Precipitation is one of the key components of the global water and energy cycles, and a robust constellation of precipitation-related satellite sensors could provide reliable global distributions of precipitation at distinct spatiotemporal scales [1
]. Following the launch of the Tropical Rainfall Measuring Mission (TRMM) satellite in November 1997, satellite-based precipitation estimation techniques received an unprecedented boost. The TRMM satellite carried the first space-borne active microwave radar, the Ku-band (13.8 GHz) Precipitation Radar (PR), to provide a three-dimensional structure of the tropical precipitation. A nine-channel conically scanning passive microwave radiometer, namely, the TRMM Microwave Imager, was paired with the PR and placed in a unique non-sun-synchronous orbit to capture the diurnal variability of the tropical precipitation [4
]. Several global or quasi-global multi-satellite precipitation products were developed and made available to users during the TRMM-era [5
]. These precipitation products take relative advantages of the passive microwave imagers onboard the low-Earth orbiting satellites and infrared sensors onboard the geostationary satellites. In addition, these multi-satellite precipitation products provide precipitation information at uniform spatial and temporal scales even over the regions where rain gauge observations are unavailable or meager [7
]. Some of the popular TRMM-era multi-satellite precipitation products are TRMM Multisatellite Precipitation Analysis (TMPA [8
]), Climate Prediction Centre Morphing (CMORPH [9
]), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN [10
]), and Global Satellite Mapping of Precipitation (GSMaP [11
]). These multi-satellite precipitation products have rather large uncertainties over several regions of the globe at multiple timescales [6
]. However, among the TRMM-era multi-satellite precipitation products, TMPA was generally shown to be superior to other products at global and regional scales [12
The TRMM satellite was decommissioned in April 2015, after 17 years of uninterrupted service. In order to continue the objectives of the TRMM satellite with some further advancement, the Global Precipitation Measurement (GPM) Core Observatory was launched in February 2014. This satellite carries Dual-frequency Precipitation Radar (DPR) paired with a 13-channel passive microwave radiometer, namely, GPM Microwave Imager (GMI), which enables more accurate precipitation estimation and its phase detection [14
]. Both TRMM and GPM are collaborative missions between the United States (US) National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA). After the successful launch of the GPM Core Observatory, two GPM-based multi-satellite precipitation products, namely, Integrated Multi-satellite Retrievals for GPM (IMERG [15
]) by NASA and GSMaP version 6 by JAXA were released. There are three kinds of IMERG products (e.g., Early, Late, and Final Runs) available depending upon their applications and latency times. Early (e.g., IMERG-E) and Late (e.g., IMERG-L) Runs are available in near-real-time, whereas Final Run (e.g., IMERG-F) is a research product available in post-real-time and includes rain gauge observations over land. IMERG products are available at finer spatial and temporal resolutions (0.1°/half-hourly) as compared to TMPA (0.25°/three-hourly).
Several research studies showed that the GPM-era multi-satellite precipitation products (e.g., IMERG) usually perform better than the TRMM-era products (e.g., TMPA). IMERG products show better performance than the TMPA-3B42 product in the estimation and detection of extreme precipitation over China [17
], India [18
], and Nepal [19
]. IMERG was also shown to be marginally better than TMPA over the southeastern United States by Tan et al. [20
], and they noticed better precipitation detection and reduction in errors when scaled up to larger area (from 0.1° to 2.5°) and longer time periods (from 0.5 h to 24 h). Sunilkumar et al. [21
] evaluated IMERG-F estimates against rain-gauge-based gridded rainfall dataset (e.g., APHRODITE-2) over Japan, Nepal, and Philippines regions for 2014–2015. They showed that IMERG is able to capture diurnal to intraseasonal variability of precipitation and be improved in the detection of extreme precipitation events compared to the TMPA-3B42. The differences between IMERG and TMPA precipitation products are larger over the ocean as compared to land due to similar gauge adjustment [22
]. Furthermore, the GPM-based GSMaP precipitation product showed similar performance as IMERG [23
]. Based on statistical and hydrological assessments, Yuan et al. [25
] demonstrated that IMERG-F (V05B) is better than TMPA-3B42 and GSMaP products over Myanmar. IMERG is also shown to be better than TMPA in orographic precipitation estimation over the Tibetan Plateau at multiple timescales [26
]. These studies revealed the superiority of IMERG products over other contemporary or TRMM-era multi-satellite precipitation products.
Three Runs of IMERG products (with different release versions) were extensively evaluated over several parts of the globe. For instance, the IMERG-F V6 product was shown to be in good agreement with the radar-based Stage IV product in the representation of seasonal spatial distribution and diurnal cycle of mesoscale convective systems over the central and eastern US for the period of 2014–2016 [27
]. Although the amplitude of the precipitation diurnal cycle has been overestimated by the IMERG suite, the near-real-time IMERG-L product has shown better performance than the IMERG-F product in the representation of precipitation diurnal cycle over Brazil during 2014–2018 as compared against 1261 rain gauge observations [28
]. An evaluation of IMERG-F V6 product over China at a daily scale for 2014–2018 revealed that the multi-satellite product has limited capability in the detection of light rainfall of less than 5 mm day–1
, which becomes further worse over the regions with complex winter precipitation phase due to large miss bias [29
]. A comparison of three Runs of IMERG over the Sichuan basin of China for 2016–2018 revealed that all three Runs perform better during summer precipitation than autumn precipitation; however, IMERG-E underestimated wet precipitation substantially [30
]. A comprehensive analysis of versions 5 and 6 products of the IMERG suite over Iran for June 2014 to June 2018 against 76 rain gauge observations showed an improvement in V6 than V5, especially for near-real-time products [31
]. The IMERG-E product unexpectedly showed a higher correlation with rain gauge observations than IMERG-F over the arid regions of the United Arab Emirates for the period of 2015–2017 [32
]. Better performance of IMERG-E than IMERG-L and IMERG-F was also reported for a tropical storm “Imelda” over the southeast coastal regions of Texas in the US when compared with Stage-IV radar precipitation estimates [33
]. In addition, it was observed that IMERG-F V5 was better than V6 over the global mountainous regions in the estimation of light and heavy precipitation because V6 utilizes total column water vapor to derive a motion vector [34
]. Although the IMERG-F V6 product was shown to be one of the best multi-satellite precipitation products over the Hindu Kush mountains of Pakistan, it has large uncertainty in the detection of light and moderate precipitation events [35
However, there are limited studies to comprehensively evaluate the IMERG precipitation products over India owing to distinct topographical characteristics (e.g., Figure 1
a). IMERG-F showed notable improvement over TMPA-3B42 in systematic error at basin scale over India across all precipitation intensities [36
]. In addition, IMERG-F showed better performance than TMPA-3B42 at a sub-daily scale across India [37
]. However, similar to other satellite precipitation products, IMERG has a rather larger bias over the orographic regions of the Western Ghats and foothills of the Himalayas [38
]. But, all these studies over India utilized an earlier version of IMERG products (e.g., V4 or V5) for a limited period (e.g., one monsoon season or few specific rain events). IMERG products were upgraded to V06B in 2020, and IMERG-F was retrospectively processed for TRMM-era as well. As IMERG V06B supersedes all prior IMERG versions [39
], it becomes imperative to assess the accuracy of the recent version of IMERG products for the Indian monsoon precipitation. Hence, the objectives of this study are as follows:
To quantify error characteristics of V06B near-real-time (IMERG-E and IMERG-L) and research (IMERG-F) products;
To assess the changes in error characteristics of the IMERG-F product from V05B to V06B;
To assess the consistency of error characteristics of IMERG-F V06B for pre-GPM and GPM periods.
This study is carried out over India at different spatial scales such as at the grid scale, sub-regional scale, and country scale for the southwest monsoon season spanning from June to September.
4. Summary and Conclusions
India receives about 80% of its annual rainfall from the southwest or summer monsoon spanning from June to September, which exhibits substantial spatiotemporal variability. In addition, diverse topography and a fairly good rain gauge network make India a good test-bed to evaluate any satellite-based precipitation product. However, very few studies dealt with the evaluation of IMERG products for the southwest monsoon season over the country. In this study, near-real-time (e.g., V6 of IMERG-E and IMERG-L) and research products (IMERG-F V6 and V5) of IMERG were comprehensively evaluated against the IMD rain-gauge-based dataset over India at a daily scale for the southwest monsoon period. The evaluation was also carried out for pre-GPM and GPM periods. The spatial distributions of different error metrics across the country showed similar performance by both IMERG-E and IMERG-L estimates in precipitation estimation with marginally better performance by IMERG-L over IMERG-E. However, near-real-time products had rather larger errors than IMERG-F V6 estimates. IMERG-F V6 showed distinct bias patterns from IMERG-E and IMERG-L estimates over the west coast associated with complex terrain and precipitation processes. Bias in all-India daily mean rainfall in the near-real-time IMERG products was about 2–3 times larger than the research product. However, there was no considerable change in error metrics in IMERG-F V6 observed compared to the IMERG-F V5 product. Both near-real-time and research products showed similar characteristics in the detection of rainy days. IMERG-F V6 exhibited an overall better performance in precipitation estimation and detection of rainy days during the GPM period (2014–2017) than the pre-GPM period (2010–2013). Better performance of IMERG during the GPM period than during the pre-GPM period might be due to the availability of the DPR after the launch of the GPM Core Observatory, which provides better calibration than the TRMM-PR for IMERG estimates.
Furthermore, IMERG products were evaluated at all-India and sub-regional scales for different precipitation intensity intervals. The contributions of different rainfall intensity intervals to the total monsoon rainfall were captured well by the IMERG V6 estimates (e.g., IMERG-E, IMERG-L, and IMERG-F), but they underestimated the frequency of light to very heavy rainfall intensities over all-India, central India, and west coast regions. Additionally, results indicated that IMERG V6 products under-detected and overestimated light rainfall intensity, which needs to be improved in the next release. Results of this study clearly revealed that the error characteristics of IMERG products differ with region, topography, precipitation process, and product release version.The results of this study would be useful for both end-users and algorithm developers. Nevertheless, there is a need for an extensive evaluation of IMERG products at a sub-daily scale using radar and automatic weather station datasets over India in order to assess their capabilities in the diurnal cycle representation. The integration of ground-based weather radars and surface parameters such as soil moisture and terrain elevation in the multi-satellite precipitation product would further enhance its accuracy for hydrological applications.