*3.2. Initial Classification Result*

Before a self-learning procedure was employed, an initial classification was first performed using the initial training data. The qualitative and visual assessment of the initial classification results was conducted using time-series Landsat images (Figure 5). Two subareas were identified as an over-estimation of sorghum and a clustered pattern of winter wheat. The clustered pattern of winter wheat was attributed to the inclusion of more training data than the other class types in the western part of the study area. Confusion between winter wheat and alfalfa, which showed similar temporal NDVI variations in winter, could have also contributed to the clustered pattern of winter wheat in the western part. In addition, sorghum and soybean, which are the typical summer crops in Kansas, showed similar temporal NDVI variations, which led to an over-estimation of sorghum. Confusion between grain/hay and grass was also observed in the initial classification result

**Figure 5.** Initial classification result, and its visual comparison with Landsat images in two subareas.
