6 Derive discharge from multispectral images (RD- multispec) 6.1 Context The method for estimating river discharge from multispectral images is based on the studies of Tarpanelli et al. (2013), Filippucci et al. (2022) and Filippucci et al. (2025), in which the differences between the passive response of the reflectance signal from the soil and that from the water are used to identify a change in the land area near the river channel that is shown to be strongly correlated with river discharge. An increase in river discharge produces an increase in wetted area, and the area near the river changes its reflectance response, which decreases.
For an area near the river that is not affected by water, reflectance remains almost constant (except for changes in vegetation cover). Its relationship with the reflectance of the wetted area is used to more accurately determine the estimate of changes in hydrologic forcing, compared to the wetted area alone. Consequently, in the case of flooding, the reflectance ratio between the dry pixel (called the calibration pixel, C) and the wet pixel (called the measurement pixel, M) is sensitive to the increase of water in the wet pixel and, therefore, is directly related to the increase of river discharge. With respect to the first study by Tarpanelli et al.
(2013) where the reflectance ratio C/M has been extracted from a temporal series of seven years of almost daily images of MODIS over four stations along the Po River, the analysis by Filippucci et al. (2022) demonstrated that the role of sediments and vegetation in the formulation was important to correct the reflectance ratio C/M during flood events. This was possible with the use of finer resolution images from Sentinel-2 and the new approach was tested over two Italian rivers, Po and Tiber. In Filippucci et al.
(2025), the selection of reflectance indices based on local hydrological conditions is carried out evaluating the best procedure to obtain river discharge proxies according to the specific flow regimes and climatic conditions across 54 sites worldwide. 6.2 Reflectance indices definition In the project, we tested several algorithms for the estimation of the reflectance index. Such algorithms come from the combination of pixels: C for calibration, W for the sediments, V for the vegetation, M for measurements of the variation. The main steps for the application of the procedure are listed below: - The study area is predetermined as a square of fixed side (e.g.
0.04 degree for Sentinel-2 data, 0.06 degree for Landsat data and 0.15 degree for MODIS data) around the selected station. - The collection of the desired data in the chosen area is then obtained for the available period. - Cloud products are considered to mask the cloud presence in the single images. After this masking, the total number of valid pixels in each image is computed. If the fraction of valid pixel was less than 0.2, the full image was discarded. Similarly, the fraction of valid reflectance value was calculated for each pixel during the study period:
if this value was below 5%, the pixel was deemed invalid and removed by the analysis, to avoid its selection in the mask calculation. - Snow or ice presence is masked through other products available in the collection of data. The fraction of snow pixels is calculated for each available image and the resulting time series is averaged for each day of the year, to obtain a standard year probability of snow presence. A windowed moving average filter of 14 days was then applied to reduce the noises.
- Water area pixels are obtained using JRC Global Surface Water Mapping Layers for coarse resolution sensors (MODIS and Sentinel-3) and the application of a threshold to the 5 th percentile of NDVI timeseries for the remnant products. For each image, if the fraction of valid pixels in the water area was less than 0.2, the full image was discarded. - The different categories of pixels (vegetation, bare soil and field) are classified through masks derived by NDVI index, coefficient of variation (NIR standard deviation divided by NIR average), mean and standard deviation. D.3 River Discharge ATBD. Reference: RD_CCI-0006-ATBD - Issue 2.0 – 31/07/2025 Open/Public © 2019 CLS. All rights reserved.
Proprietary and Confidential. 23/31