Obtaining global land-surface broadband emissivity from MODIS collection 5 spectral albedos using a dynamic learning neural network

Surface broadband emissivity (BBE) is a key parameter for estimating surface radiation budget, but it is treated crudely in land-surface models because of a lack of global-scale observational BBE data. In this study, the non-linear relationship between the BBE that is calculated from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) emissivity product and the seven Moderate Resolution Imaging Spectroradiometer (MODIS) narrowband albedos was established individually for bare soils, transition areas, and vegetated areas using a dynamic learning neural network (DLNN). The trained DLNN was tested using a vast array of independent samples, and the results are robust with a bias and root-mean square error (RMSE) of –1e−4 and 0.012 for bare soils, 2e−4 and 0.012 for transition areas, and 7e−4 and 0.010 for vegetated areas. Two independent field-measured emissivity data sets that were measured over sand dunes were used to validate the DLNN. With respect to the BBE that was calculated from the field-measured emissivities, the bias was 0.019. Ultimately, we introduced the strategy of generating a global land-surface BBE product and presented an example of a global BBE map.