結合遙測及共享單車時空數據於都市土地使用分類及模擬

共享單車的時空資料記錄了豐富的人文活動資訊，得以反映不同的土地使用類別在時空上相應的交通流動模式，由此確立了土地使用和共享單車使用模式間相互印證的可行性。土地使用分類一直是都市規劃和土地利用管理中的重要問題，能夠準確地進行都市土地分類可以幫助決策者更好地制定都市發展計劃和政策。近年來隨著大數據普遍容易取得和人工智慧的快速發展，許多研究利用機器學習或是深度學習結合衛星影像，進行大面積土地使用分類。雖然衛星影像在區分光譜特性差異明顯的建物、水體、植被下有相當好的表現，但若以使用分區對建物做細分，只靠衛星影像可能無法提供足夠的辨識資訊，而社會感知資料可反映人類的活動模式，藉以補足衛星影像的不足。因此，本研究以共享單車騎乘數據為社會感知資料，結合遙測影像，利用隨機森林以像素單位之進行都市土地使用分類推估，結果以資料內容而言，考量遙測資料結合社會感知資料的分類表現最佳，整體精度為0.88。本研究也發展都市發展時空模擬模型，藉由隨機森林得到商業區和住宅區的適合度，模擬未來5年後和10年後商業區和住宅區的發展位置。本研究印證了共享單車站點租借和歸還人次於時空分布之特性和都市土地分類的空間分布有密切關係，都市土地使用分類可說是依據人文活動來劃分使用分區，利用共享單車騎乘資料的時間與空間特性進行都市土地使用分類預測，可以為都市規劃和土地管理提供客觀真實之案例應用。

Shared bicycle data provides valuable information about human activities, reflecting the corresponding transportation flow patterns in different land use categories over time and space. This establishes the feasibility of verifying the interrelation between land use and shared bicycle usage patterns. Land use classification has been a significant issue in urban planning and land management. Accurate urban land classification can help decision-makers better formulate urban development plans and policies. Recently, with the widespread availability of big data and the rapid development of artificial intelligence, many studies have used machine learning or deep learning in conjunction with satellite imagery for large-scale land use classification. Although satellite imagery performs well in distinguishing between different spectral characteristics of buildings, water bodies, and vegetation, it may not provide sufficient recognition information for finer land use zoning. Social sensing data can reflect human activity patterns, thus complementing the shortcomings of satellite imagery. This study uses shared bicycle riding data as social sensing data, combined with remote sensing imagery to employ random forests for urban land use classification at the pixel level. According to the results, the combination of both data yields the best performance, with an overall accuracy of 0.88. This study also develops a spatiotemporal simulation model for urban development to simulate the development locations of commercial and residential areas in t he future. The study confirms that the spatial distribution characteristics of shared bicycle station rentals and returns are closely related to urban land classification, indicating that urban land use classification can be based on human activities to delineate usage zones. Using the temporal and spatial characteristics of bicycle riding data for urban land use classification prediction can provide practical case applications for urban planning and land management.