Synergy of hyperspectral and microwave remote sensing data to derive bio- and geo-physical parameters for land applications

Retrieval of bio/geophysical variables from remote sensing imagery is an important research field. Due to the development and availability of different EO satellites, especially the ones with hyperspectral sensors, the combined use of radar and optical images is expected to provide the chance for better, more reliable and continuous retrieval of land parameters and to improve the accuracy of retrieval algorithms due to the combined use of complementary€ observation principles. Synergy of optical and radar data became a challenging task in Remote Sensing. This thesis addresses some of its main aspects. To this aim the thesis has several goals to reach: i) To develop and to evaluate possible strategies for integrating different data types, with the special focus on radar and optical data synergy, and to quantify the added value of their integration. For achieving this goal one of the specific objectives is to develop an algorithm which allows retrieval of land parameters by exclusively using optical, radar and one (or few) different strategies for combining optical and radar images; ii) To study challenges in land parameters retrieval in different environmental conditions. The focus is put on crop fields and on mountain grassland: iii) To provide a detailed description of the target observed from the remote sensor at radar frequencies and to adapt (or to develop) models to simulate the interactions of the radar signal with the area of interest. Based on direct models that predict radar response the retrieval algorithms need to be developed. These inverse methods should be able to use a physical description and the knowledge obtained with the direct models. This thesis focuses on parameter retrieval at the field scale. Since in microwave domain only SAR is able to monitor soil and vegetation proprieties and their variability down to the field scale, SAR data are used in all analyses presented. The special focus is put on soil moisture retrieval. Soil moisture is an Essential Climate Variable which retrieval is very complex and often an ill-posed problem. Estimation of vegetation height and leaf area index is also addressed. The work carried out in this thesis is oriented in developing innovative and improved estimation strategies that can overcome the limitations of the existing methodologies. In particular, the following main novelties are proposed in this dissertation: i) Estimation of soil moisture and vegetation height from hyperspectral imagery with machine learning method, SVR; ii) Soil moisture retrieval in crops by joining SAR discrete radiative transfer model and kernel algorithm. Tor Vergata RTM is used as a direct model to simulate the radar response and to train SVR. The method is evaluated on real radar images; iii) Soil moisture retrieval in Alpine grassland with synergy of optical and SAR images using GPR trained by model simulations. Vegetation parameters in the RTM are extracted from optical images. One temporal dataset is used to calibrate some model parameters and to train GPR, and another to test the algorithm performance; iv)Synergy of multi-sensor observations in soil moisture retrieval in mountain areas with GPR. The retrieval is based on the inference of an experimental model from measured reference samples. For all addressed topics in this thesis, both qualitative and quantitative analysis of results achieved on sensor acquired images verify the good performances and robustness of the proposed techniques.