关键词: 时空多尺度, 带宽, 路网距离, GTWR, 住宅价格

Abstract: Scale, time, and spatial distance have always been the key to restricting the accuracy of geographically and temporally weighted regression (GTWR) models. Based on the Euclidean distance and road network distance constraints, this research extends the spatio-temporal multiscale to the modeling of geographically and temporally weighted regression methods to test the improved performance of the multiscale GTWR model and to verify the superiority of the road network distance constraint in the multiscale GTWR model. Herein, the commercial housing communities during the time period of 2015—2018 in the main urban area of Chengdu are considered as the case object, and the multiscale GTWR and GTWR are compared in terms of goodness of fit, residual sum of squares (RSS), and Akaike information criterion (AIC). Experimental results show that compared with GTWR, the multiscale GTWR provides a more effective explanation for the independent variables affecting housing prices, and improves the rationality of the model by using the road network distance.The goodness of fit based on the Euclidean distance and road network distance constraints has been improved by 0.123 and 0.208, respectively, and the RSS and AIC have been effectively reduced. Compared with the GTWR and multiscale GTWR models based on the Euclidean distance constraint, the goodness of fit of both GTWR and multiscale GTWR models based on the road network distance constraint has been improved by 0.007 and 0.092, respectively. Furthermore, the calculation results based on the road network distance confirm the correctness of the multiscale GTWR model and prove that the multiscale GTWR model exhibiting comprehensive consideration of scale and spatio-temporal distance has good versatility. Results obtained herein are expected to provide an important reference for urban planning in spatial-temporal multiscale modeling.

Key words: spatio-temporal multiscale, bandwidth, road network distance, GTWR, house prices