Fusing Near-Infrared Reflectance Spectrum and Dynamic Model to Estimate Vegetation Structural Parameters

Many efforts have been made to fuse reflectance spectrum signals and dynamic model to estimate land surface parameters from remotely sensed data more accurately as well as capture the continuous features of land surface parameters. This paper proposed a dynamic Bayesian networks (DBNs) data fusion algorithm to retrieve vegetation structural parameters by integrating multi-source information, such as time series observations of multi-angle spectral reflectance in near-infrared region, canopy radiative transfer model as well as the crop dynamic growth model. Using this method, leaf area index (LAI) of winter wheat was estimated using field measurement dataset. Estimated results show that the correlation coefficients between estimated and reference value is 0.965 and root-mean square error (RMSE) is 0.344. Since DBNs data fusion is a dynamic iteration algorithm, the results also confirmed the rule that the parameter state estimated from DBNs are the implicit information of observation data updated with crop growth model.