Validation is an important step of land cover mapping. To find a better sampling method for validation, stratified random sampling, simple random sampling, and use of training data for validation were evaluated using assumed true land cover data and simulated classified results. Evaluation was performed by comparing error matrices that were obtained from various sampling methods using the true error matrix. Stratified random sampling with area weight having a sample size of at least 20-40 per class is recommended from the perspective of statistical rigor and practicality.

抄録全体を表示

We propose a method for evaluating the damaged rice paddy caused by salt breeze. In this paper, we analyze highly-precise ground-truth hyperspectral imaging data of damaged rice paddies acquired in indoor experiments under controlled conditions. The ground-truth data are analyzed to establish the analytical method that can be applied to the data acquired by higher-altitude observation systems. In order to provide the damaged rice paddies by salt breeze as observation targets artificially, experiments to reenact salt-breezed damage are conducted by exposing plant pots to salt solution blow under the controlled condition. The sequential changes of the hyperspectral data in damaged leaves are observed every 24 hrs, up to 120 hrs after the saline treatment. In order to evaluate sequential change of the activation level at damaged leaves, the traditional index, the blue shift of the red edge, is applied. Based on the observation of sequential spectral change, we utilize an index : Normalized Difference Green Index (NDGI), to interpret the damage of rice paddies. The index, using small number of channels, is adopted for the purpose of analyzing multispectral data such as satellite images eventually. Evaluating the distribution of the data points in the feature space spanned by NDGI in combination with traditional vegetation index NDVI, the level of salt-breezed damage is explained precisely. At the end of this paper, we show the applicability of the method to the mixel data and multispectral imaging data that can be generally acquired by satellite or airplane.

抄録全体を表示

A lot of studies of global land cover map have been produced so far. However, there are only a few global maps representing the place and the time that large changes occurred. In this study, there were produced those areas where land cover changes may have taken place during the period between 1982 and 2000. These areas were extracted on the basis of NDVI temporal profiles derived from the NOAA-AVHRR data. Furthermore the difference in results by two types of NDVI data set was examined.
The method of producing these maps is as follows. Possible land cover change areas were extracted from the change of the annual mean of NDVI, the difference between its annual maximum and minimum values and its annual pattern. Years of land cover change were determined by large NDVI change during a certain year. In the mapping process potential land cover change was detected in the case that the NDVI value didn't recover to the previous level after it had changed. The NDVI change which was affected by the weather change or geometric misregistration was excluded from land cover change. As a result, a possible land cover change map was produced which shows large vegetation change such as the transformation of the desert to cropland, the deforestation for cropland or disappearing of wetland.

抄録全体を表示

What a kind of merit can be expected for the use of unclassified photo reconnaissance satellite images released in 1995? Such image data had been taken from 1960 to 1972 by the American KH-4/CORONA and KH-7/GAMBIT etc. The total number of the photographs is said to reach more than 800,000 sheets. Making a comparison of the recent satellite images with such old images at the same area, land cover change or some environmental phenomena can be detected. A case study was made to know the capability of the old images at two test sites, Yokohama area including Keihin industrial zone and Totsuka suburban area about 40 km from Central Tokyo southeast. The following results are obtained.
(i) Unclassified reconnaissance satellite images show us the environmental change in 1960s as evidence.
(ii) Those images show the discharged heavy plumes in the industrial area. The satellite images record the actual phenomena at those days.
(iii) More classified satellite images should be released as earlier as possible for any study in the area of non-security.

抄録全体を表示