Gaussian process was developed from bayesian neural networks with the infinite number of nodes in the hidden layer. It is also an bayesian model averaging approach which integrate the model prediction with the posterior probability of the parameters. In this paper, the basic theory of gaussian process for classifying satellite remote sensing data is introduced and experimented using the multi-temporal LANDSAT TM, JERS1 and ERS1 SAR data. The accuracies of the classifications have been compared with the maximum likelihood method and bayesian neural network method. The result shows that the gaussian process outperforms the other methods for classifying the LANDSAT/TM, JERS-1/SAR, and ERS-1/AMldata, and especially performs well for the sensor fusion data.
Vegetation-Soil-Water (VSW) index was applied for four sets of SPOT/HRV data collected during the 1997 rice-growing season. Objectives of this study were to examine relationships among V, S, and W indices in rice planting fields and develop new area estimation model with VSW index. Study area was located at the Saga Plains, Kyushu district in Japan. End-member points for the VSW index were determined from the four images, which were converted into reflectance factors with the 6S code. For planted rice fields, there was a negative linear relationship between the V and W indices on July 23 and between S and W indices on June 17. Linear regression lines were derived for these two relationships and used to estimate the area of the rice fields. Assuming the pixels for rice planting fields distributed around these regression lines, pixels were sampled around the lines. An optimal pixel sampling range was decided by the result of comparison between the estimated area and ground measurement points (5.47 ha). The result indicated that 2.9 times of root mean squared error achieved most suitable sampling range to estimate the area of rice fields. Consequently, our new developed model with VSW index showed -1.8% on the total estimated error ratio.
To get the better land-cover classification using the satellite multi-spectral data, the modified Maximum Likelihood classification method applying Genetic algorithms (termed MLG method) is newly proposed. As a major premise of the MLG method, the prior-probability terms of the discriminate function of maximum likelihood process are dealt with the "revision-terms" for classifying each category in the multi-dimensional feature space. A distinctive feature of this method is that the revision-terms of the discriminate function could be estimated though "GA operations". The required conditions in identifying the discriminate functions are not only to maximize the classification accuracy for the training data itself (division accuracy) and for the reference data, which is used to evaluate the overall accuracy (PCC : Probability of Correct Classification) in the image, but also to minimize the error rate considering both the "omission error" and the "commission error" as well. To satisfy those conditions simultaneously, then the revision-terms (prior-probability) of the discriminate function are estimated through the "GA operations". Three examination cases associated with modifying the discriminate functions are executed : Case 1) using the equal prior-probabilities, Case 2) using the prior-probabilities estimated from the training data, and Case 3) using the prior-probabilities estimated by using GA operations. Through those experiments, we conclude : The GA operations functioned well to increase both the division accuracy and PCC, as well as to decrease the error rate simultaneously. The average of division accuracy and PCC increase to 80.2% (Case 3) from 78.5% (Case 1) and to 82.1% (Case 3) from 72.2% (Case 1), respectively. Also, the average of error rate decreases to 41.1% (Case 3) from 46.9% (Case 1). These results indicate that the GA-based classification (MLG method) of Case3 is the most effective for improving the classification accuracy.
Radiative Transfer Equation (RTE) in the thermal infrared wavelength region is expressed as a Fred Holm type of integral equation and is essentially non-linear so that it is not easy to solve in general. Linearized inversion and iterative methods are introduced to solve the RTE. As results of the sensitivity analysis of aerosol particles on RTE, it is found that the most sensitive aerosol type is navy maritime followed by maritime, urban, desert, rural and tropospheric aerosols if the meteorological ranges are totally identical. In addition to that, it is also found that Skin Sea Surface Temperature (SSST) estimation error is sensitive to the altitude of the dust type of aerosl particles.
The mechanism of remote detection concerning the xanthophyll cycle pigments is described. The group of primary plant carotenoids can be divided into the oxygen-free "carotenes" and into the "xanthophylls, " which contain oxygen in different forms. Functional chloroplasts include β-carotene, lutein, violaxanthin and neoxanthin as major carotenoid components. Under highlight conditions the xanthophyll violaxanthin is deepoxidized to zeaxanthin via antheraxanthin. These conversions result in absorption shift of the xanthophyll cycle pigments. The difference spectrum associated with these pigment transitions shows a detectable sign near 531nm in the leaf reflectance.
A forest fire occurred in Nishiokoppe village, in northern Hokkaido, Japan, on 20-21 April 1998. Under strong winds, the burnt area extended over an area of 43 ha, including private forests of 28.4 ha and Hokkaido prefectural forest of 11.4 ha. This study examined the effectiveness and practicality of forest fire monitoring using satellite data. The study selected the optimum SPOT band and vegetation index for classifying forest cover type to distinguish between burnt stands. The field survey and analysis found the following. (1) The reflectance value for heavily damaged stands was lower than the value for other stands using three SPOT XS bands and the vegetation indices. (2) The vegetation index (Veg. index) of the transformed band between SPOT XS3-XS2 was useful for distinguish-ing the burnt areas of forest. (3) A forest cover type classification was produced using the supervised maximum-likelihood classifier using band XS1, the Veg. index, and NDVI, since they showed different spectral patterns. (4) The burnt area of forest was about 24 ha and the agreement rate for the burnt area with field measurements was 52.9%. This was attributed to differences in leaf phenology (unopened, emerging, open) and the fire tolerance of deciduous trees in spring. The burnt area in the satellite images was smaller than that measured in the field. (5) A system for forest disaster monitoring using satellite data in Japan was demonstrated.