The problem of classification performance has been studied in remote sensing community. Existing methods for evaluating the classification performance are applicable only pure pixels. In fact, remotely sensed data include mixed pixels, so the classification performance cannot be discussed without considering the mixed pixels. This paper deals with a performance evaluation method with four functions considering mixed pixels. Using classification degree for mixed pixels with the methods allow us to discuss the detailed classification performance. Our experimental result indicated that it was insufficient only in the evaluation for pure pixels and was necessary to consider mixed pixel like our evaluation method.

View full abstract

The thickness of the soil layers in which tree roots are able to develop freely influences forest composition and growth. Trees growing in shallow soil are usually less well supplied with water and mineral nutrients than those growing in deeper soil. A soil may be deep in an absolute sense but, because of a relatively impervious layer, such as hardpan or because of a high water-table, may be shallow in a physiological sense. Penetrability measurements have been found useful in evaluating the influence of different forest types on the physical properties of soils. Commonly the penetrability of soils can be measured by using the Hasegawa-formed soil penetrometer and can be judged as the soil softness content(SSC). Previous studies report soil with more than 1.9 cm/drop of SSC to be highly permeable and therefore roots are more likely to be extensively developed. Based upon this theory the depth of soil layer with more than 1.9 cm/drop of SSC can be defined as the Effective Depth of Soil Layer (EDSL). We examined the relationship between the Ratio Vegetation Index (RVI) and the EDSL and established a "Runoff Simulation Model (RSM)" based upon the theory of the Storage Function Model method.
The conclusions are that (1) a strong positive correlation between the RVI (ground measured) and the EDSL was given, (2) applying results of conclusion (1) to satellite analysis a similar correlation between the RVI (satellite analysis of JERS 1/OPS data) and the EDSL was observed and (3) the simulated storm-runoff hydrograph coincides with the observed one well.

View full abstract

Recently, withered tree areas have been appearing all over the world. In order to detect these areas in this incipient stage by remote sensing, we measured the spectral characteristics of leaves in different living conditions. By using the pattern decomposition method, we analyzed the relationship between the pattern decomposition coefficients and the state of withered vegetation, and defined a vegetation index to detect withering pine trees. We applied this index to satellite data, and surveyed the distribution of withered areas in Nara, Japan. By this method, tree mortality is detectable, especially in medium and heavy withering areas.

View full abstract

There are two major methods for NOAA/AVHRR-based fire detection. These methods are the threshold method and the contextual method. Threshold method detects fire pixels by comparing the radiant temperature of each pixel to predetermined threshold values. Contextual method detects fire pixels by comparing the radiant temperature of each pixel to its surrounding pixels. In the present study, these NOAA/AVHRR-based fire detection methods were compared and evaluated. And the authors improved the contextual method for Russian Far East. The processes were ; (a) First, conventional contextual method and modified contextual method were compared and evaluated for application to large-scale forest fire, (b) Secondly, modified contextual method and threshold method were compared and evaluated. A burnt area map developed from Landsat/TM data was used to assess and validate the locations of fire pixels detected by each method. The results show that ; (a) modified contextual method was able to detect fire pixels that the conventional contextual method failed to identify, but modified contextual method was inferior to conventional contextual method in fire detection accuracy, (b) the number of detected fire pixels was more and fire detection accuracy was better by the contextual methods than by the threshold method.

View full abstract

Phase functions of relatively large particle size of non-spherical aerosols containing bubbles are clculated by means of Monte Carlo simulation method. It is found that forward and backward scatterings are getting small in accordance with the size of bubbles. In this case, the side scattering increases with increasing of the size of bubbles.

View full abstract

ASTER sensor aboard Terra platform has a capability of spectral measurement in the thermal infrared (TIR) region with a spatial resolution adequate for geological applications. This paper proposes several indices for lithologic study using ASTER TIR multispectral data. Silica minerals represented by the commonest mineral on earth, quartz, have a unique spectral property in TIR. They have lower emissivity in ASTER band 10 and band 12 spectral regions than in band 11. All the common carbonate minerals on earth represented by calcite and dolomite have another unique spectral feature. They have low emissivity in ASTER band 14 spectral region, and high emissivity in ASTER band 10 to 13 spectral region. In silicate rocks, the wavelength at trough in TIR emissivity spectra moves to longer wavelength as the rock type changes from felsic to ultramafic. The emissivity in band 12 is lower than in band 13 for felsic rocks, and higher for ultramafic rocks. Using the spectral features described above, several indices are defined. They are applied to ASTER Level 1 B data in Beishan area of Gansu Province, China. The results are compared with the information from the field and the laboratory geological investigations including the TIR spectra of the rocks collected at the site and published geological maps. It indicates that the indices defined here are very useful for lithologic studies.

View full abstract

To confirm the accuracy of on-board calibration (OBC) of the thermal infrared bands of Terra ASTER and Landsat-7 ETM+, vicarious calibration (VC) and cross calibration (CC) methods were proposed and applied to field and airborne data. For VC at Salton Sea and Railroad Valley, US, ground temperature was measured in large areas with homogenious temperature and emissivity during overpasses of these satellites. Radiance-based calibration of ASTER TIR was conducted using a multispectral radiometer with the same spectral coverage as ASTER TIR. For CC at Lake Tahoe, airborne MASTER data were acquired when these satellites passed. We proposed a method in which the spectral difference between the satellite and the airborne sensor is considered. Atmospheric parameters were determined by MODTRAN4 from atmospheric profiles obtained with the radiosonde launched during experiments. Calibration results show that the radiant temperature of ASTER TIR (Level 1 B Ver. 0.4) is 0.2-1.3^°C lower than those of VC/CC at high temperature (15-40^°C). ETM+ (Level 1 R using calibration coefficients determined before December 20, 2000) estimates the radiant temperature 3^°C (except in Railroad Valley) higher than VC/CC. VC/CC accuracy was better than 1^°C (radiometric calibration requirement of ASTER TIR at 270-340K).

View full abstract

The road and the buildings collapsed by the Mt. Usu having erupted on 31 March 2000. Moreover, there was forest damage by the fallout of volcanic ash. The research purpose is to grasp of the current state of the forest around the Mt. Usu, and to make of the fallout distribution map and the forest damage classification map with satellite images combining forest GIS for practical use. These maps and information are opened to the public on the home page in Internet. Moreover, material was used for the fixation of the disaster restoration area on the site.

View full abstract