Votre recherche
Résultat 1 ressource
-
ABSTRACT A comprehensive strategy that incorporates trend analysis, machine learning (ML), and climate model review is needed to improve water resource forecasts and evaluate hydroclimatic variability. The present study effectively combined various forms of categorical and continuous performance metrics for the CMIP6 and the reanalysis datasets in the Upper Godavari Sub‐basin area (UGSB) (India). MERRA2 reanalysis datasets demonstrated the highest accuracy for precipitation forecasting, achieving a POD of 0.82 and CSI of 0.71, while JRA‐55 closely followed with a CSI of 0.69. CMIP6 models exhibited overestimation tendencies, with a mean FAR of 0.34, highlighting their limitations in capturing precipitation extremes. Thereafter, to understand the long‐term variability of the best reanalysis product, trend analysis was also performed using the Mann‐Kendall test, Pettitt's test, Van Neumann ratio (VNR), and Innovative Trend Analysis (ITA). This analysis revealed properly the spatial variability of the precipitation, showing increasing (1.5–2.3 mm/year) and decreasing rates for various stations inside the UGSB. Thereafter, the temporal frequency and the intensity were captured by the Continuous Wavelet Transform (CWT) analysis, which further identified shifts in hydroclimatic variability towards higher frequencies after 2000. Thereafter, the prediction accuracy of prediction datasets of various ML models, which included Random Forest (RF), Multi‐Layer Perceptron (MLP), Long Short‐Term Memory (LSTM), and XGBoost models, were optimised by The Harris Hawks Optimization (HHO) algorithm, and the best optimised model, RF‐HHO, showed reducing RMSE to 4.92 at Ambajogai, 4.81 at Bodhegaon, and 5.21 at Ranjni. The study highlights the importance of combining reanalysis products, trend analysis, and optimised ML models to improve future precipitation predictions and support effective water resource management.