In and end of season soybean yield prediction with histogram based deep learning

Depending on climate change and global warming, the yield obtained from agricultural activities is gradually affected. Farmers, government institutions, agricultural insurances need applications that can provide decision support in taking precautions. On the other hand, government agencies need pre-harvest yield estimates for crop planning, import and export planning, market price determination, and storage needs. Also, the post-harvest crop yield estimates are important for comparison with farmer statements for agricultural insurance and government agency support payments. Within the scope of this study, a crop yield prediction module has been developed that can provide in and end of season estimation of crop yields to be obtained from the determined regions. The module was developed as a QGIS plugin with the Python programming language. Two main deep learning yield prediction models based on CNN and LSTM are integrated into the plugin. Google Earth Engine is heavily utilized to generate MODIS Surface Reflectance, Land Surface Temperature and Daymet Daily Surface Weather histograms for each region in study area. Histograms are used as input data into deep learning algorithms that are trained to predict the yields of regions. Yield predictions obtained from prediction models are presented to the user in a Shape file format via the deveoped QGIS plugin in kg/ha unit. According to the results, although CNN and LSTM based models deliver similar results, CNN based method generally provide slightly better results for both in and end of season yield prediction.