由公共自行車延伸的可及範圍分析：基於全球定位系統資料在臺北都會區的實例分析

Abstract

公共自行車作為一種被廣泛採用的「最後一哩路」運輸服務，在全球各地產生大量關於使用型態和使用者行為的資料，使研究者得以一窺公共自行車如何拓展人們的移動範圍。相較於文獻關注公共自行車站點覆蓋與近用性（access to）的討論，本研究提出基於騎乘公共自行車移動（access by）的可及範圍分析，以理解公共自行車系統如何拓展主動運輸的可及範圍，進而討論影響可及範圍空間變異的因素。並試圖提升既有可及範圍偵測方法的精確度與解析度。 在方法上，本研究使用全球定位系統（Global Positioning System, GPS）旅次軌跡資料偵測可及範圍，確保可及範圍反映真實旅運行為與使用者的路徑偏好與旅次目的。本研究以臺北都會區的YouBike系統作為案例分析，證實本研究的可及方法偵測結果具有最低的不合理值 (0.613%) ，並在運量預測等實務應用上，具有最高的模型適配度，即最高的對數似然比 (-10539) 與最低的赤池及貝式資訊量準則 (21135, 21289) 。 基於偵測結果，本研究關注兩種可及範圍的空間差異：面積與均向性（Isotropy），以分別討論可及性在距離與方向上的空間差異。本研究使用SARAR （spatial autoregressive model with autoregressive disturbances）空間迴歸模式，處理都會區內因密集站點間重疊的可及範圍所導致的空間自相關特性。空間迴歸模式納入可及範圍的地形、運輸服務近用性、社會經濟與建成環境特徵作為自變數，結果指出對面積與均向性等依變數，可及範圍內的經濟福祉、商業與教育機會對於推展廣闊而均勻的可及性具有正面影響。與此同時，運輸服務近用性、就業機會等都會服務會導引可及範圍至特定的空間聚集區。公園與水域面積具有延展特定方向之可及性的影響，與休憩旅次的特徵相符。 本研究基於使用者行為，提出公共自行車的可及性範圍偵測方法，其顯示與文獻不同之市區-郊區差異：在市區廣闊而在不同方向間均勻分布，但在郊區緊湊而偏向特定方向。據此，本研究提出基於土地使用規劃與重力理論的解釋：郊區的都市機能與設施相對市區集中，但缺乏高品質的誘因以吸引來自不同方位的遠程旅次，使集中的目的地在空間上無法被平衡，進而促成可及範圍在市區與郊區間的差異。 本研究基於使用者行為，提出精細的可及範圍偵測方法，進而為空間規劃與公共自行車營運部門帶來貢獻。本研究在郊區觀察到較緊湊的可及範圍，顯示在方法上，過去文獻基於旅行距離的偵測方法，缺乏方向性地理脈絡的考量，故納入使用者不願前往，進而將真實世界中不可及的地方納入可及範圍。其成果在空間中存在偏差，從而使其規劃與政策建議產生偏誤。對空間規劃部門而言，本研究呈現土地使用對可及範圍有其影響，並篩選出可推廣的空間特性與設施，以推展廣闊且均勻的公共自行車可及性。在公共自行車營運部門，可及範圍可以呈現潮汐現象的空間範圍，並呈現區域型調度需求的空間尺度。最後，較緊湊的郊區可及性偵測結果顯示，在資源較受限的邊陲地帶，公共自行車營運者在分配站點資源時，應著眼於較小尺度的空間互動與使用者行為為宜。

As widely adopted last-mile transit services, bike-sharing systems generate extensive data on usage patterns and user behavior worldwide, shedding light on extended mobility through real-world usage. To understand how bike-sharing extends active mobility, a reachability analysis based on accessibility by using shared bikes is proposed, shifting the existing research focus from accessibility to bike-sharing stations for the neighboring population. This study aims to identify areas reachable by bike-sharing stations and discuss factors contributing to spatial variation in reachability. The reachable area is detected using Global Positioning System (GPS) trajectory data, which represent real-world trip activities and user preferences. In YouBike system of Metropolitan Taipei, Taiwan, the proposed detection method is proven as superior to existing methods by two validation methods: (1) achieving the least unreasonableness (0.613%) as the percentage of reachable area covering known inaccessible area, such as forests and rivers, and (2) yielding best fitting results in trip volume, with the highest log likelihood ratio (-10526), the lowest AIC (21115), and the lowest BIC (21279). Based on the detected reachable area, two forms of spatial variation are discussed: area and isotropy, representing size and evenness of reachability across directions. Bike-sharing stations in dense urban areas yield overlapping reachable areas thus, this research applies the spatial autoregressive model with autoregressive disturbances (SARAR) to deal with spatial autocorrelations. Influences from topography, transit accessibility, sociodemographic characteristics, and built environment are considered. SARAR estimations of area and isotropy show that economic well-being, commercial land-use, and educational opportunities in neighborhoods promote broader, more evenly distributed reachability, while urban utilities, including transit accessibility and job opportunities, funnel reachability to hotspots and urban nuclei. Reachability expands in directions adjacent to the waterfront and parks, corresponding to recreational behaviors. The proposed behavior-based reachability detection reveals an urban-suburban divide different from the literature: reachability in urban cores is extensive and highly isotropic (round and even), whereas it is limited and spatially skewed in the periphery. To explain this, this research proposes a gravitation-based zoning mechanism. Although suburban amenities are quantitatively concentrated within compact outlying nuclei, they lack the qualitative draw necessary to attract long-distance, multidirectional trips. Consequently, they fail to offset the spatial agglomeration of primary destinations, driving the observed disparity in bike-sharing reachability between urban and suburban contexts. This research assists both urban planners and bike-sharing operators by offering a detailed, behavior-based reachability detection method. The observed shrinkage of detected reachability in the suburbs indicates that traditional distance-based methods lack directional context. This oversight can lead to biased planning implications based on places unpreferable, and thus unreachable by cyclists, in real life. For planners, the results demonstrate that zoning affects reachability, suggesting that specific spatial features can foster more extensive and evenly distributed reachability. For operators, reachable areas indicate the potential scale of tidal effects, providing insights for optimizing the appropriate spatial scope for regional bike relocations. Finally, the compact nature of suburban reachability suggests that, under resource constraints, operators should manage resource allocation by accounting for spatial influences at a more compact scale in peripheral areas.