Enhancing food security predictions in drought-prone regions using Earth observations and machine learning algorithms

Droughts, characterized by minimal precipitation, significantly decrease agricultural yield, leading to food insecurity, particularly in vulnerable regions. The study addresses critical gaps in current drought prediction models, focusing on enhancing predictability regarding the location, magnitude, and assistance requirements during drought events. Machine learning algorithms, including Random Forest Regression, Decision Trees, Bayesian Ridge Regression, and others, are employed to predict drought impacts. Historical and real-time datasets, such as landcover, soil moisture, precipitation, agricultural yield, vegetation health, population, and foreign aid, are integrated from open data sources for comprehensive analysis.
The proposed forecast horizon of agricultural seasons, with monthly data for temporal accuracy, enables timely intervention and preparedness. This project defines a threshold for “sufficient yield of crops” to determine “food security” for the study areas by predicting the crop yield and if foreign aid will be needed. This predictive tool holds potential for early warning systems to safeguard food supply, especially in the face of increasing frequency and intensity of droughts driven by climate change.

Data validation is incorporated using high-resolution remote sensing techniques to verify model predictions against observed conditions. Climate projection models simulate future scenarios, evaluating potential changes in drought frequency and food- security severity over time. Sensitivity analyses assess the potential impacts of varying projections on food security outcomes.

The study's novelty lies in its utilization of machine learning to predict drought-induced food shortages, incorporating location-specific insights into assistance needs. By harnessing multiple Earth Observations, the study offers a comprehensive perspective on drought and food security issues, with implications for improving societal resilience and reducing malnutrition, particularly among vulnerable populations. This endeavor underscores the transformative potential of Earth observations and machine learning in addressing pressing societal challenges.