3.2. Contingency Statistics
The categorical statistical metrics of 3B42 V7 and IMERG V5 at different time scales are shown in
Table 3.
The precision of 3B42 V7 and IMERG V5 at 3 h, daily, monthly and annual scales were compared and analyzed. IMERG V05 demonstrated better performance than 3B42 V7 in detecting small time scale precipitation events, with high values of POD and CSI (0.13, 0.39 vs. 0.58, 0.82) and (0.08, 0.22 vs. 0.14, 0.18), respectively (
Table 3), as well as reasonably high values of FAR and FBI (0.79, 0.66 vs. 0.83, 0.81) and (0.65, 1.17 vs. 3.63, 4.37), respectively.
The performance of the categorical statistical measures at the monthly and annual levels are shown in (
Table 3). 3B42 V7 and IMERG V05 produced good results for rainfall estimation, with POD values and CSI values (0.99, 1 vs. 1, 1) and (0.07, 0 vs. 1.07, 1), respectively. Similarly, for the FAR and FBI, the results obtained were close to the perfect values, (0.08, 0 vs. 0.08, 0) and (1, 1 vs. 1.09, 1), respectively.
In general, IMERG V05 is better at detecting precipitation events, in particular at capturing precipitation traces and solid precipitation at a 3-hourly and daily scale, while 3B42 V7 can estimate precipitation on a large time scale.
3.3. Hydrological Evaluation of Discharge Simulation Using Two SPPs
The HEC-HMS model was used to calibrate and validate the 3 h of rainfall events from (1 September 2014) to (31 August 2017), at the level of the Zat basin, using the rainfall and runoff data from the Taferiat gauge station and satellite precipitation products. The four episodes that we chose to present are the most representative and complete of the data series.
The hydrological calibration and validation were carried out according to two different scenarios.
The obtained calibration and validation results are very satisfactory; the Nash–Sutcliffe coefficients obtained for calibration and validation are on average 88.20% and 57.50%, respectively (
Table 4 and
Table 5). The episode calibrations were performed by manual adjustment of the parameters in a way that does not lead to the deviation of the parameters from their real physical meaning, and which allows for a better understanding of each calibration parameter. This method requires a lot of time and effort to understand the behavior of each parameter and approximate it to the natural condition of the event occurrence. On the other hand, the objective function optimization method is simple and practical for function-based investigations but may ignore the real physical meaning of the parameters. Therefore, this paper combines the two methods to adjust the model parameters.
In this paper, eleven parameters were calibrated by maintaining the maximum and minimum intervals of calibration parameters based on the literature. The intervals of the calibrated parameter used are illustrated in
Table 4.
Scenario 1 calibration: simulation and calibration using precipitation from both satellite products with observed fluxes, by adjusting the model parameters values until the model results acceptably match the observed data.
Scenario 2 validation: due to the limited sample size, the model was validated using the leave-one-out resampling approach; for the n flood events, each event i is successively removed, in order to find the relationship between the root-soil moisture measured by the time domain reflectometry “TDR” tool and the two models’ most sensitive calibration parameters (Curve Number “CN”, and time of concentration “TC”). A new CN was then re-estimated (Calculate CN) using the remaining n-1 episode. The CN calibration and the Tc parameters for an event i are set to the median of the calibrated parameters for the n-1 episodes. The calculated CN values obtained by this procedure are then used to model flood event i, and the simulated discharge is compared to the observed discharge. The validation results for the 15 events are presented in
Table 6, indicating better model performance when using the SCS-CN model and taking into account soil moisture, with Nash coefficients between 0.51 and 0.82, using the leave-one-out procedure [
43].
3.3.1. Event of 20 November 2014
This event represents a torrential flood; since the flood was generated by extreme precipitation spread over more than 15 days, it is the most intense event in the data set. The maximum flow reached was (123, 75 m3/s). However, the soils were saturated, resulting in high permeability and an increase in the runoff coefficient of the watershed.
The results of the calibration of the 3B42V7 and IMERG V05 rainfall data with the observed flow illustrated in the hydrographs of the
Figure 2, show that the simulated flow curves were globally well reproduced for both products at the flood rise and the recession part, although the peaks flow were not reached by both products.
Based on the results the 3B42V7 calibration is satisfactory, as this product has a good capacity to record the high precipitation intensity during rainy episodes.
Furthermore, the IMERG V05 product does not have the ability to capture heavy precipitation; this is well demonstrated in the calibration results. The evaluation criteria are suitable RMSE = 0.5 for both products and a Nash of 79.20% and 70.10% for 3B42V7 and IMERG V05, respectively.
The validation hydrographs results in (
Figure 3) were well reproduced for both products. The rise and the recession curves were well reproduced for 3B42V7, noting a slight underestimation of precipitation, but in general, this product well estimates heavy precipitation events. The IMERG V05 was not able to reproduce the validation hydrograph of this event. The rising curve was underestimated in the first pick as it represents the heavier precipitation during this event, and the other two picks were underestimated, the peak flow was not reached. This is because of the inability of this product to properly estimate the heavy precipitation.
The evaluation criteria are acceptable with an RMSE of 0.7 for both products, and a Nash of 52.20% and 66.50%, respectively.
3.3.2. Event of 21 March 2016
This event represents the typical characteristics of a freshet caused by the melting of snowfall upstream of the Zat watershed. With the progressive increase of temperatures, the snow cover at the summit of the Atlas Mountains starts melting and feeding the streams of the mountainous basins including the study basin. This usually causes significant flooding during the occurrence of moderate rainfall episodes.
The hydrograph of
Figure 4 is well calibrated, the simulated flow curves were differently reproduced for both products. Concerning 3B42V7 SPPs, the rise and the recession curves were well reproduced, but the peak flow was underestimated due to the fact that this satellite product is not able to reproduce the low precipitation. However, IMERG V05 hydrograph is well reproduced at the rise, the recession, and the peak flow.
The evaluation criteria are good with an RMSE of 0.3 and 0.3 and a Nash of 66.5% and 83.7% for 3B42V7 and IMERGV5 products.
On the other hand, the hydrographs of validation are also differently reproduced for the two SPPs (
Figure 5).
The 3B42V7 underestimates the precipitation; as noticed this episode is generated by the effect of snowmelt, with the occurrence of light precipitation, and for this reason, the following product estimates poorly the slight precipitation as previously indicated, and consequently underestimates the value of precipitation during the event.
However, IMERGV5 has shown good performance in detecting small precipitation events; on the validation hydrograph the simulated curve is well reproduced at all levels. This is due to the fact that this product is capable of estimating small precipitation events with short time steps. This resolution has been well confirmed once more at this event.
The evaluation criteria are good with an RMSE of 0.7 and 0.6 for the product 3B42 V7 and IMERG V05, and a Nash of 56.20% and 74.20. 9%, respectively.
3.3.3. Event of 3 May 2016
The hydrological model results showed a reasonable fit between the shape of the simulated and observed hydrographs.
Figure 6 shows a chronological comparison of simulated and observed streamflow at the watershed outlet for a calibration period of 3–5 May 2016 (we limit ourselves to modeling short-duration floods for which the evapotranspiration process is negligible). Although the measured peak flow values did not exactly match the simulated peak flow values for both products, there was a slight improvement for IMERG V05. The model tended to underestimate the streamflow, since the river flow was already high due to snowmelt, and this was added to the runoff caused by the flood, which the model did not take into account.
On the other hand, the volumes were well respected and the rise and fall curves were generally well reproduced with a slight improvement on the IMERG V05 side. The results of the evaluation criteria for both products 3B42V7 and IMERG V05 are satisfactory with NSE ranging from 76.80% to 79.30%, respectively.
After following the validation procedures previously mentioned in
Section 3.3.4, the comparison of the observed and simulated hydrographs showed that the model underestimates the point flows, due to the non-conclusion of the snowmelt process. However,
Figure 7 shows a good trend in the reproduction of the observed and simulated discharge curves, with an NSE of 61.30% for 3B42V7, and 72.90% for IMERG V05, which are good results.
3.3.4. Event of 16 December 2016
The event represents a winter rain-storm characterized by liquid precipitation downstream and snow upstream of the watershed. This type of rain-storm is very frequent during the winter, especially in the high mountains of the Atlas.
The hydrographs of calibration in
Figure 8 represent a simulated flow curve quite illustrative; the rising curve and the recession were well reproduced for both products 3B42V7 and IMERG V05, contrary to the peak flow which has not been reached for the 3B42V7. However, the evaluation criteria are acceptable and represent an NSE of 63.50% and 61.90%, respectively.
On the other hand,
Figure 9 illustrates the validation graphs of each product, although the curves of the simulated flows are not well reproduced for both products, which underestimated winter precipitations since they are a mixture of rainfall and snowfall. The evaluation criteria are acceptable with an NSE of 63.20% for 3B42V7, and 57.20% for IMERG V05.
In this paper, an efficient method has been developed for the first time in a country with a Mediterranean climate, on the mountainous watersheds of the Moroccan High Atlas with low density and irregularity of precipitation and flow measurement stations. This is a good method to apply to solve the problem of deficiency of observed data in these regions.