2.2.2. Landsat Data

In this study, the classification output was at a 250 m spatial resolution, which was the same as the MODIS NDVI data. It is often difficult to collect training data for supervised classification from mid-resolution remote sensing data. Thus, Landsat data sets were used to supplement the MODIS NDVI data sets for initial training data collection. The training data were collected through visual analysis of a total of 33 Landsat-7 ETM+ and Landsat-8 OLI images obtained from April to August.

The class types and the number of training data per class are shown in Table 1. To mimic a situation where many training data were not available, only a small amount of training samples were collected, which occupied approximately 0.26% of the study area. Supervised classification was conducted using the initial training data sets of 10 class types. The main purpose of classification in this study was to accurately classify the major crops; to facilitate this, some minor crops such as alfalfa and hay, in addition to class types such as water, city, forest, and grass were merged as general grain/hay and non-crop classes, respectively, for evaluation of the classification results (Table 1). Besides the collection of the initial training data, the Landsat data sets were also used for visual comparison and confirmation of classification results.


**Table 1.** The list of class types for supervised classification and merged classes, and the number of initial training data per each class.
