Remote sensing and cutting-edge machine learning technology are shedding light on the agricultural prospects of conflict-affected South Sudan, particularly the cultivation of sorghum—a key staple crop for addressing food insecurity. With persistent conflict and economic instability hindering traditional farming practices, researchers have deployed advanced techniques to predict sorghum yield, seeking to provide actionable insights for policymakers and humanitarian actors.
This innovative study examined sorghum yields from 2018 to 2021, particularly focusing on the states of Upper Nile and Western Bahr El Gazal. Leveraging data from 2,100 local farmers and employing five machine learning models—including Random Forest, Decision Tree, Extreme Gradient Boosting, Support Vector Machine (SVM), and Artificial Neural Network (ANN)—the research aims to offer reliable yield predictions under challenging conditions.
The findings are promising. Researchers reported an average sorghum yield of 366.03 kg/ha, with considerable variability influenced by factors such as cultivated land size. The study revealed strong correlations, indicating larger cultivated areas typically result in higher yields. Specifically, the analysis determined the correlation coefficient between land size and sorghum yield was 0.75 (p < 0.001), highlighting the importance of land size as a primary factor influencing agricultural outcomes.
Despite the adverse effects of conflict on agricultural productivity, the use of machine learning provided surprisingly accurate predictions. During the model evaluation phases, Decision Tree, Random Forest, and Extreme Gradient Boosting algorithms achieved high levels of accuracy, with each showing R² values exceeding 0.70, indicating significant predictive power.
Interestingly, incorporating conflict occurrence data alongside other variables had minimal impact on yield predictions, underscoring the robustness of the machine learning approaches. One of the researchers noted, "Adding conflict occurrence data had minimal impact on yield predictions, illustrating the robustness of our machine learning approaches." This suggests potential for these models to be applied for yield forecasting even amid instability.
The research holds substantial promise for shaping future food security strategies. Reliable yield predictions equipped humanitarian organizations, farmers, and government officials with the insights needed for effective planning and resource allocation. The study explains, "Despite the challenges posed by conflict, we achieved reasonably good end-of-season sorghum yield prediction, indicating potential for informed food security plans."
Given South Sudan's climate, characterized by high variability, accurate predictions are imperative. The climatic factors proved to have considerable effects, with the study emphasizing the need for comprehensive analysis considering soil moisture, temperature, and precipitation levels. With rainfall patterns often inconsistent due to climate change, utilizing machine learning presents an adaptive framework to respond to these agricultural challenges.
Looking forward, the study recommends advancing data collection methods and improving capacity building to fully leverage these digital resources. While some limitations remain—such as the challenge of comprehensive data collection through conflict zones—this groundbreaking research suggests the integration of remote sensing and machine learning could significantly transform agricultural forecasting and, by extension, food security strategies across South Sudan and similar settings.