Abstract
Monitoring grasslands over the phenological season in the alpine environment using Synthetic Aperture Radar (SAR) backscatter data requires innovative approaches that can combine domain-specific ancillary information with advanced deep learning techniques. Alpine grasslands are characterized by complex topography, seasonal variations, and heterogeneous agricultural management, making traditional monitoring methods less effective. The Leaf Area Index (LAI) can generally be used to study the above ground biomass of grasslands, which is estimated rapidly and accurately from the Sentinel-2 multispectral data. However, the frequent unavailability of cloud-free imagery poses a significant challenge, creating temporal gaps. To address this limitation, there is growing interest in incorporating Sentinel-1 SAR backscatter data to perform temporal gap-filling, enhancing the temporal resolution of grassland LAI derived from Sentinel-2 and enabling continuous monitoring of grasslands throughout the growing season. In this paper, we propose a domain-specific transformer-based convolutional network that extracts features from SAR data and its derivatives to capture long-range dependencies, and we preliminarily investigate its application to temporal gap-filling of Sentinel-2 LAI. The effectiveness of the proposed method for predicting LAI has been validated by testing it against Sentinel-2-derived LAI. In the results, we achieved an R2 around 0.59 on the test sets, between the related predicted LAI and Sentinel-2 LAI.