https://orcid.org/0000-0002-7060-2109
ABSTRACT
Satellite-derived precipitation datasets are essential components of hydrological simulations, particularly in data-scarce regions of western China. However, a comprehensive assessment of their accuracy and reliability is required. Here, the accuracy of two high-resolution satellite-derived precipitation datasets, Integrated Multi-satellite Retrievals for GPM – Final (IMERG-F) and Gauge-Adjusted Global Satellite Mapping of Precipitation (GSMaP-Gauge), was evaluated across the Ten Tributaries region of the Yellow River Basin in western China using four quantitative metrics and three categorical scoring indicators. This evaluation sought to ascertain the retrieval accuracy of these products on both the daily scale and hourly scale of heavy precipitation events, and investigated their inversion error characteristics across various spatiotemporal scales. Both datasets effectively captured the spatiotemporal patterns of annual average precipitation within the study area. Notably, the daily-scale accuracy of these satellite-derived precipitation products surpassed their hourly and half-hourly counterparts. Both GPM-IMERG and GSMaP-Gauge adeptly reproduced most precipitation events in the Ten Tributaries region, with peak detection performance observed in the central and southern zones, providing a reliable data source for drought monitoring and hydrological modeling. Overall, compared with GPM-IMERG, GSMaP-Gauge displayed superior inversion accuracy across diverse spatiotemporal scales.
HIGHLIGHTS
Evaluated accuracy of satellite precipitation data in the Ten Tributaries region, addressing terrain and rainfall challenges, and filling assessment gaps. Spatiotemporal distribution and rainstorm-scale accuracy, aiding hydrological research, were analyzed.
Compared GPM-IMERG and GSMaP-Gauge, highlighting strengths, weaknesses, and aiding in mitigating extreme precipitation impacts.
INTRODUCTION
The accurate measurement of precipitation is crucial for hydrological modeling, particularly in data-scarce regions such as western China (Huang et al. 2016; Yu et al. 2020; Peng et al. 2021). Precipitation data are essential for climate trend analysis, water resource management, drought monitoring, and flood forecasting (Kim & Jehanzib 2020; Brunner et al. 2021). Traditional ground-based measurement methods, such as rain gauges and weather radar, often have limitations in spatial coverage and data availability. Given these limitations, satellite-derived precipitation data have emerged as a promising observational tool, providing extensive spatial and temporal coverage, especially in remote areas with complex terrain where ground observations are sparse or nonexistent (Michaelides et al. 2009; Tapiador et al. 2017; Peng et al. 2020; Belay et al. 2022).
The Ten Tributaries region in the Yellow River Basin of western China is an area that urgently requires accurate precipitation data, but faces considerable challenges in data acquisition. This region features complex terrain and variable climatic conditions, resulting in an uneven distribution of precipitation, which in turn leads to frequent droughts and floods (Deng et al. 2022; Wang et al. 2022). High-resolution and accurate precipitation data are crucial for this region because such data could improve hydrological modeling, enhance water resource management, and mitigate the impacts of natural disasters (Jiang et al. 2022; Liu et al. 2023).
Satellite-derived precipitation products, such as Integrated Multi-satellite Retrievals for GPM – Final (IMERG-GPM) and Gauge-Adjusted Global Satellite Mapping of Precipitation (GSMaP-Gauge), have shown potential in providing reliable precipitation estimates. These products integrate data from multiple satellite sensors using advanced algorithms to improve precipitation retrieval accuracy (Zhu et al. 2018; Wei et al. 2022). Evaluating the performance of these satellite-derived products is essential to understand their reliability and limitations, particularly in capturing daily precipitation amounts and specific heavy precipitation events in areas with complex climates and terrain (Yuan et al. 2017).
In recent years, satellite remote sensing has gradually emerged as a powerful technique for measuring precipitation. By utilizing visible, infrared, and microwave sensors, satellites can provide comprehensive and continuous precipitation data over large areas (Junior et al. 2021; Abegeja 2024). The importance of satellite precipitation products in climate and hydrological research has become increasingly evident, particularly in data-scarce regions. Lu (2020) evaluated the performance of GPM-IMERG V5 and GSMaP V7 products on the Tibetan Plateau, concluding that both were effective in capturing the spatial patterns of precipitation (Lu 2020). Similarly, Nepal (2021) conducted a study in Nepal, finding that while both satellite precipitation products tended to underestimate precipitation amounts, IMERG was more effective in capturing extreme precipitation events (Nepal 2021). These findings suggest that, despite certain differences, each product has unique strengths depending on geographical and climatic conditions. Wang & Yong (2020) performed a global assessment of error distribution in IMERG and GSMaP products, revealing that GSMaP struggled in high-latitude regions due to systematic biases and the impact of low precipitation intensity (Wang & Yong 2020). Conversely, in the Sanjiangyuan region of China, Wang et al. (2021) found that IMERG demonstrated superior capabilities in detecting precipitation events, particularly in areas with higher precipitation intensity (Wang et al. 2021). These results underscore the significant influence of geographical location and climatic conditions on the accuracy of satellite precipitation products.
The accuracy assessment of satellite remote sensing is an important research area in arid regions. Most studies primarily focus on arid regions such as northwest China, the Arabian Peninsula, and the United Arab Emirates (UAE), examining the relationship between various satellite precipitation products. Lu et al. (2018) evaluated the performance of TRMM 3B43V7 and GPM-IMERG in Xinjiang, China, finding that both products generally overestimated precipitation (Lu et al. 2018). However, GPM-IMERG outperformed TRMM 3B43V7, particularly after researchers corrected GPM-IMERG, significantly enhancing its accuracy in capturing precipitation zones in arid regions. These findings align with those of Li et al. (2022), who assessed various satellite precipitation products in the Tianshan Mountains and found that GPM was the most effective in capturing precipitation distribution, especially during the wet season (Li et al. 2022). Conversely, Xie et al. (2022) explored the impact of ground measurement data bias on the evaluation of satellite precipitation products, highlighting that systematic biases in ground measurements may lead to an underestimation of satellite product performance. This underscores the importance of bias correction when using ground data as a reference (Xie et al. 2022). Additionally, Helmi & Abdelhamed (2022) found significant uncertainty in the performance of satellite precipitation products in Saudi Arabia, particularly in daily-scale precipitation estimation in arid regions (Helmi & Abdelhamed 2022).
Overall, existing research has shown that different types of satellite-derived precipitation data have varying accuracies and detection performances owing to differences in data sources and retrieval algorithms. Remote-sensing satellite data sources have a common problem: for a given dataset, the data accuracy may vary greatly depending on the study area, and even in a single region, differences in the accuracies of the spatial and temporal variation characteristics of the data can be significant (Zhu et al. 2024). Meanwhile, satellite-derived precipitation data of the GPM era benefit from marked improvements in retrieval algorithms, integrating multiple sensor data, and offering higher spatiotemporal resolution compared with TRMM era products (Zhang et al. 2020). However, there is currently a lack of research on the accuracy assessment of various GPM era satellite-derived precipitation data products for extreme precipitation events in smaller regions. Therefore, enhancing rainfall observation capabilities and enriching rainfall observation data, particularly high-spatiotemporal-resolution precipitation data, is a fundamental requirement for improving flood disaster prevention theories and technologies, especially in regions with complex terrain and rainfall patterns. Conducting accuracy assessments of satellite-derived precipitation data on daily and heavy-rainfall-event scales in the Ten Tributaries region is important for improving the accuracy of water‒sediment simulation and flood disaster management in this region, and supports research on water‒sediment variation patterns of the Yellow River.
STUDY AREA AND DATA PREPARATION
Study area
Topographic map of the study area showing the locations of meteorological stations.
Topographic map of the study area showing the locations of meteorological stations.
In the upstream area of the northern edge of the Ordos Plateau, the landscape predominantly features gravelly hills, gullies, and ravines, with substantial hydraulic erosion owing to fragmented surfaces. The middle reaches encompass the Kubuqi Desert area, characterized by sandy hills extending east to west, with severe wind erosion that is particularly pronounced in the west. The downstream area comprises alluvial plains with flat terrain and substantial sand loss owing to runoff. The region as a whole is characterized by frequent heavy rainfall and flooding, severe soil erosion, sparse vegetation, and a very fragile ecological environment (Zhao et al. 2020, 2023).
The Ten Tributaries region has a typical arid continental climate, marked by dry conditions, frequent winds, and sandy environments. The average annual temperature is approximately 7 °C, with annual precipitation of 200‒400 mm, and an evaporation rate of approximately 2,200 mm/yr. Precipitation exhibits considerable inter-annual variability and an uneven intra-annual distribution, with approximately 70% of annual rainfall occurring from July to September. This period is marked by short-duration, high-intensity rainstorms, which, combined with the unique geomorphological features of the region, contribute to its role as a major sediment source for the Yellow River Ningmeng river channel.
Data preparation
To improve water resource management and flood prediction, it is crucial to carry out hydrological studies in the Yellow River Basin. However, because of the lack of rainfall stations in the basin, the spatiotemporal distribution of precipitation cannot be obtained accurately. This hinders the accurate assessment of water resources and high-quality development of cities; hence, a comprehensive assessment of the hydrological utility of different satellite-derived precipitation data products is vitally important.
Ground precipitation data
In this study, precipitation data from 10 stations in the Ten Tributaries region spanning the period 2018‒2020 were extracted from the precipitation extract table and day-by-day precipitation element table of the hydrological yearbook. These data have been subjected to strict quality control steps, and can be used as true values with which to compare and evaluate the accuracy and quality of satellite-derived precipitation data.
Six typical rainstorms that occurred between 2018 and 2020 were selected from the Hydrological Almanac table of precipitation elements at Tugrizge, Harahantu Trench, Chaiteng Trench, Gaotouyao, Longtouqian, Hantaimiao, Qingdamen, Earzit Trench, and Ringshawan stations. These extracted raw data did not match the satellite-derived precipitation data in terms of temporal resolution, so initial processing requires tools such as Python to perform unified temporal resolution processing. The rainstorm data extracted from the precipitation element table adopts the time series linear interpolation method, and the time step is 30 min.
Satellite-derived precipitation data
We used two satellite-derived high-spatiotemporal-resolution precipitation data products as our research data: GPM-IMERG and GSMaP-Gauge. The GPM product, specifically GPM-IMERG Final Run, is a delayed precipitation product finalized after adjustments using Global Precipitation Climatology Project (GPCP) gauge measurements. It has a resolution of 0.1° × 0.1°, and the period under study spans 2018‒2020, with datasets including those at daily and half-hourly scales. The datasets were downloaded from the NASA website (https://disc.gsfc.nasa.gov). The GSMaP-Gauge product is based on the standard GSMaP-MVK product and is generated via calibration with daily-scale data from over 30,000 CPC rain gauge sites under NOAA/Climate Prediction Center. The study period for this product also spans 2018‒2020, with datasets including those at daily and hourly scales. These datasets can be obtained from the Japan Aerospace Exploration Agency's Global Satellite Mapping of Precipitation (JAXA GSMaP) global satellite precipitation mapping project official website (https://sharaku.eorc.jaxa.jp/GSMaP).
Methodology
With the rapid development of precipitation measurement technology and precipitation inversion algorithms from satellite sensors over long distances, a series of satellite-derived precipitation data products with high spatiotemporal resolution are playing an increasingly important role in hydrometeorological, agricultural drought, and disaster prevention and mitigation research, and their potential as a complementary source of station-based precipitation data has been fully affirmed in previous studies. However, owing to different inversion algorithms, there are different degrees of error between actual ground-based precipitation measurements and precipitation products derived from different satellites and algorithms. Cattani et al. (2016) evaluated the performance of several commonly used monthly satellite precipitation products in East Africa. The study used various statistical indicators, such as correlation coefficient (CC) and root-mean-square-error, to compare and analyze satellite precipitation products with ground observation data. The results showed that there were significant differences in the performance of different satellite precipitation products in East Africa, and suitable products need to be selected for specific regions (Cattani et al. 2016). Cohen Liechti et al. (2012) compared and evaluated the hydrological simulation performance of several satellite precipitation products in the Zambezi River Basin, using a comparative analysis of hydrological models. The results showed that there were significant differences in the hydrological simulation effects of different satellite precipitation products in the basin, and suitable products need to be selected according to specific application scenarios (Cohen Liechti et al. 2012). These studies collectively underscore the complexity and variability in the errors associated with satellite-derived precipitation data, necessitating ongoing refinement of algorithms and validation techniques to enhance the accuracy and reliability of these essential data products for various applications. Therefore, before applying satellite-derived precipitation data to research fields such as climate change, hydrometeorology, and agricultural drought, it is important to make use of station-based precipitation data as ground truth values with which to evaluate different aspects of the accuracy of satellite-derived precipitation data. In addition, the Ten Tributaries region is affected by various climatic and geological factors, making the spatiotemporal distribution of its precipitation extremely uneven. The topography of the region consists of hills and deserts, and natural disasters, such as droughts and extreme precipitation events, occur frequently within the basin. In the context of desertification, refined and standardized precipitation data are urgently required for disaster prevention and mitigation research in the basin.
In this study, using precipitation data from ten stations in the Ten Tributaries region, GPM-IMERG and GSMaP-Gauge were simultaneously evaluated and analyzed in terms of their multi-scale spatiotemporal accuracy and applicability, with a view to obtaining their error characteristics and respective advantages and disadvantages. We used four continuous error statistics to quantify the errors of satellite-derived precipitation data in the Ten Tributaries region: CC, mean absolute error (MAE), root-mean-square error (RMSE), and relative bias (BIAS). We also used three categorical statistical indicators to characterize the ability of satellite-derived precipitation data to capture precipitation events in the watershed: hit rate or probability of detection (POD), false alarm rate (FAR), and equitable threat score (ETS). The calculation formulae of these seven indicators are shown in Table 1.
Calculation formulae for the seven statistical evaluation indicators used herein
| Indicator name | Calculation formula | Optimal value |
|---|---|---|
| Correlation coefficient |  | 1 |
| Mean absolute error |  | 0 |
| Root-mean-square error |  | 0 |
| Relative deviation |  | 0 |
| Precipitation event hit rate |  | 1 |
| Precipitation event false alarm rate |  | 0 |
| Equitable threat scores |  | 1 |
| Indicator name | Calculation formula | Optimal value |
|---|---|---|
| Correlation coefficient | | 1 |
| Mean absolute error | | 0 |
| Root-mean-square error | | 0 |
| Relative deviation | | 0 |
| Precipitation event hit rate | | 1 |
| Precipitation event false alarm rate | | 0 |
| Equitable threat scores | | 1 |
Note: n is the total number of samples; i stands for the sample data; S denotes satellite-derived precipitation data; O denotes precipitation data obtained from ground stations; A denotes the number of precipitation events observed by both satellites and stations; B denotes the number of precipitation events observed by satellites but not stations; C denotes the number of precipitation events observed by stations but not satellites; and D denotes the number of precipitation events not captured by either stations or satellites.
In detail, the CC reflects the linear correlation between the satellite-derived precipitation data and measured precipitation data from rainfall stations, with a CC value close to 1 indicating a high positive correlation between the two variables, a CC value close to −1 indicating a high negative correlation between the two variables, and a CC value of 0 indicating no correlation between the two variables.
The MAE reflects the average of the absolute value error between the satellite precipitation product and station precipitation data; the optimal ideal value of this index is 0.
The RMSE reflects the dispersion between the satellite-derived precipitation data and measured precipitation data from rainfall stations, and evaluates the stability of the error; a smaller RMSE value equates to a higher accuracy of satellite-derived precipitation data against precipitation data measured at the rainfall stations.
The BIAS reflects the ability of IMERG satellite-derived precipitation data to assess the measured precipitation data at rainfall stations; a BIAS > 0 indicates overestimation, and a BIAS < 0 indicates underestimation.
The POD reflects the ability of remote-sensing satellites to correctly detect daily precipitation events; the range of values is 0‒1, where a value closer to 1 indicates that the remote-sensing satellite has a larger hit rate on precipitation.
The FAR reflects the ability of remote-sensing satellites to incorrectly detect daily precipitation events; the range of values is 0‒1, where a value closer to 1 indicates a greater likelihood of the remote-sensing satellite misreporting precipitation.
The ETS comprehensively indicates the probability of successful monitoring of precipitation by satellite-derived precipitation; the value ranges from 0 to 1, where a larger value equates to a greater probability of successful monitoring of precipitation; this can be used as a comprehensive measure.
Without special instructions, the calculation of POD, FAR, and ETS in this study was based on a precipitation threshold of 0.1 mm/day (i.e., when the precipitation amount was ≥0.1 mm/day, we judged that a rainfall event had occurred).
RESULTS
Spatial distribution of daily precipitation
Precipitation in the Ten Tributaries region exhibited substantial spatial heterogeneity. A wide range of variability was predominantly observed along the north‒south axis of the basin. In the central-southern part of the basin, annual precipitation was generally < 200 mm, leading to this area being classified as an arid zone. Conversely, in the central-northern part of the basin, annual precipitation exceeded 400 mm, fulfilling the criteria required of a semi-humid climate zone. There was little spatial difference in annual precipitation between the eastern and western parts of the basin, with values falling in the median range among 151 grid-based annual precipitation sequences across the basin. This may be attributed to the limited and uneven distribution of stations, with stations primarily being located in the central-southern area, and a lack of stations in the eastern and northwestern parts of the basin. This resulted in interpolated grid-based precipitation data being unable to reflect spatial differences in the east and west.
Ground-measured and satellite-derived mean annual precipitation distribution maps.
Ground-measured and satellite-derived mean annual precipitation distribution maps.
Daily-scale accuracy assessment of satellite-derived precipitation data
Daily precipitation data corresponding to the locations of remote-sensing satellite measurements and ground stations were individually extracted and subjected to statistical analysis alongside actual daily precipitation measurements from the selected ground stations, permitting an overall precipitation accuracy assessment for ten sites for the period 2018‒2020, as presented in Table 2. An analysis of characteristic indicators between daily precipitation data from the two types of remote-sensing satellites and rain gauge stations, as well as an assessment of the ability of the satellites to capture daily precipitation events, led to the following conclusions.
Statistical results of the two satellite-derived precipitation data products in terms of their accuracy evaluation indices against rain gauge station data at the daily scale from 2018 to 2020
| CC | MAE (mm) | RMSE (mm) | BIAS (%) | POD | FAR | ETS | |
|---|---|---|---|---|---|---|---|
| GSMaP-Gauge | 0.64 | 0.06 | 3.33 | 39.86 | 0.89 | 0.45 | 0.43 |
| GPM-IMERG | 0.65 | 0.06 | 3.42 | 32.51 | 0.83 | 0.51 | 0.35 |
| CC | MAE (mm) | RMSE (mm) | BIAS (%) | POD | FAR | ETS | |
|---|---|---|---|---|---|---|---|
| GSMaP-Gauge | 0.64 | 0.06 | 3.33 | 39.86 | 0.89 | 0.45 | 0.43 |
| GPM-IMERG | 0.65 | 0.06 | 3.42 | 32.51 | 0.83 | 0.51 | 0.35 |
Although errors were present in precipitation data from both types of satellites, a linear correlation with rain gauge station precipitation was observed on the daily scale, indicating some degree of applicability within the study area. Despite the two remote-sensing satellites exhibiting some issues, such as missed and false reports in capturing daily precipitation events, they were able to correctly detect most daily precipitation and demonstrated a strong capability in estimating the occurrence of precipitation events, with GSMaP-Gauge performing better than GPM-IMERG.
Accuracy evaluation indices of daily precipitation data from GSMaP-Gauge and GPM-IMERG at each station in the Ten Tributaries region
| Ground station | Satellite precipitation | CC | MAE (mm) | RMSE (mm) | BIAS (%) | POD | FAR | ETS |
|---|---|---|---|---|---|---|---|---|
| Tarangaole | GSMaP-Gauge | 0.265 | 1.281 | 4.789 | 56.993 | 0.879 | 0.543 | 0.352 |
| GPM-IMERG | 0.425 | 1.102 | 4.428 | 43.393 | 0.836 | 0.57 | 0.316 | |
| Tugerige | GSMaP-Gauge | 0.631 | 0.915 | 3.78 | −5.333 | 0.852 | 0.38 | 0.482 |
| GPM-IMERG | 0.792 | 0.743 | 2.986 | −4.802 | 0.736 | 0.481 | 0.346 | |
| Halahantuhao | GSMaP-Gauge | 0.723 | 0.761 | 2.705 | 36.728 | 0.832 | 0.406 | 0.441 |
| GPM-IMERG | 0.817 | 0.733 | 2.71 | 33.566 | 0.751 | 0.471 | 0.35 | |
| Chaidenghao | GSMaP-Gauge | 0.775 | 0.751 | 2.717 | 56.92 | 0.891 | 0.364 | 0.51 |
| GPM-IMERG | 0.725 | 0.8 | 3.177 | 40.991 | 0.771 | 0.491 | 0.342 | |
| Gaotouyao | GSMaP-Gauge | 0.73 | 0.799 | 2.786 | 50.994 | 0.921 | 0.465 | 0.43 |
| GPM-IMERG | 0.754 | 0.747 | 2.663 | 37.619 | 0.866 | 0.493 | 0.385 | |
| Longtouguai | GSMaP-Gauge | 0.619 | 0.965 | 3.564 | 23.094 | 0.876 | 0.564 | 0.332 |
| GPM-IMERG | 0.593 | 0.928 | 3.866 | 22.902 | 0.883 | 0.572 | 0.325 | |
| Hantaimiao | GSMaP-Gauge | 0.796 | 0.762 | 2.777 | 19.933 | 0.921 | 0.428 | 0.469 |
| GPM-IMERG | 0.723 | 0.978 | 3.389 | 16.628 | 0.854 | 0.553 | 0.321 | |
| Qingdamen | GSMaP-Gauge | 0.709 | 0.897 | 3.081 | 148.957 | 0.878 | 0.391 | 0.475 |
| GPM-IMERG | 0.651 | 0.91 | 3.342 | 135.595 | 0.838 | 0.441 | 0.409 | |
| Erzihao | GSMaP-Gauge | 0.746 | 0.868 | 2.837 | 79.658 | 0.913 | 0.455 | 0.433 |
| GPM-IMERG | 0.606 | 0.97 | 3.605 | 70.785 | 0.861 | 0.502 | 0.369 | |
| Xiangshawan | GSMaP-Gauge | 0.663 | 1 | 3.652 | 9.523 | 0.943 | 0.477 | 0.427 |
| GPM-IMERG | 0.663 | 0.939 | 3.691 | 0.559 | 0.918 | 0.51 | 0.384 |
| Ground station | Satellite precipitation | CC | MAE (mm) | RMSE (mm) | BIAS (%) | POD | FAR | ETS |
|---|---|---|---|---|---|---|---|---|
| Tarangaole | GSMaP-Gauge | 0.265 | 1.281 | 4.789 | 56.993 | 0.879 | 0.543 | 0.352 |
| GPM-IMERG | 0.425 | 1.102 | 4.428 | 43.393 | 0.836 | 0.57 | 0.316 | |
| Tugerige | GSMaP-Gauge | 0.631 | 0.915 | 3.78 | −5.333 | 0.852 | 0.38 | 0.482 |
| GPM-IMERG | 0.792 | 0.743 | 2.986 | −4.802 | 0.736 | 0.481 | 0.346 | |
| Halahantuhao | GSMaP-Gauge | 0.723 | 0.761 | 2.705 | 36.728 | 0.832 | 0.406 | 0.441 |
| GPM-IMERG | 0.817 | 0.733 | 2.71 | 33.566 | 0.751 | 0.471 | 0.35 | |
| Chaidenghao | GSMaP-Gauge | 0.775 | 0.751 | 2.717 | 56.92 | 0.891 | 0.364 | 0.51 |
| GPM-IMERG | 0.725 | 0.8 | 3.177 | 40.991 | 0.771 | 0.491 | 0.342 | |
| Gaotouyao | GSMaP-Gauge | 0.73 | 0.799 | 2.786 | 50.994 | 0.921 | 0.465 | 0.43 |
| GPM-IMERG | 0.754 | 0.747 | 2.663 | 37.619 | 0.866 | 0.493 | 0.385 | |
| Longtouguai | GSMaP-Gauge | 0.619 | 0.965 | 3.564 | 23.094 | 0.876 | 0.564 | 0.332 |
| GPM-IMERG | 0.593 | 0.928 | 3.866 | 22.902 | 0.883 | 0.572 | 0.325 | |
| Hantaimiao | GSMaP-Gauge | 0.796 | 0.762 | 2.777 | 19.933 | 0.921 | 0.428 | 0.469 |
| GPM-IMERG | 0.723 | 0.978 | 3.389 | 16.628 | 0.854 | 0.553 | 0.321 | |
| Qingdamen | GSMaP-Gauge | 0.709 | 0.897 | 3.081 | 148.957 | 0.878 | 0.391 | 0.475 |
| GPM-IMERG | 0.651 | 0.91 | 3.342 | 135.595 | 0.838 | 0.441 | 0.409 | |
| Erzihao | GSMaP-Gauge | 0.746 | 0.868 | 2.837 | 79.658 | 0.913 | 0.455 | 0.433 |
| GPM-IMERG | 0.606 | 0.97 | 3.605 | 70.785 | 0.861 | 0.502 | 0.369 | |
| Xiangshawan | GSMaP-Gauge | 0.663 | 1 | 3.652 | 9.523 | 0.943 | 0.477 | 0.427 |
| GPM-IMERG | 0.663 | 0.939 | 3.691 | 0.559 | 0.918 | 0.51 | 0.384 |
Scatter graphs of daily precipitation data from the two satellite-derived precipitation data products plotted against observed rainfall data at ten stations in the Ten Tributaries region.
Scatter graphs of daily precipitation data from the two satellite-derived precipitation data products plotted against observed rainfall data at ten stations in the Ten Tributaries region.
Daily precipitation data corresponding to the position of each remote-sensing satellite measurement and ground station were extracted sequentially, and the daily precipitation values measured at the ground station were statistically analyzed to obtain an evaluation of the overall precipitation accuracy of each of the ten stations from 2018 to 2020.
In terms of characteristic indicators at a daily scale, the CC for GSMaP-Gauge and GPM-IMERG are 0.64 and 0.65, respectively. This indicates a certain linear correlation between the remote-sensing satellite precipitation data and the rainfall station measurements within the ten major holes. The MAE is 0.06 mm for both products, and the RMSE values are 3.33 mm for GSMaP-Gauge and 3.42 mm for GPM-IMERG. These metrics suggest a stable yet notable deviation between the satellite data and the ground measurements.
The Bias values are 39.86% for GSMaP-Gauge and 32.51% for GPM-IMERG, indicating that both satellite products tend to overestimate precipitation compared to the rainfall station data. However, this overestimation is significantly reduced when compared to the annual scale observations.
From the perspective of detection performance at a daily scale, the POD for GSMaP-Gauge is 0.89, and for GPM-IMERG it is 0.83. These high POD values demonstrate that both satellite products can accurately detect most daily precipitation events from 2018 to 2020. However, the FAR is 0.45 for GSMaP-Gauge and 0.51 for GPM-IMERG, indicating a moderate rate of false-positives in precipitation detection. The ETS values are 0.43 for GSMaP-Gauge and 0.35 for GPM-IMERG, reflecting a difference in precipitation event detection capabilities, with GSMaP-Gauge performing better.
In conclusion, although there are some discrepancies between the satellite precipitation data and the rainfall station measurements, there is a significant linear correlation at the daily scale, indicating a certain level of applicability in the study area. While both satellite products exhibit some degree of false reporting in daily precipitation events, they are generally able to accurately detect most of these events. Overall, GSMaP-Gauge performs better than GPM-IMERG in terms of detection capabilities.
Spatial distributions of accuracy evaluation metrics between the two satellite-derived precipitation data products and rain gauge station data. (a) Correlation coefficient (CC); (b) mean absolute error (MAE); (c) root-mean-square error (RMSE); (d) relative bias (BIAS); (e) probability of detection (POD); (f) false alarm rate (FAR); and (g) equitable threat score (ETS).
Spatial distributions of accuracy evaluation metrics between the two satellite-derived precipitation data products and rain gauge station data. (a) Correlation coefficient (CC); (b) mean absolute error (MAE); (c) root-mean-square error (RMSE); (d) relative bias (BIAS); (e) probability of detection (POD); (f) false alarm rate (FAR); and (g) equitable threat score (ETS).
The CC between the two remote-sensing products generally remains between 0.6 and 0.8, with some grids exceeding 0.8. This indicates that both GSMaP-Gauge and GPM-IMERG satellite precipitation data have a linear correlation with the rain gauge measurements in the Ten Dakou regions. Overall, the distribution of CC for GSMaP-Gauge is better across the entire basin, with a larger area of high-value regions.
The MAE reflects the average non-negative and negative offset errors between the satellite products and the gridded precipitation data. Although the MAE visual distribution maps showed apparent differences, the actual differences in value were minor, with GSMaP-Gauge having an MAE approximately 0.1 mm lower than that of GPM-IMERG; both products displayed small MAE values.
The RMSE values for both products ranged from 2.7 to 4 mm, but their spatial distribution maps revealed notable inconsistencies. GSMaP-Gauge exhibited an overall RMSE of 2.71 mm, while that of GPM-IMERG was 3.13 mm, indicating the superiority of GSMaP-Gauge in the context of this metric. This, combined with the MAE results, suggests low error dispersion and good stability for both products.
The BIAS of the two products exhibited similar spatial heterogeneity, being ideal in western and northern parts of the basin, and increasing toward the southeast. The highest BIAS values, indicating overestimation, occurred in the central basin, coinciding with a dense array of rainfall stations, and likely caused by complex topography.
To further compare detection accuracy, we analyzed the POD, FAR, and ETS maps under a 0.1 mm/day threshold. Both products exhibited a high POD in the central part of the study area, and lower values in the east and west. The POD of GSMaP-Gauge ranged from 0.77 to 0.94, while that of GPM-IMERG ranged from 0.66 to 0.92, with GSMaP-Gauge generally performing better. This is supported by statistical characteristics at the site scale, where GSMaP-Gauge excelled in quartile and mean comparisons. The FAR for both products was notably high in the central region, potentially due to an increased frequency of precipitation events, which influenced the performance indicators. Specifically, the FAR for GSMaP-Gauge varied between 0.22 and 0.56, whereas GPM-IMERG exhibited a range from 0.33 to 0.57, highlighting significant differences in their lower bounds. When comparing ETS, GSMaP-Gauge outperformed in identifying precipitation events. However, the narrower range of variation in POD, FAR, and ETS for GPM-IMERG indicates that it provides more consistent performance in capturing precipitation. Overall, GSMaP-Gauge outperformed GPM-IMERG in daily precipitation estimation, but GPM-IMERG exhibited better stability. The complex climate and terrain of the Ten Tributaries region pose challenges for accurate precipitation inversion, with both products overestimating precipitation in mountainous/hilly areas; hence, the flat northern plains are more conducive to satellite-based precipitation detection. Additionally, a high POD in satellite-derived precipitation data generally corresponds with a high FAR.
Rainstorm-scale accuracy assessment of satellite-derived precipitation data
Despite its predominantly arid and semi-arid climate, localized heavy rainfall events can lead to substantial flooding in the Ten Tributaries region of the Yellow River Basin. Here, we evaluated the accuracy of satellite-derived precipitation data in capturing such events, using ground observation data as a benchmark. We focused on the performance of these products under extremely low precipitation conditions and their capability to accurately detect short-duration intense rainfall. We also considered the impact of spatiotemporal resolution, particularly at the hourly scale, on data accuracy. This was done with the aim of providing recommendations for enhancing satellite-derived precipitation assessments, which are crucial for effective water resource management and flood mitigation in the basin.
Typical processes of six rainstorm events in the Ten Tributaries region.
Comparison of the heavy rainfall inversion accuracy of GPM-IMERG and GSMaP-Gauge across different SDII values.
Comparison of the heavy rainfall inversion accuracy of GPM-IMERG and GSMaP-Gauge across different SDII values.
The GPM-IMERG product performed poorly in terms of its CC for all heavy rainfall events, whereas the GSMaP-Gauge product performed better in events with larger SDII values. Overall, the GSMaP-Gauge product demonstrated better performance in rainfall inversion for heavy rainfall events, with precipitation estimates closely matching station observations, whereas the GPM-IMERG showed substantial deviations (BIAS values exceeding 100%). The MAE of both products exhibited an increasing trend with increasing SDII, and both products exhibited relatively low RMSE values. The RMSE of GSMaP-Gauge remained stable between 0.6 and 1.2 mm, while the RMSE of GPM-IMERG fluctuated to a greater degree with increasing SDII. In terms of BIAS, the GPM-IMERG product markedly overestimated precipitation across all heavy rainfall scales, whereas the GSMaP-Gauge product exhibited minimal overestimation or underestimation.
In terms of their ability to characterize heavy rainfall, the two products exhibited similar POD and FAR values, both having high hit rates and high false alarm rates, with the GSMaP-Gauge product having more stable precipitation hits and false alarms. Additionally, there was no significant trend between ETS and increasing SDII for either product, and their ability to characterize heavy rainfall was limited. Some heavy rainfall events had high hit rates but also high false alarm rates, leading to a relatively weak capability to identify precipitation events. There was no significant difference in the statistical indicators of the two satellite-derived precipitation datasets across the various heavy rainfall events at the basin scale when compared with the station scale. In terms of the CC indicator, the GSMaP-Gauge product alone showed good linear correlation with larger SDII values for two heavy rainfall events.
Overall, compared with the GPM-IMERG product, the GSMaP-Gauge product demonstrated better inversion capability for typical heavy rainfall events in the Ten Tributaries region, exhibiting superior accuracy in rainfall amount inversion and better performance in detecting heavy rainfall events.
DISCUSSION AND CONCLUSION
In this contribution, we have comprehensively evaluated the spatiotemporal accuracy characteristics of two satellite-derived precipitation datasets, GPM-IMERG and GSMaP-Gauge, in the Ten Tributaries region of the Yellow River Basin. We find that both datasets exhibit substantial overestimation of annual precipitation at the basin scale. However, this overestimation is notably reduced in daily precipitation estimates, which show strong correlations with ground-based rain gauge observations. At the annual scale, the inversion results in some central areas of the study region are less satisfactory, likely owing to cumulative errors over time. On the daily scale, precipitation data from both products display relatively small error variations across most parts of the study region. Our assessment indicates that GPM-IMERG and GSMaP-Gauge exhibit consistent trends in error statistics at the station scale but marked differences at the basin scale; GSMaP-Gauge demonstrates a broader distribution of high CC values and lower MAE and RMSE values across most grid points in the study area, indicating better consistency and performance. Both products are capable of capturing most rainfall events in the Ten Tributaries region, although the spatial distribution characteristics of their detection performances differ. Overall, GSMaP-Gauge outperforms GPM-IMERG in terms of accuracy at both the station and basin scales.
By focusing on the Ten Tributaries region, this study addresses the gap in accuracy evaluation of satellite-derived precipitation data under complex terrain and rainfall conditions, providing valuable data and support for hydrological research and characterization of the spatiotemporal distribution of precipitation in the region. We also analyzed the inversion accuracy of satellite-derived precipitation data at the heavy-rainfall-event scale in a region where precipitation primarily occurs in the summer with short duration and high intensity. The unique topography of the region leads to a high level of sediment inflow into the Yellow River during heavy rainfall events. Accurate evaluation at this scale offers a scientific basis for predicting heavy rainfall trends in the Ten Tributaries region, aiding in mitigating the impact of extreme precipitation events.
Given the limitations in the spatial distribution capacity of rainfall stations and the lack of measured precipitation data at ground stations, further research is required. Future research should aim to increase observational data from existing rainfall stations or use high-accuracy interpolated grid data and corrected remote-sensing precipitation data to further explore the accuracy of remote-sensing products in the Ten Tributaries region. The gridded surface observation precipitation data obtained through site data interpolation may introduce systematic errors owing to insufficient station data, low sampling frequency, and limitations of the interpolation method. Developing more suitable interpolation and resampling methods will provide more accurate station and satellite-derived precipitation data. With advancements in sensor technology and precipitation retrieval algorithms, numerous satellite-derived precipitation datasets are now available. This study focused on GPM-IMERGE and GSMaP-Gauge, excluding other products. Future studies should include additional satellite-derived precipitation datasets for comprehensive comparative analysis over longer periods, aligning with the growing research interest in regional hydrometeorology under a changing climate. Addressing these areas in future research will enhance the accuracy and reliability of satellite-derived precipitation datasets, ultimately improving hydrological modeling and disaster management in regions with complex terrain and variable climatic conditions.
FUNDING
This study was funded by the National Natural Science Foundation of China (U2243203) (L.R.) and the National Key R&D Program of China (2023YFC3209300) (C.Z.).
DATA AVAILABILITY STATEMENT
All relevant data are available from http://dx.doi.org/10.6084/m9.figshare.27896865.
CONFLICT OF INTEREST
The authors declare there is no conflict.
