Spatiotemporal Nonstationary Robust Modeling Between Luojia1-01 Night-Time Light Imagery and Urban Community Average Residence Price

Abstract

This article is the first to propose a novel spatiotemporal nonstationary robust modeling between high spatial resolution Luojia1-01 night-time light intensity (NTLI) and urban community average residence price (UCARP), which encodes the spatiotemporal independent variable NTLI based on a new proposed geographical coding (GeoCode) to enhance the explanatory power of NTLI and leverages geographically and temporally weighted regression (GTWR) based on a new proposed spatiotemporal anomaly detection (STAD) to remove spatiotemporal outliers and then to robustly estimate modeling result. UCARP data and Luojia1-01 NTL imagery obtained from Wuhan, China, in June, September and October 2018 were crawled and downloaded for the experiment, whose results show that GTWR performs better than geographically weighted regression and temporally weighted regression. The comparisons of GTWR with 1) original data; 2) GeoCode (GC); 3) STAD; 4) first STAD last GeoCode (STAD-GC), and 5) first GeoCode last STAD (GC-STAD) show that 1) the q values of geographical detector corresponding to the above methods are 0.055, 0.407, 0.126, 0.666, and 0.671, respectively, during September; 2) the adjusted R2 values of GTWR are 0.460, 0.488, 0.683, 0.693, and 0.697, respectively; and 3) the proposed spatiotemporal data processing scheme, i.e., GC-STAD, has the most robust and best precision. This article not only proposes a new spatiotemporal nonstationary robust modeling method between small-scale NTL and UCARP but also reveals its underlying mechanism.