Integrating BP and MGWR-SL Model to Estimate Village-Level Poor Population: An Experimental Study from Qianjiang, China

Abstract

Spatially-explicit, fine-scale mapping of poor population distribution at a village level is a necessary prerequisite for developing precise anti-poverty strategies in rural China. To address the data missing of poor population at a village scale, we proposed a modeling methodology from the perspective of spatial poverty, integrating BP and MGWR-SL (Mixed Geographically Weighted Regression model with Spatially Lagged dependent variable) that correspond to population estimation and poverty incidence estimation, respectively, to explore a more accurate and detailed village-level poor population distribution. Furthermore, we justified the accuracy, reliability, and scale effects of the model by using GIS spatial analysis and cross-validation. From the case test, we found that, the proposed model could characterize poor population distribution more accurately than other existing methods, resulting in that the errors of both population spatialization and poverty incidence for each village are less than 5% at a 500 * 500 m grid scale. It can also be inferred that the spatialization of socioeconomic data at a fine scale should take into full account of spatial heterogeneity and spatial autocorrelation for both dependent and independent variables, so as to improve the modeling accuracy. This study may provide a perspective for better understanding the detailed and accurate poverty status of data–scarce village in poverty-stricken rural areas, and serves as a scientific reference regarding decision-making in both promoting “entire-village advancement” anti-poverty harmonious development and constructing the new countryside of China.