Classification of tropical seasonal forests using time-series NDVI data with the tree model classification approach.

DOI Web Site Web Site 1 Citations 28 References

Bibliographic Information

Other Title
  • 時系列NDVIデータを用いた樹形モデル分類による熱帯季節林の分類
  • ジケイレツ NDVI データ オ モチイタ ジュケイ モデル ブンルイ ニ ヨル ネッタイ キセツリン ノ ブンルイ

Search this article

Abstract

This study aims to classify tropical seasonal forest using time-series NDVI data with the tree model classification approach. The whole of Cambodia was selected as the study area. Target classification categories were set as: Evergreen forest, Mixed Deciduous forest, Deciduous forest, Agricultural field etc. Two different existing maps and 36 scenes of NDVI of SPOT VEGETATION acquired between January and December 2000 were used in this study. Noise reduction processes involving local maximum filtering on a time-series domain and image reconstruction using lower-order principal components of the results of principal component analysis were applied to the time-series NDVI data. Seasonal change metrics were calculated using the noise reduced NDVI data. Datasets which contained two land cover information and several seasonal metrics were systematically sampled using a 0.04-degree interval geographic grid from the above-mentioned data. These datasets were separated for training data and test data. Two tree models for correspondence with the existing maps using the datasets were constructed using the training datasets. Land cover information was used as the response factor and seasonal change metrics were used as predictive variables in the tree model. Land cover classification was conducted using the tree models, and the accuracy of classification was evaluated using the test datasets. As a result, the overall accuracies of the two classifications were similar at about 70%. The variables used in both tree models were the annual mean of NDVI and growing period. To depict the growing period, one image was created as a composite of 36 images. The pixel value used in the composite image was the total value of all pixels in the 36 images that exceeded the threshold value of 0.7. It was indicated that tropical seasonal forests can be classified using several seasonal metrics derived from SPOT VEGETATION NDVI data with the tree model classification approach.

Journal

Citations (1)*help

See more

References(28)*help

See more

Details 詳細情報について

Report a problem

Back to top