Validation of TRMM Satellite Rainfall Algorithm for Forest Basins in Northern Tunisia

  • Saoussen DhibEmail author
  • Zoubeida Bargaoui
  • Chris M. Mannaerts
Conference paper
Part of the Advances in Science, Technology & Innovation book series (ASTI)


At present, a number of newly satellite-derived precipitation estimates are freely available for exploration and could benefit the hydrological community. This study aimed to evaluate the Tropical Rainfall Measuring Mission TRMM 3B42 rainfall estimate algorithm in forest basins in Northern Tunisia. We selected 77 events, with 50 mm/day heavy rainfall as selection criteria, for at least one station of the study area observed during 2007, 2008 and 2009. Rainfall stations were interpolated using the inverse distance method. Results were discussed in terms of the TRMM product accuracy in comparison with rain gauges over the forestry zone (169 stations). The Pearson’s correlation coefficient between satellite estimations and ground maps reached 0.7 for some events and were weak for others. The comparison results of TRMM algorithm over forestry zone within Northern Tunisia shows a weak difference in terms of false alarm ratio (FAR), and bias. However, it shows a better detection for the whole of Northern Tunisia in terms of correlation coefficient and the probability of detection (POD). Some uncertainties have been found, across the TRMM algorithm over forestry region. Thus, the evaluation of satellite algorithms before use as input for other models is recommended.


Rainfall Extremes Tunisia Forest Validation TRMM 3B42 

1 Introduction

For most Tunisians, heavy rainfall means more accidents, traffic jams. One of the main causes of the shortcomings of flood monitoring is the cruel lack of observations that hinders the characterization of the hazard. To this end, it is clear that a great effort remains to be achieved in forecasting and especially alerting. In the past two decades, satellite-derived products provided a cost-effective way to estimate precipitation. However, as mentioned by Tian et al. [1] satellite data have an inherent uncertainty. Therefore, one can see the importance of evaluation. Thus the problem can be formulated as follows: Do TRMMs 3B42 detect well extreme events over the Tunisian forestry region? And is there a difference between the rainfall detection over forestry region and the whole Northern of Tunisia?

2 Data and Methods

2.1 Regional (In Situ) Rainfall and Satellite Data

The daily rainfall network covers a zone from 36′N to 37° 2′N and from 8°E to 9° 2′E. It is included as a part of the Medjerda River watershed (W-5), and a part of the northern coastal watersheds (W-3). The network size consists of 169 operational stations. An extreme daily event is assumed when at least 50 mm/day are recorded at least in one single station in the whole domain (Northern Tunisia).

TRMM product 3B42 V 7 was used with a resolution of 0.25° and 3 h. This product was a combination of TIR from geostationary satellites and microwave sensors [2].

2.2 Methodology

The moving average using an inverse distance weighting was adopted for interpolating the rainfall for the two seasons [3]. The spatial verification methods included visual verification, statistical verification (correlation coefficient (r), and ratio bias coefficient), and categorical statistics (POD, FAR, and Bias). POD, FAR and Bias range from 0 to 1, with 1 being a perfect POD, and Bias and 0 being a perfect FAR.

3 Results

3.1 Evaluation of TRMM over the Forestry Zone

Figure 1 illustrates the averaged values of the correlation coefficient (R), POD, FAR, and Bias, for all the selected heavy events during dry and wet seasons.
Fig. 1

Box plots of the averaged coefficients for the dry and the wet season over the forestry zone

Figure 1 shows a weak difference between the correlation average and POD coefficients of the wet and dry seasons. However, the TRMM during the wet season shows a better performance in the estimation of the FAR and the bias coefficients.

Figure 2 presents maps 10 events as estimated by TRMM and ground observations during wet and dry seasons.
Fig. 2

Some maps examples for the dry season (a) the wet season (b)

We notice that TRMM yielded a good estimation for some events such as 24/09/2007 and 18/04/2007, an underestimation of other events such as 23/02/2009 and overestimated some events such as 08/03/2007 and 12/01/2009.

3.2 Comparison of TRMM algorithm performance over the forestry zone and Northern Tunisia

To compare the performance of TRMM over the forestry zone and Northern Tunisia, we performed some statistics (Fig. 3). We notice a weak difference of the average of all the evaluation coefficients (R, POD, FAR, and Ratio Bias) over the forestry zone and Northern Tunisia.
Fig. 3

Box plots of the averaged correlation coefficients (R), POD, FAR, and ratio bias over the forestry zone (f) and the whole of Northern Tunisia (n)

4 Discussion

The oscillation of TRMM rainfall estimation between an overestimation and an underestimation against in situ data confirms the findings of many worldwide previous studies. For example in the Iberia case study, [4] found that TRMM-3B42 tends to overestimate rainfall across the whole domain in most seasons but gives an overall underestimation in the orographic area.

5 Conclusion

This study evaluated TRMM 3B42 data version 7 using 169 ground rainfall observation stations in the Northern Tunisia forestry zone for extreme events recorded from January 2007 to August 2009. The TRMM showed a good occurrence compared with the rainy ground days where 96% were captured rainy by the satellite. 20% of the studied events were well estimated by TRMM with a correlation coefficient of about 0.5. The spatial average comparison showed a similar correlation and probability of detection during the wet and dry seasons while false alarms and bias coefficients are smaller during the wet season. As we saw in this work, monthly GPCC correction for TRMM 3B42 didn’t correct well the extreme events. That is why we are planning to use some bias correction methods to ameliorate the TRMM algorithm performance over Tunisia.


  1. 1.
    Tian, Y., Liu, Y., Arsenault, K.R., Behrangi, A.: A new approach to satellite-based estimation of precipitation over snow cover. Int. J. Remote Sens. 35, 4940–4951 (2014). Scholar
  2. 2.
    Ouma, Y.O., Owiti, T., Kipkorir, E., KibIy, J., Ryutaro, T.: Multitemporal comparative analysis of TRMM-3B42 satellite-estimated rainfall with surface gauge data at basin scales: daily, decadal and monthly evaluations. Int. J. Remote Sens. 33, 7662–7684 (2012)CrossRefGoogle Scholar
  3. 3.
    Ilwis Help.: ILWIS version 2.22 (1999)Google Scholar
  4. 4.
    El Kenawy, A.M., Lopez, J.I., McCabe M.F., Vicente, S.M.: Evaluation of the TMPA-3B42 precipitation product using a high-density rain gauge network over complex terrain in northeastern Iberia. Global Planet. Change 133, 188–200Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Saoussen Dhib
    • 1
    Email author
  • Zoubeida Bargaoui
    • 1
  • Chris M. Mannaerts
    • 2
  1. 1.ENIT, Université de Tunis El Manar, Ecole Nationale d’ingénieurs de TunisTunisTunisia
  2. 2.Faculty of Geo-information Science and Earth ObservationITC: University of TwenteEnschedeThe Netherlands

Personalised recommendations