Abstract
This contribution presents the development of an EO-derived operational Grassland Production Index (GPI) designed for alpine permanent meadows in the province of Bolzano, in the North-East of Italy. The GPI was developed in collaboration with local stakeholders of the agricultural sector, and it is tailored for the estimation of yield losses associated with drought events. To this aim, a protocol was developed for the collection of ground reference data, accounting for both crop-specific growing stages and optimizing the spatial coherence between in-situ and EO data. The sampled data were used to identify an appropriate EO-derived proxy for the yield estimation, namely the biophysical variable Leaf Area Index (LAI) derived from Sentinel-2 multispectral data. It was characterized by a RMSE of 0.92 [m2 m−2] and an R2 of 0.81 when validated with ground data over one agricultural season. The GPI was then derived by coupling Sentinel-2 LAI and a Water Stress Coefficient based on meteorological data. The index was validated against yield data over one agricultural season resulting in a R2 of 0.74 at the parcel scale. In accordance with these results, the proposed index was used at the end of the agricultural season 2024 as a basis for yield loss calculations in the framework of an index-based insurance for 138 farms in the province of Bolzano. The operational usage of indices such as GPI benefits from high frequency EO-derived information, which is not guaranteed when using only multispectral data due to cloud coverage, especially in complex terrains. To overcome this limitation and increase the accuracy of the GPI, several approaches are under evaluation, including machine and deep learning techniques, for integrating data from different sensors, such as Sentinel-1 Synthetic Aperture Radar (SAR). Besides the current research challenges, the use case shows the potentialities of EO-based quantitative monitoring of drought impacts on agricultural production.