Exploring Water Reservoir Dynamics in Central Italy: A Preliminary Workflow for COSMO-SkyMed Imagery-Based Water Segmentation

Hydrological drought is one of the most critical effects of climatic change, affecting inland water body extent and level, and threatening the surrounding ecosystem and industries. Traditionally, ground instruments have been used as the primary method for monitoring reservoir water level changes. However, these measurements are generally conducted with pre-installed gauge stations, which do not provide estimates of the extent of changes and raise challenges in their set-up and maintenance, especially in remote areas. Conversely, Earth Observation technologies can remarkably reduce monitoring costs and provide frequent and regular data that can facilitate the continuous monitoring of water bodies. This work is developed within the Geomatics for Resilience Against Water scarcity (GRAW) collaborative project (Agreement n. 2023-1-HB.0) between the Geodesy and Geomatics Division (DICEA) of the Sapienza University of Rome and the Italian Space Agency (ASI), in the framework of the ASI Innovation for Downstream Preparation for Science (I4DP_SCIENCE) program. It aims to develop a methodology for the routine monitoring of water resources from the analysis of satellite imagery both in terms of inland water reservoir extent and level. Here, a preliminary workflow for automatic water segmentation (Figure 1) using Synthetic Aperture Radar (SAR) COSMO-SkyMed imagery is presented. Methodology Among the products available within the ASI COSMO-SkyMed catalogue and accessed from the institutional new MapItaly portal (ASI, 2023), orthorectified StripMap HIMAGE images with a spatial resolution of 2.5 m and HH polarization were selected. The analyses presented here refer to the area of Albano Lake (central Italy). The proposed workflow is based on a radiometric approach which takes advantage of the differences in the back-scattered SAR signal to distinguish the water pixels from the land ones. In particular, the workflow consists of two main steps: image pre-processing (normalization and filtering) and thresholding. At first, the input data were normalized to 8-bit images to stretch the intensity values. Subsequently, standard filters were used to improve the contrast of the images for finer water body detection (histogram equalization) and to remove the noise (bilateral filtering and morphological transformation). A final normalization step was applied forcing the gray-scale values to be included in the range from 0 to 1. At this primary evaluation stage, this range was chosen by trial and error to ensure that distinct intensity values represent the water basin and the land. Then, the images were segmented using a binary thresholding approach, identifying Binary Large OBjects (BLOBs); the largest BLOB corresponds to the lake and it was saved as its final mask. This procedure was replicated using different images to generate water masks for each epoch of interest. First results Three images capturing Albano Lake were processed with the developed workflow. The images were selected with the same acquisition geometry (descending orbit) and during the same period in different years (July 2017, July 2020, and July 2023). The resulting masks were compared to analyze the lake extent variation during the 3 years under investigation (Figure 1). The following preliminary results are based on only three analysis periods since the purpose of this study is to test and demonstrate the potentialities of the proposed workflow. The corresponding areas during the three periods were calculated as shown in Table 1. It was found that from 2017 to 2020, considering the period of July, Albano Lake remained almost identical in its extent, while from 2020 to 2023 it decreased by 0.3% of its total area. The results also showed that the areas most affected by variation correspond to the shoreline areas, where erosion processes and changes in water levels are likely to be most present and thus constitute an element of variation in lake contours. Figure 2 shows the consistent changes in the northern shoreline of the lake, where the retreat reaches the highest value between 2020 and 2023 (around 20 m). Nevertheless, it is important to highlight that some issues might have affected the definition of the contours of the water basin along the steepest or vegetated areas around the lake, due to the known limitations of SAR technology in these areas. For this reason, the masks obtained from the workflow should be considered preliminary and need to be validated with an accurate reference (e.g., in situ measurements provided by basin authorities). Outlook for the future This work proposes a preliminary workflow for monitoring water reservoir dynamics using high-resolution SAR imagery. The first results demonstrate the feasibility of this approach for water segmentation and water area change computation. Nevertheless, implementing an adaptive thresholding technique (e.g., Otsu), applicable to different regions and not affected by vegetation or man-made structures, could improve the segmentation results and the procedure efficacy. Future developments will therefore include further evaluation of the investigated workflow performances using a larger dataset, focusing also on the effects of different acquisition geometries. Furthermore, a validation procedure will be implemented to estimate the accuracy of the proposed methodology. References ASI. ASI – Italian Space Agency upgrades access to MapItaly data. Downloading of COSMO-SkyMed constellation images has never been so easy. 1 December 2023, https://www.asi.it/en/2023/12/asi-italian-space-agency-upgrades-access-to-mapitaly-data/