comparison to local weather station data, these were used in SWAT to model streamflow for evaluation against observed streamflow at the basin outlet. ERA5-Land data provided the best streamflow match-ups and was used to infer snowpack (SWE) dynamics at basin and sub-basin scales. Snow reference data were derived from remote sensing and modelled SWE re-analysis products: ULCA-SWE and KRA-SWE, respectively. Beyond conventional auto-calibration and single-variable approaches we undertook multi-variable calibration using R-SWAT to manually adjust snow parameters alongside observed streamflow data. Characterization of basin-wide patterns of snowpack build-up and melt (SWE dynamics) were greatly strengthened using KRA-SWE data accompanied by improved streamflow simulation in sub-basins dominated by seasonal snow cover. UCLA-SWE data also improved SWE estimations using R-SWAT but weakened the performance of simulated streamflow due to under capture of seasonal runoff from permanent snow/ice fields in the CRB. This research highlights the utility and value of remote sensing and modelling data to drive better understanding of snowpack dynamics and their contribution to runoff in the absence of in situ snowpack data in high-altitude environments. An improved understanding of snow-bound water is vital in natural hazard risk assessment and in better managing worldwide water resources in the populous downstream regions of mountain-fed large rivers under threat from climate change.