High resolution evapotranspiration map over Austria : leveraging AI and Sentinel-2

Climate change is accelerating at an unprecedented rate, profoundly impacting sectors, systems, individuals, and institutions worldwide. Adaptation to its effects has become a critical priority. In Austria, the consequences of climate change are particularly pronounced, with its existence, pace, and impacts clearly evidenced by extensive measurements and observations. Recent climate data indicate that the country’s annual mean temperature has risen at more than twice the global average, exacerbating challenges in urban areas, agriculture, mountainous and forest ecosystems (https://www.iea.org/articles/austria-climate-resilience-policy-indicator).

In response to these pressing challenges, this project aims to develop innovative services to support climate adaptation strategies. By leveraging existing satellite missions and in-situ data, the integration and monitoring of evapotranspiration (ET) can be used as a key indicator for assessing climate resilience and providing actionable insights to decision-making.

The tool developed for the FFG Project GET-ET is using the measurements of ECOSTRESS and Sentinel-2 imagery to perform high resolution estimation of evapotranspiration. ECOSTRESS measurements have been selected as reference for the modelling, the sensor provides 70m evapotranspiration daily maps. The input corresponds to a combination of multispectral bands and digital elevation model from Copernicus data. To fit the reference with the input, it has been decided to enhance ECOSTRESS measurements with a python implementation of Data Mining Sharpener, based on Leaf Area Index values obtained from Sentinel-2. A dataset has been generated over Austria between 2019 and 2025, considering the atmospheric perturbation and the time correlation of both sensors. The dataset has been fed into a Unet with ResNet blocks pre-trained with Sentinel-2 images. The perceptual loss has been used to increase the capabilities of producing precise estimation of the evapotranspiration. The model trained over Austria reached meaningful results in terms of metric: 0.91 of Structural Similarity Index Measurement (SSIM), letting a confident space for scale generalization. The service can then provide for each new Sentinel-2 image an estimation of evapotranspiration. In the context of the project, monthly aggregation over Austria is produced and integrated into the GTIF platform.

High resolution evapotranspiration maps are valuable tools in urban planning enabling the strategic design of green infrastructure to build climate-resilient cities. These maps allow the precise identification of urban heat islands (UHI), areas experiencing elevated heat stress due to the lack of green infrastructures (GI). By pinpointing these areas, planners can implement targeted green interventions, to enhance natural cooling mechanisms such as cooling corridors. Beyond heat mitigation, ET maps also support the ongoing monitoring of green spaces, such as green roofs and parks to ensure their vitality and long-term effectiveness in providing cooling benefits, therefore improving urban livability. Within this project, the ET maps are demonstrated through real-world use cases over Austria.