Digital Spatial Prediction of Rainfed Wheat Yield (Case Study: Badr Watershed, Qorveh, Kurdistan Province)

Introduction: Wheat is one of the key cereals that provides a nutrition source to millions of people around the world. By conducting applied studies, the limitations of soil and climate that reduce the yield per unit area must be understood and solutions should be provided to address these limitations. One of these strategies is a detailed study and spatial prediction of yield at points with different soil and climate characteristics. Models that predict crop yield can estimate the yield regarding climate, landscape, soil and management constraints. Considering the arid and semi-arid climate of Iran, the shortage of yield per unit area and the growing population, the country needs new research and strategies to increase yield per unit area. For this purpose, the first step is to examine the spatial variations of the yield. In the present study, the factors affecting the rainfed wheat yield in order of importance and efficiency of different methods of estimating spatial variations were investigated and the predicted yield of this crop was mapped digitally. Materials and Methods: The study area, with an area of 6700 hectares is located in Badr watershed, around Ghorveh city, Kurdistan province, west of Iran. The mean annual air temperature is 12.1oC and the average annual precipitation is 345.8 mm. The soils of the area were classified in the orders of Entisols, Inceptisols and Mollisols and in 32 soil families, according to the last version of Keys to Soil Taxonomy. Based on hypercube technique, 125 observation points were selected, soil profiles were dug and described at these points and soil samples were collected from horizons of the profiles. Some physical and chemical characteristics of the soils were determined according to the standard laboratory methods. Rainfed wheat yield was measured at each side of one soil profile in a 1m×1m quadrangle. In the present study, in addition to geomorphological data, different types of auxiliary variables such as some of the primary and secondary derivatives of digital elevation model (DEM) and Landsat satellite image data were used. To find out the affecting auxiliary topographic and plantcover data on rainfed wheat yield prediction in order of importance,ReliefAttributeEval algorithm of WEKA software was used. Artificial neural network, decision tree Analysis, discriminant analysis, and averaging k-nearest neighbors are the models that were used in this research for prediction of rainfed wheat yield. Results and Discussion: Calcium carbonate, organic carbon and coarse fragments, respectively with variability coefficients of 174.4, 62.4 and 61.3%, had the highest variation and pH, CEC and sand, respectively with 3.6, 16.9 and 20.3% variability coefficients showed the least variability in the soils of the studied area. In addition to geomorphological data, the parameters that were taken from the digital elevation model include elevation, slope percentage, slope aspect, slope curvature, slope surface curvature, longitudinal curvature, slope relative position, wetness index, multiresolution valley bottom flatness index, multiresolution ridge top flatness index, valley depth, channel network base level, modified catchment area, catchment slope, catchment slope aspect and catchment height. The environmental parameters that were taken from the Landsat 8 satellite imagery, include the normalized differential vegetation and the soil-adjusted vegetation indices. The ReliefAttributeEval algorithm in Weka software, in order of decreasing importance, identified geomorphology, relative slope position, longitudinal curvature, multi-resolution ridge top flatness index, slope, normalized differential vegetation index and soil-adjusted vegetation index as the most important factors affecting rainfed wheat production in the studied area. The amount of rainfed wheat yield was predicted by the models of artificial neural network, decision tree analysis, discriminant analysis, and averaging k-nearest neighbors. The error criteria for this prediction and a significant correlation between measured and estimated values of the rainfed wheat yield, indicate a higher accuracy for the averaging k-nearest neighbors model, compared to other models. The spatial distribution of the rainfed wheat yield, predicted by the averaging k-nearest neighbors model, was mapped. In the Badr watershed, the yields are continuously reduced towards the mountains. In this landscape, as the slope increases, depth and water storage capacity of the soil decrease mainly in the presence of Entisols. These soils are seen in the eastern, southern and western parts of the watershed. At lower elevations, the soils are deeper and are mainly Inceptisols. Rainfed wheat yield increases in the piedmont landscape, including hill, glacie and alluvial fan. Conclusion: In order of decreasing importance, geomorphology, relative slope position, longitudinal curvature, multi-resolution ridge top flatness index, slope, normalized differential vegetation index and soil-adjusted vegetation index are the most important factors affecting rainfed wheat production in the studied area. The averaging k-nearest neighbors model has a higher accuracy for rainfed wheat yield prediction, compared to other models. In the Badr watershed, the rainfed wheat yield is continuously reduced towards the mountains in the eastern, southern and western parts, where mainly Entisols are present. The yield increases in the Inceptisols, located on the piedmont landscape.