多層多類別之人口地理分布模式

Abstract

人口資料為城市、經濟、商業、交通、社會、政治、歷史、生態等方面研究基礎，其扮演影響結果的重要角色，然而以往的人口分布圖多以行政區域為單位，只呈現在特定區域內的人口密度或是人口數，這樣的人口分布圖往往有分區單元不夠細緻、研究分區不同、行政單元於時間上的重組與邊界變遷等問題，進而造成研究結果的誤差。為了改善人口分布圖的缺失，現今已有許多人口分布的推估方式，然而這些推估方式各有優缺點，以往研究大多使用單一推估方式，近來有研究提出多層式架構概念，將各種推估方式作一整合，達到截長補短之效果。 為了驗證推估結果必須要以實際值做對照，但由於隱私之考量及行政作業上之限制，人口實際位置資料之取得有實務上的困難，因此本研究以臺北市門牌資料作為實際值，將點資料加總後之門牌總數，以本研究所提之多層多類別架構進行門牌數之空間分配，研究中以邊長40m的正方形網格為單位，利用建物、土地使用分區、交通路網等圖層，配合二元、權重、可及性等方式建立一多層多類別的推估模式，每一層均使用一種人口分布推估方式，而層內更利用輔助圖層將每一分區單元劃分成不同類別，給予不同權重加以逐層推估，期望以此一模式推估門牌之空間分布，以建立人口推估之架構，改善以往人口分布圖的不足之處。 由本研究結果可以看出，相較於傳統人口密度分布圖，本研究所建立之多層多類別人口地理推估模式能較準確表現人口於空間上的的變異，推估誤差的標準偏差由初始的10.39降至9.29(第一層)、8.71(第二層)、8.71(第三層)，相較於行政分區所做出的人口分布密度誤差標準偏差值(區為9.91，里為8.87)亦有較好的結果，而在逐層推估過程中，完全推估正確的網格數逐層增加，而有誤差的網格之誤差絕對值平均也由第一層的10.42、第二層8.65到第三層8.57逐層改善。 研究中也發現以多層多類別人口地理分布模式逐層推估的過程中，第一層所改善的情況為最佳，之後逐層推估雖均有改善的情況，然而進步的幅度卻相對降低，每一層推估結果均有靠近山區範圍高估而都市中心低估的情況，此為區域居住型態不同所造成之影響，但整體而言仍能大致掌握了人口分布的特性。

The population distribution plays a crucial role in many research fields. However, the map of the population distribution is usually based on administrative divisions, and only shows the population density or the number of population in the region. Traditional maps of population distribution have problems like inadequate area size that is too large, the inapplicability of the administrative division unit in many applications and the change of the administrative boundaries over time. Many estimation methods of the population distribution have been proposed in literatures to improve the defects of the population distribution maps. However these proposed methods have its advantages and disadvantages. Most researches use a single estimate method. Some recent researches proposed the concept of the multi-layered estimation which integrates various estimation methods in the framework. A multi-layer and multi-class framework is proposed in this study to improve the accuracy in estimating the population distribution. Grid data structure with 40m of resolution was used. Each layer (building, land-use, and traffic accessibility layers) uses individual estimate methods (binary, multi-class, and accessibility) to estimate populations in each cell for better capture of spatial distribution of regional populations. The proposed framework can better capture the true population distribution in comparison with the traditional population density maps based on administrative divisions. The standard deviation of estimation errors decreases from 10.39 to 9.29 in the first layer, to 8.71 in the second layer, and 8.71 in the third layer. The results are also better than decomposing the total population based on administration units, the standard deviation of errors is 9.91(town) and 8.87(village) respectively. The numbers of cell without errors also increases and average errors in the erroneous cells decrease with improvements through layers. The mean error is 10.42 in the first layer, 8.65 in the second layer and 8.57 in the third one. It is also found that the improvement is most significant in the first layer and diminished in the following ones.