*3.3. TRMM Data*

The Tropical Rainfall Measuring Mission (TRMM), launched by NASA (National Aeronautics and Space Administration) and JAXA (Japan Aerospace Exploration Agency) in 1997, provided precipitation estimates within the latitude 50◦ S to 50◦ N (see [40,41]).

The TRMM satellite carried several instruments to detect precipitation, including the Visible Infrared Radiometer (VIRS), TRMM Microwave Imager (TMI), Cloud and Earth Radiant Energy Sensor (CERES), Lightning Imaging Sensor (LIS) and the first spaceborne precipitation radar (PR). Several precipitation retrieval algorithms have been developed based on observations from the sensors on board the TRMM satellite such as the TMPA (TRMM Multi-satellite Precipitation Analysis). The TMPA algorithm combines observations from satellite-based microwave and infrared sensors and ground rainfall gauge analyses, and produces 3-hourly rainfall estimates at a spatial resolution of 0.25◦ × 0.25◦ with a quasi-global coverage (50◦ N-S) [42]. In 2015, the TRMM mission came to an end, with the instruments turned off and the spacecraft re-entering the Earth's atmosphere. However, the multi-satellite TMPA products continue to be produced using input data from other satellites in the constellation. Indeed, the TMPA algorithms are still being run using other calibrators to produce data in parallel with GPM IMERG [43].

The Level 3 TRMM 3B43 data, also called TMPA product, were chosen for our analysis [42]. In particular, TRMM 3B43 (v7) data for the period from April 2014 to June 2018 were used.

#### *3.4. Methods*

In order to perform the evaluation of IMERG and 3B43V7 products relative to the reference rain gauges data, several indices including Pearson Correlation Coefficient (R), mean error (Bias), relative Bias (rBias), Root Mean Square Error (RMSE) and mean absolute error (MAE) were computed (see Table 1). Pearson correlation coefficient (R) is a dimensionless statistical index used to assess the linear correlation between the reference ground-based data and the satellite precipitation estimates. Mean error (Bias) represents the systematic error of satellite precipitation estimates, a measure of the overestimation or underestimation of the gauge data. Relative Bias (rBias) estimates the relative difference (in percentage) between the two data sources (satellite estimates and rain gauges). RMSE quantifies the average error magnitude (mm/time) between the satellite estimates and the rain gauge data. Mean absolute error (MAE) reflects the magnitude and extent of the mean error of satellite precipitation estimates. For seasonal analysis, the year was divided into four seasons: winter (December to February); spring (March to May); summer (June to September); and autumn (October to November).


**Table 1.** Summary of statistical indices used to evaluate the satellite precipitation products (Si: satellite estimates, O*i*: observations).

Figure 2 shows the number of stations that are distributed within the grid cells of each Satellite Precipitation (SP) dataset. The number of grid cells for the study area was 19 for the TRMM 3B43 data and 61 for IMERG data, respectively. On the one hand, the TRMM 3B43 grids show a notable variation of the available number of gauge stations residing (e.g., from 1 to 23 stations per grid cell). It should be also noted that 42% of the available gauge stations (57 of 136) were located within only three cells, although these 3B43 cells are not the ones with the maximum correlation with the corresponding gauge

values. We notice that the distribution of the available gauge station within each IMERG grid cell was more balanced (1 to 5 stations per grid cell).

**Figure 2.** Distribution of rain gauge stations within Tropical Rainfall Measuring Mission (TRMM) 3B43 (**a**) and Integrated Multi-satellitE Retrievals (IMERG) (**b**) grids.
