Machine learning-based grain size mapping from satellite images
- 1Free University of Bozen, Faculty of Science and Technology, Bolzano-Bozen, Italy (email@example.com)
- 2Department of Geosciences, University of Padova, Padua, Italy
- 3CNRS UMR5600-EVS, University of Lyon, Lyon, France
- 4Water Protection Department, Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA), Rome, Italy
- 5Department of Geography, Durham University, Durham, UK
A comprehensive understanding of river dynamics requires the quantitative knowledge of the grain size distribution of bed sediments and its variation across multiple temporal and spatial scales. Several techniques are already available for grain size assessment based on field and remotely sensed data. However, the existing methods permit to cover small areas and short time scale, thus the operational measurement of grain size distribution of river bed sediments at the catchment scale remains an open problem. A solution could be the use of satellite images but very limited investigations have been carried out so far on the use of satellite-based sub-pixel mapping of river characteristics relevant to ecohydraulic processes.
In this study, we propose a new approach to retrieve sub-pixel scale grain size classes information from Sentinel 2 imagery building upon a new image-based grain size mapping procedure. Four Italian gravel-bed rivers featuring different morphology were selected to conduct UAV acquisitions and extract ground truth grain size data from the near-ground images, by photo-sieving techniques. We generated grain size maps at the resolution of 2 cm on river bars in each study site by exploiting image texture measurements, and subsequently resampled and co-registered the grain size maps with Sentinel 2 data resolution.
Relationships between the grain sizes measured and the reflectance values in Sentinel 2 imagery - available in 11 bands super resolved at 10 m resolution – were analyzed. Based on these, our first results show statistically significant predictive models (cross validation results: MAE of 3.38 ± 13.4 mm and R2=0.48) by using a machine learning framework (Support Vector Machine) to retrieve grain size classes from reflectance data.
Our proposed approach based on freely available satellite data calibrated by low-cost automated drone technology can provide reasonably accurate estimates of surface grain size for bar sediments in medium to large river channels, over lengths of hundreds of kilometers. Moreover, the proposed methodology is easily replicable to other natural environments where an extensive grain size distribution assessment is crucial to understand geomorphic processes, thus providing a new technique for collecting such precious data and support studies of landscape evolution.
How to cite: Marchetti, G., Bizzi, S., Belletti, B., Lastoria, B., Mariani, S., Casaioli, M., Comiti, F., and Carbonneau, P.: Machine learning-based grain size mapping from satellite images , EGU General Assembly 2021, online, 19–30 Apr 2021, EGU21-14945, https://doi.org/10.5194/egusphere-egu21-14945, 2021.