Semi-automatic classification method for mapping the rice-planted areas of Japan using multi-temporal Landsat images

The rice-planted areas of Japan have decreased continually with reduced rice consumption in the country. Remote sensing techniques are indispensable for providing up-to-date, public data for surveying the locations of paddy fields that have been abandoned or converted to upland crop farming. This study proposes a simple method, using multi-temporal Landsat data, for determining the threshold water index to consistently monitor the spatial distribution of the rice-planted area of Japan without using complex interactive manual operations. This method is much simpler compared to previous methods that semi-automatically or automatically determined the threshold water index based on statistical analyses using manually selected training data or the difference in histogram distribution information between flooded and non-flooded pixels. After comparing six water indices, the Normalized Difference Water Index calculated from green and short-wave infrared reflectance performed best when identifying the rice-planted area with the lowest estimation error (RMSE = 203 ha). It was also found that the proposed method had a higher sensitivity to small-size paddy fields than the previous methods, especially when using Landsat data that included many narrow rice-planted areas distributed along the valley regions. The proposed technique enabled us to consistently check the state of utilization of paddy fields across multiple Landsat images covering large regions of Japan. It was confirmed that the proposed method could accurately guide us to paddy fields that had been abandoned or were being used for upland crop farming, which is not possible from other available land use data, such as the National Land Numerical Information issued by the government.