December 20, 2021 | By: Karisma Yumna, M.Tech, Department of Hydrology, IIT Roorkee

Merging of satellite precipitation products offers new perspective

When we start any hydrological analysis, the first question in our mind is “Is our data accurate?” or “Do we have consistent long term data?”. The quality of the data we used impacts our results and hence our interpretation of many hydrological processes.

Having stated that, I think precipitation is the most important variable in any hydrological process. From measuring by a rain gauge, we have shifted to the advanced method of indirect precipitation measurement, namely through the use of different infra-red and microwave sensors. For decades, a variety of satellite precipitation products (SPPs) such as the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), Climate Prediction Technology MORPHing technique (CMORPH), Tropical Rainfall Measuring Mission (TRMM), Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG) etc. However, despite many of such efforts to provide high resolution global data, these indirect precipitation measurements are inaccurate in capturing the ground precipitation.

Many researchers have evaluated the reliability of these products in at different regions and climatic zones. One SPP may capture medium rainfall intensities but may fail to do so for the high rainfall intensities, and vice-versa for another SPP. Keeping this in mind, the Quantile based Bayesian Model Averaging (QBMA) was proposed to merge three SPPs (TRMM, PERSIANN and CMORPH) over a coastal river basin in India (Vamsadhara river), in our recently published article in the Journal of Hydrology. Several possibilities of merging the SPPs during the monsoon season were discussed.

Results indicated that Bias-corrected QBMA schemes reproduced the spatial pattern and magnitudes of the monsoon rainfall across the river basin. In particular, the linear scaling approach with the PERSIANN sampling product (QBMALSp) outperformed all the other bias-corrected products. On monthly evaluation, it is observed that all the products perform better during July and September than that in June and August. The QBMA approaches do not have any significant improvement over the Simple Model Averaging approach in terms of Probability of Detection. However, the bias-corrected QBMA products have lower FAR. The developed QBMA approach with bias-corrected inputs outperforms the IMERG product in terms of RMSE.

In summary, the study provides a new perspective of improving the quality of the indirect precipitation measurements, which will provide new insights to the future researchers in these area. These findings are described in the article entitled “Quantile-based Bayesian Model Averaging approach towards merging of precipitation products”, recently published in the Journal of Hydrology.


Figure 2: Boxplots of the CC and RMSE during the calibration period (2001 -2013) for the original rainfall products (blue band), merged products using traditional methods (grey), merged precipitation products of QBMA scheme (red band), Linear scaling bias-corrected (green band). The yellow colored box plots represents TRMM and TRMM sampling based products, green boxes for the PERSIANN-CDR and PERSIANN-CDR based sampling based products, and the magenta boxes represent the CMORPH and CMORPH sampling based products. The median values of a) Correlation coefficient are shown at the right side of the plot, while b) Root Mean Square Error, is shown on the left side.