Climate variability impacts on crop yields and agriculture contributions to gross domestic products in the Nile basin (1961–2016): What did deep machine learning algorithms tell us?

This study utilized the machine learning algorithm called "Bagging" to gain a better understanding of the relationships between climate variability, crop yields, and the agricultural contributions to the gross domestic product (agrigdp) in the Nile basin. The multiple stepwise regression model (MSRM) was also used as a comparison. The time series data collected from 1961 to 2016 for four main crops (groundnuts, sorghum, wheat, and sugarcane), rainfall, air temperature (minimum, mean, and maximum), and agrigdp were divided into 70% for training and 30% for testing using R packages. The results indicated that the Bagging algorithm improved crop yield predictions by 18%—63% compared to the MSRM. All crops demonstrated a greater sensitivity to climate variability, resulting in a low probability of achieving the highest attainable crop yield: 18% – 33% for groundnut, 18%—38% for sorghum, 21%—41% for wheat, and 22%—56% for sugarcane. The climate variability also led to a basin-wide loss of 35 million US$ in the agrigdp, with a basin-wide probability of 29%—61% for attaining the highest agrigdp. The algorithm clearly identified the ideal climatic conditions for reaching the highest attainable crop yields and agrigdp in the Nile Basin. The central factor was found to be air temperature, particularly the mean temperature, with anomalies of ± 0.5 °C responsible for 17%—69% of crop yield losses. Strategies for adapting to and managing risks associated with climate variability should focus on on-farm air temperature management practices, technologies such as greenhouses, and plant breeding.