Climate-Resilient Farming: Advancing Short-Term Weather Predictions for Sustainable Agriculture

Abstract

The agricultural sector of Pakistan is vital but at the same time under threat of significant challenges due to imminent water scarcity and a dire need for sound water management practices. The accuracy of the reference evapotranspiration estimation is critical in irrigation scheduling and water resources management; therefore, this research is aimed at realizing that objective. In this research, we follow an ensemble approach, including RF, GB, and BiLSTM, in the improvement of ETo calculations over three important agricultural regions in Pakistan: Bahawalpur, Multan, and Hyderabad. To make sure that the models exhibit strong robustness under practical applications, training, and evaluation were carried out for different input combinations in various scenarios. The analyzed scenarios span from the full suite of meteorological data to reduced feature sets to account for varying data availability. The study utilizes a comprehensive dataset, combining meteorological data from airport weather stations, Pakistan Council of Research in Water Resources Automated Weather Stations (AWS), and satellite daily values collected from NOAA RDA to overcome the limitations of data scarcity and quality. The methodologies employed are based on the Penman-Monteith (PM) equation, adapted to accommodate localized climatic inputs. Our research demonstrates significant improvements in the precision of ETo estimations for Pakistan under full (RMSE = 0.13 - 0.17 mm/day, MAPE = 2.2% - 3.1%, R2 = 0.99) and limited input parameters especially upto scenario 3 (MAPE = 2.73% - 5.52%, R2 = 0.99), facilitating more accurate irrigation scheduling and water management strategies. The practical applications of this enhanced ETo estimation extend to water conservation, drought management, and optimization of crop yields which are fundamental to the sustainability of agricultural practices in Pakistan.