Impact of climatic factors and poverty on wheat yields via machine learning in a semi-arid region across West Asia.

Abstract

Winter-sown wheat is particularly vulnerable to high temperatures in spring. However, little is known concerning the impact of high temperatures and other climatic and non-climatic factors in combination on winter-sown wheat yields in arid and semi-arid regions. Therefore, in this study we investigate the impacts of high temperatures and other climatic indices, including evapotranspiration and aridity index during spring months, and non-climatic factors, including technology improvement and poverty ratio on irrigated and dryland wheat yields. We employed Random Forests a machine learning technique to our long-term data (1980-2010) and annual wheat yields (irrigated and dryland) for 27 provinces of Iran. The results show that technology improvement and poverty ratio are the most important variables which explain irrigated wheat yields variability. Temperatures above 31°C, evapotranspiration and aridity index are the other variables that alter irrigated whet yields in the region during the period 1980-2010. Dryland wheat yields mainly explained by temperatures above 31°C. Poverty ratio, technology improvement, evapotranspiration and aridity index ranked after high temperatures above 31°C, respectively. This study demonstrates that, although, in arid areas technological improvement and irrigation may avoid yield reduction due to high temperatures in spring months, in poor arid areas and dryland systems, high temperatures significantly reduce wheat yields.