| Abstrakt: |
Rainfed agriculture is very sensitive to agro-meteorological factors such as temperature, humidity, sunshine and radiation or hygrometry and, therefore, the evolution of crops is necessarily affected by changing climatic conditions. The main expected expressions of climate change in Morocco include reduced rainfall, increased temperatures and intensification of extreme risks such as drought, heat waves, frost, floods, etc. Therefore, the development of methods for modeling and early prediction of crop yields in the Gharb plain, located in northwestern Morocco, is essential. Machine learning is a tool that can help make key decisions regarding crop yield prediction. The effects of climate change on the spread of autumn rainfed cereals (durum wheat, soft wheat and, barley), an important agro-food source in the Gharb plain, were assessed in this study. A multiple linear regression (MLR), artificial neural network (ANN) and, random forest (RF) model based on the integration of drought indices derived from satellite imagery and reanalysis meteorological data. Model accuracies were calculated and compared. Correlation values suggested that yield decreases with decreasing crop and soil moisture in early spring and with higher thermal conditions near harvest. In general, all drought indices showed the best effect during the stages when yield is most photo-synthetically dynamic (spring), in contrast to the early stages of the vegetative cycle (autumn and winter). The strength of the statistical relationships found by the MLR, ANN, and RF methods were quite similar, with some improvements found by RF. A large number of true positives (hits) of yield loss occurrence with hit rate (HR) values above 90% was obtained. |
Full text from SpringerLink