Spatio-temporal model based on Markov random fields and its application to forest ecology

点击翻译按钮获取中文摘要

Markov random field models are fitted to 12 years data of canopy heights of neo-tropical forests of Barro Corollado islands, Panama. Canopy heights are binarized, under and over 20m. From various experiments taking account of neighbouring patterns of trees to potential functions, it is verified that Markov random field model can model binary canopy patterns fairly well using 1-bocly, 2-body and 4-body patterns simultaneously. Model parameters are first estimated by the maximum pseudo-likelihood method and they are refined by a MQMC Newton-Raphson refinement clue to Huang and Ogata. Among 12 years data, 8 data shows fairly good fittings and 2 years good fittings. 2 years data cannot show good fittings, probably due to temporal drastic changes of forest systems. But it is seen that temporal unstableness recovers quickly, showing the ecological stableness of the forest. As a result, it is verified that Markov random field model can represent such forest, system. The analysis 43 under the process of submitting to an appropriate journal. Also the head investigator published a first Japanese book on spatial statistics including Markov random field in 2001.