Evaluation and projection of extreme precipitation using CMIP6 model simulations in the Yellow River Basin Open Access

Extreme precipitation events are greatly impacted by climate warming (Tabari 2020), and observations indicate that each 1 °C rise in global temperature could cause a 7% increase in extreme daily precipitation events (IPCC 2021). The influence of extreme precipitation on human life, economy, and society is severe, as it can lead to disasters such as floods, mudslides, and landslides (Konapala et al. 2020; Liu et al. 2020). It is widely recognized that extreme precipitation will become more frequent and intense under climate warming (O'Gorman 2015; Yu et al. 2023), and therefore, accurate projection of future extreme precipitation is crucial.

The global climate models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6) are the major tools for projecting future climate (Li et al. 2022a; Feng et al. 2023). Many studies have been found to project extreme precipitation using CMIP6 GCMs. Li et al. (2021) projected the future trends and spatiotemporal distributions in extreme precipitation globally. Wang et al. (2023) analyzed changes of extreme precipitation in China and the sensitivity of extreme precipitation to climate change across various regions. Furthermore, Zhang et al. (2023) projected the future trends and cycle characteristics in extreme precipitation for the Jialing River Basin under the four shared socioeconomic pathway (SSP) scenarios. Current studies found that extreme precipitation exhibited spatial heterogeneity across regions, making it imperative to research the future changes in extreme precipitation on subregional and regional scales (Jin et al. 2023).

The capability of GCMs to simulate the climate of historical periods is an important aspect of the credibility of their future projections (Wang et al. 2023). Therefore, the simulation capability of GCMs needs to be evaluated before using their outputs for projecting future extreme precipitation. Previously, various studies have evaluated the capability of GCMs to simulate extreme precipitation. Ali et al. (2023) used the Kling Gupta efficiency metric to evaluate the capability of GCMs to replicate the extreme precipitation indices during the historical period over Pakistan. Lei et al. (2023) evaluated the capability of 33 CMIP6 GCMs to simulate extreme precipitation indices across Central Asia using interannual variability scores (IVS), Taylor diagrams, and comprehensive evaluation indices. The study found that GCMs performed differently in regions with different topographic and climatic characteristics. Liu et al. (2022) found that GCMs had different performances in capturing spatial patterns of extreme precipitation in different seasons. Consequently, the historical simulated performances of GCMs need to be evaluated before projecting in the future.

The Yellow River Basin (YRB) is of vital importance to the economic, social, and environmental well-being of China (Wang et al. 2022). Nevertheless, the YRB is being severely impacted by extreme precipitation events. For example, in September 2021, the average precipitation (179.0 mm) was 1.7 times higher than that of the same period of the normal year in the YRB, which was the highest in the same period of history since 1961, resulting in 4,986 km² of crop damage and a direct economic loss of 15.34 billion RMB. According to Zhang et al. (2014) and Li et al. (2022a), precipitation in the upstream of the YRB is increasing at a rate of 8.1 mm/a, and the number and index of heavy precipitation processes are also increasing. The midstream and downstream of the YRB are situated in the eastern monsoon region, where evident seasonal changes are notable in the extreme precipitation. The extreme precipitation indices exhibited a low spatial pattern in the northwest and a high pattern in the southeast, with significant spatial variability in the YRB (Niu et al. 2020). However, there are fewer studies projecting the spatial and temporal distribution of subregional extreme precipitation under different SSP scenarios. Therefore, it is imperative to project and analyze potential variations in extreme precipitation in the YRB and subregions and implement measures to reduce the associated hazards.

In this study, 23 CMIP6 GCMs were evaluated for their capabilities to simulate extreme precipitation indices in the YRB and its eight level-II water resource regions during the historical period (from 1961 to 2010). The SSP2-4.5 and SSP5-8.5 scenarios were selected to project the temporal variability and spatial distribution of extreme precipitation for the future period (from 2021 to 2050). The study could provide a reference for future impact evaluation of extreme precipitation.

The Yellow River originates from the Bayan Kara Mountains on the Qinghai–Tibet Plateau, with a total length of 5,464 km. It flows from west to east through nine provinces (autonomous regions) of China and finally flows into the Bohai Sea. The Yellow River has a basin area of 795,000 km² and an average annual precipitation of 446 mm. The precipitation generally decreases from southeast to northwest. The midstream and upstream of the Yellow River are dominated by mountains, whereas the midstream and downstream are dominated by plains and hills. The YRB is mainly characterized by a continental monsoon climate.

The observations are the daily precipitation of 126 meteorological stations from 1961 to 2010, which are obtained from the China Meteorological Sharing Network (http://data.cma.cn). The precipitation series were relatively complete and of good quality (Wang et al. 2020; Guan et al. 2022), and the missing values were interpolated linearly to complete the series. The inverse distance interpolation method (Chen & Liu 2012) was used to interpolate into 431 grids with a resolution of 50 km × 50 km in the YRB.

The climate projection scenarios of the Scenario Model Intercomparison Project (ScenarioMIP) are rectangular combinations of different SSPs and radiative forcing. Among them, SSP2-4.5 represents the medium development and the radiative forcing stabilizes at 4.5 W/m² in 2100, and SSP5-8.5 represents the general development and the radiative forcing stabilizes at 8.5 W/m² in 2100.

Because SSP2-4.5 scenario has the highest likelihood of occurring in the future compared to the other emission scenarios (Ning et al. 2022), SSP5-8.5 scenario represents the most pessimistic scenario in 2100. We collected the available outputs of 23 CMIP6 GCMs under the SSP2-4.5 scenario and 21 CMIP6 GCMs under the SSP5-8.5 scenario (Table 1) for the historical (from 1961 to 2010) and future (from 2021 to 2050) (https://esgf–node.llnl.gov/projects/esgf–llnl) to project future extreme precipitation. Considering the different spatial resolutions of the models, a bilinear interpolation method (Mastyło 2013) was used to standardize all models to a 50 km × 50 km grid resolution.

Six extreme precipitation indices, as defined by the Expert Team on Climate Change Detection and Indicators (PL et al. 2002), were selected taking into account the climatic characteristics of the region and the statistics of extreme precipitation events (Table 2). Since the statistics show an increase in the frequency of short-duration heavy precipitation events in the YRB, RX1day and 5-day maximum precipitation (RX5day) were selected to capture these types of precipitation events. Meanwhile, considering that YRB is located in arid and semi-arid areas where droughts are common, CDD represents exactly the conditions of very low precipitation. In addition, R95p, as an indicator of the intensity of extreme precipitation, is essential for understanding and predicting the impacts of climate change related to extreme precipitation. Simultaneously, R10mm is a common indicator of the frequency of extreme precipitation, which is directly relevant to agricultural planning and the development of flood control measures. A simple daily intensity index (SDII) shows changes in the distribution of precipitation, which is very helpful in assessing overall changes in precipitation patterns. These indices have the characteristics of weak extremity, small noise, and large significance, and are widely used in the study of extreme precipitation (Dong & Dong 2021; Zhu et al. 2023).

The spatial biases of YRB and its subregions are the average of the absolute values of all grid biases within the YRB or subregions.

The distribution function for precipitation used a hybrid function consisting of a step function and a two-parameter gamma distribution function due to the interpolated diffusion problem associated with corrections using empirical functions.

The overall spatial bias of R95p is greater at 47.96%, followed by RX5day (21.04%) and CDD (18.91d) (Table 3). Meanwhile, R10mm exhibits a smaller overall spatial bias at 5.66d. Compared to other subregions, Regions I and II exhibit the greatest regional spatial biases for the indices, particularly for RX5day (37.79 and 40.65%), R95p (97.27 and 83.22%), R10mm (14.49d and 9.05d), and CDD (27.76d and 32.75d). In Region VII, the regional spatial biases are especially high for RX1day and SDII, with biases of 27.91 and 28.47%, respectively.

The spatial performances of RX1day (Figure 2) and SDII (Figure 7) show the predominance of negative biases. Specifically, 69.6 and 74.5% of the total number of grids in the basin exhibit negative biases for RX1day and SDII, respectively. This indicates that the GCMs are lower than the observations. Moreover, the spatial performance of CDD (Figure 6) has a negative bias in more than 99.9% of the grids. The positive biases dominate the spatial performance of RX5day (Figure 3), R95p (Figure 4), and R10mm (Figure 5). For RX5day, 71.2% of the grids in the basin have positive biases, while in R95p and R10mm simulations, more than 80% of the grids have positive biases. For the single model, the extreme precipitation indices' results differed between models. CanESM5 only shows a negative bias for CDD and positive biases for the other five indices. Meanwhile, EC–Earth3, EC–Earth3–Veg, and EC–Earth–Veg–LR predominantly exhibit negative biases for six indices. It has been determined that INM–CM4–8 and INM–CM5–0 have comparable performance in replicating the six extreme precipitation indices, while GCMs from the same institution may have similar results due to similarities in the parameterization process, among other factors.

In Regions I and II, biases generally perform similarly for the other indices, except negative biases in the simulation of CDD. However, the simulations of SDII perform slightly differently, with SDII showing significant positive biases with biases higher than 10% in Region I, while the simulations of SDII show negative biases near the Datong River Basin in Region II. Nevertheless, in Region III, changes in extreme precipitation indices are relatively complex. RX1day, CDD, and SDII show negative biases, while RX5day, R95p, and R10mm show positive biases in Region III except for the Dahei River Basin. In Region IV, only the R95p exhibits positive biases. The extreme precipitation indices perform similarly in Regions V and VI. RX1day, CDD, and SDII exhibit negative biases of over 5%, while R95p and R10mm exhibit a positive bias of more than 10% and over 3d, respectively. In Region VII, six extreme precipitation indices exhibit negative biases, particularly in the Yellow River estuary. R95p and R10mm exhibit positive biases, while the other indices have negative biases in Region VIII.

GCMs perform better in Regions III and VIII, with the IVS of five extreme precipitation indices being less than 0.2. However, the GCMs perform poorly in Regions I, II, V, and VII, with the IVS of most indices being greater than 0.2 and the IVS of R95p being greater than 1.

The GCM simulations show the strongest spatial distribution of extreme precipitation indices in Region VIII, with the TSS greater than 0.8. In Region IV, five indices exhibit the TSS greater than 0.8, resulting in the second highest simulation capability. The simulation capacity of GCMs is inadequate in Region VII, with RX1day and SDII with the TSS less than 0.4. This implies that GCMs do not effectively replicate the spatial distribution of extreme precipitation indices in Region VII.

The temporal variation and spatial distribution of extreme precipitation indices were projected under the SSP2-4.5 and SSP5-8.5 scenarios from 2021 to 2050, with 1981 to 2010 as the base period. The median of GCMs on each grid was used as a multimodel ensemble to project future changes in extreme precipitation indices.

Two GCMs display outliers: CNRM–ESM2–1 shows outliers in 49.0% (RX1day), 46.5% (RX5day), 72.8% (R95p), 3.2d (R10mm), and −6.8d (CDD). The other GCM, CMCC–ESM2, has outliers in 55.1% (RX1day) and 17.5% (SDII), which are approaching or exceeding three times the median. This may be due to the characteristics or assumptions inherent in the GCMs or to the uncertainty of the GCMs in addressing particular variables or conditions.

Compared to the SSP2-4.5 scenario, the relative variability of the indices increases significantly under the SSP5-8.5 scenario, with an increase in extreme precipitation on the temporal scale.

Most GCMs produce projection results that align with the multi-model median projection results. However, some models exhibit different behaviors, for example, the projection results of INM–CM4–8 and INM–CM5–0 are opposite to other GCMs. The projections of other indices except for CDD projected by INM–CM4–8 and INM–CM5–0 show negative biases in the downstream of Region I, Huangshui River Basin of Region II, Dahei River Basin near Region III, and Fen River Basin of Region V. The results differ from the simulation results of most models.

Overall, there is an increasing trend in extreme precipitation changes from the west to the east. Extreme precipitation indices change greatly in Region IV and the downstream of Regions V, VII, and VIII. This is especially true for the Wuding River Basin in Region IV and parts of Region V. The Jing River Basin and Luo River Basin in Region V have relatively minor changes (Figures 21 and 22).

Although the increase in most indices is greater in the MED, the change in the range of a single model is different. For example, in the case of RX1day, compared to the SSP2-4.5 scenario, the CMCC–ESM2 increases less in the SSP5-8.5 scenario and the CanESM5 increases more, especially in Region I where the increase is very pronounced.

CMIP6 GCMs have differences in simulating extreme precipitation indices. Our study found that most GCMs performed poorly in the spatiotemporal simulation of R95p and R10mm, which aligns with the findings of Lei et al. (2023). Due to the complex topography of the YRB, the low resolution of GCMs makes it difficult to simulate the changes in extreme precipitation thresholds. You & Ting (2023) pointed out that improving the resolution of the models could better simulate extreme precipitation, particularly in regions with complex topography. However, it does not mean that the higher resolution, the better the GCMs simulation, as Xiao et al. (2023a) showed that some low-resolution models may have better performance in simulating extreme climate events. Therefore, the resolution of the GCMs does not entirely reflect their simulation performance but also needs to consider the model's parameterization scheme, boundary conditions, and other relevant considerations.

The different altitudes may lead to different performances of GCMs in the YRB when simulating extreme precipitation indices. Ombadi et al. (2023) found that extreme precipitation events at high altitudes were anticipated to intensify by 15% for every Celsius degree increase in temperature, and the increase in extreme precipitation at high altitudes was much greater than average. Therefore, it is probable that there will be a greater increase in extreme precipitation in the future for Region I, located in the northeastern part of the Tibetan Plateau at higher altitudes, compared to Region VI, situated in the plains. In addition, Zhang & Zhou (2019) showed that the monsoon region experiences the most severe extreme precipitation. The YRB is located in the northern part of the East Asian sea–land monsoon region, with the western portion of the midstream and upstream of the YRB also affected by the plateau monsoon. Consequently, extreme precipitation is expected to differ in various regions.

The BCC–CSM2–MR and EC–Earth3–Veg have great spatial simulation capability, while the CMCC–ESM2 and FGOALS–g3 have strong temporal simulation capability, which is consistent with the findings of Tebaldi et al. (2015). Wu et al. (2021) concluded that the BCC–CSM2–MR has improved the global temperature and precipitation climate distributions, atmospheric radiation, and oceanic heat balance, among others, and the model has greatly enhanced the simulation ability of the annual mean precipitation climate distribution in China. Shi et al. (2022) showed that FGOALS–g3 had better simulation performance, and the innovative model boasts improved atmospheric composition resolution, as well as enhanced modeling of atmospheric physical processes and more.

The multimodel ensemble can improve future projection accuracy, and the multimodel median ensemble method can reduce the influence of a few outliers, which is a commonly used ensemble method (Srivastava et al. 2020; Moradian et al. 2023). Certain models use the same parameterization process or belong to the same institution, which leads to nonindependence and correlation between models, leading to similarity problems in the simulation results of some models (Xiao et al. 2023b). Therefore, the ensemble method based on model independence is also an issue that we need to pay attention to in our future research.

The extreme precipitation indices except CDD in the YRB would like to increase from 2021 to 2050 compared with 1981 to 2010 under the SSP2-4.5 and SSP5-8.5 scenarios, and the trend of the extreme precipitation indices is consistent with the finding of Ayugi et al. (2021). In addition, it was found that the spatial trend of extreme precipitation change amplitude is generally increasing from west to east. It is projected that extreme precipitation will become more frequent in the YRB in the future, compared to its incidence in recent years. Meanwhile, Zhu et al. (2021) also showed that the temporal distribution of precipitation in the YRB will become more heterogeneous in the future, especially under the high SSP scenario.

The increase in extreme precipitation in the YRB in the future has implications for water resources, agriculture, and infrastructure. The probability of heavy rainfall and flooding disasters increased as the annual frequency of extreme precipitation events increased (Li et al. 2023). Moreover, extreme precipitation can impact agricultural production by affecting soil moisture and causing flooding on farmland (Long et al. 2022; Fu et al. 2023). In addition, extreme precipitation can lead to changes in water and sediment (Xu et al. 2021), which can affect integrated watershed management and adversely affect infrastructure (Carnacina et al. 2019) within the basin. Therefore, it is essential to improve the forecasting and early warning capabilities of extreme climate events. Simultaneously, the management and deployment of water resources should be strengthened, ensuring their effective utilization.

In this study, we evaluate the spatiotemporal simulation capability of six extreme precipitation indices in the YRB and eight subregions using observations and 23 CMIP6 GCMs, and project the changes of extreme precipitation from 2020 to 2050 relative to 1981 to 2010 under the SSP2-4.5 and SSP5-8.5 scenarios, with the following main conclusions:

• (1)

  The overall spatial biases of the six extreme precipitation indices are 15.3% (RX1day), 21.04% (RX5day), 47.96% (R95p), 5.66d (R10mm), 18.91d (CDD), and 16.55% (SDII). RX1day, CDD, and SDII exhibit negative biases in 69.6, 99.9, and 74.5% of the total number of basin grids, respectively. Conversely, RX5day, R95p, and R10mm have positive biases in more than 70% of the total number of grids in the basin. CDD exhibits negative biases in all subregions, and SDII exhibits positive biases only in Region I, but negative biases in the other regions. In contrast, R95p and R10mm show predominantly positive biases in all subregions except Regions III and VII, while RX1day and RX5day exhibit inconsistent performance across subregions.

• (2)

  The GCMs can capture the temporal variability and spatial distribution of extreme values (RX1day and RX5day) and durations (CDD) well (the IVS less than 0.191 and the TSS greater than 0.8), while the spatiotemporal simulation of superthreshold indices (R95p and R10mm) is relatively poor (the IVS greater than 0.2 and the TSS less than 0.8) and can capture the temporal variability of the intensity index (SDII) temporal variability but cannot simulate the spatial distribution well. CMIP6 GCMs perform better in the spatiotemporal simulation in Regions III and VII. Specifically, CMCC–ESM2 and FGOALS–g3 demonstrate superior temporal variability of indices. In addition, BCC–CSM2–MR and EC–Earth3–Veg display better spatial simulation.

• (3)

  The extreme precipitation indices would increase in the YRB from 2021 to 2050 relative to 1981 to 2010 under the SSP2-4.5 scenario, except for the CDD, which decreases by 2.1d. R95p experiences a more significant increase of 30.3%. Under the SSP5-8.5 scenario, the temporal variation and spatial distribution of extreme precipitation indices are consistent with that of SSP2-4.5 scenario, but with more significant variations. The overall spatial trend of extreme precipitation change is gradually increasing from west to east, with RX1day, RX5day, and R95p locally exceeding 20% from Hekou Town to Longmen (Region Ⅳ) and below Huayuankou (Region Ⅶ). The risk of extreme precipitation may further increase in the future, especially in Regions III, IV, and VIII.

This work was supported by the Key Scientific and Technological Project of Henan Province, China (Grant No. 222102320286), Science and Technology Research and Development Program Joint Fund Project of Henan Province (Grant No. 232103810102), the Research Fund of Key Laboratory of Water Management and Water Security for Yellow River Basin, Ministry of Water Resources (under construction) (Grant No. 2023–SYSJJ–04), and National Key Research and Development Program project (Grant No. 2023YFC3006603).

The authors declare there is no conflict.

All relevant data are available from an online repository or repositories: http://data.cma.cn and https://esgf-node.llnl.gov/projects/esgf-llnl.