Has IMERG_V07 Improved the Precision of Precipitation Retrieval in Mainland China Compared to IMERG_V06?

Abstract

Integrated Multi-satellitE Retrievals for GPM (Global Precipitation Measurement) (IMERG) is the primary high spatiotemporal resolution precipitation product of the GPM era. To assess the applicability of the latest released IMERG_V07 in mainland China, this study systematically evaluates the error characteristics of IMERG_V07 from the perspective of different seasons, precipitation intensity, topography, and climate regions on an hourly scale. Ground-based meteorological observations are used as the reference, and the performance improvement of IMERG_V07 relative to IMERG_V06 is verified. Error evaluation is conducted in terms of precipitation amount and precipitation frequency, and an improved error component procedure is utilized to trace the error sources. The results indicate that IMERG_V07 exhibits a smaller RMSE in mainland China, especially with significant improvements in the southeastern region. IMERG_V07 shows better consistency with ground station data. IMERG_V07 shows an overall improvement of approximately 4% in capturing regional average precipitation events compared to IMERG_V06, with the northwest region showing particularly notable enhancement. The error components of IMERG_V06 and IMERG_V07 exhibit similar spatial distributions. IMERG_V07 outperforms V06 in terms of lower Missed bias but slightly underperforms in Hit bias and False bias compared to IMERG_V06. IMERG_V07 shows improved ability in capturing precipitation frequency for different intensities, but challenges remain in capturing heavy precipitation events, missing light precipitation, and winter precipitation events. Both IMERG_V06 and IMERG_V07 exhibit notable topography dependency in terms of Total bias and error components. False bias is the primary error source for both versions, except in winter, where high-altitude regions (DEM > 1200 m) primarily contribute to Missed bias. IMERG_V07 has enhanced the accuracy of precipitation retrieval in high-altitude areas, but there are still limitations in capturing precipitation events. Compared to IMERG_V06, IMERG_V07 demonstrates more concentrated error component values in the four climatic regions, with reduced data dispersion and significant improvement in Missed bias. The algorithm improvements in IMERG_V07 have the most significant impact in arid regions. False bias serves as the primary error source for both satellite-based precipitation estimations in the four climatic regions, with a secondary contribution from Hit bias. The evaluation results of this study offer scientific references for enhancing the algorithm of IMERG products and enhancing users’ understanding of error characteristics and sources in IMERG.

1. Introduction

Precipitation is an important component of the earth’s water cycle and plays a crucial role in the atmosphere, hydrosphere, and biosphere. Due to its significant spatiotemporal heterogeneity, accurate and rapid acquisition of information on the spatiotemporal distribution of precipitation has become the key to understanding and predicting weather and climate changes [1,2,3]. With the rapid development and improvement of satellite remote sensing technology, satellite precipitation products have become the mainstream technique for obtaining high spatiotemporal resolution precipitation information with a short latency.

As TRMM’s successor, the Global Precipitation Measurement (GPM) mission was launched in March 2014, symbolizing a graceful transition from the TRMM era to the GPM era. The GPM program is mainly implemented by the National Aeronautics and Space Administration (NASA) and the Japanese Aerospace Exploration Agency (JAXA). GPM represents the next generation of global precipitation products with an advanced radar/radiometer measurement system [4]. The GPM Core Observatory carries the dual-frequency phased array Precipitation Radar (DPR) and a multichannel GPM Microwave Imager (GMI) to provide more accurate estimates for light rain (<0.5 mm/h), solid precipitation, and the microphysical properties of precipitating particles [5]. The Integrated Multi-Satellite Retrievals for GPM (IMERG) is one of the mainstream precipitation products during the GPM era. IMERG entered the V06 era on 13 March 2019. Based on algorithm differences and varying time lags, three algorithm-based products have been generated: IMERG-Early, IMERG-Late, and IMERG-Final. The Early Run (with a latency of 4 h) provides relatively quick results for flood analysis and other short-fuse applications by only employing forward morphing. The Late Run (with a latency of 12 h) employs both forward and backward morphing with later data and is appropriate for daily and longer applications, such as crop forecasting. The Final Run (with a latency of 3.5 months) introduces monthly precipitation gauge analyses, providing more accurate results in regions with gauge information. The Final Run is regarded as a research-grade product [6,7].

In recent years, numerous studies have demonstrated the advantages of IMERG_V06 in precipitation retrieval, confirming its crucial role in water resource management and early disaster warning. However, shortcomings have been identified in the accuracy of IMERG_V06’s precipitation retrieval when it comes to different seasons, precipitation intensities, terrains, and climates. Studies have focused on the relationship between the performance of GPM and precipitation intensity, especially for light [8,9] and extreme precipitation events [10,11]. Some previous results found that IMERG_V06 identified a persistent overestimation of precipitation and slightly overestimates the proportion of light precipitation events [5,12,13,14,15,16]. Furthermore, studies have found that the performance of IMERG_V06 remains unsatisfactory in certain regions due to the impacts of complex topography and adverse climatic conditions on precipitation structure, as well as the associated challenges with satellite retrievals from space [5,17,18,19]. Additionally, due to interference from ice particles in the winter atmosphere or ice cover over land surfaces, the Passive Microwave (PMW) retrievals are affected, while the Infrared (IR) input with morphing technique directly infers from cloud top temperature, mitigating the impact of seasonal variation on retrieval results. Furthermore, IR is less susceptible to the challenge of distinguishing between precipitation and frozen surfaces, which is a specific issue for microwave sensors. Thus, this leads to better performance in the inversion of summer precipitation compared to winter precipitation by IMERG_V06 [6,18].

IMERG_V07 was released in July 2023 [7]. It incorporates a wide range of changes to all aspects of the algorithm, many of which were implemented in response to feedback on V06. Notably, IMERG_V07 introduces PMW estimates over frozen surfaces for the first time. In previous versions, PMW retrievals were omitted for grid boxes with estimated frozen surface types. To address the issue of underestimation of precipitation at high latitudes, a climatological Fuchs adjustment was implemented in V07 over Eurasia north of 45° N, while the Legates–Willmott scheme continues to be used for other land areas. The maximum permitted precipitation rate in V07 is 200 mm/h, an increase from 50 mm/h in V05 and a combination of 50 mm/h (for IR) and 120 mm/h (for PMW) in V06. There is a known problem with low bias in the Goddard Profiling (GPROF) at high latitudes due to the DPR’s low sensitivity to light rain and snow. Consequently, in Version 07, both the Combined and the GPROF libraries have been modified to give more weight to GMI data in these passes. Hourly-scale evaluation provides more timely and precise precipitation information, enabling the capture and description of short-duration, high-intensity precipitation events, and facilitating better responses to extreme weather events. However, there is currently limited research on systematically evaluating the accuracy of IMERG_V07 at the hourly scale, taking into account the typical error characteristics observed in existing V06 products. Additionally, the understanding of error characteristics of IMERG products in relation to different geographical factors and climate features is also limited. The effects of IMERG_V07 improvements on orographic, frozen, and light precipitation aspects, for example, remain unclear at present.

Therefore, in order to assess the retrieval accuracy of the latest version of IMERG, IMERG_V07, this study conducted a comprehensive and quantitative comparative evaluation of the error characteristics between IMERG_V06 and IMERG_V07. The evaluation was performed based on hourly precipitation data from 2166 meteorological stations across mainland China. Various methodologies, including continuity evaluation metrics, classification evaluation metrics, and error component decomposition, were employed. The evaluation encompassed different seasons, precipitation intensities, terrains, and climate regions. The research findings provide valuable insights into the error characteristics, revealing the improvements in IMERG_V07 compared to previous versions. This contributes to the continuous enhancement and refinement of the IMERG satellite precipitation algorithm. Through in-depth analysis of the error characteristics, the study identifies error sources and proposes potential strategies for improvement, serving as a beneficial reference for the advancement and application of satellite precipitation algorithms. Additionally, the results offer practical guidance for users of IMERG data in mainland China.

The remaining sections of this paper are structured as follows: Section 2 provides a detailed description of the materials and methods employed in this study, including the study area, datasets, and methodologies utilized. Section 3 presents a comprehensive evaluation of the error characteristics of IMERG_V06 and IMERG_V07 in various aspects such as different seasons, precipitation intensities, topography, and climate regions. Finally, Section 4 summarizes the main research findings of this paper.

2. Data and Methodology

2.1. Study Area

Mainland China is located between 73°33′E–135°05′E and 3°51′N–53°33′N (Figure 1a). It exhibits a complex distribution of terrain, characterized by a gradual west-to-east decrease in elevation, forming distinct stepped patterns. The eastern region is relatively flat, while the northwestern region gradually becomes more geographically intricate. Mainland China is located in the western Pacific region and is influenced by complex terrain and unique land–sea distribution. As a result, precipitation exhibits significant spatiotemporal heterogeneity. The annual precipitation shows a gradual decrease trend from the southeastern coastal areas towards the northwestern inland regions. This study divided mainland China into four climatic regions based on annual precipitation amounts [20] (Figure 1a): (1) Humid region, covering the southeastern part of China, experiences annual average precipitation exceeding 800 mm. (2) Semi-humid region extends from northeastern China to the southwestern Qinghai–Tibet Plateau area and has an annual average precipitation ranging from 400 to 800 mm. (3) Semi-arid region includes the Tianshan Mountain range and a narrow strip extending from the southwestern Qinghai–Tibet Plateau to northern Inner Mongolia. It experiences an annual average precipitation ranging from 200 to 400 mm. (4) Arid region encompasses the main areas of Xinjiang and the southern part of Inner Mongolia. It has an annual average precipitation below 200 mm. Figure 1b illustrates the spatial distribution of hourly surface meteorological stations in mainland China. From the figure, it can be observed that the distribution of surface stations exhibits a denser concentration in the eastern regions compared to the sparser distribution in the western regions. Within the study area, the southeastern part shows the highest density of station distribution, while the distribution becomes sparser in the northwestern regions and the geographically complex Qinghai–Tibet Plateau area.

2.2. Datasets

2.2.1. Ground-Based Observations

This article utilizes the hourly surface meteorological observation data provided by the China Meteorological Administration (CMA) as the ground reference data. The data are sourced from the official website of the CMA (http://data.cma.cn). The precipitation data from all meteorological stations have undergone stringent quality control: extreme value examination, internal consistency assessment, and spatial consistency evaluation [21]. These rigorous procedures ensure the accuracy and reliability of the data. Notably, this dataset has been widely employed as a benchmark for evaluating systematic errors in numerical weather prediction models [1,22,23].

In order to ensure the scientific rigor and credibility of the study, we excluded stations with missing precipitation data during the period from September 2019 to September 2021. Ultimately, this study selected hourly precipitation data from 2166 meteorological stations, which had a data availability rate of 100% during the study period, as the reference data.

2.2.2. GPM IMERG

IMERG (Integrated Multi-Satellite Retrievals for GPM) is a representative product of the Global Precipitation Measurement (GPM) mission. It provides users with high spatiotemporal resolution global precipitation data. The temporal resolution is 30 min, and the spatial resolution is 0.1° × 0.1° [24]. IMERG incorporates multiple sensors, including microwave, infrared, and precipitation radar, to leverage the complementary advantages of various data sources. In order to enhance the inversion accuracy of IMERG, several algorithmic improvements and data optimizations have been implemented in IMERG_V07 compared to IMERG_V06 [7]. These improvements encompass algorithmic enhancements, extreme precipitation error correction, solid precipitation error correction, frequency error correction, and more. IMERG_V07 introduces the CORRA V07 and GPROF V07 algorithms, addressing spatial offset issues in the gridding process. Furthermore, data biases are corrected by refining the intercalibration process of PMW data. The data sources for calculating motion vectors have been updated to ensure more reliable results. To improve the accuracy of infrared precipitation retrieval, the PERSIANN Dynamic Infrared–Rain Rate (PDIR-NOW) algorithm has been introduced, along with automated infrared quality control. Additionally, the Scheme for Histogram Adjustment with Ranked Precipitation Estimates in the Neighborhood (SHARPEN) combines instantaneous PMW detection results with propagating precipitation, further enhancing the continuity and accuracy of the results. Several other improvements have been made as well. The upper limit of precipitation rate has been increased to 200 mm/h to better capture extreme precipitation events. The joint Fuchs-Legates gauge correction method has been implemented to address underestimation issues in precipitation analysis. Grid-based variables have been utilized to refine precipitation phase determination, improving the accuracy of precipitation type classification. For solid precipitation, PMW estimates have been incorporated to improve the estimation of precipitation on frozen surfaces. For detailed information on the algorithmic improvements, please refer to the technical documentation of GPM IMERG_V07 available at https://gpm.nasa.gov/data/, accessed on 21 February 2024.

The IMERG Satellite-based Precipitation Estimations (SPEs) consist of three versions: IMERG Early Run, IMERG Late Run, and IMERG Final Run. However, for this study, we only considered the IMERG Final Run since the Early and Late versions had not been officially released when we started the current study.

2.3. Methods

2.3.1. Continuous Statistical Indices

Several widely used continuous statistical indices were selected to quantitatively evaluate the accuracy and error of IMERG SPEs against the ground observations of precipitation amounts, including Pearson correlation coefficient (CC), relative bias (RB), root-mean square error (RMSE), and BIAS (

Table 1. Continuous statistical indices.

Metric Categories	Statistical Metrics	Formula	Optimal Value
Continuous Metrics	Correlation Coefficient (CC)	$CC={\displaystyle \frac{{\sum }_{i=1}^N\left(G_i-\bar{G}\right)\left(S_i-\bar{S}\right)}{\sqrt{{\sum }_{i=1}^N{\left(G_i-\bar{G}\right)}^2}\sqrt{{\sum }_{i=1}^N{\left(S_i-\bar{S}\right)}^2}}}$	1
	Relative Bias (RB)	$RB={\displaystyle \frac{{\sum }_{i=1}^N\left(S_i-G_i\right)}{{\sum }_{i=1}^N{G}_i}}\times 100\mathrm{\%}$	0
	Root Mean Square Error (RMSE)	$RMSE=\sqrt{{\displaystyle \frac{1}{N}}{\sum }_{i=1}^N{\left(S_i-G_i\right)}^2}$	0
	BIAS	$BIAS={\displaystyle \sum _{i=1}^N}\left(S_i-G_i\right)/N$	0

Note: N denotes the number of samples; i denotes the ith data; G_i refers to the precipitation estimated by gauge observations and $\bar{G}$ is the average of the precipitation estimated by gauge observations, S_i and $\bar{S}$ represent the satellite precipitation estimates and their average.

). CC describes the agreement between the SPEs and ground observations; The RB describes the systematic deviation between SPEs and ground observation data and RMSE measures the average magnitude of the error; BIAS describes the total bias between the SPEs and the ground observations.

2.3.2. Categorical Statistical Indices

To evaluate the ability to detect precipitation events and gain a deeper understanding of error sources, this study employed three categorical statistical indicators: Probability of Detection (POD), Misses (MIS), and False Alarm Ratio (FAR) [24]. POD indicates the percentage of precipitation events accurately detected by SPEs. FAR represents the percentage of precipitation events identified solely by satellites while not observed by gauges. MIS describes the proportion of precipitation events captured by gauges but missed by satellites. All categorical indicators have values ranging from 0 to 1. To account for the temporal resolution of 1 h and mitigate the impact of light precipitation, a threshold of 0.1 mm/h was set to determine whether precipitation has occurred or not.

$$\mathrm{POD}\left(\mathrm{\%}\right)={\displaystyle \frac{N\left\{\left(S_i\ge t\right)\mathrm{and}\left(G_i\ge t\right)\right\}}{N\left\{\left(G_i\ge t\right)\right\}}}\times 100$$

(1)

$$\mathrm{FAR}\left(\mathrm{\%}\right)={\displaystyle \frac{N\left\{\left(S_i\ge t\right)\mathrm{and}\left(G_i<t\right)\right\}}{N\left\{\left(S_i\ge t\right)\right\}}}\times 100$$

(2)

$$\mathrm{MIS}\left(\mathrm{\%}\right)={\displaystyle \frac{N\left\{\left(S_i<t\right)\mathrm{and}\left(G_i\ge t\right)\right\}}{N\left\{\left(G_i\ge t\right)\right\}}}\times 100$$

(3)

where i denotes the ith data, S_i represents the hourly precipitation estimated by satellite-based precipitation products, and G_i corresponds to the gauge observations. N denotes the total number of hours under different conditions and t signifies the threshold value for valid hourly precipitation.

2.3.3. Error Decomposition

In order to assess the accuracy of precipitation estimation, we employed an error component decomposition method [25] to decompose the Total bias into three distinct components: Hit bias, Missed bias, and False bias. Hit bias occurs when both SPEs and gauge observations detect a precipitation event, but the estimated precipitation differs from the actual situation. It can manifest as either positive or negative bias. Missed bias refers to precipitation events that go undetected by SPEs, resulting in a negative bias as it represents the amount of missed precipitation. Additionally, SPEs may sometimes exhibit false detections of precipitation features [26,27], erroneously identifying precipitation when no corresponding measurements are captured by gauges. This type of precipitation is known as False bias and consistently presents as a positive value. It is worth mentioning that these biases have the potential to counterbalance each other, resulting in a total bias that is smaller in magnitude than the individual components. To assess the accuracy of precipitation amount estimation, an error decomposition scheme is employed [17]. The error components can be calculated using the following equations:

$$\mathrm{Total}\mathrm{bias}\left(\mathrm{m}\mathrm{m}\right)={\sum }_{i=1}^N\left(S_i-G_i\right)/N,$$

(4)

$$\mathrm{Hit}\mathrm{bias}\left(\mathrm{m}\mathrm{m}\right)={\sum }_{i=1}^N\left(S_i-G_i\right)/N,\mathrm{when}\left(S_i\ge t\right)\mathrm{and}\left(G_i\ge t\right)$$

(5)

$$\mathrm{Missed}\mathrm{bias}\left(\mathrm{m}\mathrm{m}\right)=\sum _{i=1}^N\left(S_i-G_i\right)/N,\mathrm{when}\left(S_i<t\right)\mathrm{and}\left(G_i\ge t\right)$$

(6)

$$\mathrm{False}\mathrm{bias}\left(\mathrm{m}\mathrm{m}\right)={\sum }_{i=1}^N\left(S_i-G_i\right)/N,\mathrm{when}\left(S_i\ge t\right)\mathrm{and}\left(G_i<t\right)$$

(7)

where i denotes the ith data, S_i represents the hourly precipitation estimated by satellite-based precipitation products, G_i corresponds to the gauge observations, t signifies the threshold value for valid hourly precipitation, and N denotes the number of simples.

3. Results

3.1. Error Characteristics of IMERG V6 and V7

3.1.1. Characterization of Precipitation Amount Errors

Figure 2 presents the spatial distribution of daily precipitation amounts for meteorological stations, IMERG_V06, and IMERG_V07 across different seasons. From the figure, it can be observed that the spatial-temporal distribution of precipitations in IMERG_V06 and IMERG_V07 closely resemble the ground-based observations. On a seasonal scale, both IMERG_V06 and IMERG_V07 products effectively capture the characteristics of abundant summer precipitation, reduced winter precipitation, and moderate precipitation in spring and autumn. Additionally, they accurately reflect the seasonal pattern of higher spring precipitation compared to autumn precipitation in humid regions. In terms of spatial distribution, both IMERG_V06 and IMERG_V07 successfully capture the gradient feature of higher precipitation in the southeastern part and lower precipitation in the northwestern part of mainland China. However, it should be noted that there is a noticeable overestimation of precipitation in coastal areas, particularly in the southeastern region (southeastern part of mainland China), in the IMERG products (Figure 2). This overestimation issue is observed in all four seasons, with the most significant impact observed during the summer and winter.

Figure 3 illustrates the spatial distribution of different accuracy evaluation metrics for IMERG_V06 and IMERG_V07 at an hourly scale. The Bias metric represents the difference between IMERG SPEs and the precipitation measured by meteorological stations. From Figure 3a–e, it is evident that both SPEs exhibit significant Bias in the humid region, particularly in the southeastern coastal areas. In the humid region, IMERG_V07 shows a higher level of overestimation (Bias = 0.24 mm) compared to IMERG_V06 (Bias = 0.18 mm). In the semi-humid, semi-arid, and arid regions, both SPEs demonstrate similar performance with relatively smaller overestimation of precipitation. Despite some degree of overestimation errors, the overall consistency of IMERG_V07 with the station observations is noticeably better than that of IMERG_V06, as indicated by the higher CC of IMERG_V07 compared to IMERG_V06. Particularly in the northwestern region (northwestern part of mainland China), the CC of IMERG_V07 has improved from around 0.1 in IMERG_V06 to 0.3–0.4. In the humid region, the overall CC value of IMERG_V07 (CC = 0.41) is higher than that of the arid region (CC = 0.31).

In terms of RB, both satellite precipitation products tend to overestimate precipitation, which is consistent with the distribution of Bias errors. In humid, semi-humid, and semi-arid regions, the RB of both SPEs falls within the range of 0–30%. However, in arid regions, the RB for both versions of IMERG exceeds 80%, with IMERG_V07 (83.5%) having a lower RB value than IMERG_V06 (98.6%). This change may be attributed to the updated algorithm in IMERG_V07, which improved the retrieval accuracy for light precipitation [7]. As for RMSE, both SPEs exhibit a decreasing trend from the southeast to the northwest, consistent with the distribution of precipitation. This indicates that as the precipitation amount decreases, the RMSE values gradually decrease. RMSE represents the average deviation between meteorological stations and SPEs. Comparing the two precipitation products, it is observed that IMERG_V07 has smaller RMSE values than IMERG_V06 in all four climate regions, particularly in the humid region.

Overall, IMERG_V07 demonstrates better consistency with ground station data, further validating the positive impact of algorithm improvements in IMERG_V07 by the algorithm developers.

To visually understand the precipitation retrieval accuracy and errors of the two versions of IMERG SPEs in different seasons, a Taylor diagram (Figure 4) is used to display the overall performance at an hourly scale based on three indicators, including correlation coefficient, standardized standard deviation, and root mean square error. As shown in Figure 4, IMERG_V07 consistently exhibits higher CC values than IMERG_V06 in all seasons, especially in winter. The CC value of IMERG_V06 is only around 0.1 in winter, while the CC value of IMERG_V07 is approximately 0.3–0.4. This is consistent with the spatial distribution of IMERG_V07 in Figure 4, which shows higher correlation coefficients. It indicates that IMERG_V07 has significantly improved data consistency. Although IMERG_V07 has slightly lower RMSE values than IMERG_V06 in summer and autumn, its performance is inferior to IMERG_V06 in winter and spring. IMERG_V06 has a smaller standard deviation (STD) than IMERG_V07 in spring and winter, while both precipitation products perform similarly in summer. In autumn, IMERG_V07 shows better performance. Overall, IMERG_V06 and IMERG_V07 exhibit their respective advantages in different seasons. IMERG_V07 has shown significant improvement in data consistency compared to IMERG_V06. It performs better than IMERG_V06 in summer and autumn but falls short in winter and spring.

3.1.2. Comparative Analysis of Precipitation Frequency Error Characteristics

The ability to capture precipitation events is an important aspect of evaluating the precipitation retrieval accuracy of SPEs. In this section, the detection capability of IMERG_V06 and IMERG_V07 for precipitation events at the hourly scale is systematically evaluated using three metrics: POD, FAR, and MIS. Figure 5 shows the spatial distribution of the classification statistics indicators for IMERG_V06 and IMERG_V07. There are significant regional differences in the ability to capture precipitation events between IMERG_V06 and IMERG_V07, similar to the characteristics of precipitation error. Both versions exhibit a decreasing accuracy in capturing precipitation events from the southeastern coastal areas to the northwestern inland regions. Benefiting from changes in the data sources of the sensors and algorithm updates, IMERG_V07 shows significant improvement in the detection capability of precipitation events compared to IMERG_V06.

In terms of POD, both IMERG_V06 and IMERG_V07 demonstrate strong precipitation event detection capabilities in the humid region and the semi-humid region, with POD values exceeding 50%. Moreover, the POD in the semi-humid region is higher than that in the humid region. However, in the semi-arid and arid regions, the POD of IMERG_V06 remains around 34–38%, with a few stations below 10%. On the other hand, IMERG_V07 achieves a POD of 43.6–44.5% in these regions.

In terms of MIS, Figure 5b–e indicates that IMERG_V07 shows significant improvement in missed precipitation events, particularly in the northwestern region. For several stations in the northwestern region, IMERG_V06 had a high MIS of up to 90%, while IMERG_V07 reduced the MIS to around 60%. This improvement may be attributed to “IMERG_V07 increased PMW estimates over frozen surfaces [7]”. For the entire mainland of China, the regional average MIS of IMERG_V06 is 49.8%, while that of IMERG_V07 is 45.8%, representing an overall decrease of 4%. The MIS values of both SPEs are lowest in the semi-humid region, followed by the humid region, and highest in the arid region. IMERG_V06’s performance in FAR is less satisfactory, with values increasing from the humid region to the arid region. Although the FAR in the humid region is relatively low, individual stations still reach up to 80%. In the northern regions of China, the FAR of IMERG_V06 reaches 80–100%. In contrast, IMERG_V07 maintains an hourly scale FAR between 50% and 60% in regions other than the arid region, with a uniform distribution and stable performance. In the arid region, although IMERG_V07 still has a relatively high FAR (70.9%), it is greatly reduced compared to IMERG_V06.

In summary, IMERG_V07 demonstrates an overall improvement in the ability to capture precipitation events compared to IMERG_V06, with a 4% increase in regional average accuracy. The improvement is particularly significant in the northwestern region, which may be attributed to the sensor updates and algorithm improvements in IMERG_V07.

3.1.3. Comparative Analysis of Error Components

Due to interdependencies among various metrics, relying solely on conventional indicators to evaluate the performance of SPEs has limitations. For instance, bias can be influenced by the offsetting of positive and negative errors, which can result in the presence of significant absolute errors even when the bias values are small. Ushio et al. [28] emphasized the crucial importance of assessing error sources for improving SPE algorithms. To enable a more objective error characterization and identify error sources, Tian Yudong [29] proposed an effective error decomposition method called the Error Component Procedure. It decomposes the Total bias into three error components: Hit bias, Missed bias, and False bias, facilitating further analysis of precipitation errors resulting from the detection of different precipitation events. Figure 6 illustrates the spatial distribution of Total bias and the three error components for IMERG_V06 and IMERG_V07.

Firstly, from Figure 6a–e, it can be observed that the values of Total bias are relatively small (ranging from 0 to 0.3 mm) and significantly smaller than the sum of the three error components. This indicates the presence of offsetting positive and negative errors among the error components. Therefore, relying solely on the conventional total precipitation bias to evaluate the performance of SPEs is not sufficiently accurate, and the analysis of precipitation amount errors should consider the characteristics of precipitation events [30,31]. Both IMERG_V06 and IMERG_V07 exhibit an overestimation of precipitation in mainland China, particularly in the humid regions.

The Hit bias can have positive or negative values. Based on Figure 6b–f, it can be observed that the Hit bias of IMERG_V07 is positive in 50% of mainland China, ranging from 0 to 0.2 mm. On the other hand, for IMERG_V06, the portion with positive Hit bias accounts for only 31% and is mainly concentrated in humid regions. From the spatial distribution of Hit bias, it seems that IMERG_V06 performs better in terms of Hit bias.

In relation to the Missed bias, IMERG_V07 exhibits the benefits of algorithm enhancement. The Missed bias values in IMERG_V07, namely 0.69 mm, are slightly lower in comparison to IMERG_V06, which recorded 0.71 mm. This indicates that IMERG_V07 offers more precise estimations regarding missed precipitation, showcasing improved performance resulting from algorithm enhancements. This improvement can be attributed to the incorporation of frozen precipitation retrieval in the IMERG_V07 algorithm.

Among the three error components, False bias is the primary source of error, with values ranging from 0.32 to 1.95 mm. It gradually decreases from humid to arid regions. Overestimating precipitation has been consistently associated with IMERG SPEs since their release [5,32,33,34,35]. The IMERG_V07 algorithms have taken several measures to correct overestimation by intercalibrating PMW datasets, which mainly include [7], firstly, the mean GMI/TMI precipitation rate is not constant along the scan, so the GMI/TMI-constellation calibration was modified to use the full-swath GMI/TMI matched to the narrower swath of CORRA-G/T. Secondly, the GPROF-GMI, GPROF-TMI, and CORRA-G/T footprint sizes differ, so CORRA-G/T is averaged on 0.2° × 0.2° and 0.3° × 0.3° grid box templates to approximately match the GPROF-GMI and GPROF-TMI footprint sizes, respectively. Finally, the GPCP adjustment to CORRA in V06 was unrealistically raising winter precipitation over land and not capturing its longitudinal variability, so the GPCP adjustment is not applied over land in V07 (https://gpm.nasa.gov/resources/documents/imerg-v07-release-notes, accessed on 21 February 2024). However, the effects of these measures are still not satisfactory, especially for correcting the False bias in the humid (1.04 mm), semi-humid (0.93 mm), and semi-arid regions (0.83 mm), compared to IMERG_V06 (0.99 mm, 0.82 mm, and 0.74 mm, respectively).

Overall, IMERG_V06 and IMERG_V07 exhibit similar spatial distributions of error components, with only slight differences. IMERG_V07 shows a noticeable improvement in Missed bias, but there is still room for improvement in Hit bias and False bias.

3.2. Dependency of Error on Precipitation Intensity

In the previous section, we observed differences in error metrics of IMERG SPEs across different climate regimes, which we speculate may be influenced by precipitation intensity. This indicates that there is variability in satellite identification of precipitation events with different precipitation intensities. Based on this, in this section, we assess the impact of different precipitation intensities on the identification of precipitation events by IMERG SPEs in four seasons, as shown in Figure 7. We classify precipitation into four thresholds corresponding to light rain, moderate rain, heavy rain, and rainstorm [36,37]. It is important to note that the determination of precipitation intensity thresholds relies on ground-based precipitation observations. As a result, the calculation of FAR (false alarms of IMERG SPEs when a ground-based site observed no precipitation event) for different precipitation intensities is not feasible. Therefore, this section focuses solely on the analysis of POD and MIS.

In general, according to the results shown in Figure 7, there is a strong correlation between the precipitation-capturing capability of IMERG SPEs and the precipitation intensity. The POD of IMERG SPEs is generally below 40%. Among them, IMERG_V06 and IMERG_V07 exhibit the highest POD in summer, followed by autumn, and the performance is least satisfactory in winter (Figure 7a–d). As the precipitation intensity increases, the recognition ability of IMERG SPEs shows an increasing trend followed by a decreasing trend. The POD of IMERG SPEs is highest for moderate rain, with all values above 30%, while it is lowest for rainstorm events. Although IMERG SPEs can effectively capture precipitation events, there is still a significant misjudgment in capturing precipitation intensity. However, this may also be related to the limited number of samples of heavy precipitation events. It is worth noting that the POD of IMERG SPEs is highest for moderate rain, indicating its strong recognition capability for this category. Compared to IMERG_V06, IMERG_V07 shows a significant improvement in POD, especially for light rain and moderate rain events. By observing Figure 7e–h, it can be seen that with the improvement in IMERG_V07’s ability to capture light precipitation events, its MIS also increases accordingly. The MIS of IMERG SPEs is lowest in summer, with all values below 35%. The MIS is highest in winter, with IMERG_V07 reaching a maximum of 66.2% and IMERG_V06 reaching a maximum of 63.9% (Figure 7h), which is associated with snow cover on the ground during winter. As the precipitation intensity increases, the MIS of IMERG SPEs decreases, indicating a lower false-negative rate for strong precipitation events. Except for winter, IMERG_V07 performs best in terms of missed events in rainstorm events, slightly outperforming IMERG_V06. However, these minor improvements do not fully compensate for the missed precipitation events in light rain, moderate rain, and heavy rain categories of IMERG_V07.

Overall, compared to IMERG_V06, IMERG_V07 has made some improvements in capturing precipitation events. However, there is still room for improvement in accurately capturing rainstorm events, reducing false negatives for light precipitation, and improving the accuracy of capturing precipitation events in winter.

3.3. Topographical Dependency of Error

The spatial distribution analysis of accuracy metrics discussed earlier indicates that topographical factors significantly influence the accuracy of precipitation retrieval by IMERG SPEs. To further investigate the impact of terrain on precipitation retrieval accuracy, we have presented line graphs in Figure 8. These graphs illustrate the influence of topographical factors on the capture accuracy of the precipitation amount and precipitation events in each season. Both IMERG_V06 and IMERG_V07 demonstrate distinct dependencies on elevation, as reflected in the RMSE, POD, MIS, and FAR metrics.

From Figure 8b,d,f,h, it can be observed that as the elevation increases, both IMERG_V06 and IMERG_V07 exhibit a decreasing trend in RMSE values. The lowest RMSE values are observed in winter (0.05–0.40 mm/h), followed by autumn (0.32–0.80 mm/h), while the highest RMSE values are found in summer (0.73–1.80 mm/h). This pattern is consistent with the spatial distribution shown in Figure 2 and is associated with the seasonal variation in precipitation. In spring, within the elevation range of 0–600 m, IMERG_V06 has lower RMSE values compared to IMERG_V07. However, beyond an elevation of 600 m, IMERG_V07 demonstrates higher accuracy in precipitation retrieval. IMERG_V07 performs less satisfactorily in summer, with consistently higher RMSE values compared to IMERG_V06. In autumn and winter, IMERG_V06 has lower RMSE values at lower elevations, while IMERG_V07 exhibits lower RMSE values at higher elevations.

Based on the results from the line graphs (Figure 8a,c,e,g), it is also evident that the capture capability of precipitation events by IMERG SPEs decreases with increasing elevation. This can be attributed to the influence of complex terrain on the distribution and intensity of precipitation, thereby affecting the retrieval results of IMERG SPEs. Within the elevation range of 0–2400 m, both IMERG_V06 and IMERG_V07 show similar performance in capturing precipitation events. However, when the elevation exceeds 2400 m, the retrieval accuracy of IMERG_V07 is lower than that of IMERG_V06.

This indicates that the measures taken to improve the accuracy of precipitation retrieval in high-elevation areas for IMERG_V07 have improved the accuracy of precipitation amount retrieval. However, there are still limitations in capturing precipitation events in areas with elevations exceeding 2400 m.

Based on the observed elevation-dependent characteristics of the error components discussed earlier, we further evaluated the terrain dependency of IMERG_V06 and IMERG_V07 SPEs. In this section, we computed the corresponding error components (Total bias, Missed bias, Hit bias, and False bias) based on seasonal average precipitation and analyzed the dependency of IMERG SPEs’ error components on different elevations (Figure 9).

According to the results in Figure 9, we can observe that both types of SPEs exhibit clear elevation dependency in Total bias and error components, with the magnitude of errors varying with altitude. In comparison, the absolute value of Total bias is relatively small and significantly lower than the sum of the absolute values of the three error components. This is due to the phenomenon of positive and negative error components offsetting each other. This conclusion aligns with the spatial distribution of error components presented in Figure 6. It demonstrates that relying solely on Total bias to evaluate the error characteristics of SPEs may result in an overestimation of their performance.

Based on the dependency of error components on elevation shown in Figure 9a–h, we observe that the error components of IMERG_V06 and IMERG_V07 exhibit distinct seasonal characteristics at different elevations. In spring, both IMERG_V06 and IMERG_V07 have a Total bias of around 0–0.2 mm. Most of IMERG_V06’s Hit bias is negative, while IMERG_V07’s Hit bias changes from positive to negative at an elevation of 1800 m.

In summer, the absolute values of all error components are significantly higher compared to spring, which is attributed to higher precipitation levels, consistent with the precipitation distribution shown in Figure 2b,f,j. Both precipitation products exhibit an overestimation of precipitation in summer. Specifically, the False bias and Missed bias of IMERG_V06 and IMERG_V07 exhibit roughly symmetric distributions (Figure 9c,d). However, False bias remains the dominant error source for both IMERG_V06 and IMERG_V07 in summer, as after positive and negative offset, their Total bias is positive, and Hit bias is smaller than Total bias, indicating that False bias has the largest contribution to the error components.

In autumn, the error components decrease compared to summer, ranging from −0.98 to 0.85 mm. Both IMERG_V06 and IMERG_V07 have positive Total bias, indicating a prevalent overestimation of precipitation. The False bias decreases with increasing altitude, while Missed bias shows the opposite trend.

In winter, the absolute values of False bias and Missed bias of both IMERG_V06 and IMERG_V07 decrease with increasing elevation. Hit bias decreases with elevation in the range of 0–2700 m and then changes from positive to negative, with its absolute value increasing with elevation. Within the elevation range of 0–1200 m, both precipitation products have positive Total bias, which then changes to negative. Therefore, it can be inferred that when the elevation exceeds 1200 m, Missed bias becomes the main error source for both IMERG_V06 and IMERG_V07 in winter.

In conclusion, both types of SPEs exhibit elevation dependency in Total bias and error components. In spring, summer, and autumn, False bias is the main error source for IMERG_V06 and IMERG_V07, while in winter, when the elevation exceeds 1200 m, Missed bias becomes the primary error source for both IMERG_V06 and IMERG_V07.

3.4. Dependency of Errors on Climate Type

Next, we divided the entire mainland of China into four climate regions: humid, semi-humid, semi-arid, and arid regions (Figure 1a), and studied the sources of errors in these regions. Overall, the error components of IMERG_V06 and IMERG_V07 decrease with increasing aridity, which is consistent with the objective reality.

By observing Figure 10a,e,i,m, it can be found that in the semi-humid region, IMERG_V07 shows significant improvement in Total bias compared to IMERG_V06, decreasing from 0.175–0.19 mm to 0.025 mm. IMERG_V06 and IMERG_V07 perform almost the same in terms of Total bias in the arid region. However, in the humid and semi-arid regions, IMERG_V06 performs better than IMERG_V07, and IMERG_V07 exhibits higher data dispersion in the arid region. In all four climate regions, both IMERG_V06 and IMERG_V07 have a Total bias greater than 0, indicating an overestimation of precipitation, especially in the humid region, which is consistent with the analysis results in the previous figure (Figure 6).

In terms of Hit bias (Figure 10b,f,g,n), except for the arid region, the performance of IMERG_V06 and IMERG_V07 in Hit bias values is nearly the same. The main difference is that the Hit bias of IMERG_V07 is mostly positive, while the Hit bias of IMERG_V06 tends to be negative, especially in the semi-arid region. In the arid region, IMERG_V07 shows improvement, with enhanced numerical values and dispersion compared to IMERG_V06. The improvement in the accuracy of precipitation inversion in the arid region by IMERG_V07 may be attributed to the adoption of more advanced precipitation estimation algorithms and techniques. The skill of the GPROF V07 retrieval algorithm varies with surface types. The GPROF V07 algorithm used to retrieve precipitation from all PMW inputs for IMERG has made progress in handling difficult surface types that tend to yield lesser quality results, including frozen surfaces, orographic areas, and coastal zones. By introducing new precipitation estimation methods, IMERG_V07 can more accurately capture precipitation in arid regions, thereby improving consistency with ground station data.

Figure 10c,g,k,o shows the situation of Missed bias for IMERG_V06 and IMERG_V07. By comparison, we found that the overall performance of Missed bias in the four climate regions is better for IMERG_V07 than for IMERG_V06. This may be associated with algorithm improvements in IMERG_V07, including the inclusion of PMW estimates over frozen surfaces [7]. Specifically, the Missed bias of IMERG_V06 ranges from 0 to −2.20 mm with high data dispersion, while the absolute values of Missed bias in IMERG_V07 are relatively small, mainly concentrated in the range of 0.5–1.0 mm.

Based on Figure 10d,h,l,p, it can be seen that the main source of error for both IMERG_V06 and IMERG_V07 in the four climate regions comes from False bias. In the humid region, the False bias of both SPEs is mainly concentrated in the range of 0.8–1.2 mm, but the data for IMERG_V06 is more dispersed, and the data dispersion of IMERG_V06 is higher in the semi-humid region (0.54–1.74 mm). In the semi-arid region, the False bias of IMERG_V06 is better than that of IMERG_V07, mostly distributed in the range of 0.6–0.8 mm, while the False bias of IMERG_V07 is mainly concentrated in the range of 0.8–1.0 mm.

Only by using Total bias, can it be inferred that the performance of IMERG_V06 is better than IMERG_V07 in the humid, semi-arid, and arid regions. However, by comparing the performance of each error component in the four climate regions, it is found that the actual situation is not so simple. This fully demonstrates the necessity of decomposing error components.

In summary, the error components of IMERG_V07 in the four climate regions are relatively more concentrated, with smaller data dispersion, especially in terms of Missed bias. The algorithm improvements in IMERG_V07 are most evident in the arid region. False bias is the main source of errors for both SPEs in the four climate regions, and the secondary contribution comes from Hit bias.

4. Conclusions

In this study, we conducted a systematic evaluation of IMERG_V06 and the newly released IMERG_V07 from various perspectives, including precipitation intensity, topography, and climate regions. Feedback on IMERG_V06 is analyzed, highlighting the improvements made in IMERG_V07. The main conclusions of this study are summarized as follows:

(1).

IMERG_V07 shows smaller RMSE values compared to IMERG_V06 in all four climatic regions, indicating improved accuracy in estimating precipitation amounts. IMERG_V07 also exhibits better consistency with ground station data and shows an overall improvement of 4% in capturing regional average precipitation events compared to IMERG_V06. The spatial distribution of error components between IMERG_V06 and IMERG_V07 is similar, both IMERG_V06 and IMERG_V07 suffer from the issue of overestimating precipitation.

(2).

IMERG_V07 has shown significant improvement in capturing precipitation events of different intensities. However, there are still issues to address, such as higher missing rates for light precipitation and winter precipitation events, as well as lower detection rates for heavy precipitation events.

(3).

Both IMERG_V06 and IMERG_V07 exhibit a dependency on topography, with False bias being the main error source in most cases. In winter, Missed bias becomes the primary error source at elevations exceeding 1200 m. IMERG_V07 has improved precipitation retrieval accuracy in high-altitude areas but has limitations in capturing precipitation events.

(4).

IMERG_V07 demonstrates higher correlation coefficients and reduced data dispersion in error components across the four climatic regions, with significant improvements in arid regions. False bias remains the primary error source, with Hit bias contributing as a secondary factor in the four climatic regions.

This study provides a systematic quantitative analysis of the error characteristics of various versions of IMERG products at an hourly scale in mainland China. However, further investigation is required to explore the error characteristics in other geographical regions. Additionally, there is a need for in-depth analysis of the error and identification of sources regarding the capturing capability of IMERG products in terms of light precipitation, extreme precipitation, and other factors. We intend to prioritize and address this issue in our future research endeavors.