Retrieving spectral and biophysical parameters of land vegetation by the Earth Observation Land Data Assimilation System

In this thesis, a new methodology for retrieval of land spectral and biophysical parameters from optical remote sensing data has been designed and used. The result of the work was a physically based methodology for Fraction of Photosynthetically Active Radiation (FAPAR) and Leaf Area Index (LAI) retrievals, simulation of hyper-spectral information and estimation of associated uncertainties. The presented methodology is based on the generic Earth Observation-Land Data Assimilation System (EO-LDAS). In the course of the work it was found that EO-LDAS can be used for daily estimation of FAPAR and associated uncertainties without any in-situ information and when the number of available observations is low. The results were in line with the field measurements with r2 varying from 0.84 to 0.92 and Root Mean Square Error (RMSE) from 0.11 to 0.16. This was the highest rate among compared products (Two Stream Inversion Package - JRC-TIP, Medium Resolution Imaging Spectrometer - MERIS FR and Moderate Resolution Imaging Spectro-radiometer - MODIS MCD15). It was shown, that using MISR information, EO-LDAS temporal regularization and generic dynamic prior, it was possible to stabilize results of the retrieval and to obtain better results than MERIS FAPAR or JRC-TIP MISR. In addition, inclusion of generic static and dynamic prior information, decreases posterior uncertainties and can increase accuracies compared to in-situ data. The results showed that proper estimation of LAI and soil parameters were sufficient to simulate a hyper-spectral signal between 400 and 1000 nm with acceptable precision: best RMSE is equal to 0.03 for real data and less than 0.008 for synthetic data. This implies that in case of the given experimental set-up, LAI and soil parameters are the major mechanisms controlling spectral variations in the visible and near infrared regions.