| Soil moisture is an important medium for water and energy exchange between the atmosphere and land surface,it plays an important role in controlling the hydrologic cycle such as precipitation,evapotranspiration and runoff,especially in arid and semi-arid areas.It is of great significance to obtain accurate soil moisture data at regional scale for ecosystem protection,sustainable use of land,and scientific management and effective utilization of water resources.At present,there are three main approaches to obtain regional soil moisture data:cosmic-ray neutron sensor（CRNS）method,remote sensing measurement（e.g.,Soil Moisture Active and Passive,SMAP）and land surface model（e.g.,Community Land Model,CLM）simulations.These methods have their own advantages and disadvantages,and the adaptability of these methods vary in different areas.The agro-pastoral ecotone of Northwest China（APENC）is a typical fragile ecological zone,facing ecological problems such as water scarcity,desertification and grassland degradation.In recent years,the APENC has been experiencing drastic land use changes resulting in complex land cover patterns.and heterogeneous distribution of soil moisture.In addition,soil moisture data are scarce and difficult to obtain.Therefore,it is essential to acquire accurate regional soil moisture data to understand the spatial distribution of the soil moisture in this region.In this study,soil moisture products from different data sources were compared and evaluated and data fusion was carried out by BP（back propagation）neural network algorithm,to provide a suitable method for regional soil moisture acquisition in the study area.The study area is the the agro-pastoral ecotone of Northwest China.The research sites include both selects the Ordos Observation Station in a grassland of Ordos,Inner Mongolia Autonomous Region and the Yanchi Observation Station in a farmland of Yanchi County,Ningxia Autonomous Region,First of all,based on the soil moisture data measured by ECH2O,soil moisture simulations by both CLM 4.5 and CLM 5.0 in the Yanchi Observation Station and Ordos Observation Station were compared and evaluated on three typical surfaces:grassland,rain-fed farmland,and irrigation farmland.The simulated soil moisture of Ordos from 2017 to 2019 was compared with the field-scale soil moisture measured by CRNS.Subsequently,the CLM simulated soil moisture data were compared with the SMAP soil moisture data and fused soil moisture data.The main conclusions are as follows:（1）CLM 4.5 and CLM 5.0 have better simulation performance on the 0 to 15 cm soil moisture in grassland and rain-red farmland,but its simulation performance decreases significantly with the 15-50 cm soil layers.Compared with CLM 4.5,CLM5.0 introduces a dry soil layer into soil evaporation,which reduces the evaporation simulation but increases the simulated soil moisture content.The dry soil layer introduced by CLM 5.0 is not suitable for the agro-pastoral ecotone of Northwest China,and needs to be improved.（2）The automatic irrigation trigger mechanism and calculation methods for irrigation depth and irrigation amount have been improved in CLM 5.0.In the irrigated farmland,CLM 5.0 is more reasonable than CLM 4.5 in simulating the irrigation amount,significantly improving the simulation performance from April to August.The improvement of the irrigation mechanism in CLM 5.0 has improved the soil moisture simulation in irrigated farmland,but the accuracy of the simulation needs to be further improved.（3）CLM 4.5 can well simulate the soil moisture trend in the region and has higher simulation accuracy than CLM 5.0 in the region.It also has large simulation error at low temperature and overestimates the soil moisture content during the freezing and thawing period.Unaffected by soil temperature,the SMAP soil moisture data are suitable in the study area.However,its data accuracy is worse than CLM 4.5,and the data are discontinuous with gaps.（4）The correlation coefficient,root mean square error of the soil moisture data set fused by the SMAP and CLM soil moisture data sets are 0.778,0.015 mm3·mm-3,indicating good trend fitting and high data accuracy,and no obvious overestimation and underestimation trend.Compared with CLM and SMAP data,the accuracy is significantly improved,which can represent the regional average soil moisture in the study area.When CLM and SMAP data are relatively complete,the soil moisture data set fused by the SMAP and CLM soil moisture data sets through the BP neural network algorithm shows improved accuracy,indicating it as an alternate method to obtain regional soil moisture data in the study area. |
