Editorial: Assessment of Climate Change Impact on Water Resources Using Machine Learning Algorithms | Journal of Water and Climate Change (2024)

Journal of Water and Climate Change jwc2024002.

Search Dropdown Menu

Understanding how hydroclimatic variables change over time considering climate change impacts is crucial. Nguyen et al. (2023) evaluated two ML models, i.e., convolutional neural network (CNN) and long short-term memories (LSTM), for estimating hydroclimatic variables at the 3S River Basin. For assessing climate change impacts, three climate models, i.e., CMCC-CMS, HadGEM-AO2, and MIROC5, and two climate scenarios, i.e., Representative Concentration Pathways (RCPs) 4.5 and 8.5, were considered for three future periods. An increase in the mean annual temperature and fluctuations in the annual precipitation were detected. Furthermore, ML-based future projections yield a rise in the streamflow in the Srepok and Sesan Rivers, a reducing trend of streamflow in the Sekong, and increasing flood risk in the Sekong and Sesan basins.

Climate change may impact the quantity and quality of water in rivers. By considering two different climate datasets, Syed et al. (2023) modeled the water balance of Chitral River using three hydrological methods: (i) soil and water assessment tool (SWAT), artificial neural network (ANN), and the hybrid SWAT–ANN. Their comparative analysis revealed that ANN achieved the best values of accuracy metrics followed by the hybrid SWAT–ANN for the reconciled gridded climatology data, while ANN was found to be more sensitive to the input data. Hence, it may not be the best option when the data at disposal entails significant uncertainty or is limited.

Nandi & Reddy (2024) simulated streamflows of the past (1999–2010) and future (2019–2040) of the Bhima River basin, India using an improved water balance model incorporating storage structures. The future climate data were adopted from 19 CMIP5 GCMs considering RCP4.5 and RCP8.5 scenarios. The results of historical data showed an approximate 30% improvement in terms of the Nash–Sutcliffe criterion, while the projection results revealed that climate change has a negative impact on precipitation, yielding drought conditions in the Upper Bhima River basin. Furthermore, a significant reduction in the monsoon streamflow, which have a negative influence on the productivity of agriculture in the watershed, is estimated. Their findings can be useful for the future management of water, agriculture, and hydropower in the study area.

Goodarzi et al. (2024) assessed the impacts of climate change on the streamflow in the Dez basin, Iran. They downscaled the daily precipitation and temperature of two CMIP6 GCMs, i.e., ACCESS-ESM1 and BCC-CSM-MR models, using the LARS-WG technique under the SSP5.8.5 and SSP2.4.5 climate scenarios. Consequently, they developed a dataset spanning from 2015 to 2043. It was utilized as an input to simulate streamflow using IHACRES, which is a hydrological model. Based on their results, more rainfall varying from 20 to 33%, depending on the climate model and scenario, is expected. Furthermore, the variation of the estimated river discharge relies on the model. To be more specific, in comparison with the base period, a decrease of 15% and an increase of 16% are expected for the ACCESS-ESM1 and BCC-CSM2-MR models, respectively. Such climate change impact assessments can provide useful information for water resources management in the river basin.

Kodihal et al. (2024) studied variations of runoff depth under a changing climate for the Jaipur city in the state of Rajasthan, India. They used a statistical downscaling model (SDSM) to downscale the precipitation of one CMIP6 GCM, i.e., NorESM. Under the SSP 4.5 and SSP 8.5 scenarios, the projected rainfall has a decreasing trend of 50% and 24% between 2030 and 2042, while a rising trend of 7% and 19% is forecasted for the same period, respectively. By developing a curve number map, the runoff was computed using the curve number method. Their results indicated a reducing trend in the surface runoff in the future (2030–2050) in the region studied.

Garg et al. (2023) applied multivariate multi-step LSTM to forecast the flood runoff in the Godavari River Basin, India using 18 years of data. The performance of the proposed model was compared with those of Auto-Regressive Integrated Moving Average, Prophet, and Neural Prophet, while LSTM outperformed others. Considering various features as input data, their comparative analysis showed that ML models performed better than statistical ones for the case under investigation.

Climate change can have an influence on the frequency of hydrological extreme events occurrence. Amichiatchi et al. (2024) studied trends of four extreme precipitation indices (flood and drought) in several basins in Côte d'Ivoire considering historical rainfall data ranging from 1976 to 2017 and futuristic projections between 2020 and 2050 under climate scenarios. The indices were (i) a drought-related index, which is the consecutive dry days (CDD), and (ii) three flood-related indices, including maximum annual rainfall (Pmaxan), very wet day (R95p), and maximum 5-day rainfall (Rx5days). They exploited the distribution mapping method and the modified Mann–Kendall's test for correcting the bias of regional climate models (RCMs) and trend analysis, respectively. According to their results, a downward trend in the flood-related indices was detected in the historical data for almost all watersheds under investigation. Moreover, a significant upward trend was identified for the future projections. Finally, their findings can provide a better perspective of flood and drought events.

One of the inevitable challenges of using climate models for local applications is their spatial resolutions. Basically, GCMs have a coarse resolution, while RCMs have a relatively finer resolution. For assessing climate change impacts, the outputs of GCMs are required to be downscaled and bias-adjusted so that they can be more reliable for site-specific applications. Jamal et al. (2023) exploited the ensemble empirical mode decomposition method to adjust the bias of temperature data of seven GCMs for the Upper Indus Basin, which is covered with glaciers and snow. Their research forecasted a decrease between the minimum and maximum daily temperatures in the future under the CMIP6 scenarios. For all climate scenarios, the duration of melting glaciers and snow is expected to expand by 1 or 2 months over the catchment due to the increase in projected temperatures.

Water supply systems may be vulnerable to climate change impacts when severe droughts, as a climate-related risk, are expected in the future. Ma et al. (2023) assessed applications of GCMs by conducting a three-fold analysis to evaluate the future status of water supply systems in Korea, particularly the Boryeong Dam basin. Their evaluations include (i) analysis of basic characteristics of rainfall events (like the number of rainy days and rainfall intensity, etc.), (ii) analysis of the occurrence and persistence characteristics of dry periods using annual rainfall data, and (iii) rainfall–runoff simulation considering reservoir operation. The future climate data indicated no water supply problem in the case study, which seems to be in contrast with many other studies forecasting severe droughts in the future. Thus, they concluded that the GCM-based rainfall data may not be suitable for estimating the water supply shortage in the Boryeong Dam basin.

ACKNOWLEDGEMENTS

REFERENCES