Impact of cumulus parameterization schemes on summer extreme precipitation simulation in the Yellow River Basin: the 2018 case

According to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, extreme weather and climate occurrences have reportedly become more frequent and intense in recent years (Huang et al. 2021; Mo et al. 2022; Yang et al. 2023). Precipitation, particularly extreme precipitation, has previously been demonstrated to be very sensitive to weather and climate change (Bao et al. 2015). Changes in the frequency and intensity of extreme precipitation usually result in extreme hydrological events, which can have detrimental effects on the societal environment and the natural environment (Gimhan et al. 2022; Zhao et al. 2023). Researchers have conducted studies on precipitation over numerous locations throughout the world, particularly summer extreme precipitation (Tian et al. 2021; Bieniek et al. 2022). Previous studies concluded that summer extreme precipitation occurrences, particularly in complex topographic regions (Newman et al. 2021; Pervin & Gan 2021), have increased in frequency and intensity during recent years (Tramblay & Somot 2018; Akinyemi & Abiodun 2019). For example, Shang et al. (2020) observed that extreme precipitation events occur much more frequently in summer in central-eastern China. Nguyen et al. (2020) found that the flow and convergence of air and moisture from the Swiss Plateau to the Swiss Alps is very prone to the initiation of extreme precipitation events due to the complex orography of the Bernese Alps. Michel et al. (2021) pointed out that 78.5% of the daily extreme precipitation events in southwestern Norway were linked to atmospheric rivers during the period 1979–2018, and this percentage decreased to 59% in the more northern coastal regions and 40% in the inland regions. To disclose more specific characteristics of summer extreme precipitation episodes at a smaller spatial scale, summer extreme precipitation should be studied from a watershed viewpoint.

One of the most recent regional climate models is the Weather Research and Forecasting (WRF) model (Skamarock et al. 2008), which is frequently used in simulations of regional precipitation. Sun & Zhao (2003) considered that the distribution and movement of precipitation is simulated by WRF successfully. Qiu et al. (2017) used the WRF model to downscale the European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA)-Interim data and showed that the precipitation data, which were optimized by the WRF model, showed more accuracy in some sub-regions of Central Asia. Patel et al. (2019) used the WRF model to generate fine-scale rainfall estimates in a coastal urban environment and provided evidence that WRF simulations have shown better precipitation estimates than the Global Forecasting System forecasts over Mumbai, India. Qiu et al. (2020) investigated the long-term climate simulations of very high resolution in South Korea using the WRF model and concluded that WRF simulations show reasonable performance in capturing the general characteristics of summer extreme temperature and precipitation and show good agreement with observation. Hamouda & Pasquero (2021) used the WRF model to estimate high-resolution precipitation extremes. They explicitly revealed that WRF simulations succeed in correcting the failure of ERA-Interim reanalysis to capture the positive trends over recent decades of European extreme precipitation in summer and transition seasons. The WRF model has been validated to have a certain ability in simulating precipitation in China. Yu et al. (2010) described a precipitation simulation over China using the WRF model, which showed that the improvement in precipitation simulation is perceptible with the WRF model.

The WRF model offers a variety of parameterization schemes. The appropriate combination of schemes will allow for a more accurate simulation of the meteorological condition and the physical mechanism that leads to precipitation events. Studies have shown that different parameterization schemes, particularly cumulus parameterization schemes (CPSs), have a significant impact on precipitation simulation. Jankov et al. (2005) noted that changes in CPSs notably impacted the WRF simulation of the system average rain rate in eight international H₂O Project cases. Li et al. (2014) found that the WRF simulations for summer precipitation in the southeastern United States are most sensitive to CPSs. Remesan et al. (2015) pointed out that a better choice of CPSs was good at predicting rare, high-intensity events over the Yorkshire region.

Previous studies have not reached a consensus on how to select CPSs for simulating precipitation in a region with complex orography. Qiu et al. (2020) found that the Kain–Fritsch (KF) scheme showed reasonable performance in capturing the general characteristics of summer precipitation in Korea, while Lee et al. (2017) pointed out that the temporal and spatial patterns of observed precipitation fields were well reconstructed using the WRF model simulated with the Betts–Miller–Janjic (BMJ) scheme in Korea. In addition, the majority of current studies on extreme precipitation events based on the WRF model have neglected to analyze factors that may have an impact on the performance of the WRF model in favor of assessing the sensitivity of regional extreme precipitation events to various parameterization schemes.

Owing to the impact of global climate change, summer extreme precipitation events in the YRB have occurred frequently in recent years. The average precipitation of the YRB in the summer of 2018 was 41.5% more than that of the average, ranking eighth since 1961 and second since 2000. During the summer of 2018, two floods occurred in the Upper Yellow River Basin (UYRB) and the Weihe River (the largest tributary of the Yellow River). Several tributaries between Lanzhou and Tuoketuo had the largest flow since the hydrological station was constructed. The Tongguan station had the largest flood since the flood season came in the Middle Yellow River Basin (MYRB) (Liu et al. 2022). The examination of this event can contribute to the understanding and management of YRB flooding in the context of climate change. The YRB is characterized by fewer meteorological stations with uneven spatial distribution. Reanalysis datasets have been widely used in the study of areas where data availability is limited. However, the precipitation data of reanalysis datasets might not accurately represent the complex regional-scale processes and phenomena occurring in the YRB due to their relatively low spatial resolution and oversimplified physical parameterization. Obtaining precipitation data with high spatial resolution poses a challenge, so downscaling the precipitation data of reanalysis datasets is necessary to improve their accuracy and spatial resolution.

The supply and transport of moisture have a direct impact on precipitation variation (Wu et al. 2016). Huang et al. (2022) pointed out that such an increase in summer precipitation was due to significant changes in moisture contribution from external moisture sources. Investigating the relationship between the occurrence process of precipitation and moisture transport can enhance forecasting and warning systems for summer extreme precipitation, thereby reducing human and property damage. Precipitation simulations may benefit from understanding the moisture source and the transport process (Wang et al. 2018). The response pattern of extreme summer precipitation to moisture transport in the complex watershed, therefore, must be discussed. While previous studies have focused mainly on the source area of the YRB (Meng et al. 2016; Huang et al. 2022), there is a lack of research on the entire YRB.

As a case study, the summer (June, July, and August) extreme precipitation event in the YRB during 2018 was selected. The scientific goals of this study were twofold: (1) assess the capability of CPSs in the WRF model to simulate summer extreme precipitation in the YRB and (2) explore the atmospheric circulation background of this case and the factors contributing to discrepancies between WRF-simulated precipitation and observed precipitation. The organization of this study is as follows: Section 2 describes the model setup, data used for model simulation, and a detailed overview of the study area. Section 3 focuses on comparing WRF-simulated precipitation with the stations' observations. Section 4 delves into the comparisons between WRF-simulated moisture and the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP/NCAR) reanalysis data. Finally, concluding remarks are made in Section 5.

Therefore, conducting reasonable simulations of precipitation is helpful for more targeted water resource management in the YRB to cope with increasing climate extremes.

To investigate the effects of different CPSs on the summer extreme precipitation simulation over the YRB, the regional climate model used in this study was the Advanced Research WRF model with version 3.9.1. With this aim, the control experiments were executed using five CPSs (as shown in Table 1) in this study, consisting of KF, BMJ, Grell–Devenyi (GD), Grell Three-Dimensional (G-3D), and the Simplified Arakawa–Schubert (SAS). The main characteristics of each CPS are compared in Table 1.

The same scheme setup, except the CPS, is applied to all experiments. Other main schemes were considered: the Thompson scheme for microphysics, the Rapid Radiative Transfer Model scheme for General circulation models (RRTMG) for both short-wave and long-wave radiation, the Noah Land Surface Model for land surface, the revised fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) scheme for the surface layer, and the Yonsei University scheme for planetary boundary layer parameterization. This combination of parameterization schemes shows good applicability in the simulation of precipitation in China (Yang et al. 2021).

The initial and lateral boundary conditions were provided by the National Centers for Environmental Prediction (NCEP) Global Final Analysis (FNL) with a horizontal resolution of and a time interval of 6 h. Ensemble simulations of three months were initialized at 0000 UTC 1 June 2018 and integrated continuously to 0000 UTC 1 September 2018. The WRF model was centered at 31.5°N/114.7°E and configured with one-way double-nested domains (as shown in Figure 1(a)). The outer domain (D01) had a horizontal grid spacing of 90 km, which covered most mainland of China and extended to the oceans. D01 provided the large-scale circulation background for the inner domain (D02) and reduced the influence of lateral boundary conditions on D02 simulation. The inner domain (D02) had horizontal grid spacing of 30 km, which covered the whole YRB, and D02 was the key region where sensitivity simulations were performed and analyzed in detail. Yu (2019) found that WRF can capture the surface geographic variation much better with 30 km horizontal resolution, particularly in areas of complex orography.

There are 114 meteorological stations operated by the China Meteorological Administration that are publicly available in the study domain (D02, as shown in Figure 1(b)). Daily records of average precipitation from June to August 2018 at these stations were obtained from the China Meteorological Data website (http://data.cma.cn) and used as the observed data in this study.

The observed precipitation amount (as shown in Figure 2(a)) showed a spatial pattern that the summer precipitation in the YRB during 2018 increased from northwest to southeast. The center of precipitation was located in the middle of the Lower YRB (LYRB), and the precipitation amount exceeded 10 mm. The WRF-simulated precipitation amount (as shown in Figure 2(b)–2(f)) indicated that all the CPSs were capable of simulating the approximate direction of the rain bands in the YRB during the summer of 2018. However, there were large deviations between the WRF-simulated precipitation amount and the observed precipitation amount in the value of precipitation amount and the center of precipitation. The KF scheme (as shown in Figure 2(b)) had two centers of precipitation, with the upstream center of precipitation located in the Heihe River basin and the midstream center of precipitation located in Hekou town (as shown approximately in Figure 1(b)). Compared with the observed precipitation amount, the value of the KF-simulated precipitation amount in the UYRB and the MYRB was slightly larger, while that in the LYRB was the same. Compared with the observed precipitation amount, the BMJ scheme's center of precipitation (as shown in Figure 2(c)) was the same, and the value of the BMJ-simulated precipitation amount in the MYRB was slightly larger, while that in the UYRB and LYRB was the same. The GD scheme (as shown in Figure 2(d)) had two centers of precipitation, with the upstream center of precipitation located in the Taohe River basin and the midstream center of precipitation located in Longmen town. Compared with the observed precipitation amount, the value of the GD-simulated precipitation amount in the UYRB and the MYRB was slightly larger, while that in the LYRB was the same. The SAS scheme (as shown in Figure 2(e)) had two centers of precipitation, with the upstream center of precipitation located in the Taohe River basin and the midstream center of precipitation located in Hekou town. Compared with the observed precipitation amount, the value of the SAS-simulated precipitation amount in the UYRB and the MYRB was larger, while that in the LYRB was the same. The G-3D scheme (as shown in Figure 2(f)) had two centers of precipitation, with the upstream center of precipitation located in the Heihe River basin and the downstream center of precipitation located in the Yiluo River basin. Compared with the observed precipitation amount, the value of the G-3D-simulated precipitation amount in the MYRB was slightly larger, while that in the UYRB and the LYRB was the same.

The possible reason for this deviation was that the WRF model is limited by grid resolution and struggles to fully replicate the actual orography and landforms. The YRB has complex orography, with higher elevations in the northwest and lower elevations in the southeast. Additionally, the WRF model tends to overemphasize the uplift effect of orography on a small–medium scale (Zhang & Duan 2021), which may result in the southeast (located in the plain at the foot of the mountain) experiencing regional precipitation due to the blocking and uplifting of moisture by the mountain.

The evaluation results of TS, BS, FAR, and MAR are shown in the graded evaluation diagram (as shown in Figure 3(b)). The daily mean precipitation amount is classified into three classes in this paper: light precipitation P_L (4 mm > P_L > 0 mm); moderate precipitation P_M (7 mm > P_M ≥ 4 mm); heavy precipitation P_H (P_H ≥ 7 mm). The MAR of all five CPSs was less than 0.1 in the light precipitation class. The MAR of the KF, BMJ, SAS, and G-3D schemes was less than 0.1 in the moderate precipitation class, with the MAR of the GD scheme being the largest at 0.19. The MAR of the KF, BMJ, and G-3D schemes was close to 0 at the heavy precipitation class, with the GD scheme being the largest at 0.4. The FAR of all five CPSs was less than 0.4 in the light precipitation class, with the FAR of the BMJ scheme being the smallest at 0.2. The FAR of all five CPSs was less than 0.6 in the moderate precipitation class, with the FAR of the BMJ scheme being the smallest at 0.42. The FAR of the GD, SAS, and G-3D schemes was around 0.8 at the heavy precipitation class, with the FAR of the KF scheme being the smallest at 0.6. Except for the G-3D scheme, the TS of the other four CPSs was larger than 0.7 in the light precipitation class, with the TS of the BMJ scheme being the largest at 0.81. Except for the GD scheme, the TS of the other four CPSs was larger than 0.4 in the moderate precipitation class, with the TS of the BMJ scheme being the smallest at 0.56. The TS of the GD, SAS, and G-3D schemes was larger than 0.1 in the heavy precipitation class, with the TS of the KF scheme being the smallest at 0.38. Except for the BMJ scheme, the BS of the other four CPSs was larger than 1.25 in the light precipitation class, with the BS of the BMJ scheme being the smallest at 1.2. In the moderate precipitation class, all five CPSs had BS values that were comparable to those in the light precipitation class. Except for the BMJ scheme, the BS of the other four CPSs was larger than 1.67 in the moderate precipitation class, with the BS of the BMJ scheme being the smallest at 1.62. The BS of all five CPSs was larger than 2.5 in the heavy precipitation class, with the BS of the KF scheme being the smallest at 2.73. The simulation of light and moderate precipitation was in line with expectations, but the simulation of heavy precipitation fell short of expectations. The possible reason for this issue is representativeness errors (Jiménez & Dudhia 2012), which leads to suboptimal simulation results. In summary, by comparing the simulation performance of the five CPSs in terms of spatial distribution and evaluation methods, it was consistently concluded that the BMJ scheme had the best capability in simulating the daily mean precipitation amount in the YRB during the summer of 2018.

In Figure 4(b), the 500 hPa geopotential height field displayed a ‘ + −+ ’ pattern in the mid–high latitudes of Eurasia (Tan & Sun 2004). A low-pressure trough formed over the regions of central and west Siberia (60°N–70°N, 90°E–120°E), lake Baikal (51°N–55°N, 103°E–110°E), and Xinjiang (35°N–50°N, 75°E–95°E). Polar cold air continuously flowed southward along this trough, entering the YRB. The Summer Western North Pacific Subtropical High (SWNPSH) exhibited a larger extent and intensity in 2018. The typical ridge position of the SWNPSH (shown by the white dashed line in Figure 4(b)) is usually near 27°N. However, the ridge position of the SWNPSH in 2018 (shown by the solid white line in Figure 4(b)) was notably located north of 30°N, with its western boundary extending further west compared with previous years. The westward and northward extension of the SWNPSH facilitated the northward transport of warm and moist air from the Pacific Ocean. This air converged with cold air from mid–high latitudes, thereby resulting in an excess of extreme precipitation in the YRB.

The result of the precipitation simulation from Section 3.1 showed that the WRF model had overestimated the precipitation amount in the 2018 case. Widespread, continuous precipitation requires a constant flow of moisture. Therefore, this study further investigated the impact of the WRF model on precipitation simulation from the perspective of WRF-simulated moisture.

In Figure 6(a), the moisture transport showed strong meridional characteristics in the YRB during the summer of 2018. The moisture transport paths were mainly manifested as three branches: southwest moisture came from the Bay of Bengal–Central south Peninsula, southeast moisture came from the western Pacific Ocean, and northwest moisture came from Siberia–Mongolia. The three branches of moisture flowed in from the western, northern, and southern parts of the YRB, respectively, and converged in the LYRB before flowing out from the eastern part of the YRB. Compared with the southern moisture transport, the northwest one was significantly smaller, which was consistent with the conclusion of the previous study (Li et al. 2012). As shown in Figure 6(b)–6(f), all five CPSs showed fine capability in simulating the transport process over the YRB during the summer of 2018. The moisture transport characteristics simulated by the WRF model exhibit similarities to the moisture transport characteristics estimated by NCEP-R1 data in terms of transport direction.

The simulation results of all five CPSs were larger than the NCEP-R1 data in terms of moisture magnitude, with the BMJ scheme being the closest to the NCEP-R1 data. This implied that more WRF-simulated moisture flows through the YRB, which was consistent with the conclusion in Section 3.1 that the daily mean precipitation amount simulated by the WRF model was larger than the daily mean precipitation amount observed at meteorological stations.

In Figure 7(a), the moisture was dominated by convergence in the southwestern part of the YRB and the LYRB during the summer of 2018, while that in the rest of the YRB showed weak divergence. This was consistent with the spatial distribution of the observed precipitation amount (as shown in Figure 2(a)). In Figure 7(b)–7(f), the simulation results of moisture flux divergence were similar for all CPSs. All CPSs could simulate the area of moisture convergence well, but all the intensity of moisture divergence simulated by the WRF model was larger than that estimated by NCEP-R1 data. This was consistent with the previous conclusion that the simulation result of all five CPSs is larger than the estimation result of the NCEP-R1 data in terms of moisture magnitude (as shown in Figure 6).

There was a significant difference between all CPSs and the NCEP-R1 data in the UYRB. Specifically, all CPSs exhibited convergence, while the NCEP-R1 data exhibited divergence. A possible reason for this issue is that the NCEP data lack details at the center of the YRB (Xu et al. 2021), and the impact of orography on moisture may therefore not be accurately reflected. Additionally, the WRF model was highly sensitive to orography, as shown in Figure 5, which showed the significant orography variation in the UYRB. The obstruction of moisture by high mountains in the northwest region may have led to this issue.

Since the moisture transport situation in the YRB during the summer of 2018 varies greatly, to further understand the dynamic water resource characteristics, the moisture budget at each boundary of the YRB during 2018 was calculated in this study, and the estimations based on the NCEP-R1 data and the simulation result of the different CPSs are shown in Table 3.

It can be seen from the estimation result of the NCEP-R1 data that the moisture flowed in from the western, southern, and northern boundaries and out from the eastern boundary in the YRB during the summer of 2018. This was consistent with the moisture transport conclusion (as shown in Figure 6(a)). The total moisture budget was positive and showed a surplus of moisture. The simulation result for all CPSs indicated that the moisture budget was larger than the estimation result based on NCEP-R1 data at the western and southern boundaries, while it was similar at the northern and eastern boundaries.

This was consistent with the results of the moisture flux (as shown in Figure 6). The CC between the precipitation amount and the total moisture budget (Q_T) in the YRB during the summer of 2018 was 0.57 at a 95% confidence level, implying that the increase of input moisture may be the main reason for the increase in summer precipitation. The CCs between the precipitation amount and the moisture budgets at the four boundaries (Q_W, Q_E, Q_N, and Q_S) were, respectively, 0.63, 0.28, −0.03, and 0.19, of which only the correlation for the western boundary passed the 95% confidence level. This means that the moisture input from the western boundary may be the main reason for the extreme precipitation in the YRB during the summer of 2018.

This study investigated the effects of various CPSs on summer extreme precipitation events and their response to moisture transport in a complex watershed using the WRF model. To this end, five CPSs (KF, BMJ, GD, SAS, and G-3D) at a resolution of 30 km were applied to simulate the summer (June, July, and August) extreme precipitation event in the YRB during 2018. The findings of our analysis are summarized as follows:

• (1)

  All of the tested CPSs were capable of simulating the approximate direction of the rain bands in the YRB during the summer of 2018. However, there were significant differences between WRF-simulated and observed daily mean precipitation in the region of highest precipitation, with the WRF-simulated values tending to be larger than the observed values. Upon further evaluation using seven different methods, it was determined that the BMJ scheme demonstrated the best performance in simulating that summer's extreme precipitation event.

• (2)

  The complex orography of the YRB has a significant influence on moisture transport. During the summer of 2018, moisture fluxes were higher than the annual average in some areas. The summer moisture transport in the YRB during 2018 exhibited strong meridional characteristics, with moisture flowing in from the western, northern, and southern parts of the basin and converging in the LYRB before being transported out of the eastern part of the basin. The input of moisture from the western boundary may be a major contributing factor to the extreme summer precipitation events observed in the YRB during 2018.

• (3)

  During the summer of 2018, moisture convergence dominated the southwestern part of the YRB and the LYRB, while the rest of the region exhibited weak divergence. While all of the tested CPSs could accurately simulate the areas of moisture convergence, the intensity of WRF-simulated moisture divergence was higher than observed values according to NCEP-R1 data. It was suggested that this overestimation of simulated moisture by the CPSs in the WRF model may be responsible for the overestimation of simulated precipitation.

The overestimation issue of this study could be attributed to the limitations of the used parameterization schemes, meteorological data, and initial and lateral boundary conditions. There are numerous choices of parameterization schemes for the WRF model, such as microphysical, radiation, and boundary layer. These choices allow users to optimize the model for specific geographies. In addition to CPS, other parameterization schemes also have an impact on precipitation simulation. Therefore, combining different parameterization schemes to further improve the precision of precipitation simulation in the YRB is a direction for future research. Furthermore, the initial and lateral boundary conditions also play an important role in the precision of precipitation simulation. It is suggested that future research could optimize the initial and lateral boundary conditions through radar data assimilation to further improve the precision of precipitation simulation.

To conclude, this study assessed the ability of the WRF model to simulate summer extreme precipitation events in a complex watershed. In addition, we aimed to explore the sensitivity of extreme precipitation simulation to CPSs in the YRB during the summer of 2018 and the atmospheric circulation background of this case. Furthermore, the factors, contributing to discrepancies between WRF-simulated precipitation and observed precipitation, were also analyzed.

B.M. and S.C. were responsible for the initial project proposal. S.C. and X.L. were responsible for the model configuration and data preparation. S.C. performed the analysis and drafted the manuscript. B.M., S.C., and X.L. modified and improved the manuscript. All authors read and approved the manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.