日本地球惑星科学連合2014年大会

講演情報

インターナショナルセッション(口頭発表)

セッション記号 H (地球人間圏科学) » H-TT 計測技術・研究手法

[H-TT07_28AM1] GIS

2014年4月28日(月) 09:00 〜 10:45 422 (4F)

コンビーナ:*小口 高(東京大学空間情報科学研究センター)、村山 祐司(筑波大学大学院生命環境科学研究科地球環境科学専攻)、柴崎 亮介(東京大学空間情報科学研究センター)、吉川 眞(大阪工業大学工学部)、座長:村山 祐司(筑波大学大学院生命環境科学研究科地球環境科学専攻)

10:15 〜 10:30

[HTT07-06] ENSEMBLE-CELLULAR AUTOMATA (CA) MODELS FOR IMPROVING FOREST COVER CHANGE SIMULATION

*KAMUSOKO Courage1 (1.Asia Air Survey Co., Ltd)

キーワード:Adaboost, Random forests, Cellular automata, Transition potential, Forest cover changes

Reliable spatial simulation models are a prequisite for understanding temporal and spatial forest cover changes. However, spatial simulation models require accurate transition potential maps, which represent the probability of change from one forest cover class to another. Previous studies have shown that conventional methods such as logistic regression, weights-of-evidence and neural networks fail to adequately model forest cover transition potential. The objectives of this study are to: (1) evaluate the performance of adaboost (AB) and random forests (RF) algorithms for computing transition potential maps, and (2) simulate forest cover changes using the computed transition potential maps and cellular automata (CA) model. Our results show that adaboost-CA and random forest-CA models produced better simulation accuracy than logistic regression/ weights of evidence-CA models. These results provide valuable insights, which can be used to improve transition potential modeling and forest cover change simulation in complex landscapes.