| The spatial distribution pattern of land cover and its dynamic changes have an important effect on the ecological balance of the Loess Plateau.Understanding the biophysical drivers of land cover change is of great significance for ecological restoration in the context of climate change.Traditional methods for assessing the importance of driving forces of land cover change are mainly statistical methods,which focus on the impact of human activities and ignore the role of natural factors s in the land cover change.In particular,traditional statistical methods fail to quantitatively assess the importance of biophysical drivers of land cover change.Random forest method is a relatively mature machine learning method,which has the advantages of fast learning process,multi-factor modeling,and the ability to score the importance of input variables.In recent years,it has been widely used in ecology and environmental science and other fields.Therefore,using random forest method to quantitatively evaluate the importance of biophysical driving factors on land cover change has great practical significance for rational allocation of land resources in the Loess Plateau.In this study,land use type data,annual NDVI,temperature,precipitation,evapotranspiration data and slope data from 2001 to 2018 were used to develope a random forest model based on the importance assessment of driving forces of land cover change,with temperature,rainfall,evapotranspiration and slope as input variables and land cover type as output variables.This paper systematically analyzed the change trend and spatial distribution difference of land cover and natural factors on the Loess Plateau from 2001 to2018,analyzed the effects of several natural driving factors such as precipitation and temperature on land cover change,and grasped the relationship between natural driving factors and land cover change.The results of this study are of great importance to the restoration of vegetation,restoration of local ecological environment and adjustment of water resources redistribution on the Loess Plateau,which will contribute to the quantitative assessment of the importance of driving factors of land cover change.It will also provide scientific reference for the ecological restoration work of the government and relevant administrative departments.The main research conclusions are as follows:1.From 2001 to 2018,the NDVI of the Loess Plateau showed an overall upward trend,and the vegetation recovered and became green as a whole.The main vegetation types are grassland and deciduous forest,and the main crop types are cereal crops.2.Temperature,precipitation and evapotranspiration increased gradually in arid,semi-arid,subhumid and humid regions of the Loess Plateau.Compared with precipitation and evapotranspiration,temperature was relatively stable during the 18 years.Evapotranspiration and precipitation have poor stability.Most of the regions show an increasing trend,and the overall trend of warming and drying.3.Based on key biophysical variables(such as temperature,precipitation,evapotranspiration,and slope)and land cover types,we developed a random forest model.The model performed quite well,achieving an accuracy of over 0.8 on the test set and being able to assess the importance of each biophysical driving force.4.Evapotranspiration is the most important driving factor in the Loess Plateau and the four climatic regions in the past 18 years(IS > 0.2),the slope was the least important(IS <0.05).The importance of temperature and precipitation showed regional heterogeneity.5.Evapotranspiration is mainly affected by temperature and water content.In the process of ecological restoration,not only the local temperature,precipitation and topography should be considered,but also the evapotranspiration of land cover type should be included.Under the background of global warming,excessive evapotranspiration and insufficient precipitation are difficulty to maintain the regional water balance and sustainable development. |
PDF Full Text Request