Ensemble intelligence algorithms and soil environmental quality to model economic quantity of land resource allocation and spatial inequality

With the increasing concern on soil pollution in context of land market reform, it’s an emerging topic to discuss whether soil pollution can cause land economic value substantially depreciated at geospatial scale. This study proposed a data science approach by synthesizing machine learning and deep learning algorithms to establish the land economic equivalent model (LEEM) with best prediction ability. The results indicated that Light Gradient Boosting Machine (LightGBM) and Random Forest (RF) performed well in estimating residential/commercial and industrial land value, respectively. Spatial variation of soil environment quality indeed largely affected economic productivity of residential/commercial land with contribution of 13.82%. For industrial land, population density was strongly related to the land price (11.77%), of which importance score was much higher than other 10 variables except distance to center business district (CBD, 28.75%) and transaction time (28.61%). Another important finding was that soil protection measures could generate extra benefits equivalent with a mean of 22.46% and 28.92% for residential/commercial and industrial land value in current status, respectively, and high-value areas could also be predicted for effective decisions on land allocation and trading. In terms of sustainable land management, the regression tree-based LEEM is an effective tool in assisting decision makers with urban land reclamation, planning and pricing.