- The final step consists in the selection of the periodically wet area. When observed data concurrent to satellite overpass are available, four algorithms of the reflectance indices are calculated to be compared with the in situ river discharge: the pixels with highest Spearman’s correlation are selected as M. When observations are not available, two uncalibrated procedures are adopted, one based on the correlation between the pixels within the water area and the different categories, and one based on the JRC Global Surface Water Mapping “Occurrence” information. Different formulations of the proxy are adopted according to the characteristics of each river reach (e.g.
application of vegetation and sediment correction; use of multiple water mask selection algorithms) and satellite pixels may be aggregated for stations where the river width is considerably larger than the satellite spatial resolution (especially when we use Landsat and Sentinel-2). Moreover, because the analysis involved long temporal period, we need to consider the natural evolution of the river morphology. This means that in some areas a static temporal analysis that considers the same pixels for the entire period is not sufficient to describe the dynamic of the river. In such cases it is necessary to perform the analysis for brief periods.
Therefore, the procedure was applied two years in two years, maintaining one year in common between a period and the following one, in order to avoid big changes in the transition period. The pixel categories masks were obtained combining the ones of the single periods. The procedure, called “multi-year”, is used along with the “full period” for evaluating the benefit to separate the period with respect to the static investigation. Based on what we describe above, the total number of algorithms to be considered in the analysis is 24 to be applied to the 11 satellite products.
6.3 Multi-mission reflectance time series A total of 24 calibrated reflectance indices were obtained for each station considering all formulations (CM, CMW1, CMW2), aggregation kernels (one small and one large), vegetation correction and calibration type (full or multiyear). A first analysis is then necessary to identify the best algorithm for the definition of river discharge. This discrimination is carried out based on a stepwise maximization of the Spearman correlation coefficient, to consider the non-linearity between the reflectance index and the river discharge.
Because the procedure includes several combinations, we proceed to distinguish for each satellite products the following sub-cycles: 1- the best temporal method between full period and multi-year; 2- the best spatial resampling of the images with the aggregation; 3- the best formulation between the simple C/M or the ingestion of the sediments with CMW; 4- the best formulation with the inclusion of the vegetation This stepwise procedure allowed to obtain two results:
first, the best calibrated procedure was obtained for each station, second, the analysis of the results combined with the stations characteristics allowed to obtain a heuristic procedure to obtain a valid algorithm for each station in the uncalibrated procedure. The application of the above methodology to the 10 datasets investigated in this project allows to obtain 10 different timeseries related to river discharge in 54 and 26 stations using the uncalibrated and calibrated procedure, respectively. The indexes were then merged together to obtain a single consistent signal from multispectral sensors (one for the calibrated and one for the uncalibrated methodology).
This procedure is complicated by the fact that most of the sensors have different spatial resolution, coverage period and revisit time. A summary of the different dataset characteristics is shown below. 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. 24/31