Impacts of Climate Extremes on Crop Production: Modelling Yield Loss using Machine Learning Algorithms in the Sudanese and North-Guinean Climatic zones of Mali

Abstract

Climate change is a global phenomenon of climate transformation of the planet. Climate change causes environmental degradation and decreases in agricultural production and incomes and therefore poses a significant threat to food security and sustainable livelihoods, as well as to socio-economic stability. The approach to delineating impacts of climate extremes on crop production is complex and may involve the use of crop simulation models and, in some cases, the use of statistical techniques of equal complexity. The main objective of this study was to assess the impact of agroclimatic extremes on crop yields. In this research, two machine learning (ML) algorithms, namely logistic regression and Random Forest models, were used to assess the yield loss as a result of agroclimatic extremes. The input data included observed yields of cotton, maize, and millet and meteorological data (1990-2017) from the Sudanian and North Guinean zones of Mali. The growth of the digital age has made almost all human activities the source of ever-increasing amounts of information. machine learning techniques for data analysis can be understood as a problem of pattern recognition or, more informally, knowledge discovery and data mining. The agroclimatic extremes considered in this study are the late onset of the cropping season, early cessation of the cropping season, shorter duration of the rainy season, intra-seasonal heat waves, and seasonal rainfall deficit. While yield loss was the predictand, these agroclimatic extremes were considered as the predictors. When taken individually, a simple linear regression does not describe the relationship between the predictors and the predictand. When considered altogether in ML, such as random forest regression (RF) and logistic regression (LR) modelling, the relationship can be depicted as yield loss as a result of agroclimatic extremes. Our results showed that LCS and cropping season were dominant factors affecting yield. The LCS and cropping season were the most correlated indices. Predicting the occurrence of these agroclimatic extremes have the advantage of identifying suitable agricultural inputs and avoiding certain risks. However, RF showed much better performance compared to LR. Therefore, ML is useful and very robust tool to predict yields loss as a result of extreme climate events, especially when large-scale datasets are available.