Evaluation of spaceborne and airborne hyperspectral data for inland water applications

In recent decades and in the current global warming scenario, the quality of inland waters has been severely challenged and often dramatically degraded [1]. In this context, the monitoring of water quality in different types of water bodies (e.g., lakes and lagoons) can be optimally performed by integrating traditional in-situ measurements with remote sensing techniques. In particular, a new generation of spaceborne and airborne hyperspectral sensors is contributing to the water resources studies and monitoring. The aim of the present study is to evaluate hyperspectral spaceborne data (PRISMA and DESIS, both with pixel size=30m) and hyperspectral airborne data (AVIRIS and HYSPEX, with pixel size=5m and 2m respectively) of inland waters characterized by different gradients of trophic levels and turbidity in order to contribute to global water quality mapping. A comparison of imagery data with in-situ measurements, in terms of Remote Sensing Reflectance (Rrs) was firstly performed. A total of ten matchups were analyzed for the following water bodies: four Italian lakes (Lake Garda, Lake Mulargia, Lake Trasimeno and Lake Varese), Lake Peipsi in Estonia and Curonian Lagoon in Lithuania. Since the AVIRIS and the HYSPEX images were affected by sunglint, the "DeGlint processor" implemented in the SNAP software within the Sen2Coral v1.0.0 plugin and based on [2] was tested. Then, to classify the spectra retrieved from each image, the dominant wavelength (?d) was identified by performing a chromaticity analysis. This process transforms hyperspectral reflectance values into a human-perceived color space; ?d thus represents the color of the visible spectrum, instead of the wavelengths of remote sensing reflectance spectra. The results showed that the sunglint correction improved by 70% (R2), 92% (RMSE) and 76% (Spectral Angle (SA)) the statistical agreement between airborne and in-situ data. Chromaticity analysis allowed the dataset to be divided into three categories from the clearest to the most turbid waters, here called as C1, C2 and C3. The result demonstrated that watercolor can be seen as a water quality attribute and, when combined with satellite remote sensing, it may inform macrosystems ecology in different waterbodies, e.g., lakes [3]. The results of the matchups (C1: R2=0.75, RMSE=44.6%, SA=19.4; C2: R2=0.87, RMSE=35.1%, SA=11.9; C3: R2=0.94, RMSE=20.5%, SA=9.1) showed the improvement in the agreement between the data when moving to more turbid waters (i.e., C3). Subsequently, the dataset was grouped by sensor type to rank their performance. The results obtained (spaceborne data: R2=0.86, RMSE=32.4%, SA=13.6; airborne data: R2=0.96, RMSE=18.3%, SA=6.7) could be explained by the higher spatial resolution of airborne sensors data. In addition, to generate more accurate water quality products, the Signal-to-Noise Ratio (SNR) was calculated to identify the noisiest bands (usually located in the blue region of the spectrum) and remove them. The maps (in terms of Chlorophyll-a and Total Suspended Matter) generated using the BOMBER bio-optical model [4] highlighted the potential contribution that hyperspectral data could provide for global water monitoring. In conclusion, hyperspectral data showed good radiometric quality and, in synergy with multispectral data, could provide a valuable contribution to the monitoring of inland waters, which are in urgent need of protection due to rising temperatures (which alter water quality and support the growth of Harmful Algal Blooms) and droughts that are lowering lake levels.