Comparison analysis of six purely satellite-derived global precipitation estimates

Journal Pre-proofs Research papers Comparison Analysis of Six Purely Satellite-derived Global Precipitation Estimates Hanqing Chen, Bin Yong, Yan Shen, Jiufu Liu, Yang Hong, Jianyun Zhang PII: DOI: Reference:

S0022-1694(19)31111-4 https://doi.org/10.1016/j.jhydrol.2019.124376 HYDROL 124376

To appear in:

Journal of Hydrology

Received Date: Revised Date: Accepted Date:

16 August 2019 19 October 2019 18 November 2019

Please cite this article as: Chen, H., Yong, B., Shen, Y., Liu, J., Hong, Y., Zhang, J., Comparison Analysis of Six Purely Satellite-derived Global Precipitation Estimates, Journal of Hydrology (2019), doi: https://doi.org/10.1016/ j.jhydrol.2019.124376

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier B.V.

Comparison Analysis of Six Purely Satellite-derived Global Precipitation Estimates Hanqing Chena,b,c, Bin Yonga,b,*[email protected], Yan Shend, Jiufu Liue, Yang Hongf, Jianyun Zhange aState

Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering,

Hohai University, Nanjing 210098, China bSchool cKey

of Earth Sciences and Engineering, Hohai University, Nanjing 210098, China

Laboratory for Digital Land and Resources of Jiangxi Province, East China

University of Technology, Nanchang 330013, China dNational

Meteorological Information Center, China Meteorological Administration,

Beijing 100081, China eNanjing fSchool

Hydraulic Research Institute, Nanjing 210029, China

of Civil Engineering and Environment Sciences, University of Oklahoma,

Norman, OK 73019, USA *Corresponding

author.

Highlights 

Six satellite-only precipitation products (SPPs) were evaluated over globe and mainland China.



The error sources of five SPPs over mainland China were revealed.



IMERG-Late is the best one of six evaluated SPPs.



A power function is observed between the root mean squared error (RMSE) and

logarithm of precipitation intensity in all six SPPs. 

GPM-based SPPs in light rainfall still exhibit large errors.

Abstract We executed a comprehensive evaluation and intercomparison between six purely satellite-derived precipitation estimates (i.e., IMERG-Late, IMERG-Early, GSMaPNRT, GSMaP-MVK, TMPA-RT and PERSIANN-CCS) at global and regional scales for the period from February 2017 to January 2019. The results show that IMERG-Late exhibits the best performance among six evaluated products, while the worst performance was found in GSMaP-NRT and GSMaP-MVK. The root mean squared error (RMSE) has a power function to the logarithm of precipitation intensity in all six satellite products. On the basis of our findings, the RMSE of all products in rainfall events with intensity exceeding 32 mm/day (or 8 mm/h) accounts for beyond 30% of the corresponding precipitation intensity, which might result in a significant impact on the detectability and forecast of flash floods simulated by satellite precipitation. Additionally, both IMERG and GSMaP overestimate the proportions of light rainfall occurrences, and also display relatively larger errors in light precipitation (0.2-0.4 mm/h or 1-2 mm/day) with the RMSE values exceeding 0.5 mm (or 2 mm) at hourly (or daily) time scale. As for the error analysis, we decomposed the total bias of each product into hits, misses and false biases at hourly and 0.1° resolution over mainland China except for TMPA-RT. We found that the false bias is the dominated error sources for these five products in cold season over semi-humid areas despite that the hit bias

accounts for a non-negligible proportion for GSMaP suite. The missed precipitation is the dominated error sources of PERSIANN-CCS both in two seasons over most of humid regions, and meanwhile is one of major error sources for other four products. We expect that the findings of this study not only provide some valuable feedbacks for algorithm developers to improve the GPM-based satellite precipitation retrievals, but also provide some guidance for data users across the world. Keywords: Satellite-only precipitation products; Light rainfall; RMSE; Error source; Mainland China

1. Introduction Precipitation is one of the important variables coupling the water and energy cycles, and it is also one of the conundrums to estimate due to its strong spatiotemporal heterogeneity (Hou et al., 2014; Chen et al., 2019b). Since the development of satellite techniques and retrieval algorithms, a growing number of satellite-based precipitation datasets have been released to the public. In the Tropical Rainfall Measurement Mission (TRMM) era, there are four widely used satellite precipitation products, including TRMM Multi-satellite Precipitation Analysis (TMPA; Huffman et al., 2007), Global Satellite Mapping of Precipitation (GSMaP; Kubota et al., 2007), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN; Sorooshian et al., 2000) and Climate Prediction Center MORPHing technique (CMORPH; Joyce et al., 2004). The Global Precipitation Measurement (GPM) Core Observatory satellite was launched on February 28, 2014, which means

that satellite precipitation has entered the GPM era (Hou et al., 2014). In this era, satellite-based precipitation datasets mainly include IMERG (Integrated Multi-satellite Retrievals for GPM; Huffman et al., 2019) and GSMaP, which have been widely used in the hydrometeorological field. These operational satellite precipitation products aforementioned above have played an important role in the research, community, flood and drought prediction, water source management at local, regional, and even global scales (Hou et al., 2014; Yong et al., 2015; Maggioni et al., 2016; Chen et al., 2019a). Compared with gauge-adjusted satellite precipitation datasets, satellite-only precipitation products (SPPs) without correction from ground-based observations have the advantage of near-real-time which have been widely used in near-real-time applications, such as crop forecasting, flood forecasting, drought monitoring and landslides detection, etc. However, the performance of SPPs normally performed worse than that of gauge-adjusted satellite precipitation datasets over most regions. Previous studies indicate that SPPs are prone to generate larger random errors and systematic errors, especially for light or heavy rainfall events, which can lead to some negative impacts on various near-real-time applications (Tapiador et al., 2012; Maggioni et al., 2016; Deng et al., 2018; Hong et al., 2019). Currently, many researchers have done a great deal of assessments for revealing the error structure and error characteristics of SPPs in order to help data developers improving the quality of SPPs. It has been found that several global evaluation studies (e.g., Tian and Peters-Lidard, 2011; Yong et al., 2015; Gehne et al., 2016; Liu et al., 2016; Beck et al., 2017; Wang et al., 2018) and considerable number of regional evaluation (e.g., Yang et al., 2014; Yong et al., 2010,

2013, 2014, 2016; Tang et al., 2016; Guo et al., 2016; Xu et al., 2017; Tan et al., 2017; Wang et al., 2017; Sun et al., 2018; Zhao et al., 2018; Deng et al., 2018; Su et al., 2018; Lu and Yong, 2018; Beck et al., 2019; Wu et al., 2019) have executed to examine and understand the error characteristics of satellite-based precipitation datasets for data users and algorithm developers. However, many of these studies did not use the latest satellite-based precipitation datasets (e.g., Tian and Peters-Lidard, 2011; Yang et al., 2014; Yong et al., 2010, 2013, 2014, 2015; Gehne et al., 2016; Beck et al., 2017; ) or evaluated only one or two types of datasets (e.g., Yong et al., 2016, Tan et al., 2017; Wang et al., 2018; Deng et al., 2018; Lu and Yong, 2018; Zhao et al., 2018; Wang et al., 2018). With the emerge of new products and versions, the advantages and limitations of these SPPs have been widely concerned by users and developers. Consequently, it is necessary and timely to comprehensive analyze these latest satelliteonly products and versions at global and regional scales. A larger number of studies have examined the relationship between the performance of SPPs and precipitation intensity. Most of these studies attempted to analyze the potential links between systematic bias (e.g., Xie et al., 2011; Deng et al., 2018; Zhu et al., 2018) or detection capability (e.g., Guo et al., 2016; Qi et al., 2016; Xu et al., 2016; Manz et al., 2017) and precipitation intensity. However, they couldn’t reflect the accuracy of SPPs under different precipitation intensity. In particular, it is important for data users to clearly understand the accuracy of SPPs in estimating the heavy rainfall events based on empirical function models before using SPPs to force the flood monitoring and forecasting systems. In the recent GPM era, the retrievals of

light rainfall also attracted the interests of many researchers because the Dual-frequency phased array Precipitation Radar and a multichannel GPM Microwave Imager (DPR and GMI) carried the GPM Core Observatory satellite can provide more accurate estimation for light rainfall and snow. Although some previous studies utilized probability of density function (PDF) to detect the estimation sensitivity and precipitation intensity for SPPs (e.g., Kirstetter et al., 2013; Chen et al., 2013; Su et al., 2018; Tang et al., 2016; Wang et al., 2018), few studies focused on the GPM-based purely satellite-derived products. Furthermore, the accuracy of SPPs in retrieving the light rainfall was often neglected in these studies. Separating the total errors into components is a meaningful effort that can help data developers better tracing the error sources and reducing them (Tian et al., 2009). Several reported works have provided some valuable information on error sources for real-time TMPA (TMPA-RT), gauge-calibrated IMERG, and gauge-calibrated GSMaP over mainland China (e.g., Yong et al., 2016; Su et al., 2018). However, a systematic study aiming to the latest satellite-only precipitation products in operation over mainland China is still lacking. In this study, the current popular satellite-only precipitation products, including IMERG-Late, IMERG-Early, Global Satellite Mapping of Precipitation Microwave-IR Combined Product (GSMaP-MVK), Global Satellite Mapping of Precipitation in NearReal-Time (GSMaP-NRT), TMPA-RT, and Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Cloud Classification System (PERSIANN-CCS), are statistically investigated over the globe and mainland China.

Our aim is mainly to answer four key questions as follows. 1.

With the emerge of new products and versions, how do these SPPs perform, especially for light precipitation?

2.

What is the relationship between retrieval accuracy and precipitation intensity?

3.

What is the major error source for these mainstream SPPs in mainland China?

4.

Are IMERG products satisfactory enough to support TMPA-RT stopping?

2. Study area, datasets and methodology 2.1 Study area In global analysis, the study areas of all SPPs cover the latitude band 60°N/S except for TMPA-RT (50 ° N/S). Fig. 1(a) presents the spatial distribution of mean daily precipitation for Climate Prediction Center (CPC) unified. It is seen that the heavy precipitation mainly distributed in the countries closest to the equator and in regions controlled by the monsoon climate, such as Malaysia, Indonesia, Columbia, Japan, southeastern China, India, and Bangladesh and so on (Liu et al., 2016; Wang et al., 2018). Of regional analysis, mainland China is chosen as regional study areas. The reason is that mainland China has multiple types of climate features and complex topography which may expose more potential issues of SPPs. The spatial distribution of mean hourly precipitation for Merged Precipitation Analysis (MPA) over mainland China is presented in Fig. 1(b). Overall, the precipitation decreases from southeast to northwest in mainland China (Tang et al., 2016; Chen et al., 2019b). In addition, mainland China can be divided into four climatic regions, namely the humid, semi-humid, semi-arid,

and arid regions (refer to Chen et al., 2019b), as shown in Fig. 1b. It is obvious that humid regions have a large amount of precipitation because of the impacts of subtropical monsoon climate, and those continental domains mainly include Pearl River basin, the middle and lower reach of the Yangtze River and Huaihe River basin. The semi-humid, controlled by the monsoon in middle latitudes, covers the Yellow River basin and the eastern Tibetan Plateau. The semi-arid with a relatively complex topography includes the Tianshan mountains and middle part of the Tibetan Plateau. Finally, the residual parts belong to arid regions mainly including the Tarim basin and the western Tibetan Plateau. 2.2 Datasets 2.2.1 Six purely satellite-derived precipitation datasets In this study, six SPPs are employed in the evaluation, i.e., IMERG-Late V6, IMERG-Early V6, GSMaP-MVK V7, GSMaP-NRT V7, TMPA 3B42 RT V7 and PERSIANN-CCS. The parameters of these six SPPs are listed in Table 1. The IMERG suite including IMERG-Late V6 (hereafter referred to IMERG-Late) and IMERG-Early V6 (hereafter referred to IMERG-Early), which merged available data from GPM constellation satellites for getting accurate precipitation estimates (Hou et al., 2014). Firstly, the DPR and GMI data are applied to intercalibrate and merge precipitation estimates from several passive microwave (PMW) satellites. Then, the PERSIANN-CCS algorithm is applied to compute microwave-calibrated infrared (IR) precipitation rates. Further, the morphing technique using motion vectors computed from Goddard Earth Observing System model (GEOS) Forward Processing (FP) data

to forward/backward propagate microwave maps is applied to produce global gridded precipitation estimates at fine resolution. Finally, a Kalman filter is used to combine these morphing-based estimates and the microwave-calibrated IR estimates into a weighted estimate. Note that the IMERG-Early is forward-only in morphing process. The GSMaP suite, including GSMaP-MVK V7 (hereafter referred to GSMaPMVK) and GSMaP-NRT V7 (hereafter referred to GSMaP-NRT), is used in our study. GSMaP-MVK integrates PMW data with IR data in order to produce high spatiotemporal resolution (0.1°, 1hour) global precipitation estimates. Similarly, it also uses a Kalman filter to refine the precipitation rate propagated based on the atmospheric moving vector derived from two successive IR images. Unlike IMERG system, the GSMaP system propagates and applies the Kalman filter only for the precipitation estimates identified by the PMW algorithms. It indicates that the Kalman filter used in GSMaP system strongly depend on the PMW-based precipitation estimates (Ushio et al., 2009). Also, this reflects that the IR-based information is not fully utilized in GSMaP system. Another difference between IMERG and GSMaP is that the IR data is not calibrated by PMW data using pertinent algorithms. Besides, the GSMaP system only uses DPR data as a calibration standard for PMW retrievals, not including GMI data. As for GSMaP-NRT, the precipitation estimates of this near-real-time product are based on the GSMaP-MVK algorithm, but some processes are simplified in order to keep operability and latency in near-real-time. The major differences between GSMaPNRT and GSMaP-MVK in processing are as follows: (1) Japan Meteorological Agency (JMA) forecast data is used as atmospheric information; (2) Latest available JMA

Merged satellite and in situ data Global Daily Sea Surface Temperature (MGDSST) data is used as sea surface temperature information; (3) Forward-only cloud movement is used in algorithm. Real-time TMPA 3B42 V7 (hereafter referred to TMPA-RT) estimates integrates most PMW data available from low earth orbit satellites and IR measurements from geostationary platforms. This near-real-time product mainly uses IR-based retrievals to fill the PMW coverage gaps. Unlike IMERG, TMPA is calibrated to the TRMM combined instrument (i.e., Precipitation radar (PR) and TRMM Microwave Imager (TMI)) until the demise of TRMM satellite. After that, TMPA-RT uses climatological satellite calibrations, so continue despite the loss of quality for TMPA estimates (Huffman et al., 2018). As for PERSIANN-CCS, this product mainly uses PERSIAN-CCS algorithm to estimate global rainfall in near-real-time and at higher spatial resolution (0.04°) despite using IR data as the sole input (Sorooshian et al., 2000; Hong et al., 2004). Since its release, PERSIANN-CCS has been widely used in the research, flood and drought prediction, and more (Nguyen et al., 2018). Due to the data period of GSMaP-NRT V7 starting from January 2017, these six purely satellite-derived precipitation datasets from February 2017 to January 2019 (2 years) are selected to evaluate for obtaining new insights for data users and algorithm developers. 2.2.2 Two ground reference precipitation datasets In our study, CPC unified and gauge-satellite merged gridded hourly MPA product

are regarded as the benchmarks in the evaluation over the globe and mainland China, respectively. CPC unified developed by the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center was used as “ground truth” in global evaluation, which combines gauge-based observations from more than 17,000 gauges over the globe. The Optimal Interpolation (OI) with orographic consideration is applied to produce a real-time gauge-based precipitation product (Xie et al., 2007; Chen et al., 2008). However, CPC unified shows poor performance in tropical Africa and Antartica. Meanwhile, its quality varies with the gauge network density. Despite that, this dataset performed the best performance by intercomparing among the 22 precipitation datasets over the globe (Beck et al., 2017). The spatial distribution of gauge density for CPC unified is presented in Fig. 1c. Another reference data is gauge-satellite merged MPA, which developed by the National

Meteorological

Information

Center

of

the

China

Meteorological

Administration, and is used in the evaluation over mainland China. MPA combines the gauge-based observations from more than 30,000 automatic weather stations (AWS) with CMORPH. Firstly, the rain gauge observations go through strict quality control (refer to Ren et al., 2010). Then, the CMORPH estimates are corrected by using the probability density function to reduce the systematic errors. Finally, the OI method is applied to combine quality-controlled gauge observations and gauge-corrected CMORPH producing the final gauge-satellite merged MPA dataset (Shen et al., 2014). Overall, MPA has lower bias which is less than 10%, and outperforms other gauge-

based precipitation products over mainland China (Shen et al., 2013). Additionally, the AWS used in MPA are mainly concentrated in the humid and semi-humid areas (see Fig.1d). Note that only liquid rainfall occurrences are investigated in the evaluation over mainland China because of the AWS observations used in MPA only suiting for liquid rainfall events. Thus, we use the flag information of gauge numbers within one grid cell in the MPA system to strictly match the satellite estimates and liquid rainfall occurrences detected by AWS to exclude the snowfall events. To ensure the robustness of the evaluation results, only grid cells with at least one gauge in these two reference datasets are chosen to execute evaluation. 2.3 Methodology The performance of SPPs is basically behaved in detection capability and accuracy. The evaluation metrics used in this study are listed in Table 2. Probability of detection (POD) and false alarm ratio (FAR) are used to examine the detection capability of SPPs. POD gives how often the satellite correctly detects precipitation, while FAR gives how often the satellite detects precipitation when the rain gauge does not detect precipitation. The correlation coefficient (CC) shows the degree of linear correlation between the reference values and satellite estimates, and it is able to reflect the capability of reproducing the temporal variations (Wang et al., 2018). While relative bias (BIAS) describes the systematic bias of satellite-only precipitation estimates, and it is able to reflect the underestimation (BIAS < 0%) or overestimation (BIAS > 0%) of ones. Tian et al. (2009) proposed an error decomposition scheme to decompose the BIAS (also known as total bias (T)) into hit bias (H), miss bias (M) and false bias (F). The

relationship between total bias and its three error components can be defined by T = H - M + F. The detailed calculation formulas of total bias and its three independent components are shown in Table 2. As for RMSE, this metric describes the errors of SPPs, and it reflects the accuracy of SPPs because it is used to measure the deviation between the estimates and the observations. It is worth noting that the rain /no-rain threshold is set to 0.2 mm for hourly time scale and 1 mm for daily time scale.

3. Results 3.1 Global evaluation and intercomparison between six SPPs The results of our global analysis associated with six SPPs are organized into two parts in this section: general analysis of six SPPs over the whole globe; and the spatial analysis of six SPPs over the globe. These two aspects will show a relatively comprehensive picture of the performance for each SPP. [Note that all SPPs are aggregated precipitation estimates to CPC unified scales 0.5° and daily before the evaluation metrics calculated]. 3.1.1 Global general analysis Table 3 presents the summary of evaluation metrics for six SPPs at 0.5° and daily resolution over the whole globe. Almost all results of evaluation metrics indicate that IMERG-Late is slightly better than the other five SPPs over the globe. One can see that IMERG suite, including IMERG-Early and IMERG-Late, shows the best performance in detection capability over the whole globe, with higher POD value of 0.75. While the PERSIANN-CCS and TMPA-RT reveal inadequacies in representing areal rainfall

patterns relative to other SPPs. In terms of CC, IMERG suite has higher CC value of exceeding 0.6, performs better than other four SPPs, suggesting that IMERG suite can better reproduce the temporal dynamics over the globe. In the BIAS column, we found that all SPPs overestimate precipitation over the whole globe except for TMPA-RT. Obviously, the GSMaP suite dramatically overestimates precipitation with higher BIAS values of exceeding 34% over the whole globe. Conversely, the TMPA-RT underestimates precipitation with the BIAS value of -13.16%. Lastly, the accuracy of IMERG suite outperforms that of other four SPPs over the whole globe, with lower RMSE values of less than 7.25 mm. The evaluation results show that the IMERG suite estimates have been slightly improved building upon the success of the TMPA-RT ones, especially for detectability. 3.1.2 Global view of spatial analysis Fig. 2 shows the spatial distribution of POD and FAR for six SPPs at 0.5⁰ and daily resolution over the globe. It is seen that the comparable spatial distribution patterns of POD and FAR are found between these four SPPs except for TMPA-RT and PERSIANN-CCS. Overall, the worse performance in detecting capability for six SPPs is found in the western conterminous United States (COUNS), southern Canada, eastern Brazil, southwestern Mongolia, western China, southeastern Australia, Andean Mountains in South America etc. Our results are consistent with previous studies (e.g., Tang et al., 2016; Xu et al., 2017; Tan et al., 2017; Wang et al., 2018; Lu and Yong., 2018; Su et al., 2018; Beck et al., 2019; Chen et al., 2019b). Evidently, PERSIANNCCS and TMPA-RT perform relatively worse detection capability over most parts of

the world. The error feature of six SPPs is another research target. Fig. 3 presents the spatial distribution of BIAS and RMSE for six SPPs at 0.5⁰ and daily resolution over the globe. As shown in Figs. 3a-f, the negative and positive BIAS values of six SPPs are scattered over the globe. Overall, IMERG suite and TMPA-RT perform relatively well with slight underestimation or overestimation over the globe. While GSMaP suite and PERSIANN-CCS perform relatively worse, dramatically overestimate precipitation with their BIAS over 100% in the most of COUNS, western China, Mongolia, Japan, India, Kazakhstan etc. Finally, the spatial distribution of RMSE values for six SPPs are presented in Figs. 3g-l. Obviously, these six SPPs share the considerable similarities in spatial distribution. The most remarkable common feature is that higher RMSE values are distributed in these areas which have ample medium and heavy rainfall events (see Fig.1a), such as eastern COUNS, most of South America, Japan, southeastern China, India, Bangladesh etc. It seems that the RMSE values are related to the precipitation intensity. 3.2 Evaluation of five SPPs in Mainland China As is known to all that TMPA-RT will be stopped when IMERG products are satisfactory (Huffman, 2018). On the basis of the global analysis, the performance of IMERG suite is generally better than that of TMPA-RT over the globe. Thus, the five SPPs except for TMPA-RT will be evaluated at finer spatial resolution (0.1°) over mainland China to essentially reveal their error features. 3.2.1 Routine analysis

Table 4 presents the summary of evaluation metrics for five SPPs at hourly and daily time scales over mainland China. Almost all results of metrics indicate that IMERG suite shows relatively better performance than other three SPPs at hourly and daily time scales over mainland China. It is worth noting that GSMaP-NRT and GSMaP-MVK have fairly good performance in detection capability. However, they have the highest BIAS and RMSE values both at hourly and daily time scales, suggesting that GSMaP-NRT and GSMaP-MVK can better detect the rainfall events correctly but their precipitation amount is not accurate enough. In contrast, PERSIANN-CCS performs relatively worse performance in detection capability, but better in accuracy of precipitation amount. The spatial analysis of five SPPs in mainland China is given. Fig. 4 shows the spatial distribution of POD and FAR for five SPPs at hourly and daily time scales. The comparable spatial distribution patterns appear in most SPPs (except for PERSIANNCCS) both at hourly and daily time scales. Compare with other SPPs, PERSIANN-CCS has lower POD values (<0.5) in most mainland China at hourly time scale, and its FAR values are exceeding 0.5 (Figs. 4q and s). Thus, not surprisingly, its POD value is as low as 0.20 while its FAR value is up to 0.73 over the whole mainland China (Table 4). Meanwhile, all SPPs are relatively good at detecting the heavy precipitation because of their higher (or lower) POD (or FAR) values appearing in these regions which have ample heavy precipitation. Though Fig. 4 reveals the detection capability of five SPPs, it couldn’t reflect the error feature for each SPP. The spatial distribution of BIAS and RMSE for five SPPs at

hourly and daily time scales over mainland China is presented in Fig. 5. In terms of BIAS, GSMaP suite exhibits significant overestimates over semi-humid and northern humid areas (Figs. 5c-d and h-i) compared with other three SPPs. In addition, PERSIANN-CCS have remarkable underestimates in humid areas, and overestimates the precipitation in the residual parts of mainland China (Figs. 5e and j). For the RMSE, GSMaP suite has higher RMSE values in humid areas relative to other three SPPs both at hourly and daily time scales. It is clear that the RMSE values of IMERG suite and PERSIANN-CCS gradually decrease from southeast to northwest. Also, five SPPs have relatively higher RMSE values in humid regions both at hourly and daily time scales. It seems that the RMSE is related to the precipitation intensity. This further confirms the necessity of investigating the relationship between RMSE and precipitation intensity. 3.2.2 Tracing the error sources for five SPPs in mainland China Firstly, we decomposed the total bias of each SPP into hit bias, miss bias and false bias at hourly time scale over mainland China, and then accumulated the total bias and three hourly error components (i.e., hit bias, miss bias and false bias) into cold season (including these months of two-year study period: from November to December and January to April) and warm season (including these months of two-year study period: from May to October), respectively. In this section, our analysis is mainly concentrated in the humid and semi-humid areas because of having dense rain gauge networks over these areas. The Figs.6 and 7 show the spatial distribution patterns of the total bias and its three error components for the cold and warm seasons, respectively.

For cold season (Fig. 6), the spatial distribution patterns of five SPPs in total bias and three error components share considerable similarities. The most obvious common feature is that the total biases of five SPPs is exceeding 150% in most semi-humid regions. One can see that the false error is the dominant error components causing the overestimates for IMERG suite and PERSIANN-CCS over these areas, while GSMaP suite’s overestimates are attributed primarily to the false biases, but the hit biases cannot be neglected over these areas. Also, it cannot be ignored that the miss error is one of major error components for PERSIANN-CCS over these areas. On the other hand, we also found that some differences exist between these five SPPs. GSMaP suite and PERSIANN-CCS overestimate precipitation in the northern humid regions because of the larger false biases (>150%). In addition, PERSIANNCCS seriously underestimates precipitation in the southern humid regions. This is due primarily to missed precipitation (< -80%) over these areas. For warm season (Fig. 7), all SPPs display lower biases relative to the cold season, which is consistent with previous results (e.g., Yong et al., 2016; Guo et al., 2016; Chen et al., 2019b). IMERG-Early and IMERG-Late (or GSMaP-NRT and GSMaP-MVK) almost show a high degree of consistency in spatial distribution patterns of the total bias and three error components, except for the miss error. However, IMERG-Late has lower miss biases than IMERG-Early over the large areas of semi-humid and humid regions. Correspondingly, GSMaP-MVK shows lower miss biases than GSMaP-NRT but only for the large areas of semi-humid regions. Meanwhile, IMERG suite and GSMaP suite exhibit slight underestimates or

overestimates over humid regions. These underestimates (or overestimates) can be traced to miss (or false) biases. Additionally, the main error sources for GSMaP suite in semi-humid come from false and hit biases differing from IMERG suite for their error sources coming from misses and false biases. Finally, the total errors of PERSIANN-CCS are dominated by the miss errors (< -60%) in the large areas of mainland China. 3.3 Evaluation of estimation sensitivity and precipitation intensity The PDF is a metric used to evaluate and compare different SPPs in terms of estimation sensitivity and precipitation intensity, and it is also a good method to describe the performance of SPPs to measure light precipitation (Kirstetter et al., 2013; Chen et al., 2013; Tang et al., 2016; Wang et al., 2018). Fig.8 displays the variations of PDF values for six SPPs with precipitation intensity both at daily and hourly time scales. In global analysis, it is obvious that more than 75% precipitation is under 1 mm/day, as shown in Figs. 8a-b. The majority of SPPs display slight underestimation in detecting precipitation with intensity less than 1 mm/day except for TMPA-RT. The significant differences of PDF exist between these six SPPs in the precipitation intensity range (>1 mm/day). PERSIANN-CCS shows poorer performance with overestimating rainfall events in the precipitation intensity range (1-40mm/day). There are some differences between IMERG-Early and IMERG-Late (or GSMaP-NRT and GSMaP-MVK), which is possibly caused by the different retrieval algorithms and input sources. Overall, TMPA-RT exhibit better detecting rainfall events than other five SPPs in most precipitation intensities.

The variations of PDF values for five SPPs with precipitation intensity share considerable similarities in regional and global analyzes (see Figs. 8a-d). The most significant common feature is that the PDF values of five SPPs have a clear decreasing tendency with precipitation intensity. Just as a coin has two sides, some significant differences exist between these two spatial scale analyses. For example, compared with global analysis, five SPPs in regional analysis overestimate more proportions in the precipitation intensity range (>1 mm/day) and reduce proportions in precipitation with intensity less than 1 mm/day. For the hourly scenario, IMERG suite shows evident overestimates in most precipitation intensities, while GSMaP suite significantly overestimates rainfall events in the precipitation intensity range (1-5mm/h). Surprisingly, PERSIANN-CCS have similar proportions with MPA dataset in most precipitation intensities.

4. Discussion Based on global and regional results, it seems that the RMSE (accuracy) is related to the precipitation intensity. Additionally, the accuracy of these SPPs in heavy precipitation is closely associated with the detectability and prediction for flood events (Kucera et al., 2013; Hou et al., 2014; Wu et al., 2014; Zhang et al., 2015). Thus, it is a meaningful effort that investigating the relationship between RMSE and precipitation intensity, which is helpful for data users to understand the accuracy of SPPs under different rainfall events. Overall, there is a power function relationship between RMSE and logarithm of precipitation intensity which implies the errors of six SPPs having a stronger dependency on precipitation intensity, as shown in Fig. 9. Furthermore, we

found that all SPPs display better detectability in flood-prone areas which have ample heavy precipitation (see Figs. 1a-b, 2 and 4). However, SPPs in these areas are unsatisfactory because of their higher RMSE values (see Figs. 3g-l, and 5k-t). It can be seen that the RMSE of six SPPs in precipitation with intensity exceeding 32 mm/day (8 mm/h) accounts for more than 30% of corresponding precipitation intensity (see Fig. 9). The poor accuracy of these six SPPs in such rainfall events may produce a negative impact in hydrologic studies to predict flood events. As is known to all that SPPs have been widely applied in flood detection and prediction (Khan et al., 2011; Hong et al., 2019). For example, two important global flood systems, i.e., Global Hydrological Prediction System (GHPS; Wang et al., 2011) and Global Flood Monitoring System (GFMS; Wu et al., 2014), are all driven by the TMPA and GPM products. Obviously, the power function relationship can provide a method to quickly obtain the RMSE values of different rainfall events for SPPs, which is helpful for data users knowing the accuracy of SPPs when using SPPs to force the flood prediction system. The two more advanced instruments, including DPR and GMI, are aim to more accurate instantaneous precipitation estimates, particularly for light and solid precipitation. In particular, the precipitation sensitivity of spaceborne radar increased from 0.5mm/h for PR to 0.2mm/h for DPR. The performance of SPPs in light rainfall events in GPM era has attracted wide attention of algorithm developers and data users. The majority of these six SPPs overestimate the proportions of light rainfall events (12 mm/day) across the globe, especially for IMERG suite (see Fig. 8b). In regional analysis, the proportions of light rainfall events (0.2-0.4 mm/h) for IMERG suite show

more significant overestimates with the bias of exceeding 68% at hourly time scale (see Fig, 8f). The similar phenomenon also appears at daily time scale (see Fig. 8d). We noted that there are some studies investigating the performance of light precipitation for satellite precipitation products. For example, Chen et al. (2013) found that the TMPA-RT underdetects light rainfall. Also, gauge-corrected IMERG V3 and V5 underestimate the proportions in light rainfall, while gauge-corrected IMERG V4 overestimates ones (Wang et al., 2018). Prakash et al. (2018) even found that the proportions of light precipitation were underestimated in the gauge-corrected IMERG V3 and GSMaP products. On the other hand, investigating the accuracy of light precipitation for SPPs is still lacking in previous studies. In our study results, all six SPPs have large RMSE values greater than 2 mm (or 0.5 mm) at daily (or hourly) time scale in the light rainfall events. It means that the current satellite precipitation retrievals in light rainfall still exhibit poor performance in estimation despite introducing DPR data to improve retrievals. Consequently, we suggested that future efforts for algorithm developers still need to be focused on the detectability and estimation of light rainfall. The uncertainty of results is an issue in the evaluation. We have examined the RMSE as a function of different rain gauge density over humid regions of mainland China, as shown in Fig. 10. Overall, there are having a decreasing trend of RMSE values with the increased rain gauge density, indicating that using a sparse rain gauge network might underestimate the performance of SPP. For example, the bad performance of the SPPs appears in the most areas of Africa, central Australia, western China and so on, where the rain gauges for CPC unified or MPA are sparsely and unevenly distributed

in these areas. We though that the bad performance of the SPPs in these areas with sparse rain gauges is not actual performance of SPPs but of the evaluation approach itself. Also, previous studies have confirmed that the uncertainty of the evaluation for SPPs is associate with rain gauge density (e.g., Tian et al., 2018; Tang et al., 2018; Prakash et al., 2019). All these studies revealed that a dense rain gauge density network used in the evaluation can better reflect the actual performance of SPPs. Besides, the reference data including satellite information is one of sources for the uncertainty of the evaluation. One can see that the input sources of reference data MPA are overlapping with those of these six SPPs to some extent, which might cause some potential uncertainties in the evaluation. The data developers hope that TMPA-RT stops when IMERG products are satisfactory (Huffman et al., 2018). Our evaluation results show that the IMERG suite including IMERG-Early and IMERG-Late generally outperforms TMPA-RT especially for detectability. However, the performance of IMERG suite is still unsatisfactory in some areas, such as western and northern COUNS, western China, central Australia etc. The cause is primary contributable to the impacts of complex topography and bad climate situations on the precipitation structure and associated challenges with satellite retrievals from space (Daly et al., 2008; Guo et al., 2017; Chen et al., 2019b). Generally speaking, the equivalent IMERG suite outperforms the TMPA-RT, and is satisfactory at this time. We expect that our study results can provide a feedback for data developers to decide whether TMPA-RT can stop or not. On the basis of our findings, we proposed some valuable suggestions for data

developers and data users: (1) Overall, IMERG-Late should be given priority in nearreal-time applications because of its excellent performance among the six satellite-only precipitation products. In addition, data users should give priority to using IMERGEarly in the applications with strong timeliness due to its 4-hr date latency. (2) Though GSMaP suite also uses a Kaman filter to combine the forward and backward estimates into a weighted estimate, its accuracy is obviously worse than IMERG suite’s. We speculate that one reason is that the Kalman filter used in GSMaP retrieval system only combines the precipitation estimates identified by the PMW algorithms. Another reason may be that the weights used in Kalman filter are not precise enough. Thus, we suggested that the data developers should further improve the GSMaP retrieval system in determining the weights and making full use of IR-based information. Finally, separating the errors into components is helpful for algorithm developers to trace error and reduce them (Tian et al., 2009). We revealed the error sources for these mainstream SPPs over mainland China. However, whether the evaluation results for cold and warm seasons from mainland China can be extended to other regions of the world still needs to be further investigated.

5. Conclusion This study is to comprehensively evaluate and compare the performance of six purely satellite-derived precipitation datasets, i.e., IMERG-Early, IMERG-Late, GSMaP-NRT, GSMaP-MVK, TMPA-RT and PERSIANN-CCS. On the basis of global and regional evaluation results, we answered four questions that both users and developers widely concerned for satellite precipitation estimates. Our major

conclusions are summarized as follows. 1. Generally speaking, IMERG-Late is the best one of six satellite-only precipitation data sets over the globe. While the GSMaP suite, including GSMaP-NRT and GSMaP-MVK, shows the worst performance with relatively larger values of BIAS (>28%) and RMSE (>1.39 mm at hourly time scale and >9.41 mm at daily time scale). Compared with TRMM-based TMPA-RT, the GPM-based IMERG estimates including both IMERG-Early and IMERGLate significantly reduce the precipitation underestimates over the most parts of the world. Although IMERG generally seems superior to TMPA, its performance is still unsatisfactory in some specific regions, particularly over complex topography, such as western and northern COUNS, western China, etc. 2. The RMSE has a power function to the logarithm of precipitation intensity in all six purely satellite-derived precipitation data sets. On the basis of this functional relationship, we found that the RMSE values of six satellite-only precipitation data sets account for over 30% of corresponding precipitation intensity in rainfall events exceeding 32mm/day (or 8mm/h). It is predictable that these SPPs with lower precision in heavy rainfall events may produce the negative impacts in hydrologic studies to predict flood events when using these SPPs to force the flood monitoring and forecasting system. 3. The IMERG suite overestimates the proportions of light rainfall events (1-2 mm/day) over the globe. Furthermore, IMERG and GSMaP display more

significant overestimates over mainland China especially for IMERG with bias over 68% at hourly time scale. As for the accuracy, all SPPs in light precipitation is poor with the RMSE values exceeding 2 mm and 0.5 mm for daily and hourly time scales, respectively. 4. The five purely satellite-derived precipitation products, including IMERGEarly, IMERG-Late, GSMaP-NRT, GSMaP-MVK and PERSIANN-CCS, share some similarities and differences in spatial distribution patterns of total bias and its three components in cold and warm seasons. For the similarity, these products display the large overestimates in semi-humid areas in cold season, and these overestimates are primarily attributed to the false errors (>150%). On the other hand, some remarkable differences exist between these five satellite-only precipitation products. PERSIANN-CCS shows serious underestimates in southern humid regions both in two seasons, which is primarily attributed to missed precipitation. While GSMaP suite evidently overestimates the precipitation in semi-humid areas in warm season relative to other three SPPs. These overestimates mainly come from the hits and false biases over these areas.

Acknowledgments We are very grateful to the satellite-only precipitation dataset developers and ground observation providers. This work was sponsored by National Key Research and Development Program of China (2018YFA0605402) and National Natural Science Foundation of China (91647203). In addition, this work is partially supported by Key

Laboratory for Digital Land and Resources of Jiangxi Province, East China University of Technology (DLLJ201907).

Conflicts of interest The authors claim no conflicts of interest.

References Beck, H. E., Pan, M., Roy, T., Weedon, G. P., Pappenberger, F., van Dijk, A. I., Huffman, G. J., Adler, R. F., Wood, E. F., 2019. Daily evaluation of 26 precipitation datasets using Stage-IV gauge-radar data for the CONUS. Hydrol. Earth Syst. Sci. 23(1), 207-224. Beck, H. E., Vergopolan, N., Pan, M., Levizzani, V., van Dijk, A. I., Weedon, G. P., Brocca, L., Pappenberger, F., Huffman, G. J., Wood, E. F., 2017. Global-scale evaluation of 22 precipitation datasets using gauge observations and hydrological modeling. Hydrol. Earth Syst. Sci. 21(12), 6201-6217. Chen, H., Lu, D., Zhou, Z., Zhu, Z., Ren, Y., Yong, B. 2019a. An Overview of the Evaluation of Satellite Precipitation Products for Global Precipitation Measurement (GPM) (in Chinese). Water Resour. Prot. 35(1): 27-34.

Chen, H., Yong, B., Gourly, J. J., Liu, J., Ren, L., Wang, W., Hong, Y., Zhang, J., 2019b. Impact of the Crucial Geographical and Climatic Factors on the Input Source Errors of GPM-based Global Satellite Precipitation Estimates. J. Hydrol. 575: 116, doi: https://doi.org/10.1016/j.jhydrol.2019.05.020. Chen, M., Shi, W., Xie, P., Silva, V. B., Kousky, V. E., Wayne Higgins, R., Janowiak, J. E., 2008. Assessing objective techniques for gauge ‐ based analyses of global daily precipitation. J. Geophys. Res.-Atmos. 113(D4). Chen, S., Hong, Y., Cao, Q., Gourley, J. J., Kirstetter, P. E., Yong, B., Tian, Y., Zhang, Z. X., Shen, Y., Hu, J. J., Hardy, J., 2013. Similarity and difference of the two successive V6 and V7 TRMM multisatellite precipitation analysis performance over China. J. Geophys. Res.-Atmos. 118(23). Daly, C., Halbleib, M., Smith, J. I., Gibson, W. P., Doggett, M. K., Taylor, G. H., Pasteris, P. P., 2008. Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. Int. J. Climatol. 28(15), 2031-2064. Deng, P., Zhang, M., Guo, H., Xu, C., Bing, J., Jia, J., 2018. Error analysis and correction of the daily GSMaP products over Hanjiang River Basin of China. Atmos. Res. 214, 121–134. Gehne, M., Hamill, T. M., Kiladis, G. N., Trenberth, K. E., 2016. Comparison of global precipitation estimates across a range of temporal and spatial scales. J. Clim. 29(21), 7773-7795. Guo, H., Bao, A., Ndayisaba, F., Liu, T., Kurban, A., De Maeyer, P., 2017. Systematical

Evaluation of Satellite Precipitation Estimates Over Central Asia Using an Improved Error‐Component Procedure. J. Geophys. Res.-Atmos. 122(20), 10-906. Guo, H., Chen, S., Bao, A. M., Behrangi, A., Hong, Y., Ndayisaba, F., Hu, J. J., Stepanian, P. M., 2016. Early assessment of integrated multi-satellite retrievals for global precipitation measurement over China. Atmos. Res. 176–177:121-133. Hong, Y., Hsu, K. L., Sorooshian, S., Gao, X., 2004. Precipitation estimation from remotely sensed imagery using an artificial neural network cloud classification system, J. Appl. Meteorol. 43, 1834–1853. Hong, Y., Tang, G., Ma, Y., Huang, Q., Han, Z., Zeng, Z., Yang, Y., Wang, C., Guo, X., 2019. Remote Sensing Precipitation: Sensors, Retrievals, Validations, and Applications. Observation and Measurement of Ecohydrological Processes. 107128. Hou, A. Y., Kakar, R. K., Neeck, S., Azarbarzin, A. A., Kummerow, C. D., Kojima, M., Oki, R., Nakamura, K., Lguchi, T., 2014. The global precipitation measurement mission. Bull. Am. Meteorol. Soc. 95(5), 701-722. Huffman, G. J., Bolvin, D. T., Nelkin, E. J., Tan, J., 2019. Integrated Multi-satellitE Retrievals for GPM (IMERG) Technical Documentation. NASA/GSFC, pp. 1. Huffman, G. J., Bolvin, D. T., Nelkin, E. J., Wolff, D. B., Adler, R. F., Gu, G. J., Nelkin, E. J., Bowman, K. P., Hong, Y., Stocker, E. F., Stocker, E. F., 2007. The TRMM multisatellite precipitation analysis (TMPA): Quasi-global, multiyear, combinedsensor precipitation estimates at fine scales. J. Hydrometeorol. 8(1), 38-55. Huffman, G.J., 2018. The Transition in Multi-Satellite Products from TRMM to GPM

(TMPA to IMERG). NASA/GSFC, pp. 1. Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P., 2004. CMORPH: a method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeorol. 5 (3), 487–503. Khan, S., 2011. Satellite remote sensing and hydrologic modeling for flood inundation mapping in Lake Victoria basin: Implications for hydrologic prediction in ungauged basins. IEEE Trans. Geosci. Remote Sens. 49, 85–95. Kirstetter, P. E., Hong, Y., Gourley, J. J., Schwaller, M., Petersen, W., Zhang, J., 2013. Comparison of TRMM 2A25 products, version 6 and version 7, with NOAA/NSSL ground radar–based National Mosaic QPE. J. Hydrometeorol. 14(2), 661-669. Kubota, T., Shige, S., Hashizume, H., Aonashi, K., Takahashi, N., Seto, S., Hirose, M., Takayabu, Y. N., 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. IEEE Trans. Geosci. Remote Sensing. 45(7), 2259-2275. Kucera, P. A., Ebert, E. E., Turk, F. J., Levizzani, V., Kirschbaum, D., Tapiador, F. J., Loew, A., Borsche, M., 2013. Precipitation from space: Advancing Earth system science. Bull. Amer. Meteorol. Soc. 94(3), 365-375. Liu, Z., 2016. Comparison of ntegrated Multi-satellite Retrievals for GPM (IMERG) and TRMM Multi-satellite Precipitation Analysis (TMPA) monthly precipitation products: initial results. J. Hydrometeorol. 17(3): 777-790.

Lu, D. K., and Yong, B., 2018. Evaluation and hydrological utility of the latest GPM IMERG V5 and GSMaP V7 precipitation products over the Tibetan Plateau. Remote Sens. 10(12), 2022. Maggioni, V., Meyers, P. C., Robinson, M. D., 2016. A review of merged highresolution satellite precipitation product accuracy during the Tropical Rainfall Measuring Mission (TRMM) era. J. Hydrometeorol. 17(4), 1101-1117. Manz, B., Páez-Bimos, S., Horna, N., Buytaert, W., Ochoa-Tocachi, B., LavadoCasimiro, W., Willems, B., 2017. Comparative ground validation of IMERG and TMPA

at

variable

spatiotemporal

scales

in

the

tropical

Andes. J.

Hydrometeorol. 18(9), 2469-2489. Nguyen, P., Shearer, E. J., Tran, H., Ombadi, M., Hayatbini, N., Palacios, T., Huynh, P., Braithwaite, D., Updegraff, G., Hsu, K., Kuligowski, B., Logan, W. S., Sorooshian, S., 2019. The CHRS Data Portal, an easily accessible public repository for PERSIANN global satellite precipitation data. Sci. Data. 6, 180296. Prakash, S., Mitra, A. K., AghaKouchak, A., Liu, Z., Norouzi, H., Pai, D. S., 2018. A preliminary assessment of GPM-based multi-satellite precipitation estimates over a monsoon dominated region. J. Hydrol. 556, 865-876. Prakash, S., Seshadri, A., Srinivasan, J., Pai, D. S., 2019. A New Parameter to Assess Impact of Rain Gauge Density on Uncertainty in the Estimate of Monthly Rainfall over India. J. Hydrometeorol. 20(5), 821-832. Qi, W., Zhang, C., Fu, G., Sweetapple, C., Zhou, H., 2016. Evaluation of global fineresolution precipitation products and their uncertainty quantification in ensemble

discharge simulations. Hydrol. Earth Syst. Sci. 20(2), 903-920. Ren, Z., Zhao, P., Zhang, Q., 2010. Quality control procedures for hourly precipitation data from automatic weather stations in China (in Chinese), Meteorol. Mon., 36, 123–132. Shen, Y., Pan, Y., Yu, J., Zhao, P., 2013. Quality assessment of hourly merged precipitation product over China (in Chinese). Trans Atmos Sci. 36 (1), 37–46. Shen, Y., Zhao, P., Pan, Y., Yu, J., 2014. A high spatiotemporal gauge‐satellite merged precipitation analysis over China. J. Geophys. Res.-Atmos. 119(6), 3063-3075. Sorooshian, S., Hsu, K. L., Gao, X., Gupta, H. V., Imam, B., Braithwaite, D., 2000. Evaluation of PERSIANN system satellite-based estimates of tropical rainfall. Bull. Amer. Meteorol. Soc. 81(9), 2035-2046. Su, J., Lü, H., Zhu, Y., Wang, X., Wei, G., 2018. Component analysis of errors in four GPM-based precipitation estimations over Mainland China. Remote Sens. 10(9), 1420. Sun, Q., Miao, C., Duan, Q., Ashouri, H., Sorooshian, S., Hsu, K. L., 2018. A review of

global

precipitation

data

sets:

data

sources,

estimation,

and

intercomparisons. Rev. Geophys. 56(1), 79-107. Tan, J., Petersen, W. A., Kirstetter, P. E., Tian, Y., 2017. Performance of IMERG as a function of spatiotemporal scale. J. Hydrometeorol. 18(2), 307-319. Tang, G., Behrangi, A., Long, D., Li, C., Hong, Y., 2018. Accounting for spatiotemporal errors of gauges: A critical step to evaluate gridded precipitation products. J. Hydrol. 559, 294-306.

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. Tapiador, F. J., Turk, F. J., Petersen, W., Hou, A. Y., García-Ortega, E., Machado, L. A. T., Angelis, C. F., Salio, P., Kidd, C., Huffman, G.J., De Castro, M., 2012. Global precipitation measurement: Methods, datasets and applications. Atmos. Res. 104, 70-97. Tian, F., Hou, S., Yang, L., Hu, H., Hou, A., 2018. How Does the Evaluation of the GPM IMERG Rainfall Product Depend on Gauge Density and Rainfall Intensity? J. Hydrometeorol. 19(2), 339-349. Tian, Y., Peters‐Lidard, C. D., 2010. A global map of uncertainties in satellite‐based precipitation measurements. Geophys. Res. Lett. 37(24). Tian, Y., Peters‐Lidard, C. D., Eylander, J. B., Joyce, R. J., Huffman, G. J., Adler, R. F., Hsu, K. L., Turk, F. J., Garcia, M., Zeng, J., 2009. Component analysis of errors in satellite‐based precipitation estimates. J. Geophys. Res.-Atmos. 114(D24). Ushio, T., Sasashige, K., Kubota, T., Shige, S., Okamoto, K. I., Aonashi, K., Inoue, T., Takahashi, N., Iguchi, T., Kachi, M., Oki, R., Morimoto, T., Kawasaki, Z. I., 2009. A Kalman filter approach to the Global Satellite Mapping of Precipitation (GSMaP) from combined passive microwave and infrared radiometric data. J. Meteorol. Soc. Jpn. Ser. II, 87, 137-151. Wang, C., Tang, G. Q., Han, Z., Guo, X., Hong, Y., 2018. Global intercomparison and regional evaluation of GPM IMERG Version-03, Version-04 and its latest

Version-05 precipitation products: Similarity, difference and improvements. J. Hydrol. 564, 342-356. Wang, J., Hong, Y., Li, L., Gourley, J. J., Khan, S. I., Yilmaz, K. K., Adler, R. F., Policelli, F. S., Habib, S., Irwn, D., Limaye, A. S., Korme, T., Limaye, A. S., 2011. The coupled routing and excess storage (CREST) distributed hydrological model. Hydrol. Sci. J. 56(1), 84-98. Wang, Z., Zhong, R., Lai, C., Chen, J. C., 2017. Evaluation of the GPM IMERG satellite-based precipitation products and the hydrological utility. Atmos. Res. 196: 151-163. Wu, H., Adler, R. F., Hong, Y., Tian, Y., Policelli, F., 2012. Evaluation of global flood detection

using

satellite-based

rainfall

and

a

hydrologic

model. J.

Hydrometeorol. 13(4), 1268-1284. Wu, H., Adler, R. F., Tian, Y., Huffman, G. J., Li, H., Wang, J., 2014. Real‐time global flood estimation using satellite‐based precipitation and a coupled land surface and routing model. Water Resour. Res. 50(3), 2693-2717. Wu, Y., Guo, L., Zheng, H., Zhang, B., Li, M., 2019. Hydroclimate assessment of gridded precipitation products for the Tibetan Plateau. Sci. Total Environ. 660, 1555-1564. Xie, P., Chen, M., Yang, S., Yatagai, A., Hayasaka, T., Fukushima, Y., Liu, C., 2007. A

gauge-based

analysis

of

daily

precipitation

over

East

Asia. J.

Hydrometeorol. 8(3), 607-626. Xie, P., Xiong, A. Y., 2011. A conceptual model for constructing high‐resolution gauge

‐ satellite merged precipitation analyses. J. Geophys. Res.-Atmos. 116, D21106, doi: 10.1029/2011JD016118. Xu, R., Tian, F., Yang, L., Hu, H., Lu, H., Hou, A., 2017. Ground validation of GPM IMERG and TRMM 3B42V7 rainfall products over southern Tibetan Plateau based on a high‐density rain gauge network. J. Geophys. Res.-Atmos. 122(2), 910924, doi:10.1002/2016JD025418. Yang, Y., Luo, Y., 2014. Evaluating the performance of remote sensing precipitation products CMORPH, PERSIANN, and TMPA, in the arid region of northwest China. Theor. Appl. Climatol. 118(3), 429-445. Yong, B., Chen, B., Gourley, J. J., Ren, L., Hong, Y., Chen, X., Wang, W. G., Chen, S., Gong, L. Y., 2014. Intercomparison of the Version-6 and Version-7 TMPA precipitation products over high and low latitudes basins with independent gauge networks: Is the newer version better in both real-time and post-real-time analysis for water resources and hydrologic extremes?. J. Hydrol. 508, 77-87. Yong, B., Chen, B., Tian, Y., Yu, Z., Hong, Y., 2016. Error-component analysis of TRMM-based multi-satellite precipitation estimates over mainland China. Remote Sens. 8(5), 440. Yong, B., Liu, D., Gourley, J. J., Tian, Y., Huffman, G. J., Ren, L., Hong, Y., 2015. Global view of real-time TRMM multisatellite precipitation analysis: Implications for its successor global precipitation measurement mission. Bull. Am. Meteorol. Soc. 96(2), 283-296. Yong, B., Ren, L., Hong, Y., Wang, J. H., Gourley, J. J., Jiang, S., Chen, X., Wang, W.,

2010. Hydrologic evaluation of Multisatellite Precipitation Analysis standard precipitation products in basins beyond its inclined latitude band: A case study in Laohahe

basin,

China. Water

Resour.

Res. 46,

W07542,

doi :10.1029/2009WR008965. Yong, B., Ren, L., Hong, Y., Gourley, J. J., Tian, Y., Huffman, G. J., Chen X., Wang W. G., Wen, Y. X., 2013. First evaluation of the climatological calibration algorithm in the real‐time TMPA precipitation estimates over two basins at high and low latitudes. Water Resour. Res. 49(5), 2461-2472. Zhang, Y., Hong, Y., Wang, X., Gourley, J. J., Xue, X., Saharia, M., Ni, G., Wang, G., Huang, Y., Chen, S., Tang, G., 2015. Hydrometeorological analysis and remote sensing of extremes: Was the July 2012 Beijing flood event detectable and predictable by global satellite observing and global weather modeling systems?. J. Hydrometeorol. 16(1), 381-395. Zhao, H., Yang, B., Yang, S., Huang, Y., Dong, G., Bai, J., Wang, Z., 2018. Systematical estimation of GPM-based global satellite mapping of precipitation products over China. Atmos. Res. 201, 206-217. Zhu, Z., Yong, B., Ke, L., Wang, G., Ren, L., Chen, X., 2018. Tracing the Error Sources of Global Satellite Mapping of Precipitation for GPM (GPM-GSMaP) Over the Tibetan Plateau, China. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 11 (8), 2181–2191. https://doi.org/10.1109/JSTARS.2018.2825336.

Fig.1. (a) Spatial distribution of mean daily precipitation for CPC unified over the globe; (b) Spatial distribution of mean hourly precipitation for Merged Precipitation Analysis (MPA) over mainland China; (c) density maps of gauges used in CPC unified over the globe; (d) density maps of gauges used in MPA over mainland China. Fig. 2. Spatial distribution of POD (a-f) and FAR (g-l) for six satellite-only precipitation products (SPPs) at daily and 0.5 ° resolution over the globe for the period from February 2017 to January 2019. Fig. 3. Same as Fig. 2, but for BIAS and RMSE: (a-f) BIAS, and (g-l) RMSE. Fig. 4. Spatial distribution of POD and FAR for five SPPs over mainland China for the period from February 2017 to January 2019: POD of five SPPs at hourly scale (first column), POD of five SPPs at daily scale (second column), FAR of five SPPs at hourly scale (third column), and FAR of five SPPs at daily scale (fourth column). Fig. 5. Same as Fig. 4, but for BIAS and RMSE: BIAS of five SPPs at hourly scale (first line), BIAS of five SPPs at daily scale (second line), RMSE of five SPPs at hourly scale (third line), RMSE of five SPPs at daily scale (fourth line). Fig. 6. Total bias and its three components of five SPPs for the cold season over mainland China for the period from February 2017 to January 2019. Fig. 7. Same as Fig. 6, but for warm season. Fig. 8. The variations of PDF values for six SPPs with precipitation intensity: (a-b) at daily and 0.5 ° resolution over the globe; (c-d) at daily and 0.1 ° resolution over

mainland China; (e-f) at hourly and 0.1° resolution over mainland China. Fig. 9. Same as Fig. 8, but for RMSE. Note that the x axis is in log scale. Fig. 10. The variations of RMSE values for five SPPs with rain gauge density over humid regions of mainland China.

Table 1. the parameters of six satellite-only precipitation datasets.

Data Product

Resolution

Coverage

Explanation

Reference

4-hr date

Huffman et

source(s) IMERG-

PMW, 0.1°/30min 90°S to 90°N

Early

IR

latency

al., (2019)

IMERG-

PMW,

12-hr date

Huffman et

Late

IR

latency

al., (2019)

GSMaP-

PMW,

4-hr date

Kubota et

0.1°/30min 90°S to 90°N

0.1°/1h

60°S to 60°N

NRT

IR

latency

al., (2007)

GSMaP-

PMW,

3-day date

Ushio et al.,

IR

latency

(2009)

PMW,

8-hr date

Huffman et

latency

al., (2007)

3-day date

Sorooshian

latency

et al.,

0.1°/1h MVK

TMPA-RT

0.25°/3h

60°S to 60°N

50°S to 50°N

IR PERSIANNIR CCS

0.04°/1h

60°S to 60°N

(2000); Hong et al., (2004)

Table 2. The list of the evaluation metrics used in this study a

Perfect Statistic metrics

Equation value

Probability of Detection POD =

H H+M

1

FAR =

F H+F

0

(POD) False Alarm Ratio (FAR) 𝑛

Correlation Coefficient CC = (CC) Relative Bias (BIAS)

∑𝑖 = 1(𝐺𝑖 ― 𝐺)(𝑆𝑖 ― 𝑆) 𝑛

𝑛

∑𝑖 = 1(𝐺𝑖 ― 𝐺)2 × ∑𝑖 = 1(𝑆𝑖 ― 𝑆)2 BIAS =

∑(𝑆 ― 𝐺)

∑(𝑆𝐻 ― 𝐺𝐻)

Hit bias

Hit bias =

Miss bias

Miss bias =

False bias

False bias =

Root Mean Squared Error (RMSE)

× 100%

∑𝐺

∑𝐺

× 100%

∑(𝑆𝑀 ― 𝐺𝑀) ∑𝐺 ∑(𝑆𝐹 ― 𝐺𝐹) ∑𝐺

1 RMSE = 𝑛

∑

𝑛

0

0

× 100%

0

× 100%

0

(𝑆𝑖 ― 𝐺𝑖)2

𝑖=1

1

0

aNotation:

H represents the number of hit rainfall events for satellite-only precipitation

datasets; M denotes the number of miss rainfall events for satellite-only precipitation datasets; F indicate the number of false rainfall events for satellite-only precipitation datasets; n is the number of sample sizes for satellite-only precipitation datasets; G represents precipitation observations from gauge; S is precipitation estimates from satellite, and the subscripts H, M, and F denote estimates from satellite/gauge in the three rainfall events.

Table 3. the calculated values of evaluation metrics for six purely satellite-derived precipitation datasets at 0.5° and daily resolution over the whole globe.

Product

POD

FAR

CC

BIAS (%)

RMSE (mm)

IMERG-Early

0.75

0.34

0.60

26.61

7.20

IMERG-Late

0.75

0.32

0.62

26.55

7.25

GSMaP-NRT

0.66

0.29

0.48

44.71

12.37

GSMaP-MVK

0.68

0.29

0.57

34.18

9.41

TMPA-RT

0.60

0.32

0.53

-13.16

7.24

PERSIANN-CCS

0.68

0.46

0.44

23.06

8.19

Table 4. the calculated values of evaluation metrics for five satellite-only precipitation datasets at 0.1° resolution over the whole mainland China.

Product

Timescale

POD

FAR

CC

BIAS

RMSE

IMERGEarly IMERGLate GSMaPNRT

Hourly

GSMaPMVK PERSIANNCCS IMERGEarly IMERGLate GSMaPNRT GSMaPMVK PERSIANNCCS

Daily

(%)

(mm)

0.51

0.57

0.46

17.93

0.83

0.54

0.54

0.50

18.64

0.80

0.43

0.50

0.23

35.78

1.74

0.47

0.51

0.30

28.46

1.39

0.20

0.73

0.24

-19.27

0.92

0.74

0.39

0.67

19.71

7.00

0.74

0.36

0.69

21.27

7.14

0.74

0.37

0.46

40.28

12.73

0.74

0.38

0.52

32.60

10.64

0.66

0.56

0.39

-21.47

7.99