Optical remote sensing provides invaluable data for monitoring the Earth and its vegetation. However, cloud cover hinders the acquisition of images and leads to data gaps. While radar remote sensing can penetrate clouds, different sensing principles and data characteristics prevent a direct data transfer between the two modalities. This thesis aims to bridge this gap by translating synthetic aperture radar (SAR) data into the most commonly used optical vegetation index, the normalized difference vegetation index (NDVI). First, the relationship between SAR backscatter and NDVI values is explored as a basis for a potential translation. This is done for three globally distributed agricultural study areas covering a range of environmental conditions. The analysis includes data from several SAR sensors with different frequencies, including C-, S-, X- and L-band data. The investigation reveals a notable relationship between S- and C-band data, but also demonstrates the influence of numerous factors on this relationship, limiting generalization. Building on the previously established relationship, the next step is to demonstrate the estimation of NDVI images from SAR backscatter data. For this purpose, a U-Net, a deep learning model, is trained. To allow a global application, a comprehensive dataset is created, named SEN12TP, consisting of close temporal pairs of Sentinel-1 and Sentinel-2 images from over 1200 different areas with a balanced distribution considering land cover, climate, and seasonality. The evaluation demonstrates the low error and good spatial detail of the trained U-Net. Further, it is shown that the model is globally applicable, outperforming a region-specific model. Time series can be generated from the SAR-estimated NDVI images, but their utility would be considerably enhanced by integrating them with the available sparse optical data. Consequently, a flexible approach to fuse remotely sensed time series is presented as the third and final aspect of this dissertation. This approach is based on an RNN. For training purposes, a dataset consisting of 1.5 years of data and regions from the SEN12TP dataset was created. The results demonstrate the successful fusion of SAR-estimated and optical NDVI time series. A low error is achieved for both short and long gaps while allowing for the global application of this method. Overall, this thesis presents a comprehensive framework to overcome the inherent limitations of optical, cloud-affected vegetation indices. This is achieved by augmenting the indices with information derived from SAR data.    «

Optical remote sensing provides invaluable data for monitoring the Earth and its vegetation. However, cloud cover hinders the acquisition of images and leads to data gaps. While radar remote sensing can penetrate clouds, different sensing principles and data characteristics prevent a direct data transfer between the two modalities. This thesis aims to bridge this gap by translating synthetic aperture radar (SAR) data into the most commonly used optical vegetation index, the normalized difference...    »