| 摘要: | 都市的發展影響了環境並改變了地景，都市擴張是影響土地利用變遷的主要因素之一，其是由人口、交通路網、社會經濟、政府政策等多方面因素交互作用的結果。為了了解都市擴張的過程，使用土地利用變遷模型可分析土地變遷的原因與結果，並預測未來土地的發展。馬可夫細胞自動機(CA-Markov)模型廣泛應用在都市成長與土地利用變遷上，其考量土地轉移機率以評估各種土地利用面積變化量、利用影響因子評估土地潛在圖和使用土地的規則鄰域單元空間狀況分析土地變遷情形，並預測未來的土地發展。然而土地常以坵塊的方式出現，且土地利用變遷受影響土地發展的因子影響存在著空間的異質性，故本研究為解決此問題，提出以CA-Markov模型為概念的土地覆蓋模型─以坵塊為基礎的細胞自動機模型(Patch-based CA-Markov model, PBCA-Markov model)，其考量土地以坵塊基礎的鄰域結構狀況，並利用機器學習法評估土地與影響土地因子的空間複雜關係。
近年來，桃園地區人口成長快速，且在桃園航空城計畫的發展下，使得未來的都市發展狀況值得關注。故本研究選擇桃園地區為研究區域，利用衛星影像以支持向量機(Support Vector Machine, SVM)的方法進行影像分類，以建立多時期的土地覆蓋圖資和監測土地變遷的概況。透過模型模擬與預測桃園的土地發展，並比較本研究提出的模型PBCA-Markov與CA-Markov模型模擬結果的差異。研究結果顯示，PBCA-Markov模型較CA-Markov模型有更符合真實情況的土地預測結果，且PBCA-Markov模型能預測都市土地坵塊的分布，而CA-Markov模型主要預測都市擴張的現象，故本研究提出的模型PBCA-Markov能應用在土地覆蓋的預測上，並提供未來土地規畫之參考。
;Urban growth can be caused by the concentration of population, the development of transportation network and government policies. The urban development is one of the major forces to drive land use and land cover change (LUCC), and to funtionalize this change, LUCC models are tools to analyze and predict the consequences of land use change. The Cellular Automata Markov (CA-Markov) model is the most commonly used one, which simulates the land cover change with a set of linearly combined driving factors and cell-based expansion process. However, land cover with patches is commonly observed, and the driving forces can be spatially heterogeneous. This study aims to develop a new predictive model (Patch-based CA-Markov model, PBCA-Markov model), by integrating patch-based transition rules of land cover change and a machine-learning algorithm to assess the non-linear land development processes. The Taoyuan City, the sixth major city of Taiwan, is selected as the study area, because the city has the highest population growth rate in Taiwan since 1984, and also has experienced a significant increase of urban land in recent years. Therefore, a rapid development of the city could reshape the landscape in near future.
This study analyzed the land cover changes of Taoyuan City from 1984 to 2014, with using Landsat TM and OLI data, and different land cover types were classified and mapped through support vector machine classifier. The land cover results are used for model calibration and validation, and results show the PBCA-Markov generates higher overall accuracies than the CA-Markov. This study focuses on the prediction of urban areas and found that some urban land patches can be only successfully predicted by PBCA-Markov, while CA-Markov mainly predicts expansion areas along the urban fringe. This study suggests the proposed PBCA-Markov can be useful for the region urban planning and decision making practices. |
