Prediction of Crop Planting Map Using One-dimensional Convolutional Neural Network and Decision Tree Algorithm

The crop type planting prediction map is an essential agro-geoinformation data source to explore and quantify agriculture cultivation distribution in the coming year, implying crop planting change tendency. This paper validates the feasibility of crop type prediction using a one-dimensional convolutional neural network (1D CNN) and decision tree algorithm. To construct the ID CNN model, we encode and stack the historical Cropland Data Layer (CDL) into a 3D time series location matrix as the training dataset. According to the validation for the 2021 crop planting map in Cass County of Iowa, the prediction result owns high overall accuracy (0.927) and kappa coefficient (0.857). The major crop types, corn and soybean, have high prediction producer accuracy (0.9 - 0.95) and user accuracy (0.91-0.94). The minor crop alfalfa has lower accuracy (0.55-0.73). This approach provides an option to predict major crop type's planting maps for the next year.