Eq.12c With: A B C D The results of this analysis will be exploited to obtain the correct S t to be applied in the uncalibrated procedure, also ensuring the selection of at least one daily timeseries (MODIS AQUA, TERRA or S3) or 3 different products. If this condition is not obtained, all the products of the selected combination will be selected for the merging. Once the products to be merged have been selected, the first operation to be done is rescale all the satellite data to a reference scale. The product with the highest amount of available data is chosen as reference. All the products are hence scaled to this one. First the number of overlapping data is calculated.
If this is greater than 5, the minimum and 75th percentile of the selected product and the reference one is calculated, and the product is rescaled according to (Eq.9): Eq.13 Where P i is the i-th element of the selected product, P s i is the same element scaled to the reference, p75 and m are the minimum and the 75th percentile of the selected product, respectively, calculated during the overlapping days, and p75 r and m r are the same for the reference product. In case less than 5 overlapping days between the reference and the selected datasets are available, the minimum and 75th percentile are calculated for the whole period for both the selected and reference products.
It is worth underlining that it was decided to use the 75th percentile instead of the more common maximum value to avoid selecting potential outlier due to errors in the cloud masking or presence of cloud shadow. Potential outliers are then eliminated by removing all the data index greater than 10 or less than 0. The merging procedure is finally carried out: all the data is interpolated at daily steps. A weight is assigned to each daily data, proportional to the distance from the sensing date: the weights are assigned according to the gaussian distribution, fixing w=1 in the sensing date, w=0.5 at three days from the sensing date (figure 6).
All the interpolated data at more than 6 days from the sensing date are fixed to NaN. Then, the potential presence of contradictory information from the different sensor is accounted for by calculating the difference between the maximum and minimum value of the index for each day of the analysis (considering all the products). The distribution of this variability is assumed as gaussian (excluding data in which only one sensor is available, and the max-min difference is equal to 0): the average and the standard deviation of the max-min difference is calculated and all the dates outside the 3 times standard deviation range are excluded from the analysis.
Finally, the weighted sum is performed by maintaining only the dates in which one or more products has w=1 (at least one product is obtained in that date). An exponential filter with T=6 is then applied to remove possible noises. 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. 26/31