<![CDATA[Meteorological drought monitoring is a pivotal action in everyday humankinds’ activities around the globe. It evaluates atmospheric conditions using weather observation instruments to measure atmospheric variables. Due to the highly sophisticated atmospheric environment, errors in drought monitoring and uncertain observation have been observed. Therefore, this research paper develops a lightweight Machine Learning (ML) and Deep Learning (DL) framework to forecast medium term meteorological drought in North West, South Africa using Standardized Precipitation Evapotranspiration Index at 3 -months (SPEI-3) timescale. This time scale reflects moisture deficits directly impacting agricultural production, early warning decisions and water management. The dataset used in this research study was obtained from South African Weather Services through a formal data request submission and not publicly accessible over a period of 10 years. Furthermore, the dataset consists of 20085 data entries and 11 data columns collected from 10 weather stations. The proposed models include SVR-RF, and, CNN-LSTM-ANN, compared to benchmark models, such as SVR, RF, CNN, LSTM, ANN, CNN-LSTM evaluated using statistical metrics, such as MSE, MAE, and . The results demonstrated irregular drought patterns during the defined period with SPEI-3 values clustered below normal conditions. Similarly, validation results showed that SVR demonstrated strong predictive performance with competitive MSE of 0.28, low MAE of 0.34 and  of 0.86. Although, the proposed CNN-LSTM-ANN and SVR-RF models did not exhibit competitive performance compared to benchmarking models, the result provides valuable comprehension, data collection, distribution, architecture, and computational power]]>
Copyrights © 2025
