Spatiotemporal inter-comparison of satellite precipitation products over India

doi:10.21203/rs.3.rs-1508855/v1

Region

Satellite Precipitation Products (SPPs) were statistically evaluated for their usefulness at different spatial, temporal, seasonal, altitude, and rainfall intensities scales over India.

Study Focus

The study evaluates the performance of World Meteorological Organization (WMO) Space-based Weather and Climate Extremes Monitoring Demonstration Project (SEMDP) Global Satellite Mapping of Precipitation (GSMaP) Gauge Adjusted Near Real Time Product (GNRT), Integrated Multi-satellitE Retrievals for GPM (IMERG) and Climate Prediction Center morphing method (CMORPH) for three years (2015-2017) using daily gridded gauge data from Indian Meteorological Department (IMD). The statistics used are visual interpretation (histograms, CDF plots), yes/no dichotomous [(Bias score (B), probability of detection (POD), false alarm ratio (FAR), probability of false alarm detection (POFD), accuracy (ACC), threat score (TS) and skill score (SS)] and continuous variable verification statistics [correlation coefficient (CC), relative bias (RB), mean absolute difference (MAD) and root mean square error (RMSE)].

New Meteorological Insights for the Region

The study found that GNRT is better than other SPPs at different temporal, spatial, altitudinal and rainfall frequency bands. The spatial variability of the metrics are similar but maximum CC >0.8 and least RMSE <10mm was recorded for GNRT over Central India. Additionally, all SPPs underestimate the precipitation amount in the Northeast and overestimate the same in Southern India. The precipitation detection score is good except in the Himalayan foothills and the Southern coast. All the SPPs are able to represent well the CDF for the precipitation intensities. Among all SPPs, the elevation has the least impact on GNRT, however significant elevation error has been found for CMROPH. At first glance, the gauge adjusted GNRT and IMERG_F perform better than the other SPPs. The results of the study can be utilized for the disaster risk and hydrological modelling and irrigation system analysis in the rain gauge data poor regions of the country.

GPM

GSMaP

IMERG

CMORPH

Spatio-temporal

India

High-resolution in space and accurate long-term precipitation data are mandatory for analysis of extreme events (floods and droughts) for the management of disasters. Obtaining accurate high spatio-temporal resolution rainfall data is a challenge in developing countries like India where rain-gauge networks are sparse (Kumar Singh et al., 2019). Moreover, the situation gets worse in the trans-boundary regions due to limited data sharing (Thu and Wehn, 2016). In this context, satellite precipitation products (SPPs) can be very useful particularly in trans-boundary and data sparse basins where there is large data latency (Kumar et al., 2016).

Several multi-satellite precipitation retrieval algorithms from NASA/JAXA missions such as Tropical Rainfall Measuring Mission-Multi-Satellite Precipitation Analysis (TRMM- TMPA) (Huffman et al., 2007), the Precipitation Estimation from Remotely Sensed Imagery Using Artificial Neural Networks (PERSIANN) (Hsu et al., 1997, Sorooshian et al., 2000) and Climate Prediction Centre morphing technique (CMORPH) (Joyce et al., 2004). Among all of the SPPs developed so far, TRMM TMPA has been extensively used for hydrological, metrological and water resources management studies over tropical and subtropical regions world-wide (Lakshmi et al., 2018; Sun et al., 2018). The TRMM TMPA has been used in several studies over United States (Hashemi et al., 2020, 2017), India (Mondal et al., 2018). However, TRMM mission has come to end on 8th April 2015 after over 17 years extensive services of productive data gathering.

The Global Precipitation Measurement (GPM) Core Observatory (GPM CO) spacecraft has been launched by National Aeronautics and Space Administration (NASA) and Japanese Aerospace Exploration Agency (JAXA) on 28th February 2014 as the successor to the TRMM mission. GPM CO carries the microwave imager (GMI) and latest Ku/Ka Doppler dual-frequency precipitation radar (DPR) in its main satellite. GPM inherits the advantage of TRMM satellites to detect the tropical precipitation and believed to provide enhance detection for micro and solid precipitation (0-1mm/d) because of its DPR and GMI (Draper et al., 2015). Thus, GPM CO is believed to provide more precise and accurate global precipitation than TRMM mission for global hydrological and meteorological analysis. Also, “GPM CO functions in a non-sun-synchronous orbit with an inclination angle of 65°, thus it is able to sample precipitation across all hours of the day from the tropics to the Arctic and Antarctic circles and for observing hurricanes and typhoons as they transition from the tropics to the midlatitudes”(Skofronick-Jackson et al., 2017). The GPM era has various SPPs: Integrated Multi-Satellite Retrievals for GPM (IMERG) by NASA (Hou et al., 2014) and Global Satellite Mapping of Precipitation (GSMaP) by JAXA (Kubota et al., 2007). The IMERG has mainly three SPPs: IMERG Early Run (IMERG_E), IMERG Final Run (IMERG_F) and IMERG Final Run (IMERG_F).

Similarly, GSMaP has three SPPs: near real time product (GSMaP_NRT), Moving Vector with Kalman filter product (GSMaP_MVK) and gauge-calibrated standard products (GSMaP_Gauge). To analyse and evaluate the extreme events over Southeast Asia and the Pacific, the World Meteorological Organization (WMO) initiated the space-based weather and climate extremes monitoring demonstration project (SEMDP) which has also produced SPPs using GSMaP inputs. SEMDP SPP is produced by Earth Observation Research Center/Japan Aerospace Exploration Agency (EORC/ JAXA). “SEMDP project provides mean precipitation estimates that has been derived from GSMaP at hourly, daily (00–23 UTC), pentad (5 days), weekly (Monday–Sunday), 10-day, and monthly temporal scale. It has spatial resolution of 0.1° (~ 10km) and covers the geographical area between 40°N to 45°S and 50°E to 160°W”(Yuriy Kuleshov et al., 2012). The key precipitation product of SEMDP is gauge adjusted near real time precipitation product (SEMDP_GSMaP_GNRT) which is being abbreviated as GNRT in this manuscript.

Each SPP has a unique algorithm to estimate the amount of rainfall despite similar satellite data input from GPM CO and GSMaP. Not only that, each SPP has their own specific gauge adjustment techniques to reduce the error. Due to the specific input data, estimation algorithms and gauge correction techniques for each SPP, the accuracy of these products is dissimilar in space and time (Kumar et al., 2016; Sunilkumar et al., 2015). Therefore, the quantitative and qualitative evaluation of considered SPP at different time and space is critical before considering them for hydro-meteorological applications.

Recently, many researchers have focused on the validation of SPPs in India and other parts of the world to see the usefulness of the data at different Spatio-temporal scale. Most of the findings of these studies have highlighted the superiority of the GPM (IMERG) SPPs over those from the TRMM era. (Dezfuli et al., 2017; Mitra et al., 2018; Zhou et al., 2020b). Also, many researchers have evaluated the TRMM, GSMaP, and CMORPH SPPs (Kumar et al., 2016; Prakash et al., 2016) over India for its ability to detect the rain and its magnitudes. (Prakash et al., 2016) have evaluated the performance of TRMM Multisatellite Precipitation Analysis (TMPA)-3B42 (V7 and RT) and GSMaP (MVK and NRT) against IMD gridded gauged precipitation data over India during monsoon. They interpreted that all SPPs are able to represent the large-scale spatial monsoon features but observed regional biases. The TMPA-3B42RT over estimate and under estimate the rainfall by 21% and 22% respectively. Also, TMPA-3B42V7 has rated the best among all in terms of skill score and Categorical verification. Finally, they concluded that TRMM-TMPA products can be used with much confidence than GSMaP been over monsoon India. (Mitra et al., 2018) evaluated the GPM IMERG (daily merged satellite gauge (DMSG)) and INSAT-3D SPPs (Hydro-Estimator Method (HEM), INSAT Multi-spectral rainfall (IMR) and Quantitative Precipitation Estimation (QPE)) with the IMD daily gridded gauge precipitation data for India. They found that the HEM has good correlation (r > 0.7) and skill score (> 0.8) in detecting heavy rainfall events with an accuracy of ± 20mm/hr and also, suggested that HEM, IMR and QPE are not able to detect orographic enhancement at higher elevations. They concluded that the “HEM is better for heavy rainfall and could be used in the future for meteorological and hydrological models while QPE and IMR have better skills for light to moderate rainfall which can be used for agriculture and ground water recharge”. Venkata Lakshmi Kumar et al., 2019 have validated TRMM-TMPA and IMERG SPPs against IMD gridded data for 20 years (1998–2017) and found that “Satellite rainfall data sets have very less bias with relation to the IMD over the monsoon core region of India, also during the El Niño and La Niña years, satellite rainfall could show better features than IMD when the normalized power is being considered”. Thakur et al., 2020 compared the IMERG final run precipitation data with the IMD gridded data over monsoon India and found that “IMERG is a potential source for adequately reflecting the ground gauge-gridded data of categorical rainfall amounts, from very light rain (trace–2.4 mm/day) to very heavy rain (about 115.6–204.4 mm/day?), however, the IMERG does not capture satisfactorily the extreme heavy rain events (≥ 204.5 mm·day–1)”. (Reddy et al., 2019) compared different SPPs from GSMaP-V6 (NRT & moving vector with Kalman filter (MVK)), IMERG-V4 (near real time-NRT and final run-FNL products), INSAT3D (Indian National Satellite System (INSAT) and National Centre for Medium Range Weather Forecasting (NCMRWF) Merged product from IMD with gridded gauged precipitation data from IMD at daily, monthly and seasonal scale. They found that IMD-NCMRWF merged and IMEGRG_FNL products have much better agreement with IMD gridded gauge precipitation data than GSMaP products. Also, they found that GPM based GSMaP and IMERG products have better ability of detection than INSAT3D. Mondal et al. 2018 studied the spatio-temporal trend in the rainfall data over India using Multisatellite High Resolution Precipitation Products (HRPPs):Tropical Rainfall Measurement Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) V7, Climate Prediction Center Morphing (CMORPH) V1.0, Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), and Multi-Source Weighted-Ensemble Precipitation (MSWEP) version 1.1. They used Mann–Kendall (MK) and modified Mann–Kendall (MMK) tests are applied for analyzing the data trend and the change is detected by Sen's Slope using all datasets for annual and seasonal time periods. The spatio-temporal trends for the HRPPs have been validated using IMD 0.25^o gridded data. They concluded that that the TMPA product is best in terms of accuracy and PERSIANN in terms of annual and monsoon trend.

Several attempts have been carried out to evaluate the performance of IMERG, GSMaP SPPs with other SPPs viz. TRMM, CMROPH, INSAT3D etc. over India and most of the studies found the weakness in the GSMaP SPPs (NRT, MVK and Gauge) for extreme events such as floods and droughts. The World Metrological Organization (WMO) initiated a space-based weather and climate extremes monitoring demonstration project (SEMDP) initially for two years (2018–2019) and introduced SEMDP SPP (GNRT V6). The SEMDP SPP (GNRT V6) uses cloud-motion advection method by utilizing IR images to derive cloud motion vectors, which are then used with Passive Microwave based precipitation estimates. Therefore, the suitability and superiority of GNRT V6 product over the other SPPs needs to be evaluated before considering it for other uses.

Until now, most of the studies were focused on comparison of TRMM and GPM. However, the comparison between the IMERG, GSMaP (especially GNRT) and CMORPH at different spatiotemporal scale over India has yet to be attempted. In this study, four SPPs [two pure satellite estimate (CMORPH, and IMERG_E) and another two gauge adjusted satellite estimates (IMERG_F and GNRT)] based on differing algorithms and satellite inputs are adjudged for their accuracy based the visual verification, yes/no dichotomous and continuous variable verification method over the India. The results of the study will help end users to choose the best SPP for their specific application, location, elevation bands, rainfall frequency, temporal and seasonal scales.

2.1. Study area

India, with a total geographical area 32.87x10⁵ Sq.km (2.4% of the world’s land area), is the seventh largest country in the world. It is situated on the northern part of the Indo-Australian plate in the Indian subcontinent (8^o4^'-37^o6^'N and 68^o7^'-97^o25^'E). Also, India has more than 1.3 billion humans including 8% of the global biodiversity. Straddling the Palearctic and Indo-Malayan realms, India is 17th mega-diverse country of the world having Western Ghats and the Himalayan range globally known biodiversity hotspots. The vast geography of India is divided into six physiographic regions viz. northern Himalayan mountains, Peninsular Plateau [(mountains-Arawali, Vindhayachal and Satpura ranges, ghats-eastern and western ghats) and plateaus (Chhota Nagpur, Malwa, Southern garanulite, Deccan and Kutch Kathiawar)], Indo-Gangetic Plain, Desert, Costel Plains and Islands (Andama and Nicobar Islands, Lakshadweep). The Indian mainland has maximum elevation ~ 8600m somewhere in the Himalayan range and minimum − 70m towards south coast (Fig. 1). The range of physiographic regions and varying topography makes India as one of the most important regions in south-east Asia for SPPs validation. In the presents study, district centroid gauge points (590) are considered for the validation of SPPs using the IMD gridded data.

2.2. Description of Satellite Precipitation Products (SPPs)

In the SEMDP, WMO requested to provide the near-real time (NRT)-basis Gauge adjusted GSMaP SPP to monitor the extreme weather and climatic events from space over Southeast Asia and the Pacific (Yuriy Kuleshov et al., 2012). With that, JAXA has developed GSMaP_Gauge_NRT version 6 (GNRT V6) in collaboration with Tokyo Metropolitan University (Profs. Ushio and Mega). In this method, estimate from the GSMaP_NRT are adjusted using optimization model with the parameters \([C, {\sigma }_{v} ({\sigma }_{w}, {\mu }_{w}\left)\right]\)calculated from GSMap_Gauge (Gauge adjusted standard version) over the past 30 days (Kubota et al., 2019; Mega et al., 2014). The spatial resolution of the parameters that are used in the optimization model to compute GNRT SPP, is 5^o. But later it was realized that 5^o of the grid resolution for the parameters is not suitable for coastal regions including Malay Peninsula. The product is available for the SEMDP members since Oct. 2018 and then to public since Dec. 2018 Via JAXA’s ftp site (ftp://semdp:[email protected]/GNRT/DATA/). The 24 hourly GSMaP SPP granules are cumulated (from 4UTC of Previous day to 3 UTC of current day in 24-hour UTC time) to get local daily (8:30 IST) precipitation estimates.

Multi-satellite-gridded precipitation product IMERG Level 3 is an advanced newly developed data set of GPM mission. GPM produces orbital and gridded data sets at three levels of data processing. The calibrated brightness temperature (1C-GMI) orbital product is level-1 data, however combined GMI-DPR precipitation estimates are classified as level-2 orbital dataset (Rios Gaona et al., 2016). Level-3 products are gridded that combines GMI-DPR rainfall average (3-CBM) or rainfall estimates combined from all active and passive microwave datasets in the GPM constellation, and this is referred to as IMERG (Huffman et al., 2019, 2014). There are mainly three products provided by IMERG viz. 1. near real time ‘Early Run’ (IMERG_E) uses only forward morphing 2. ‘Late Run’ (IMERG_L) uses both backward and forward morphing and 3. ‘Final Run’ uses backward and forward morphing along with monthly gauge analysis. The half hourly IMERG products (IMERG_E & IMERG_F) are obtained from site(https://disc.gsfc.nasa.gov/datasets/GPM_3IMERGHH_06/summary?keywords=%22IMERG%20final%22). The 48 half-hourly IMEERG SPP granules (3:30UTC previous day to 3:00UTC of current day in 24-hour UTC time) are added to find local daily (8:30 IST) precipitation estimates. Then, the daily accumulated precipitation amounts are multiplied by factor 0.5 to convert the half hourly rain rates into mm/h.

The CMORPH algorithm produces very fine resolution precipitation estimates with near global coverage. “This technique uses rainfall estimates exclusively derived from low orbit satellite microwave observations and whose features are transported using spatial propagation information that is obtained entirely from geostationary satellite infrared (IR) sensors”(Joyce et al., 2004). At present, “CMORPH incorporates precipitation estimates derived from the passive microwave sensors − 14 and 15 (SSM/I), 16, 17 and 18 (AMSU-B), NOAA-15, DMSP-13, AMSR-E, and TMI sensor onboard in the TRMM spacecraft but technique is extremely flexible to incorporate precipitation estimates from any microwave satellite sensors” (Kumar et al., 2016). The three-hourly precipitation estimates are downloaded from ftp site (https://www.ncei.noaa.gov/data/cmorph-high-resolution-global-precipitation-estimates/access/hourly/0.25deg/). Then, 24 hourly CMORPH SPP granules are cumulated (from 4UTC of Previous day to 3 UTC of current day in 24-hour UTC time) to get local daily (8:30 IST) precipitation estimates.

2.3. Gauge Precipitation Data

As a benchmark, daily gridded-gauge rainfall data from Indian Metrological Department (IMD) having spatial resolution of 0.25^ox0.25^o over India (Pai et al., 2014) is used in this study. This product uses the daily rainfall records from 6995-gauge stations across country following various quality controls. On an average 3100-numbers of daily gauge precipitation data points are used to develop this product. It can show many features of rainfall, including spatial gradient in orographic rainfall (Pai et al., 2015).

The data for district centroid points (as shown in Fig. 1) for each SPPs and benchmark gridded-gauge rainfall is extracted using nearest neighbourhood method. Thus, each considered grid point particular granules has been compared with nearest gridded gauge points of the benchmark IMD data set. The details of the spatial/temporal resolution and coverage period of the data sets used in this study is presented in Table 1.

2.4. Methods

In this study, SPPs are compared with IMD gridded gauge data on daily, monthly and seasonal basis for three years from 1st January 2015 to 31st December 2017. To evaluate the performance of SPPs across different precipitation intensities, precipitation intensity is categorized according to the IMD/WMO with minor adjustments as: (a) 0-0.1 mm/d - trace precipitation, (b) 0.1–2.5 mm/d - Light precipitation, (c) 2.5–7.5 mm/d - Moderate precipitation, (d) 7.5–50 mm/d - heavy precipitation, (e) > 50 mm/d - Very heavy precipitation. The trace precipitation are removed from IMD gridded-gauge data and corresponding SPPs before calculation. The interpretations are done on the basis of rainfall intensities listed above however, the calculation bins are evenly spaced, starting from zero with 5mm increment per bin. Also, SPPs are evaluated for different seasons viz. rainy (July-August-September), winter (October-November-December), spring (January-February-March) and summer (April-May-June).

(Murphy, 1993) defined three types of ‘goodness’ to verify SPPs, viz., consistency, quality, and value. (Wilks, 2006) defined “a partial list of scalar attributes of forecast qualities: accuracy, bias, reliability, resolution, discrimination and sharpness”. (Stanski et al., 1989) divided verification of forecast into (a) visual verification (b) dichotomous or categorical variable verification, and (c) continuous variable verification categories. In this study, the SPP verification methods as proposed by Stanski et al. (1989) are used, which is successfully applied for SPPs verification in recent past in the different parts of the world including India (Dezfuli et al., 2017; Kumar et al., 2016; Prakash et al., 2016; Thiemig et al., 2012; Zhou et al., 2020b).

2.4.1. Visual verification

Visual verification examines the difference between SPPs and gauge data. It is a method to find instantaneous decision on the error between SPPs and gauge data. Researchers often use the graph plots such as histogram, time series, probability distribution function (PDF) and cumulative distribution function. Among these, few fail to find either discrete or continuous error however, CDF is utilized to adjudge both continuous and discrete error between the SPPs and gauge data and therefore, it is used in this study for visual verification.

2.4.2. Yes/no dichotomous

Wilks (2006) has proposed the yes/no dichotomous (categorical variable verification) based on the statistical quantification of scaler attribute of the SPP (forecast). In the yes/no dichotomous, ‘yes’ implies that the event will happen and ‘no’ implies that the event will not happen. The verification may be done in the thresholds of rain events (1, 2, 5, 10, 20, 50 mm/day) based on the time scale of the dataset WMO/TD-No. 1485 (2008). In this study a minimum threshold of 0.1 mm/day is considered to distinguish ‘yes’ and ‘no’ events as per the Lu and Panmao, 2012 A contingency table for the query of ‘yes’ and ‘no’ estimation between SPP and gauge data is prepared based on the (Brown et al., 1997) and it is presented in Table 2.

From the Table 2, four combinations of chances are looked between the gauge and SPPs estimates. These are as follows-

(i) Hit: both SPP and gauge believe that the event will occur

(ii) Miss: gauge views event as occurring, but SPP does observe the event

(iii) False alarm: SPP views the event as occurring but gauge does not observe the event

(iv) Null or correct negative: both SPP and gauge do not observe the event

Based on the ‘yes/no’ dichotomous a variety of statistics were derived by Wilks (2006) and WMO/TDNo. 1485 for the validation of SPPs. The statistical indicators used in this work are presented in the Table 3.

2.4.3. Continuous variable verification

The continuous variable verification measures how the value of SPP estimates differs with the value of gauge observations. This can be done by various summary scores. The summary score indicators used for this study are presented in Table 3.

3.1. Performance of SPPs at multiple temporal scales

3.1.1. Annual Visual Interpretation

The spatial distribution of annual precipitation of IMD, GNRT, CMORPH, IMERG_E and IMERG_F and differences in annual precipitation of SPPs from IMD data over India are presented in Fig. 2. The annual precipitation shows variation in rainfall from east to west and north to south and among all SPPs. The ground based IMD shows highest precipitation in the north eastern states and along the west coast, while minimum in the western deserts of Rajasthan. GNRT has shown similar annual precipitation map similar to IMD in the high rainfall regions of north-east India and lower rainfall regions of the desert in Rajasthan. IMERG products had low predictive accuracy for high rainfall regions of north-east, however have shown good annual rainfall prediction for southern-central India. Other side, CMORPH had singular annual rainfall zone > 950mm throughout the country and thus it completely failed to depict the rainfall variability of the country.

The difference of annual precipitation \((SPP Annual-IMD Annual)\) has also been calculated for the studied period. GNRT shows overestimation the high rainfall zones (> 1500mm) and underestimated in the eastern coast, however it has better estimation (annual differences − 100 to 100mm) over 80% of the flat regions in the country. Among IMERG products, IMERG_F has lower difference than IMERG_E, while both have over-estimated over 80% of the area especially in northern part of the country. CMORPH have shown that the product has less annual difference from IMD in the very north and central part of the country. Overall, GNRT have shown the least annual differences with IMD as compared to other SPPs.

3.1.2. Daily Scale

For visual verification at daily scale, basin-wide scatter-density plot between gauge precipitation and SPPs are presented in Fig. 3. From the figure, it is evident that GNRT vs gauge precipitations have minimal outlier than the others SPPs and therefore GNRT seems closer to rain gauge observations. Apart from GNRT, IMERG_F run is better than IMERG_E and CMORPH.

The statistical metrics at basin-wide daily scale is presented in the Table 4. These correlation results show that the GNRT performs better than the other SPPs. GNRT is having highest CC (0.92), corresponding to the lowest MAD (0.93), MSE (2.30), RMSE (1.51) and RB (12.36%). This exhibits a closer alignment of GNRT with the gauge observation. IMERGE_F [CC (0.68), MAD (2.11), MSE (12.35), RMSE (3.51) and RB (-11.28%)], IMERG_E [CC (0.65), MAD (2.25), MSE (16.12), RMSE (4.01) and RB -19.67%)], and CMORPH [CC (0.33), MAD (3.81), MSE (75.27), RMSE (8.67) and RB 27.94%)] are ranked in second, third and fourth places respectively, using per daily basin wide-statistics. The RB statistics reveals that overall IMERG SPPs are underestimating the precipitation estimates at daily scale, however the gauge adjusted IMERG_F has slightly improved the results over IMERG_E.

3.1.3. Monthly Scale

Figure 4 presents the time-series for visual verification and continuous variable verification statistics. Figure 4 (a) shows the temporal trend of the SPPs and gauge precipitation at monthly scale from January 2015- December 2017. Each SPP captured the temporal trend similar to the gauge precipitation. The precipitation mainly occurred in the monsoon months (June-September) and also each statistic displays a symmetric distribution along the timeline indicating that the SPPs have an annual periodic variation. In addition, the statistics of each SPP also presents the seasonality similar to the gauge precipitation.

Figure 4 (b) shows that the GNRT is positively correlated with the gauge data. However, the IMERG_E and IMERG_F are negatively correlated with the gauge data. The negative CC value of the IMERG SPP reveals that when intensity of gauge precipitation increases then the intensity of IMERG SPP decreases and vice-versa and thus, it is underestimating the precipitation for higher intensities at daily scale (as seen in section 3.1.1). The RMSE, RB and MAD are minimum for SEMDP’s GNRT, which shows that it has best agreement with the gauge precipitation. These statistics are minimum for IMERG_F (Fig. 4c-e). GNRT and CMORPH are characterized by overestimation, while the IMERG SPPs are characterized by underestimation. We observe that the GNRT gauge adjusted algorithm provides the best precipitation estimate among the evaluated SPPs. All SPPs show better agreements in the monsoon season (June-September) and present relatively higher errors in the dry-season (October-May) but GNRT have predicted precipitation well throughout the year.

Figure 5 shows the temporal pattern of ‘yes/no’ dichotomous (categorical variable verification) statistics. The POD_rain is > 0.6 for the GNRT throughout the year and close to 0.8 in the monsoon season, while other SPPs are consistently below the GNRT except for few deviations, for example IMERG_F in the monsoon period. We observed that GNRT can probably detect rainy events in more than 80% cases over the country during monsoon season and fairy good throughout rest of the year. Similarly, POD_no−rain graph for GNRT is always above other SPPs and close to 1, and it indicates that the GNRT is the best choice to detect no-rain events among the SPPs. The FAR_rain and FAR_no−rain is always < 0.4 for the GNRT, while the other SPPs are having greater value for these statistics. This indicates that the GNRT has the least number of false alarms for both rain and no rain events than other SPPs. Also, another statistics POFD (probability of false alarm detection, best value equals zero) is < 0.2 for GNRT, which is an indicator of the efficiency of the product to detect the rain events. The accuracy (ACC) of GNRT ranges from 0.75–0.95 (a perfect accuracy score equals 1) and the plot is above the other SPPs, thus, GNRT estimation has better accuracy. The higher value of TS for GNRT compared to the others revealed its better agreement with the gauge data. The SS score for GNRT is always above 0.5 while for other SPPs it is less than 0.2, which indicates the capability of GNRT for skilful detection over other SPPs. The SS = 1 indicates perfect detection of gauge precipitation by satellite precipitation. SS > 0 indicates a skilful detection better than the reference, whereas SS < 0 indicates a less than skilful detection (lower than reference detection). The BIAS score (perfect value is 1) for GNRT is found to range between 0.9-1.0, indicates near to perfect detection of number of rainfall events. Overall, GNRT shows better precipitation detection ability in terms of yes/no dichotomous statistics than other SPPs.

It is worth mentioning that yes/no dichotomous statistics of GNRT present different trends from other SPPs throughout the study period. It is obvious that the POD and CC of GNRT are much higher than those for the other SPPs. This could be due to the nature of the GNRT algorithm which calculates the calibration parameters \([C, {\sigma }_{v} ({\sigma }_{w}, {\mu }_{w}\left)\right]\) from GSMap_Gauge during past 30 days. The gauge corrected products GNRT and IMERG_F have effectively reduced the error than simple satellite estimates CMORPH and IMERG_E. The possible inferiority of gauge adjusted IMERG_F than gauge adjusted GNRT may be due to the monthly gauge data from GPCC used in IMERG_F. This also reveals that the introduction of finer spatiotemporal gauge data in the gauge calibration of SPPs can effectively reduce the error.

3.1.4. Seasonal scale

This study also includes the assessment of seasonal performance for considered SPPs. The seasonal performance of SPPs in terms of continuous variable verification statistics viz. MAD, MSE, RMSE, RB and CC are presented in Table 5. The GNRT has better performance, followed by IMERG_F, IMERG_E, and CMORPH, respectively. Among all SPPs, GNRT achieves the best estimate with CC between 0.65–0.7.0, MAD between 0.80-7.0 mm, MSE between 16–237 mm, RMSE between 4–15 mm and absolute value of RB between − 8–20%. The results also show that IMERG SPPs underestimate the precipitation in rainy, winter and summer seasons, and overestimates during the spring season. CMOROH overestimates the precipitation amounts for all seasons. GNRT’s nature of over estimating in the rainy season and under estimating in winters does not prove detrimental to the studies related to flood and drought during respective seasons. GNRT is found to perform best during rainy and summer seasons (monsoon months) however, its performance reduces during winter and spring (no-monsoon months). It implies that the GNRT is less capable of capturing the rainfall due to westerns disturbances and monsoon which occurs during winters in the southern coastal states of the country.

The performance of ‘yes/no’ dichotomous verification statistics for each season is presented in Fig. 6. Compared with other SPPs, the SEMDP based GNRT shows better precipitation detection capability in terms of higher value of metrics Bias score (B), POD (rain, no-rain), ACC, TS, SS and lower value of metrics FAR (rain, no-rain), PODF. It implies that GNRT has detection of rain events detection and lower false alarms. In contrast, CMORPH, IMERG_E and IMERG_F have lower value of Bias score (B), POD (rain, no-rain), ACC, TS, SS and higher value of metrics FAR (rain, no-rain), PODF, reflecting the incapability of these SPPs in detecting the rain and no rain events. Although the GNRT performs well in general but during spring IMERF_F displays a small advantage over GNRT in terms of highest value of metrics Bias score (B), POD (rain, no-rain), ACC, TS, SS and lowest value of metrics FAR (rain, no-rain), PODF. This shows that GNRT has better capability for detecting rain and no rain events during monsoon months, while relatively poor detection rate during non-monsoon months.

3.2. Spatial variation in the performance of SPPs

The spatial distribution of continuous variable verification statistics for SPPs can be useful to understand the error propagation in hydrological and metrological studies. Figure 7 presents the spatial distribution of CC, MAD, MSE, RMSE and RB over India. It is observed that the SPPs have significant variation in the statistical metrics over the area. Among all SPPs the GNRT has shown very good correlation (> 0.7) in the flat/low elevation areas (approximately 80% of the country’s geography), however it reduces towards the very north (Uttarakhand, Jammu & Kashmir, Himanchal) and north-east states (specifically - Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim and Tripura). Apart from GNRT, IMERG_E and IMERG_F have also shown good correlation (> 0.6) for central-eastern states (West Bengal, Odisha, Andhra Pradesh and Karnataka). Also, from the correlation plot (Fig. 7), it is observed that IMERG SPPs have better CC in a small strip (south-west to north-east), and the reason behind it needs to be determined to improve the efficiency of this estimation algorithm for other part of the country. The spatial performance of SPPs are similar for metrics MAD, MSE, RMSE, which is fairly low for GNRT and high for other SPPs. For about 95% stations GNRT achieves RMSE value < 10mm/day and for only ~ 5% stations, in the high-altitude regions of very north and north-east, has RMSE value > 10mm/day. Similar to RMSE, the RB for GNRT is low (\(\pm\)5mm/day) for ~ 95% of the stations, while it increases towards north-east. Among the SPPs, IMERG products are over estimating the precipitation amount in the north-east portion (~10%) and under estimates for rest of the area (~90%). Although, GNRT has significantly reduced the error, but still has the tendency to slightly over estimate the precipitation for more than ~ 60% of the stations.

The spatial variability of ‘yes/no’ dichotomous variable verification metrics for all stations is presented in the Fig. 8. The metrics have considerable variation spatially over the country. BIAS score of rain event detection is fairly good (> 0.9) for GNRT throughout the country however, IMERG and CMORPH SPPs have higher BIAS score for north-west part of the country. The POD_rain (probability of rain event detection) for GNRT is close to 0.8 for the whole country except for the western coastal belt. IMERG and CMORPH SPPs does well in detecting the rain events in the central part of the country and are unable to detect events for the south-west coastal states and in the north-east where rainfall is mainly affected due to western disturbances and orography. The spatial map of POD_no-rain exhibits the results similar to POD_rain. The FAR_rain and FAR_no-rain do not perform well (> 0.3) for the coastal and Himalayan foothill regions (north and north-eastern parts of the country). The POFD of GNRT is very good (~ 0) throughout the country while IMERG and CMORPH show good POFD for south-west and poor POFD towards north-west portions of the country. The metrics measuring the skilful based on the dichotomous (ACC, TS and SS) have shown similar results to each other viz. skilful detection (SS > 0.7) for GNRT through the country, while reasonably good skilful (SS > 0.5) detection for IMERG and CMORPH SPPs over the Central India. It is evident from ‘yes/no’ dichotomous metrics that all the SPPs have limitations to detect the rain and no rain events in the western coastal region and northern Himalayan region of the country where the rainfall due to western-disturbances and orography plays an important role. Therefore, the weakness of the SPPs is evident for western disturbance and orographic regions similar to that of Tropical Rainfall Measuring Mission (TRMM3B42 V7) (Bharti and Singh, 2015; Hunt et al., 2018).

3.3. Performance of SPPs in different intensity classes

Cumulative density function (CDF) is superior over the probability density function (PDF) since it can be used for discrete as well as for continuous variables (Progênio and Blanco, 2020). Therefore, CDF is used to evaluate the agreement of rain-rate occurrence for SPPs with gauge precipitation. Figure 9 shows CDF of daily precipitation for gauge and SPPs events. The CDF is suitable to evaluate the performance of SPPs (Progênio and Blanco, 2020) divided into different intensity bins for each season over India. Results show that all SPPs follow the CDF pattern of gauge precipitation for entire period except CMORPH. Also, each of the SPPs slightly underestimate the precipitation events in this bin other than GNRT which tends to overestimate it. The GNRT’s CDF plot is closer to gauge CDF and hence GNRT agrees well with number of gauge precipitation events per bin for entire duration. During entire period, about 70% of the GNRT precipitation events fall inside light precipitation intensity band (< 4mm/day), closely matching the gauge precipitation events with very slight over estimation. The over estimation of gauge events by GNRT widens for moderate intensity bins (5-10mm/day) where around 15% of precipitation events during entire period take place. Again, for intensity bins > 10mm/day, the GNRT and gauge rain events are in good agreement with each other.

The performance of SPPs in different intensity bins also varies seasonally. Throughout the country light intensity events are absent in the rainy and summer seasons. During rainy season the rain events that occur mostly correspond to moderate to very heavy intensity. In this season, GNRT CDF plot is close to gauge CDF plot, however it slightly over estimates the number of rain events for bins < 15mm/day. IMERG_F CDF plot is matching well with the gauge CDF plot for intensities 6-10mm/day however, for intensity bins > 10mm/day, it is underestimating the number of rain events. During the winter season, CDF plot of GNRT closely matches the gauge CDF plot except for a very little underestimation for intensity bins < 6mm/day however, other SPPs are completely mismatched to the gauge CDF. In spring, CDF plot of GNRT closely matches the CDF plot of gauge precipitation where as other SPPs are completely mismatched to the gauge CDF. In summer season the all SPPs do not match the gauge CDF plot, however CDF plot of GNRT follows the pattern of gauge CDF plot but over estimates the number of rain events in each intensity bin. Overall, GNRT is matches the number of rain events well for each intensity bin with very slight overestimation. In another hand, IMERG_F follow the CDF pattern of gauge precipitation but greatly underestimates the number of rain events in each intensity bin.

The continuous variable verification metrics of SPPs at different intensity bins over India is presented in Fig. 10. GNRT shows best agreement with the gauge precipitation among all SPPs in terms of lowest MAD, RMSE, RB and highest CC value at daily scale. Followed by GNRT, IMERG_F has better statistical metrics than IMERG_E and CMORP. IMERG_F has comes close to GNRT with highest CC (> 0.5), lowest RB and MAD at lower intensity bins (> 5mm/day). Except for GNRT, each SSP has decreasing and increasing trend for CC and RMSE, respectively with increase in the rainfall intensities. From RB plot [Fig. 10(d)], it is clearly evident that all SPPs overestimates the rainfall amount for intensities > 10mm/day however, underestimates the rain amount for intensities < 10mm/day. But among all SPPs, GNRT has least RB for every precipitation intensity bins which shows that it has the capability to capture the rain amounts for each rainfall intensity class.

The “yes/no” dichotomous (categorical variable verification) metrics (in mm) for different precipitation intensities at the daily time scale are presented in Fig. 11. GNRT shows better precipitation event detection rate in all intensity bins when compared the other SPPs, more specifically for precipitation intensities > 5mm/day. However, it has little underestimation error for rain events < 5mm/day. The metrics POD, ACC, TS and SS increase with increase in the intensities revealing that heavy rain events are more likely to be observed as compared to light rain events. This is also affirmed by the BIAS which increases with the increase in the precipitation intensities. In general, GNRT among all SPPs considered here has the best categorical variable verification metrics and hence shows greatest capability to detect the rain and no rain events for all intensity bins. The above comparison suggests that the SEDMP project-based gauge adjusted GNRT SPPs has improved rain events detection capability than other SPPs.

3.4 Evaluation of SPPs for different the elevation bands

The elevation dependence of error is evaluated with respect to “yes/no” dichotomous and continuous variable metrics. The continuous variable verification statistics for different elevation bins are presented in Fig. 12. In general, GNRT has highest CC and lowest MAD, RMSE and RB for each elevation bin than the other SPPs. Figure 12 (c) shows that for GNRT, CC has decreasing trend with respect to increase in elevation. The RB and MAD increase with increase in elevation for GNRT, indicating it is overestimating with increase in elevation. Interestingly, other SPPs have higher error than GNRT but no clear relation with elevation over the geographical area. Among IMERG products, IMERG_F has higher CC, and lower RB and MAD clearly indicating the improvement in the product due to gauge adjustment. However, for IMERG products, RB value displays a decreasing trend with elevation < 250m and a increasing trend for elevations greater than 250m. This indicates that IMERG products underestimate the rainfall magnitude in regions with elevation less than 250m and overestimate in the regions with elevation greater than 250m.

Figure 13 presents the “yes/no” dichotomous or categorical variable verification metrics with respect to elevation bins for India. It is evident from the results of ACC, TS and SS [Figure 13(g, h, i)] that the accuracy of rain events detection increases with increase in elevation for each SPPs and it is best for GNRT among all. The turning point for increasing predictive accuracy was observed for elevation greater than 750m. The bias score is close to 1 (best score) for GNRT but it starts reducing for elevation > 3500m. Similar to the bias score, the trend for POD_rain and POD_no−rain exhibited a decreasing trend with elevation greater than 3500m. The FAR_no−rain [Figure13 (f)] with respect to elevation bands shows that the false alarm of no rain events increases for elevation bands > 1500m. The POFD (best score 0) is close to 0.1 for GNRT for regions with elevation greater than 750m, increases to more than double for regions with elevation greater than 750m. Apart from GNRT, IMERG_F had second best values for TS, POD and FAR in regions with elevation greater than 750m. In overall, GNRT has presented the lowest FAR and POFD and highest POD for rain and no rain events this exhibits superiority of GNRT over other SPPs to detect the rain events in the elevation bins.

4.1. Effect of gauge calibration on the performance of the SPPs

The results of SPPs over India show that gauge adjusted products viz. GNRT and IMERG_F perform better than the pure satellite-based products (IMERG_E, CMORPH). This is mainly due to the inclusion of gauge-calibration based on the high quality observation as in case of GNRT (Kubota et al., 2019; Mega et al., 2014).

The continuous variable verification metrics also indicate the superiority of gauged-SPPs (especially for GNRT) at different temporal, spatial and elevation scales in-terms of improved CC and reduced RB, MAD and RMSE. It indicates that the gauged-SPPs produce more accurate precipitation estimates at different temporal and spatial scales. Similarly, “yes/no” dichotomous (categorical verification) metrics indicate that the gauged-SPPs (especially for GNRT) have effectively detected the rain events and omitted the false alarms and miss reporting of the rain events.

Overall, SEMDP based GNRT is superior than other SPPs. Monthly gauge-adjusted GSMaP data used for parameter derivation at 5^o resolution could be one of the main reasons. GNRT lags compared to IMERG_F, and the reason could be monthly GPCC gauge analysis employed by IMERG rather than daily data as used in GNRT. Recently, GPCC has started computing daily gauge-adjustments based on the station’s precipitation data and therefore IMERG_F may further get improved in the future versions.

4.2. Effect seasons on the performance of SPPs

The performance of SPPs vary with the seasons and GNRT has better metrics in general compared to the others. During winter-spring, GNRT has minimum CC and negative RB, indicating the underestimation error for this duration. This could be because of incapability of GNRT to detect and estimate the lower intensity rainfalls occurring due to receding monsoon and winter thunderstorms. In summer-rainy seasons, GNRT has detected rain events well and better than other SPPs, reveals the capability of GNRT to detect and estimate the heavy rain events and thus, can be used for the flood related studies in the south-east Asia. IMERG_F has shown slight advantages during springs over GNRT (in terms of CC) and thus, IMERG_F can be utilized for drought stress monitoring for Rabi crops grown during this time.

4.3. Spatial impact of the on the performance of SPPs

The metrics plotted spatially over the country shows that the performance of SPPs is good throughout the country except in the north-east Himalayan belt and coastal regions. Decline in the performance of SPPs could be due to the fact “Himalayan ranges create more of a sinusoidal terrain formation and also act as an elevated heat source capable of producing small cumulus clouds to large mesoscale convective systems (Bharti and Singh, 2015)”. Also in coastal region “in the windward side of the Western Ghats, SPPs are unable to resolve the heavy orographic rainfall amounts and over the leeward side the rainfall amounts in the immediate rain-shadow region are overestimated (Nair et al., 2009)”

4.4. Performance of SPPs in intensity bands

Most of the precipitation occurs in the precipitation band < 2 mm/day. Thus, the among the SPPs one which detects the light precipitation events more accurately is going to be the best SPP. Unfortunately, except GNRT other SPPs are either aggressively underestimating or overestimating the number of rain events for traces to 2mm/day of precipitation band. The reason could be the over-correction of the correction algorithm in the satellite products (Zhou et al., 2020b) and thus, SPPs might have identified non-rain events as rain events mistakenly. Therefore, overestimation and underestimation of the precipitation for traces to 2mm/day (shown in Fig. 9) proves over hypothesis and this can be noted for further improvement of algorithms by the developers. Among all SPPs, GNRT have shown better agreement with IMD for higher rainfall intensity bands > 10mm/day. Additionally, the presented analysis is based on the daily rainfall accumulations, however the precipitation is a time scale dependent phenomenon. Sub-daily scale analysis of rainfall statistics is ignored, which is very relevant for the SPPs (Zhou et al., 2020a).

4.5. Elevation impact on the performance of SPPs

SPPs perform better in low elevation regions compared to high elevation. This is reasonable since low elevation regions mostly receive moderate to heavy precipitation intensities and have better statistics (visual verification, ‘yes/no dichotomous’ and continuous variable verification), and these results are consistent with the findings of other researchers (Tang et al., 2016; Zhou et al., 2020b). For all SPPs, rain detection capability is poor towards north and north-east (Himalayan foot hills) which is basically due to Indian summer monsoon and also has plateau atmospheric circulation system towards Ladakh and Tibet. In addition, SPPs are poor in depicting the costal atmospheric circulation affected by western disturbances.

With increasing elevation, GNRT has shown better precipitation detection capability than others with best score for indices POD, FAR, POFD and SS. Although IMERG_F has shown skilful scores for lower intensities and specially in the flat central India but it is still lagging compared to GNRT for higher intensities and in elevated terrains, suggesting that IMERG_F has poor gauge adjustment compared to GNRT. Pure satellite-based CMORPH and IMERG_E are completely out of scores in different elevation and therefore the authors do not recommend to use the satellite-based precipitation products without gauge adjustments. In the determination of continuous evaluation statistics, gauge adjusted SPPs can well improve the by reducing the relative error and improving the correlation coefficient. It indicates that that gauge calibration is beneficial for achieving more accurate spatial and temporal results. Gauge adjusted SPPs are also superior to pure satellite based SPPs in terms of precipitation event detection. Also, gauge-adjusted SPPs have effectively reduced the false alarms for rain and no-rain events and more importantly, miss reporting of satellite products to precipitation events, especially in the winter.

Authors have seen significantly negative correlations between change in elevation and performance of GNRT. The negative correlation indicates that GNRT may be influenced by topographic variations. However, the elevation dependence has been seen least in GNRT than IMERG_F, IMERG_E and CMORPH in order respectively.

In this study, the authors have comparatively evaluated four SPPs namely GSMaP_NRT (GNRT), IMERG_E, IMERG_F and CMROPH over India. Among four SPPs, two are gauge adjusted (GNRT and IMERG_F) and other two (IMERG_E and CMORPH) are pure satellite-based products. The evaluation for SPPs has been done using visual verification, yes/no dichotomous, and continuous variable verification method for 590-gauge district centroid points based on the IMD gridded dada for three years from 1st January 2015 to 31st December 2017.

Based on the analysis, the study reveals the following conclusions-

1. Among all SPPs, GNRT has the least number of outliers than the others for different statistical indicators at daily, monthly and yearly scale throughout the country. Other than GNRT, IMERG_F also has better statistical metrics specially in the dry seasons and for low rainfall intensities. IMERG SPPs underestimates the precipitation whereas CMORPH overestimates the precipitation. GNRT although overestimates the precipitation a bit for medium intensities but it has fairly good agreement with gauge rainfall for large intensities, making it a potential precipitation product for flood related studies.

2. As per visual verification method, all SPPs are able to follow the pattern and seasonality of gauge precipitation except CMORPH. Among SPPs, GNRT has best delineated the pattern and seasonality as its CDF plot is very close to gauge CDF plot.

3. SPPs are also found to vary differently for the various seasons. During monsoons, most of them have fairly good agreement with the gauge precipitation but have weakness while estimating it during dry seasons. GNRT has performed better than other SPPs across the seasons but IMERG_F has an advantage in central part of the country during dry seasons.

4. Categorical verification statistics show that GNRT has consistently good (> 0.6 POD_rain) and extremely good (> 0.9 POD_no-rain) rain and no-rain events detection rate respectively. The BIAS score > 0.9 (close to perfect value 1) for GNRT shows its best rain events prediction rate in terms of ‘yes/no dichotomous’ than the other SPPs.

5. Spatially, GNRT has shown better statistics (visual, yes/no dichotomous, and continuous variable verification metrics) throughout the country than others, except northern Himalayan foothills and southern costal boundaries where none of them perform well. On the other hand, IMERG and CMORPH SPPs failed to do so except IMERG_F which has shown fairly good metrics in central part of the country.

6. Regarding the statistics in the different precipitation intensity bands, GNRT plots are close to the gauge precipitation plots except for the low intensities (< 5mm/day) for which it slightly overestimates the precipitation amount. The indices improve for larger intensities which reflects the inability of SPPs to estimate the lower intensity rainfall rates.

7. The rain estimation by SPPs is also influenced by elevation. The SPPs do fairly well to estimate the rain rate and rain events respectively at low elevation (> 3500m from msl). In the different elevation bands, gauge adjusted SPPs specially GNRT do better than non- gage adjusted SPPs.

The study concludes that among all SPPs, GNRT has best estimated the rainfall spatially, temporally, seasonally, in elevations, and for intensity bands as compared to the other SPPs. Also, GNRT is best for flood related studies since its accuracies are best in high rainfall intensities. For dry seasons, IMERG_F has better estimation for rain events and rain rates over the central part of India and hence IMERG_F may be utilized for drought studies in the drought prone Central Indian states.

Availability of data and material:

Authors have provided the url links for each data set in the manuscript.

Competing interests:

There is no any competing interest.

Funding:

Authors have not received any funding for this research.

Authors' contributions:

Authors have validated the satellite precipitation products on pan-India scale and found out their usefulness for different conditions.

Acknowledgements

Authors would like to acknowledge agencies for their datasets.

Bharti V, Singh C (2015) Evaluation of error in TRMM 3B42V7 precipitation estimates over the Himalayan region. J Geophys Res Atmos 120:12458–12473. https://doi.org/10.1002/2015JD023779
Dezfuli AK, Ichoku CM, Huffman GJ, Mohr KI, Selker JS, van de Giesen N, Hochreutener R, Annor FO (2017) Validation of IMERG precipitation in Africa. J Hydrometeorol 18:2817–2825. https://doi.org/10.1175/JHM-D-17-0139.1
Draper DW, Newell DA, Wentz FJ, Krimchansky S, Skofronick-Jackson GM (2015) The Global Precipitation Measurement (GPM) microwave imager (GMI): Instrument overview and early on-orbit performance. IEEE J Sel Top Appl Earth Obs Remote Sens 8:3452–3462. https://doi.org/10.1109/JSTARS.2015.2403303
Hashemi H, Fayne J, Lakshmi V, Huffman GJ (2020) Very high resolution, altitude-corrected, TMPA-based monthly satellite precipitation product over the CONUS. Sci Data. https://doi.org/10.1038/s41597-020-0411-0
Hashemi H, Nordin M, Lakshmi V, Huffman GJ, Knight R (2017) Bias correction of long-term satellite monthly precipitation product (TRMM 3B43) over the conterminous United States. J Hydrometeorol. https://doi.org/10.1175/JHM-D-17-0025.1
Hou AY, Kakar RK, Neeck S, Azarbarzin AA, Kummerow CD, Kojima M, Oki R, Nakamura K, Iguchi T (2014) The global precipitation measurement mission. Bull Am Meteorol Soc 95:701–722. https://doi.org/10.1175/BAMS-D-13-00164.1
Huffman GJ, Bolvin DT, Braithwaite D, Hsu K, Joyce R, Kidd C, Nelkin EJ, Sorooshian S, Tan T, Xie P (2014) NASA Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG), Algorithm Theoretical Basis Document (ATBD)
Huffman GJ, Bolvin DT, Nelkin EJ (2019) Integrated Multi-satellitE Retrievals for GPM (IMERG) Technical Documentation. IMERG Tech Document
Hunt KMR, Turner AG, Shaffrey LC, Hunt KMR (2018) The evolution, seasonality, and impacts of western disturbances: On Western Disturbances. Q J R Meteorol Soc Q J R Meteorol Soc. https://doi.org/10.1002/qj.3200
Kubota T, Shige S, Hashizume H, Aonashi K, Takahashi N, Seto S, Hirose M, Takayabu YN, Ushio T, Nakagawa K, Iwanami K, Kachi M, Okamoto K (2007) Global precipitation map using satellite-borne microwave radiometers by the GSMaP project: Production and validation, in: IEEE Transactions on Geoscience and Remote Sensing. pp. 2259–2275. https://doi.org/10.1109/TGRS.2007.895337
Kubota T, Tashima T, Oki R (2019) Progress Report from JAXA: Current status of the GSMaP in the SEMDP, 201908_SEMDP_JAXA_report. https://doi.org/https://www.wmo.int/pages/prog/sat/meetings/SG-SEM
Kumar B, Patra KC, Lakshmi V (2016) Daily rainfall statistics of TRMM and CMORPH: A case for trans-boundary Gandak River basin. J Earth Syst Sci 125:919–934. https://doi.org/10.1007/s12040-016-0710-1
Kumar Singh A, Tripathi JN, Singh KK, Singh V, Sateesh M (2019) Comparison of different satellite-derived rainfall products with IMD gridded data over Indian meteorological subdivisions during Indian Summer Monsoon (ISM) 2016 at weekly temporal resolution. J Hydrol 575:1371–1379. https://doi.org/10.1016/j.jhydrol.2019.02.016
Lakshmi V, Fayne J, Bolten J (2018) A comparative study of available water in the major river basins of the world. J Hydrol 567:510–532. https://doi.org/10.1016/j.jhydrol.2018.10.038
Lu WU, Panmao Z (2012) Validation of daily precipitation from two high-resolution satellite precipitation datasets over the Tibetan plateau and the regions to its east. Acta Meteor Sin 26:735–745. https://doi.org/10.1007/s13351-012-0605-2.1
Mega T, Ushio T, Kubota T, Kachi M, Aonashi K, Shige S (2014) Gauge adjusted global satellite mapping of precipitation (GSMaP-Gauge), in: 2014 31th URSI General Assembly and Scientific Symposium, URSI GASS 2014. Institute of Electrical and Electronics Engineers Inc., pp. 1–4. https://doi.org/10.1109/URSIGASS.2014.6929683
Mitra AK, Kaushik N, Singh K, Parihar A, Bhan S, S.C (2018) Evaluation of INSAT-3D satellite derived precipitation estimates for heavy rainfall events and its validation with gridded GPM (IMERG) rainfall dataset over the Indian region. Remote Sens Appl Soc Environ 9:91–99. https://doi.org/10.1016/j.rsase.2017.12.006
Mondal A, Lakshmi V, Hashemi H (2018) Intercomparison of trend analysis of Multisatellite Monthly Precipitation Products and Gauge Measurements for River Basins of India. J Hydrol. https://doi.org/10.1016/j.jhydrol.2018.08.083
Nair S, Srinivasan G, Nemani R (2009) Evaluation of multi-satellite TRMM derived rainfall estimates over a western state of India. J Meteorol Soc Japan 87:927–939. https://doi.org/10.2151/jmsj.87.927
Pai DS, Sridhar L, Badwaik MR, Rajeevan M (2015) Analysis of the daily rainfall events over India using a new long period (1901–2010) high resolution (0.25° × 0.25°) gridded rainfall data set. Clim Dyn 45:755–776. https://doi.org/10.1007/s00382-014-2307-1
Pai DS, Sridhar L, Rajeevan M, Sreejith OP, Satbhai NS, Mukhopadhyay B (2014) Development of a new high spatial resolution (0.25° × 0.25°) long period (1901–2010) daily gridded rainfall data set over India and its comparison with existing data sets over the region. Mausam 65:1–18
Prakash S, Mitra AK, Rajagopal EN, Pai DS (2016) Assessment of TRMM-based TMPA-3B42 and GSMaP precipitation products over India for the peak southwest monsoon season. Int J Climatol 36:1614–1631. https://doi.org/10.1002/joc.4446
Progênio MF, Blanco CJC (2020) Cumulative distribution function of daily rainfall in the Tocantins–Araguaia hydrographic region, Amazon, Brazil. Nat Resour Model 33. https://doi.org/10.1111/nrm.12264
Reddy MV, Mitra AK, Momin IM, Mitra AK, Pai DS (2019) Evaluation and inter-comparison of high-resolution multi-satellite rainfall products over India for the southwest monsoon period. Int J Remote Sens 40:4577–4603. https://doi.org/10.1080/01431161.2019.1569786
Rios Gaona MF, Overeem A, Leijnse H, Uijlenhoet R (2016) First-year evaluation of GPM rainfall over the Netherlands: IMERG day 1 final run (V03D). J Hydrometeorol 17:2799–2814. https://doi.org/10.1175/JHM-D-16-0087.1
Skofronick-Jackson G, Petersen WA, Berg W, Kidd C, Stocker EF, Kirschbaum DB, Kakar R, Braun SA, Huffman GJ, Iguchi T, Kirstetter PE, Kummerow C, Meneghini R, Oki R, Olson WS, Takayabu YN, Furukawa K, Wilheit T (2017) The global precipitation measurement (GPM) mission for science and Society. Bull Am Meteorol Soc 98:1679–1695. https://doi.org/10.1175/BAMS-D-15-00306.1
Stanski HR, Wilson LJ, Burrows WR (1989) Survey of common verfication methods in meteorology -World Weather Watch Technical Report, Atmospheric Research
Sun Q, Miao C, Duan Q, Ashouri H, Sorooshian S, Hsu KL (2018) Rev Geophys 56:79–107. https://doi.org/10.1002/2017RG000574. A Review of Global Precipitation Data Sets: Data Sources, Estimation, and Intercomparisons
Sunilkumar K, Narayana Rao T, Saikranthi K, Purnachandra Rao M (2015) Comprehensive evaluation of multisatellite precipitation estimates over India using gridded rainfall data. J Geophys Res Atmos 120:8987–9005. https://doi.org/10.1002/2015JD023437
Tang G, Ma Y, Long D, Zhong L, Hong Y (2016) Evaluation of GPM Day-1 IMERG and TMPA Version-7 legacy products over Mainland China at multiple spatiotemporal scales. J Hydrol 533:152–167. https://doi.org/10.1016/j.jhydrol.2015.12.008
Thakur MK, Kumar TVL, Narayanan MS, Kundeti KR, Barbosa H (2020) Analytical study of the performance of the IMERG over the Indian landmass. Meteorol Appl 27. https://doi.org/10.1002/met.1908
Thiemig V, Rojas R, Zambrano-Bigiarini M, Levizzani V, De Roo A (2012) Validation of Satellite-Based Precipitation Products over Sparsely Gauged African River Basins. J Hydrometeorol 13:1760–1783. https://doi.org/10.1175/JHM-D-12-032.1
Thu HN, Wehn U (2016) Data sharing in international transboundary contexts: The Vietnamese perspective on data sharing in the Lower Mekong Basin. J Hydrol 536:351–364. https://doi.org/10.1016/j.jhydrol.2016.02.035
Kumar VL, Alves Barbosa T, Kumar Thakur H, Paredes-Trejo M (2019) F., Validation of Satellite (TMPA and IMERG) Rainfall Products with the IMD Gridded Data Sets over Monsoon Core Region of India, in: Satellite Information Classification and Interpretation. IntechOpen. https://doi.org/10.5772/intechopen.84999
Wilks DS (2006) Statistical methods in the atmosphere. Academic Press
Yuriy K, IntechOpen et al (2012) p. DOI: http://dx.doi.org/10.5772/intechopen.85824. https://doi.org/DOI: http://dx.doi.org/10.5772/intechopen.85824
Zhou Z, Guo B, Su Y, Chen Z, Wang J (2020a) Multidimensional evaluation of the TRMM 3B43V7 satellite-based precipitation product in mainland China from 1998–2016. PeerJ 8:e8615. https://doi.org/10.7717/peerj.8615
Zhou Z, Guo B, Xing W, Zhou J, Xu F, Xu Y (2020b) Comprehensive evaluation of latest GPM era IMERG and GSMaP precipitation products over mainland China. Atmos Res 246:105132. https://doi.org/10.1016/j.atmosres.2020.105132
Yuriy Kuleshov et al., 2012. WMO Space-Based Weather and Climate Extremes Monitoring Demonstration Project (SEMDP): First Outcomes of Regional Cooperation on Drought and Heavy Precipitation Monitoring for Australia and Southeast Asia, in: Rainfall - Extremes, Distribution and Properties. IntechOpen, p. DOI: http://dx.doi.org/10.5772/intechopen.85824. https://doi.org/DOI: http://dx.doi.org/10.5772/intechopen.85824
Zhou, Z., Guo, B., Su, Y., Chen, Z., Wang, J., 2020a. Multidimensional evaluation of the TRMM 3B43V7 satellite-based precipitation product in mainland China from 1998–2016. PeerJ 8, e8615. https://doi.org/10.7717/peerj.8615
Zhou, Z., Guo, B., Xing, W., Zhou, J., Xu, F., Xu, Y., 2020b. Comprehensive evaluation of latest GPM era IMERG and GSMaP precipitation products over mainland China. Atmos. Res. 246, 105132. https://doi.org/10.1016/j.atmosres.2020.105132

Table 1

Description of the data sets used in the study
SPPs	Coverage Period	Spatial Coverage	Resolution (S/T)
IMERG_E	March 2014-present	Global (60°N-60°S)	0.1⁰/0.5h
IMERG_F	March 2014-present	Global (60°N-60°S)	0.1⁰/0.5h
GSMaP_GNRT	April 2000-Present	Regional: Asia & Pacific (40°N-45°S&50°E-160°W)	0.1⁰/1h
CMORPH	December 2002-Present	Global (60°N-60°S)	0.25⁰/1h
IMD (Gridded)	January 1901-Present	Regional: India	0.25⁰/1d

Table 2

Contingency table for yes/ no dichotomous between Gauge precipitation and satellite precipitation products (SPPs).
Gauge Precipitation (x_i)
		Yes	No	Total
SPPs(y_i)	Yes	a Hit	b False Alarm	a + b Satellite observed Yes
SPPs(y_i)	No	c Miss	d Correct negative	c + d Satellite observed No
Total		a + c Gauge observed Yes	b + d Gauge observed No	N Total

Table 3

The statistics used in this study for evaluations of quality of SPPs
Statistics	Formula	Best Score
Continuous Variable verification Method
1. Correlation Coefficient (Cc)	\({C}_{c}=\frac{{\sum }_{i=1}^{n}({x}_{i}-\stackrel{-}{x})({y}_{i}-\stackrel{-}{y}))}{\sqrt{{\sum }_{i=1}^{n}{\left({x}_{i}-\stackrel{-}{x}\right)}^{2}{\sum }_{i=1}^{n}{\left({y}_{i}-\stackrel{-}{y}\right)}^{2} }}\)	1
2. Relative Bias (RB)	\(RB=\frac{{\sum }_{i=1}^{n}\left({x}_{i}-{y}_{i}\right)}{{\sum }_{i=1}^{n}{y}_{i}}\times 100\)	0
3. Mean Absolute Deviation (MAD)	\(MAD=\frac{{\sum }_{i=1}^{n}\left\|{x}_{i}-{y}_{i}\right\|}{n}\)	0
4. Root Mean Square Error (RMSE)	\(RMSE=\sqrt{\frac{1}{n} {\sum }_{i=1}^{n}{\left({x}_{i}-{y}_{i}\right)}^{2}}\)	0
Yes/No Dichotomous Method
1. Bias score (B)	\(B=\frac{a+b}{a+c}\)	1
2. Probability of Detection (Hit rate)(POD_rain)	\(PO{D}_{rain}=\frac{a}{a+c}\)	1
3. Correct Negative (POD_{no rain})	\(PO{D}_{no rain}=\frac{d}{b+d}\)	1
4. False Alarm Ratio (FAR_rain)	\(FA{R}_{rain}=\frac{b}{a+b}\)	0
5. False Alarm Ratio (FAR_rain)	\(FA{R}_{no rain}=\frac{c}{c+d}\)	0
6. Probability of False Alarm Detection (POFD)	\(POFD=\frac{b}{b+d}\)	0
7. Accuracy (fraction correct)	\(ACC=\frac{a+d}{N}\)	1
8. Threat score (Critical Success Index)	\(TS=CSI=\frac{a}{N}\)	1
9. Skill Score (SS)	\(SS=\frac{A-{A}_{ref}}{{P}_{perf}-{A}_{ref}}\) \({A}_{ref}=\left(\frac{a+b}{N}\right)\times \left(\frac{a+c}{N}\right)\)+\(\left(\frac{d+c}{N}\right)\times \left(\frac{d+b}{N}\right),\) \(A=ACC\)	1

Table 4

Daily Rainfall Statistics of Different Satellite Precipitation Products (SPPs) at Basin Scale
STAT		SPPs
STAT	GSMaP_GNRT	CMROPH	IMERG-E	IMERG-F
MAD	0.93	3.81	2.25	2.11
MSE	2.30	75.27	16.12	12.35
RMSE	1.51	8.67	4.01	3.51
RB%	12.36	27.94	-19.67	-11.28
CC	0.92	0.33	0.65	0.68

Table 5

Continuous variable verification statistics: mean absolute difference (MAD), mean square error (MSE), root mean square error (RMSE), relative bias (RB) and correlation coefficient (CC) for satellite precipitation products (SPPs) for different seasons
Season	Stat	GSMaP_GNRT	CMORPH	IMERG_E	IMERG_F
Rainy	MAD	7.09	11.27	12.18	12.03
	MSE	237.74	507.82	684.7	634.02
	RMSE	15.42	22.53	26.17	25.18
	RB %	11.98	17.4	-4.82	-3.84
	CC	0.65	0.38	0.37	0.53
Winter	MAD	1.25	1.48	3.51	3.67
	MSE	41.34	49.49	240.88	283.07
	RMSE	6.43	7.03	15.52	16.82
	RB %	-8.75	134.11	-54.53	-57.5
	CC	0.70	0.24	0.3	0.27
Spring	MAD	0.79	0.98	1	1.16
	MSE	16.39	20.46	25.3	33.9
	RMSE	4.05	4.52	5.03	5.82
	RB %	20.8	109.24	104.18	39.29
	CC	0.22	0.17	0.49	0.57
Summer	MAD	3.6	5.46	6.53	7.75
	MSE	126.6	195.86	401.32	519.19
	RMSE	11.25	13.99	20.03	22.79
	RB %	19.88	25.41	-7.38	-30.82
	CC	0.67	0.4	0.43	0.52

GRaphicalPic.docx

Spatiotemporal inter-comparison of satellite precipitation products over India

Status:

Version 1

Abstract

Figures

1. Introduction

2. Materials And Methods