Abstract
While many studies have compared global precipitation datasets at national, continental, and global scales, few have evaluated these data at river basin scales. This study explored differences in precipitation estimates and trends of 12 widely applied precipitation datasets, including gauge-, satellite-, and reanalysis-based products, for the world's 6,292 river basins. Results showed that disparities between 12 precipitation datasets were considerable. A total of 3,125 river basins, with a land area of 5,989.1×104 km2, had differences in estimated annual average precipitation exceeding 500 mm yr−1, and these basins were mainly distributed in Greenland, Africa, Oceania, and West Asia. Disparities between the precipitation datasets were particularly large during the dry season when the percentage difference between the highest and lowest precipitation estimates exceeded 500% in 1,390 river basins (4,839.7×104 km2) expected due to numerical reasons. Differences in rainfall trends also varied markedly between data sources. The data products do not agree on precipitation trends for all river basins. These findings illustrate the importance of accurate precipitation data to ensure effective policy and planning in term of hydropower generation, domestic water supply, flood protection, and drought relief at river basin scales and highlight the uncertainty that exists in current global precipitation data.
HIGHLIGHTS
A comprehensive evaluation of global precipitation datasets was compared at the river basin scale.
Seasonal and annual differences in these precipitation datasets were compared.
Gives a sufficient explanation of which product is suitable for which river basin.
INTRODUCTION
The intensification of climate change has accelerated the global water cycle, causing significant spatiotemporal heterogeneity in global precipitation (Los et al. 2001; Siepielski et al. 2017; Roushangar et al. 2018; Chai et al. 2019, 2020, 2021, 2022; Zhu et al. 2023). Accurate and reliable precipitation data can reveal the mechanisms driving precipitation change, hydraulic performance of flow confluence characteristics, and the accuracy of hydrological, ecological, and atmospheric models depends heavily on the availability of good-quality precipitation estimates (Huffman et al. 1997; Bagley et al. 2014; Espinoza et al. 2016; Roushangar et al. 2020a, 2020b).
Currently, many precipitation data products based on observations exist, including gauge-, satellite-, and reanalysis-based products. Tools, including rain gauges, disdrometers, and radar are widely used to monitor precipitation near the Earth's surface, with the advantage of high precision at gauge locations (Sun et al. 2018). Gauge observations have been widely used to measure precipitation directly at the Earth's surface (Kidd 2010). However, gauge-based products have certain disadvantages: uneven distribution of gauge stations; gauge scarcity in deserts, oceans, forests, mountains, valleys, and developing countries; and a lack of long-term and continuous observations (Easterling et al. 1996; Ringard et al. 2015). Satellite-based precipitation products can overcome these deficiencies by providing precipitation fields with high space–time variability and global coverage (Thiemig et al. 2012; Derin & Yilmaz 2014). Therefore, satellite precipitation products have become essential sources of precipitation information, especially in regions where the gauged distribution is sparse and uneven (Derin & Yilmaz 2014). Satellite data have been used widely to forecast typhoons, flooding, and heavy rain, and in the analysis of atmospheric circulation characteristics (Olson et al. 1996; Liu et al. 2011). This kind of dataset has also been used in water resources management and drought monitoring (Li et al. 2018; Amini et al. 2019). However, it has been suggested that satellite-based precipitation products may include large systematic and random errors (e.g., observation errors, sample source uncertainty, and algorithm errors), especially in the mid-to-high latitudes and areas of complex terrain (Xie & Arkin 1996; Nijssen & Lettenmaier 2004; Dinku et al. 2008; Villarini et al. 2009; Pan et al. 2010). The complex terrain, with varied cloud cover, may disturb satellites to capture a more accurate estimate of precipitation (Villarini et al. 2009; Pan et al. 2010). Reanalysis precipitation products are obtained by reintegrating and optimizing various observation data (Bengtsson & Shukla 1988). In detail, reanalysis provides the most complete picture currently possible of past weather and climate, and they are a blend of observations with past short-range weather forecasts rerun with modern weather forecasting models. Reanalysis produces data that goes several decades back in time, providing an accurate description of the climate of the past. Owing to their dynamic mechanisms and physical characteristics, reanalysis products have the advantages of wide spatial coverage, long-term data availability, continuity, and consistency (Masina et al. 2011; Bellucci et al. 2013). Notwithstanding, reanalysis data are still affected by observation uncertainty, model, and assimilation errors (Karam & Bras 2008; Thorne & Vose 2010).
Before selecting and applying a precipitation product, it is necessary to evaluate the differences between the available data sources. In recent decades, researchers have focused on the inter-comparison of precipitation datasets at national, continental, and global scales (Adler et al. 2001; Tian et al. 2009; Liu et al. 2010, 2011; Gehne et al. 2016; Herold et al. 2017; Tang et al. 2018; Li et al. 2019; Zhang et al. 2019). For example, Sun et al. (2018) quantified the discrepancies between 30 precipitation products over multiple time scales and found that the difference in annual precipitation over the land surface could be as high as 300 mm year−1 (Sun et al. 2018). At the river basin scale, the reliable estimate of precipitation is closely related to the local water resources management, flood control and disaster reduction and reservoir operation. However, few studies have considered data inter-comparison at the watershed scale. In particular, a comprehensive evaluation of the available precipitation data for each river basin worldwide has not yet been attempted.
In the past five years alone, thousands of studies have applied different precipitation products to address scientific problems on the catchment scale (Kling et al. 2014). For instance, both gauge-based (Chen et al. 2014; Zhang et al. 2015) and reanalysis precipitation data (Su et al. 2017) have been used in the Yangtze River Basin to investigate the effects of precipitation on river discharge. There exist certain differences in the estimation of precipitation in the Yangtze River Basin with different precipitation products, thus it is crucial to select the appropriate precipitation products. Considering the wide application of different precipitation products at the catchment scale, a comprehensive evaluation of these products at the river basin is warranted. Hence, this study aims to identify and quantify the spatiotemporal heterogeneities of 12 different precipitation products, including gauge-, satellite-, and reanalysis-based products. To do this, we use the TFPM-MK Method and the Maximum Percentage Difference (MPD) Method to evaluate the magnitude of the difference in estimating multi-year averaged values of the precipitation at seasonal scale (dry season and wet season) and at annual scale, and in estimating the trends of precipitation, in 6,292 river basins.
DATA AND METHODS
Methods
In this paper, we used 12 precipitation data products, including gauge-, satellite-, and reanalysis-based products, to evaluate the performance of the different products. To do this, we calculate the values of annual mean precipitation, and mean precipitation in dry and flood seasons, by using Thiessen Polygons method. We further analyze the trends of precipitation in each river basin (TFPM-MK Method), and estimated the MPD of the annual precipitation across the products. To evaluate the difference in estimating precipitation, we have the bilinear interpolation method to transform all the precipitation products into the spatial resolution of 0.5° × 0.5°.
Thiessen Polygons method
Global precipitation datasets
Name . | Source . | Spatial resolution . | Spatial coverage . | Temporal resolution . | Temporal coverage . | References . |
---|---|---|---|---|---|---|
Gauge-based products | ||||||
CRU | The CRU of the University of East Anglia | 0.5° × 0.5° | Global | Monthly | 1901–2020 | Harris et al. (2014) |
GPCC | GPCC | 0.5° × 0.5° | Global | Monthly | 1891–2016 | Rudolf et al. (2010) |
PREC/L | NOAA/ESRL/PSD | 0.5° × 0.5° | Global | Monthly | 1948–2020 | Chen et al. (2002) |
UDEL | NOAA/ESRL/PSD | 0.5° × 0.5° | Global | Monthly | 1900–2017 | Willmott & Matsuura (2001) |
CPC-Global | NCEP/Climate Prediction Center | 0.5° × 0.5° | Global | Daily | 1979–2020 | Chen et al. (2008) |
Satellite-based products | ||||||
CMAP | NOAA/ESRL/PSD | 2.5° × 2.5° | Global | Monthly | 1979–2020 | Xie & Arkin (1997) |
GPCP | NOAA/ESRL/PSD | 2.5° × 2.5° | Global | Monthly | 1979–2020 | Adler et al. (2012, 2016) |
MSWEP | CPC/GPCC/CMORPH | 0.5° × 0.5° | Global | Daily/3 hourly | 1979–2020 | Beck et al. (2017) |
GPCP_PEN_v2.2 | OPI, SSM/I, GPI, MSU | 2.5° × 2.5° | Global | 5-daily | 1979–2017 | Xie et al. (2003) |
Reanalysis products | ||||||
NECP2 | NOAA/ESRL/PSD | 1.875° × 1.875° | Global | Monthly/6 hourly | 1979–2020 | Kalnay et al. (1996) Kanamitsu et al. (2002) |
ERA5 | ECMWF | 1.5° × 1.5°/0.75° × 0.75° | Global | Monthly /6 hourly | 1979–2020 | Dee et al. (2011) |
JRA-55 | NOAA | 60 km | Global | Monthly/3 hourly /6 hourly | 1958–2020 | Ebita et al. (2011) |
Name . | Source . | Spatial resolution . | Spatial coverage . | Temporal resolution . | Temporal coverage . | References . |
---|---|---|---|---|---|---|
Gauge-based products | ||||||
CRU | The CRU of the University of East Anglia | 0.5° × 0.5° | Global | Monthly | 1901–2020 | Harris et al. (2014) |
GPCC | GPCC | 0.5° × 0.5° | Global | Monthly | 1891–2016 | Rudolf et al. (2010) |
PREC/L | NOAA/ESRL/PSD | 0.5° × 0.5° | Global | Monthly | 1948–2020 | Chen et al. (2002) |
UDEL | NOAA/ESRL/PSD | 0.5° × 0.5° | Global | Monthly | 1900–2017 | Willmott & Matsuura (2001) |
CPC-Global | NCEP/Climate Prediction Center | 0.5° × 0.5° | Global | Daily | 1979–2020 | Chen et al. (2008) |
Satellite-based products | ||||||
CMAP | NOAA/ESRL/PSD | 2.5° × 2.5° | Global | Monthly | 1979–2020 | Xie & Arkin (1997) |
GPCP | NOAA/ESRL/PSD | 2.5° × 2.5° | Global | Monthly | 1979–2020 | Adler et al. (2012, 2016) |
MSWEP | CPC/GPCC/CMORPH | 0.5° × 0.5° | Global | Daily/3 hourly | 1979–2020 | Beck et al. (2017) |
GPCP_PEN_v2.2 | OPI, SSM/I, GPI, MSU | 2.5° × 2.5° | Global | 5-daily | 1979–2017 | Xie et al. (2003) |
Reanalysis products | ||||||
NECP2 | NOAA/ESRL/PSD | 1.875° × 1.875° | Global | Monthly/6 hourly | 1979–2020 | Kalnay et al. (1996) Kanamitsu et al. (2002) |
ERA5 | ECMWF | 1.5° × 1.5°/0.75° × 0.75° | Global | Monthly /6 hourly | 1979–2020 | Dee et al. (2011) |
JRA-55 | NOAA | 60 km | Global | Monthly/3 hourly /6 hourly | 1958–2020 | Ebita et al. (2011) |
Definition of the dry and wet seasons
To detect seasonal discrepancies in precipitation, the calendar year was divided into dry, normal, and wet seasons (see results in Supplementary material, Figures S2 and S3). The three successive months in which the sum of the multi-year average runoff was the lowest (highest) was considered to be the dry (wet) season (Chou & Lan 2012; Chou et al. 2013). The remaining six months were classified as the normal season. However, it should be noted that different river basins have different hydroclimatic characteristics with different duration of dry seasons. Thereby, our selection criteria of dry seasons might be improper for a small portion of river basins.
TFPM-MK method for trend analysis
The Mann–Kendall trend test method proposed by Mann (1945) and Kendall (1975) has been widely used to detect variation trends in hydro-meteorological time series data (Zareiee 2014). One assumption of this method is that the data series should be independent, and this is not true for many hydrological time series. Consequently, this Mann–Kendall trend test may overestimate the significance of both positive and negative trends (Yue & Wang 2002; Su et al. 2018). To overcome this, we applied the Mann–Kendall trend test with trend-free pre-whitening to evaluate precipitation trends.
If Ld≤rm≤Lu, the time series are independent and the trends can be evaluated using the traditional Mann–Kendall trend test method. Otherwise, the time series need to be revised to meet the pre-whitening requirements. The processes are as follows:
MPD of the multi-year average annual precipitation
Data sources
Table 1 lists 12 precipitation data products used in this study, which represent the most widely used rainfall data sources. These products include five gauge, four satellite, and three reanalysis-based products. The common product data range is 1979–2016; hence, this period was selected as the research period in this study.
RESULTS AND DISCUSSION
Discrepancies in annual precipitation
Multi-year average annual mean precipitation (mm year−1) in the 6,292 river basins in 1979–2016 based on the 12 precipitation productions. The river basins with red color represent the low annual mean precipitation, while the river basins with blue color show the high annual mean precipitation. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
Multi-year average annual mean precipitation (mm year−1) in the 6,292 river basins in 1979–2016 based on the 12 precipitation productions. The river basins with red color represent the low annual mean precipitation, while the river basins with blue color show the high annual mean precipitation. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
Difference in the multi-year average annual mean precipitation in 1979–2016 between the 12 precipitation productions. (a) The maximum difference in the multi-year average annual precipitation between the 12 precipitation productions, mm year−1; (b) presents the MPD of the multi-year average annual precipitation between the 12 precipitation productions. Please see the calculation in Section 2.1.4. The river basins with red color represent the low difference in estimating annual mean precipitation across the datasets, while the river basins with blue color show the high difference in estimating annual mean precipitation across the datasets.
Difference in the multi-year average annual mean precipitation in 1979–2016 between the 12 precipitation productions. (a) The maximum difference in the multi-year average annual precipitation between the 12 precipitation productions, mm year−1; (b) presents the MPD of the multi-year average annual precipitation between the 12 precipitation productions. Please see the calculation in Section 2.1.4. The river basins with red color represent the low difference in estimating annual mean precipitation across the datasets, while the river basins with blue color show the high difference in estimating annual mean precipitation across the datasets.
Figure 2(b) shows the percentage difference between the maximum and minimum annual precipitation for each river basin. This ratio exceeds 100% for more than half of the world's river basins (3,708 covering an area of 6,198.4 × 104 km2), mainly in Greenland, Africa, western Asia, and Oceania. Only 177 river basins (939.2 × 104 km2) had an MPD smaller than 30%, and these were concentrated in South America and Europe. The degree of variability in annual precipitation between the different data products questions the findings of previous studies that have used and applied these precipitation data. Indeed, the level uncertainty may be substantial when the percentage differences in annual precipitation between the datasets are considered.
The world's 15 largest river basins occupy 29.3% of the global land area (3,977.1 × 104 km2), support large populations, and are among the most diverse land-based ecosystems (Best 2019). We found that annual precipitation in the 13 largest river basins varied by over 200 mm year−1. Variability was especially high in the Nile (697.0 mm year−1), Tamanrasett (696.9 mm year−1), Ganges (679.6 mm year−1), and Lake Chad River Basins (699.8 mm year−1). The NECP precipitation product significantly overestimated annual precipitation in the Nile, Zaire, Parana, Niger, Tamanrasett, and Lake Chad River Basins.
Variations trends in the multi-year average annual mean precipitation based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the annual precipitation shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the annual precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
Variations trends in the multi-year average annual mean precipitation based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the annual precipitation shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the annual precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
Dry season precipitation changes and uncertainty
Multi-year average of the precipitation of the dry season in the 6,292 river basins in 1979–2016 between the 12 precipitation productions. (a–l) Multi-year average precipitation of the dry season in 1979–2016, mm. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
Multi-year average of the precipitation of the dry season in the 6,292 river basins in 1979–2016 between the 12 precipitation productions. (a–l) Multi-year average precipitation of the dry season in 1979–2016, mm. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis products.
The driest river basins were mainly in Africa, Oceania, and Central Asia. These regions have a high risk of drought, with dry season precipitation averaging less than 20 mm year−1. Conversely, Europe, eastern North America, and South America have abundant water resources, with dry season precipitation exceeding 100 mm year−1. Both satellite (Figure 4(f)–(i)) and reanalysis (Figure 4(j)–(l)) based products display a major variability. For instance, GPCP and GPCP_PEN_v2.2 (satellite precipitation products) indicate higher dry season precipitation in Europe and eastern North America compared to other satellite-based products (CMAP and MSWEP). Of the reanalysis-based products, NECP2 tends to overestimate dry season precipitation in the Southern Hemisphere, especially in Africa and Oceania. It should be noted that all the reanalysis precipitation products deviate notably from the gauge-based products and satellite-based products, especially in mountainous and coastal regions. This might be caused by the reanalysis models' inability to represent the effects of complex orography and/or sparse observational inputs for assimilations (Kim & Park 2016).
Variation trends in the multi-year average of the precipitation of the dry season in 1979–2016 based on the 12 precipitation productions. (a–l) Variation trends in the average precipitation of the dry season based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the precipitation in the dry season shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Variation trends in the multi-year average of the precipitation of the dry season in 1979–2016 based on the 12 precipitation productions. (a–l) Variation trends in the average precipitation of the dry season based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the precipitation in the dry season shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Difference in the multi-year average of the precipitation of the dry season during 1979–2016 between the 12 precipitation productions. (a) Maximum difference in the multi-year average precipitation of the dry season between the 12 precipitation productions, mm year−1; (b) the MPD of the multi-year average precipitation of the dry season between the 12 precipitation productions. Please see the calculation in Section 2.1.4. The river basins with red color represent the low difference in estimating the multi-year average of the precipitation of the dry season across the datasets, while the river basins with blue color show the high difference in estimating the multi-year average of the precipitation of the dry season across the datasets.
Difference in the multi-year average of the precipitation of the dry season during 1979–2016 between the 12 precipitation productions. (a) Maximum difference in the multi-year average precipitation of the dry season between the 12 precipitation productions, mm year−1; (b) the MPD of the multi-year average precipitation of the dry season between the 12 precipitation productions. Please see the calculation in Section 2.1.4. The river basins with red color represent the low difference in estimating the multi-year average of the precipitation of the dry season across the datasets, while the river basins with blue color show the high difference in estimating the multi-year average of the precipitation of the dry season across the datasets.
Wet season precipitation changes and uncertainty
Multi-year average of the precipitation of the wet season in the 6,292 river basins in 1979–2016 based on the 12 precipitation productions. (a–l) Multi-year average precipitation of the wet season in 1979–2016, mm. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Multi-year average of the precipitation of the wet season in the 6,292 river basins in 1979–2016 based on the 12 precipitation productions. (a–l) Multi-year average precipitation of the wet season in 1979–2016, mm. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Variation trend in multi-year average precipitation of the wet season in 1979–2016 based on the 12 precipitation productions. (a–l) Variation trends in the average precipitation of the wet season based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the precipitation in the wet season shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Variation trend in multi-year average precipitation of the wet season in 1979–2016 based on the 12 precipitation productions. (a–l) Variation trends in the average precipitation of the wet season based on the TFPM-MK Method. Note: |ZMK| > 2.32, |ZMK| > 1.96, and |ZMK| > 1.64 mean the precipitation in the wet season shows an increasing or decreasing trend with the confidence level over 99, 95, and 90%, respectively; |ZMK| < 1.64 means the precipitation had no trend. CRU, GPCC, PREC/L, UDEL, and CPC-Global are the gauge-based products. CMAP, GPCP, MSWEP, and GPCP_PEN_v2.2 are the satellite-based products. NECP2, ERA5, and JRA-55 are the reanalysis-based products.
Difference in the multi-year average precipitation of the wet season in 1979–2016 between the 12 precipitation productions. (a) Maximum difference in the multi-year average precipitation of wet season between the 12 precipitation productions, mm year−1; (b) MPD of the multi-year average precipitation of wet season between the 12 precipitation productions. Please see the calculation given in Section 2.1.4. The river basins with red color represent the low difference in estimating the multi-year average precipitation of the wet season across the datasets, while the river basins with blue color show the high difference in estimating the multi-year average precipitation of the wet season across the datasets.
Difference in the multi-year average precipitation of the wet season in 1979–2016 between the 12 precipitation productions. (a) Maximum difference in the multi-year average precipitation of wet season between the 12 precipitation productions, mm year−1; (b) MPD of the multi-year average precipitation of wet season between the 12 precipitation productions. Please see the calculation given in Section 2.1.4. The river basins with red color represent the low difference in estimating the multi-year average precipitation of the wet season across the datasets, while the river basins with blue color show the high difference in estimating the multi-year average precipitation of the wet season across the datasets.
Most previous studies evaluate the performance of precipitation products at national, continental, and global scales. However, there is a lack of a comprehensive evaluation of precipitation products at catchment scale. Our study may bring some important implications on the applications of precipitation products on water resources management, flood control and disaster reduction and reservoir operations. Consistent with the previous studies, we all concluded that the NECP precipitation dataset may largely overestimate the global land precipitation products.
CONCLUSION
An accurate representation of the historical and current climate is necessary to provide confidence in future climate projections. As a fundamental driver of the global hydrological cycle, accurate precipitation data are essential. In this study, we provided a comprehensive comparison of 12 precipitation products that have been used extensively as abundant data sources of global precipitation patterns, to quantify the uncertainty in these data at the river basin scale. Our results revealed extensive discrepancies between precipitation data products. In particular, reanalysis-based precipitation products, especially NECP, had the highest uncertainty compared to other data sources. Specifically, we found that NECP tended to overestimate dry season precipitation in the Southern Hemisphere, which is consistent with the findings of similar studies. During the wet season, NECP precipitation at the global scale was fairly accurate. While data discrepancies were large in both the wet and dry seasons, dry season rainfall distribution revealed the greatest disparity between data sources.
These very large discrepancies in precipitation estimates at the basin scale raise concerns with respect to the findings of many studies that have used these data for regional hydrological, agricultural, and climate analysis. This is particularly problematic in data-poor regions of the Southern Hemisphere, such as northern Africa, where accurate information with respect to dry season rainfall trends is needed to reduce the impact of drought. With the development and progress of future monitoring methods, the diversity and accuracy of observation methods, and the increasing number of observation sites, we are more confident to select a few better rainfall analysis products. For instance, in the future, we can do some rainfall experiments in the small watershed scale to judge which rainfall product is better based on the experimental results.
ACKNOWLEDGEMENT
GJW acknowledges the National key research and development program [grant number: 2022YFC3005404], the Youth Project of National Natural Science Foundation China [grant number: 42301018], the Independent Innovation Project of Changjiang Design Group Co., Ltd (Grant CX2020Z19) and China Postdoctoral Science Foundation funded project (2022M710490). We acknowledge Yuanfang Chai who provided the valuable writing and comments.
AUTHOR CONTRIBUTIONS
X.H. and X.Y. led the writing, designed the research and performed the data analysis. H.X., Q.G., Z.Z., and X.C. provided valuable comments.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.
CONFLICT OF INTEREST
The authors declare there is no conflict.