This study aimed to improve the accuracies of forest vegetation type classification in Daihachiga River Basin and examined the possibility in using multi-seasonal Quickbird (QB) images from phenology viewpoint and in using LiDAR data from tree canopy height viewpoint. QB images were acquired on April 12th and May 23rd 2007 and corrected geographically and topographically. Digital Canopy Height Model (DCHM) was developed using LiDAR data (Gifu prefecture, 2003). The forest was classified using QB images and LiDAR data and the classification results were compared to those of April and May QB images. First, it was classified based on April and May QB images using maximum likelihood method (MLM). Second, it was classified based on QB images and LiDAR data and the methods are as follows: 1) The area in which DCHM was higher than or equal to 3 m and Normalized Difference Vegetation Index (computed using May QB image) was greater than or equal to 0.001 was classified as forest area. 2) The forest area was divided into a forest area at altitudes less than 1,200 m (MapA) and a forest area at altitudes over 1,200 m (MapB) using Digital Terrain Model (Gifu prefecture) and each of the two areas was classified into deciduous and evergreen forest with April QB image using MLM. Deciduous forest area at altitudes over 1,200 m was classified into larch and deciduous broadleaf forest with May QB image using MLM. Finally, a forest vegetation type map was made up by combining Map A and B. 3) Classification accuracies were evaluated by selecting validation pixels using random point sampling and overall accuracies and KHAT were computed using the pixels. The results showed overall accuracy of forest and non forest classification based on QB images and LiDAR data was 96 % (KHAT: 0.90) and which was followed by the overall accuracy of forest and non forest classification based on May QB images .We suggest that the higher accuracy was achieved by avoiding misclassification between forest and non forest using LiDAR data. The overall accuracy of forest vegetation type within forest area based on QB images and LiDAR data was 93.8 % (KHAT:0.91) and evergreen forests could be classified with high accuracy(0.96), while the overall accuracy of forest vegetation type classification based on April and May QB images are 66.8 % (KHAT:0.91) and 68.8 %(KHAT:0.50). Thus, use of multi-seasonal QB images and LiDAR data improved the accuracies of forest vegetation type classification greatly.
The author developed a method to estimate paddy rice planting time using multi-temporal MODIS data for a study site of Java Island, Indonesia, which was representative rice cultivation area in the Tropical Asian Region characterized by complex pattern of rice cropping. The method employed 16-day composite MODIS data product (MOD13Q1) and indices about surface conditions of vegetation and water obtained from the dataset. Time of planting paddy rice with temporal unit of 16 days was estimated from temporal changes of these indices for the period from April 2000 to December 2011. Estimated result showed good coincidence in time of planting with statistics data, but it also showed limitation of estimating precise acreage of planted rice due to coarse spatial resolution of data which was often mixed with other land use or different cropping pattern. The method developed in this study depicted clearly time and place of paddy rice planting for the first period starting after the end of dry season and for the second period in whole Java Island. The first period which was appeared in around December to January showed more yearly variation of planting time compared to the second period appeared in around March to April. This variation was discovered to have correlation with rainfall amount in late of dry season to early rainy season. Delay of paddy rice planting time was found to be associated with less amount of rainfall especially for areas of rainfed agriculture and lower part of irrigation networks.
This report showed footprints of a scientist who tried to establish measurement of ecosystem functions using remote sensing technology. First, the author started a study of grassland ecosystems, and then shifted to agriculture, forest and satoyama ecosystems. In this process, he encountered with remote sensing technology which was making rapid advances. He tried to establish “satellite ecology”, as a new dogma. Around 2000 and after, satellite sensors attained remarkable progresses in spatially, spectrally, and temporally. In the period, both hardware as computers, and software as GIS (Geographic Information System) had also improved. As the results, ecological researchers could compare and analyze experimental data obtained in the fields and satellite data directly. In the analysis, the author tried to keep ecosystem as it was, not to divide small orders, because it may possible to extract own ecosystem functions. Remote sensing has broad base and the study of synthetic sciences. In the tenor of JASS (Japanese Agricultural Systems Society), it says that the mission of JASS may integrate the divided agricultural sciences climbing over the fence of existing studies. This idea may common for satellite ecology.