GIS- and Machine Learning—Based Modeling of the Potential Distribution of Broadleaved Deciduous Forest in the Chinese Loess Plateau

Abstract

The Chinese Loess Plateau has experienced drastic human-induced ecological degradation in the last century. The Chinese government has initiated a longterm project to restore the ecology and thus the environment. This effort requires information about the ecological potential, especially about the potential distribution of forests. This study first presented a complete procedure for spatially modeling the potential distribution of broadleaved deciduous forests (BDF) in the western part of the Chinese Loess Plateau. The procedure efficiently integrated the spatially distributed environmental variables using a machine learning approach (i.e., GARP) that extracted rules and patterns from the massive datasets. The GARP-produced composite map revealed that BDF can potentially be distributed continuously in almost all of the areas where annual precipitation is above 450 mm and BDF has discontinuous or sporadic distribution in areas where annual precipitation is below 450 mm. The discontinuously distributed BDF occurs in two different physiographic settings. The first setting is the higher-elevation mountains within the area between 450 and 400 mm isohyets, where effective soil moisture is considerably higher than low-elevation valleys due to reduced evaporation and increased precipitation. The second setting is the north-facing slopes of rocky terrain within the area north to the 450 mm isohyet, where evaporation is significantly reduced. We believe that the modeled result can effectively facilitate forestation planning for this area. In addition, the results may function as a base against which to estimate the historic changes in land cover and land use, to assess the ecological potential for carbon sequestration, and to evaluate the climatic significance of land-air interactions.