Vegetation monitoring using remotely sensed data is useful for many applications, for example crop yield prediction. Many of these applications utilize the normalized difference vegetation index (NDVI) acquired using space-borne optical sensors. Using the NDVI however has one drawback: cloud coverage prevents data acquisition. To tackle this we present a method to estimate the NDVI from cloud penetrating radar sensors using a convolutional neural network (CNN). This model is trained with a global, balanced dataset called SEN12TP consisting of temporally paired Sentinel-1 and cloud-free Sentinel-2 images together with auxiliary data. A good performance is achieved with this globally applicable model. Additionally we show that radiometric terrain correction of the radar backscatter is unnecessary if the model is provided with the elevation data.    «

Vegetation monitoring using remotely sensed data is useful for many applications, for example crop yield prediction. Many of these applications utilize the normalized difference vegetation index (NDVI) acquired using space-borne optical sensors. Using the NDVI however has one drawback: cloud coverage prevents data acquisition. To tackle this we present a method to estimate the NDVI from cloud penetrating radar sensors using a convolutional neural network (CNN). This model is trained with a globa...    »