Fractional Vegetation Cover Estimation Method Through Dynamic Bayesian Network Combining Radiative Transfer Model and Crop Growth Model

Abstract

Fractional vegetation cover (FVC) is an important parameter for describing the conditions of land surface vegetation and is widely used for Earth surface process simulations and global change studies. Regional FVC is primarily derived from remotely sensed data. However, current FVC estimation methods are mainly employed on remotely sensed data at a single time point, which can only reflect the instantaneous physical state of the land surface and ignore the important information from the vegetation growing characteristics. The vegetation growing characteristics have great potential to capture the temporal variations of FVC and, thus, can provide complementary information to improve the FVC estimation accuracy. In this paper, a dynamic Bayesian network method was proposed to estimate FVC from Moderate Resolution Imaging Spectroradiometer (MODIS) reflectance data through combining a radiative transfer model and a statistical crop growth model, which could synthetically use information from both remote sensing data and crop growing characteristics. The performance of the proposed method was investigated in a cropland area of the Heihe River Basin, covering the whole growing season of maize in 2012. The time series field FVC data were quantitatively measured using digital photography and then used to generate high-spatial-resolution FVC maps using the Advanced Spaceborne Thermal Emission and Reflection Radiometer and Compact Airborne Imaging Spectrometer data for evaluating the accuracy of FVC estimates from MODIS data. The validation results showed a satisfactory performance with a coefficient of determination R2 of 0.956 and a root-mean-square error (RMSE) of 0.057, as compared with the performance (R2 = 0.817, RMSE = 0.106) of the FVC estimates using the lookup table method, which utilized the information from remote sensing data. These results indicated that the proposed method could effectively utilize the vegetation growth information and achieve reliable FVC estimates in the cropland area.