建石 隆太郎

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.<br>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.

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.

The estimation of tree cover area at continental scale is becoming more important than before due to the needs to improve our understanding of carbon dynamics. The estimation of percent tree cover of a large area using MODIS data by regression tree method is a promising method. New points of this study are the use of QuickBird images for the collection of training data and the use of the Stepwise Linear Regression (SLR) for selecting the best subset of predictor variables. The estimation of percent tree cover of African continent was tried using 11 QuickBird images to get 195 cells as training data and 32-day composite MODIS 2003 data as predictor variables. The predictor variables consist of surface reflectance, normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), normalized difference soil index (NDSI) and land surface temperature (LST). The result shows that NDVI and surface reflectance bands are effective to estimate percent tree cover and this method is acceptable with the prediction error of 5.17%.

本研究では, 粗雑な地表面である森林火災跡とマイクロ波 (Lバンド) との関係を理論的に調査した。まず, 解析モデルは3つの媒質である自由空間, 森林火災跡, 泥炭地から構成される。ここでは, レーダの後方散乱係数と森林火災跡の厚みの関係を得るために, vv偏波とhh偏波の散乱電磁界を導出した。この解析法の計算の流れとして, まず自由空間と森林火災跡間の境界における散乱界と透過界を導出した。そして, この透過界を使用して, 森林火災跡と泥炭地間の境界における散乱界を導出した。この計算を繰り返し行い, 最後に自由空間における散乱界の総合電界強度を使って後方散乱係数を導出した。本研究の結果を検証するために, Transmission-line method (TLM) による前解析法の結果との比較検討を行った。この結果として, 地表面の粗雑度の影響によって両解析法の結果の間に誤差が生じることを確認した。また, その他の理由として, 本研究で提案した理論解析モデルには, 解析を簡単にするために, 燃焼した樹幹による散乱を無視したことが考えられる。本研究の結果はJERS-1 SAR画像を用いて, 1997年に中部ボルネオ島に生じた森林火災地域を対象として, 森林火災跡の厚みの推定に応用した。その結果, 本研究による推定結果は現地調査データの範囲 (0～0.5m) に入ったことを確認した。

本研究は, 土地被覆変化および土地劣化 (植生減少) のモニタリングを目的とした衛星データの処理手法とそのヨルダンにおける適用の研究である。1987年8月と2001年8月のLandsatTMとETM+データに対して幾何補正と放射量補正を行った。視覚判読のためのチャンネル2, 3, 4とチャンネル2, 4, 7によるカラー合成, 土地被覆変化検出のためのTMとETM+の (温度データを除いた) 6チャネルのデータによる教師付き分類, 植生減少検出のためのNDVI画像比較法 (image differencing) を行い, その処理手法の実用性を調べた。テスト地域におけるモニタリングの結果, 1987年から2001年の間に中央ヨルダン地域に大きな土地被覆変化が生じ, 農業地帯から都市地域に変化したことが確認された。