Spatial prediction of winter wheat yield gap: agro-climatic model and machine learning approaches.

This study aimed to identify the most influential soil and environmental factors for predicting wheat yield (WY) in a part of irrigated croplands in southwest Iran, using the FAO-Agro-Climate method and machine learning algorithms (MLAs). A total of 60 soil samples and wheat grain (1 m × 1 m) in 1200 ha of Pasargad plain were collected and analyzed in the laboratory. Attainable WY was assessed using the FAO method for the area. Pearson correlation analysis was used to select the best set of soil properties for modeling. Topographic attributes and vegetation indices were used as proxies of landscape components and cover crop to map actual WY in the study area. Two well-known MLAs, random forest (RF) and artificial neural networks (ANNs), were utilized to prepare an actual continuous WY map. The k-fold method was used to determine the uncertainty of WY prediction and quantify the quality of prediction accuracy. Results showed that soil organic carbon (SOC) and total nitrogen (TN) had a positive and significant correlation with WY. The SOC, TN, normalized different vegetation index (NDVI), and channel network base level (CHN) were recognized as the most important predictors and justifying more than 50% of actual WY. The ANNs outperformed the RF algorithm with an