| Soil moisture affects climate change,crop growth,energy and water cycle.The study of soil moisture is of great significance in the fields of climate prediction,vegetation growth,ecological environment and so on.Using remote sensing technology to obtain soil moisture is a common data acquisition method,but the remote sensing soil moisture data contains some errors due to the inaccuracy of terrain,soil texture,vegetation cover,meteorological elements and other data used in the inversion process.Understanding the error information of remote sensing soil moisture data is very important for correctly interpreting the remote sensing observation information of soil moisture and successfully applying it to all kinds of research.In this paper,six kinds of soil moisture products of microwave remote sensing,including advanced microwave scanning radiometer 2(AMSR2),advanced scatterometer(ASCAT),global soil data assimilation system version 2.1(GLDAS2),meteorological reanalysis data set(ERA5),soil moisture active and passive monitoring satellite(SMAP)and soil moisture and ocean salinity satellite(SMOS),are selected to evaluate the data quality with reference to the observation data of International Soil Moisture Network(ISMN),The correlation coefficient,root mean square error and deviation were used as data quality evaluation indexes,and the data quality of soil moisture data in different land cover types was analyzed.It is found that the performance of six soil moisture products is different in different regions.ERA5 and GLDAS2 soil moisture products in the eastern United States have strong correlation with stations(correlation coefficient > 0.5),small root mean square error and small deviation(approaching 0 m~3/m~3);In northern England,ERA5 has strong correlation with stations(correlation coefficient is greater than 0.5),but the root mean square error is large;In the Qinghai Tibet Plateau of China and the southeast coast of Australia,six soil moisture products have high data quality,strong correlation with stations,and small deviation and root mean square error.In general,ERA5 has the highest correlation coefficient with stations in 43.5% of the experimental grid,GLDAS2 accounts for 25.2%,SMAP accounts for 14.3%,ASCAT and SMOS are close,accounting for 8.4% and 6.3% respectively,and AMSR2 has the worst performance(accounting for 2.3%).For the root mean square error and deviation study,the number of grids to be verified with the smallest deviation between GLDAS2 and the station accounts for 21.97% of the total number of grids to be verified,and the number of grids to be verified with the smallest root mean square error between GLDAS2 and the station accounts for 27.7%.In the root mean square error,AMSR2 performed the worst,with the optimal number accounting for 6.4%.The other four kinds were SMAP accounting for 21.4%,ERA5 accounting for 19.6%,SMOS accounting for 13% and ASCAT accounting for 11.9%.Among the deviations,SMAP performed the worst,with the optimal quantity accounting for 10.65%.There was little difference among the other four optimal quantities,which were SMOS accounting for17.58%,ERA5 accounting for 17.31%,ASCAT accounting for 16.64% and AMSR2 accounting for 15.85%.The results of correlation coefficient,root mean square error and deviation analysis for different MODIS IGBP land cover types show that AMSR2 has the smallest deviation from site soil moisture on grassland,SMAP has little difference in correlation coefficient on open shrub,tropical multi tree grassland,savanna,grassland,cultivated land and bare land,and GLDAS2 has the strongest correlation and the smallest root mean square error(0.05m~3/m~3 in evergreen forest To 0.07 m~3/m~3),However,the deviation from the station is large.The root mean error of the deciduous forest was the smallest in the deciduous forest.Data fusion is a research direction that has attracted much attention.In this paper,a remote sensing product data fusion model is established by using machine learning algorithm to fuse a variety of soil moisture products.The remote sensing soil moisture data is further accurate,and the average absolute error between soil moisture data and site soil moisture is reduced to 0.05 m~3/m~3。Data fusion is a research direction that has attracted much attention.This paper uses machine learning algorithm to fuse a variety of soil moisture products,establishes a data fusion model of remote sensing products,obtains more accurate soil moisture data,and reduces the average absolute error between soil moisture data and site soil moisture to 0.05 m~3/m~3。... |
PDF Full Text Request