We developed the method for crop classification of upland crops at a plain in Hokkaido, Japan through examining seasonal changes in NDVI of multi-temporal RapidEye satellite data. We classified crops into wheat, potato, sugar beet, adzuki bean, soybean and other crops using several combinations of RapidEye data. The plot based accuracies on each combination are evaluated through comparing error matrixes. Results indicate that the overall accuracies on the combinations of less than one-month interval imageries become higher. The proposed method obtained the overall accuracy of nearly 90% on all plots over the study area.
Gully erosion, which shows ditch like feature on sloping land caused by high intensity rainfall, deteriorates agricultural productivity of land. This phenomenon can be observed widely in the world and surveyed individually its extent by measurement at ground level. However, spatial distribution composed of multiple gullies in wide area could not be characterized quantitatively due to lack of optimal technologies. High spatial resolution satellite data has been popularized in the latest decade and some researchers have attempted to utilize this to extract gully erosion. They showed potential capability of such data for rapid mapping of gully erosion in wide area in condition that more robust method would be developed. In this study, therefore, the authors examined a method of integration of image processing and GIS techniques for extracting gully eroded area from multiple geographical sources including high spatial resolution satellite data. The study site was located on hilly land in the northern part of Luzon Island of the Philippines. The major source to extract gully erosion was WorldView data observed on October 23, 2011, when most part of land was not covered by vegetation. Methodology was divided into two steps; firstly extraction of candidate of gully eroded area and secondly removal of non-gully area. For the first step, after noise reduction by Median Filter to WorldView data, Sobel Operator was performed to enhance edge features. At this step, another treatment of selecting gully candidate was enhancement of edge feature appearing in parallel to slope direction. For the second step, removal area was discriminated using image texture information obtained from Grey Level Co-occurrence Matrix (GLCM) of image data. Other removal areas were identified by the following methods; higher value part compared to average filtered data, high value of cross section convexity calculated from digital elevation data and forest area obtained from multi-temporal ALOS/AVNIR2 data. Result of extracting gully eroded area was validated by comparison with location data of head and end of major gullies surveyed by Bureau of Soil and Water Management (BSWM). Ratio of properly extracted gully was 63.4% in total average but about 80% for some gullies, while 30% for other gullies. This difference was originated presumably by the condition that vegetation cover remained considerably in low ratio case. The result also demonstrated effectively spatial distribution of gully eroded land for wide area, which was represented by watershed unit obtained from digital elevation data.