Publications

Wang, CM; Xie, QX; Gu, XF; Yu, T; Meng, QY; Zhou, X; Han, LR; Zhan, YL (2020). Soil moisture estimation using Bayesian Maximum Entropy algorithm from FY3-B, MODIS and ASTER GDEM remote-sensing data in a maize region of HeBei province, China. INTERNATIONAL JOURNAL OF REMOTE SENSING, 41(18), 7018-7041.

Abstract
The Bayesian Maximum Entropy (BME) algorithm that can model larger-scale spatial heterogeneity and integrate multiple types of data is a better spatial estimation algorithm in avoiding the circular problem and improving the estimation accuracy of parameters than original empirical geostatistic and spatial statistic methods. To obtain higher-resolution and higher-accuracy soil moisture (SM) data, this study used the BME algorithm to integrate the multiple environmental factor variables related to SM as auxiliary data (11 data sets in total) such as the Normalized Difference Vegetation Index (NDVI), Land Surface Temperature (LST), and Albedo products with 1 kilometre (km) grid resolution from the Moderate-resolution Imaging Spectroradiometer (MODIS), the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model (ASTER GDEM) satellite data, and the Slope, Aspect, Plan curvature, Profile curvature, Surface roughness, Wetness index, Relief amplitude products generated from ASTER GDEM data. First, the FengYun 3-B satellite (FY3-B) SM product with 25 km grid resolution was downscaled to 1 km grid resolution using the visible, infrared (IR), and microwave fusion method. The downscaled FY3-B SM data and auxiliary data were used to generate the weighted probability soft data (SD) under four cases, case 1: 500 sample points of SD, case 2: 450 sample points of SD, case 3: 400 sample points of SD, and case 4: 350 sample points of SD. Particularly, we used two methods, the multivariable correlation analysis method and principal component analysis (PCA) method to get the weight values of environmental factor variables, i.e. NDVI, LST, Albedo, the Digital Elevation Model (DEM), Slope, Aspect, Plan curvature, Profile curvature, Surface roughness, Wetness index, and Relief amplitude with SM. Then, thein-situSM measurements as hard data (HD) were used to calibrate the weighted probability SD in the procedure of BME SM estimation. The SM under four cases were estimated from the weighted probability SD and HD using BME algorithm. Finally, comparisons of the estimated SM using BME algorithm within-situSM measurements at maize study area were carried out in this study. Our results indicated that the accuracy of the estimated SM using BME algorithm, the root-mean-square error (RMSE) = 0.049 cm(3) cm(-3), the correlation coefficient (r) = 0.639, the unbiased RMSE (RMSEu) = 0.047 cm(3) cm(-3), and Bias = 0.002 cm(3) cm(-3), under case 1 based on PCA method was obviously better than the downscaled FY3-B SM (RMSE = 0.079 cm(3) cm(-3),r= 0.096, RMSEu = 0.069 cm(3)cm(-3), and Bias = 0.039 cm(3)cm(-3)) that was generated by the visible, IR, and microwave fusion method. We concluded that integrating auxiliary data into SM estimation using BME algorithm could further improve the downscaled FY3-B SM that was generated by visible, IR, and microwave fusion method.

DOI:
10.1080/01431161.2020.1752953

ISSN:
0143-1161