Abstract Drought significantly impacts water resources, ecosystems, and human livelihoods, particularly in northern Nigeria, where rainfed agriculture is predominant. The effects, including reduced water supply, poor water quality, crop failure, environmental hazards, and civil unrest, are worsened by climate change, which increases drought frequency and severity. Monitoring the spatiotemporal dynamics of drought—severity, magnitude, intensity, duration, and extent—is critical for early warning, prevention, and preparedness. This study evaluates the effectiveness of machine learning models—Random Forest (RF) and Support Vector Machine (SVM)—in predicting drought events in Kano State. Historical meteorological data (2012–2023) from the Visual Crossing Weather (VCW) dataset were validated using Nigeria Meteorological Agency (NIMET) data at a 0.05 confidence level. Key variables included precipitation, temperature, solar radiation, and indices such as SPI Gamma, SPEI, and Fisk Distribution. Results indicated irregular rainfall patterns, with a recovery from 2018. SPI3 Gamma and Pearson indices identified dry periods (2013–2018) and wet periods (2019–2022), while SPEI provided a more comprehensive drought assessment by integrating evapotranspiration. The RF model outperformed SVM in accuracy (66.8% vs. 65.7%) and precision (60.3% vs. 45.4%), effectively reducing false positives. Recall scores were comparable (66%), indicating both models reliably identified drought events. Random Forest (RF) and Support Vector Machine (SVM) performed well and in the absent of hydroclimatic data, the models offer valuable insights for improving climate resilience and drought preparedness in Kano State.
Oriahki et al. (Tue,) studied this question.