Mid-resolution regional terrestrial water storage simulation using WRF over Central Asia

Located in the middle part of the Asian and European continents, Central Asia (referred to as CA) covers an area of over 4,000,000 km² with complex topography and diverse climates (Qiu et al. 2022). CA is considered to be one of the driest areas in the world with extreme water shortage and a fragile ecosystem because it is far away from the oceans (Oki & Kanae 2006). Thus, the region is also among the most vulnerable and sensitive to climate change, which can have a great influence on the regional water storage and resources in CA. Both climate–ecosystem and human society are highly dependent on the distribution and variation of water resources, and over 70% of the development-related problems were caused by water shortage in CA (Severskiy 2004). Because of the extremely limited water conditions by rare precipitation and strong evapotranspiration, the water balance is particularly sensitive and vulnerable to climate change in CA. Numerous studies and reports have indicated that the water resource shortage is a very serious international water problem in CA influenced by climate change (Yao et al. 2016; Yang et al. 2017; Qiu et al. 2022).

Therefore, the water resources are closely related to regional climate change and have complex interactions in CA. During the past century, CA has experienced a significant warming period at a rate of 0.16 °C/10 yr (Chen et al. 2009), especially in the recent 30 years (Hu et al. 2017). The precipitation has also increased under the accelerated water cycle of the warmer climate, and the trend seems to be continuing in the future (Hu et al. 2017; Jiang et al. 2020; Qiu et al. 2022), which may bring more extreme climatic events in CA, along with more severe water resource conditions and more challenges to ‘The Belt and Road Initiative’.

To consider the water resource evaluation under the climate change background in CA, a comprehensive indicator that can not only directly reflect the water resource condition but is also closely connected with climate change is greatly needed for investigation in CA. The terrestrial water storage (TWS), which contains all the water content stored in the vegetation, entire soil profile, glacier, snow and underground aquifers, has been considered a better indicator for water resource assessment than the traditional single variable, e.g., soil moisture, runoff, and streamflow, in recent studies (Tapley et al. 2004; Scanlon et al. 2015; Zhang et al. 2015). Unfortunately, due to the harsh environment and expensive construction and maintenance costs, there were few ground-based TWS observations in CA. The spatial representation of the observations in CA was extremely limited because of the complex terrain, the huge mountains and the sparsely distributed sites. The detailed spatial pattern of TWS was very difficult to obtain.

In March 2002, the launch of the Gravity Recovery and Climate Experiment (GRACE) mission jointly by the US and German academicians facilitated the monitoring of the large-scale TWS directly (Swenson et al. 2003; Syed et al. 2008; Famiglietti & Rodell 2013), which can provide highly accurate monthly solutions for the Earth's gravity field at an approximately 100 km scale by a twin satellite in identical Earth orbit at a distance of ∼220 km. Then the Earth's gravity field can be solved by the change of this distance after removing the effects of nongravitational accelerations (Swenson et al. 2003). With these advantages, the GRACE-based TWS can provide a more direct, continuous, and comprehensive way to estimate TWS than ground-based observations, especially in an area such as CA. A number of related studies have show remarkable prospects and obvious effects of global/regional hydrological research and water resource assessment based on GRACE-based TWS products (Long et al. 2014, 2015; Scanlon et al. 2016, 2018; Rodell et al. 2018; Ma et al. 2021; Zheng et al. 2022; Rodell & Li 2023).

Another effective method of obtaining distributed TWS is coupled numerical regional simulation. As we know, the climate system is coupled with a variety of physical processes including microphysical, radiation, land surface, atmospheric boundary layer, and hydrological processes. Along with the rapid development of theoretical atmospheric science and computing technologies during recent years, these processes were comprehensively parameterized in global and regional coupled climate models. As an outstanding representation, the Weather Research and Forecasting (WRF) Model has been widely used in regional climate and weather simulation, land–atmosphere interaction investigation, urbanization effects’ evaluation, and hydrological processes and water resource estimation with different combinations of physical parameterizations (Cesana et al. 2017). The physical parameterizations in WRF can be divided into specific classes such as microphysics (MP), radiation, land surface model (LSM), planetary boundary layer (PBL), and cumulus scheme (CU), which can generate thousands of combinations (Cesana et al. 2017). According to previous studies, different physical parameterization combinations could contribute to considerable influences and disparities in simulation results. For example, Wang et al. (2019) made use of different radiation schemes in WRF simulation over China and found that the New Goddard and Community Atmosphere Model (CAM) schemes are suitable for longtime simulation without nesting and nudging options, and different PBL schemes showed different performances for different observation and meteorological parameters, which can have a substantial influence on land–atmosphere energy balance (Xie et al. 2012; Wang et al. 2019; Xu et al. 2019). The LSM, which connects the land surface and lower atmosphere, also plays an important role in weather forecasting (Chen & Dudhia 2001), climate prediction (Barlage et al. 2015), water cycle simulation (Li et al. 2015; Qiu et al. 2017), and regional urbanization (Wang et al. 2012).

Based on the previous studies mentioned above, water resources’ evaluation is also an important application field of the WRF model, especially in areas with scarce water resources, serious water crisis, and limited ground-based observations. Although previous studies indicated that the WRF model had good performance in areas with complex mountain terrain (Feng et al. 2015; Qiu et al. 2017; Salamanca et al. 2018; Chen et al. 2019; Constantinidou et al. 2019), uncertainties still exist in both model simulation in the water cycle and GRACE products in the area of CA. The water balance in CA is sensitive and vulnerable to climate change, and different selections and combinations of schemes in WRF may affect the regional water resource evaluation. The questions mentioned above still need more attention in water resource evaluation in CA. This paper aims to evaluate the TWS evaluation over CA from coupled climate–land surface simulation by the WRF model with different combinations of MP, CU, and PBL schemes for further inter-comparison. To give more detailed descriptions of groundwater processes, we introduced the multi-parameterizations of Noah LSM (Noah-MP) as the land surface scheme in experimental design. Compared to original Noah LSM, the greatest advantage of the Noah-MP is that it can improve simulation effects and achieve model optimization for specific regions by combining different schemes of vegetation, soil moisture, runoff, and other land surface process (Jiang et al. 2009; Niu et al. 2011; Cai et al. 2014; Gao et al. 2017; Chen et al. 2019). Here, in the experimental design, we take the combination of Noah-MP that was presented by Wang et al. (2020) in CA as the land surface scheme. For the model description, an experimental design is shown in Section 2. The results’ analyses are presented in Section 3. Finally, the conclusions are summarized and discussed in Section 4.

The GRACE-based TWS used in this study are retrieved from GRACE monthly mass grid-land datasets, which are based on the RL05 products with 1° spherical harmonics (SH) (Scanlon et al. 2015) and 0.5° mass concentration (Mascon) (Scanlon et al. 2016; Zhang et al. 2020) solutions. The GRACE-based products are obtained from the Center for Space Research (CSR) at the University of Texas at Austin, the German Research Centre for GeoForschungsZentrum (GFZ), and Jet Propulsion Laboratory (JPL) at National Aeronautics and Space Administration (NASA). These TWS products are used for the comparison and validation of WRF modeling results which intended to guide historical and future longer estimation of regional TWS in CA, which represent the TWS deviation for that month relative to the baseline average from 2004 to 2008. The RL05 products are considered more accurate than previously released GRACE products and are much less noisy than RL04 because of describing procedures applied to the data. Therefore, RL05 needs less spatial smoothing than earlier products.

In all design simulations, the Rapid Radiative Transfer Model for GCM (RRTMG) long-wave and short-wave radiation scheme and the LSM scheme Noah-MP (Niu et al. 2011) are used. To evaluate the performance of different physical schemes on WRF/Noah-MP in simulating the Earth–atmosphere interaction in CA, nine sets of experiments are conducted, namely, two PBL schemes, MYJ and YSU PBL schemes; two microphysical schemes, Thompson and WDM6 microphysical schemes; and three CUs, KF, Tiedtke (TDK), and the new Tiedtke (NewTDK). All the different physical scheme combinations are listed in Table 1.

The spatial resolution with 0.25° × 0.25° precipitation data from European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) precipitation (Hersbach et al. 2020) is used to verify the simulated precipitation.

The coupled model WRF can provide much more detailed regional information about spatial TWS distributions. Consistent with the distribution of GRACE-based products, the simulated high-value areas of TWS are still located around the Himalayas in all numerical experiments and with higher positive anomalies than GRACE-based TWS except JPL-Mascon. This could be attributed to the higher horizontal resolution of the WRF model, which can reproduce more detailed processes of water transfer and the dominant effect of terrain in lifting the water vapor and generating precipitation. The range of TWS values in most areas is between −2.0 and 2.0 cm, which is consistent with the values monitored by GRACE satellites. However, in the south of the study domain and around the Caspian Sea, there are inconsistent areas of spatial TWS where the negative anomalies are modeled by WRF while positive values are in GRACE-based products. Insufficient descriptions of the groundwater scheme in this complex topography area and shot spin-up period for the experiments are possible reasons for this phenomenon. Overall, the WRF simulation can basically reproduce the spatial characteristics of TWS in CA, especially in the eastern part of the study domain.

A box plot is a statistical chart that is often used to display data dispersion information. The dominant advantage of the box plot is that it is not affected by outlier values and can accurately and steadily depict the discrete distribution of the data. Figure 3(c) is the box plot of the TWS, the up and down box represents 75% value and 25% value, while the top line and bottom line are the maximum and minimum values. As can be seen from the figure, the reproduced TWS have larger ranges than the GRACE, either from the maximum (minimum) value or percentage values. The GRACE shows that the average value is the same as the 50th percentage value, which is a positive value. There was more water storage in CA during the period 2004–2008 compared to the baseline years’ average (2003–2009). However, the simulated average value is larger than the 50th percentage value; therefore, the simulation presents less water storage in the CA. Among the three options, the PBL has an impact on the up and down percentage values, the CUs have little effect on the 50th percentage, and the MP schemes have the least impact on the box plot.

In most atmospheric models, sub-grid scale processes that are not explicitly resolved by the model are represented by parameterization schemes. For example, cloud MP schemes are used to model the microphysical processes that govern cloud particle formation, growth, and dissipation on small scales. The effect of sub-grid scale clouds is represented by cumulus parameterization schemes. The short- and long-wave radiation schemes provide the atmospheric heating profiles and the estimation of net radiation for the ground heat budget. The surface layer schemes are used to calculate the friction velocity and exchange coefficients that enable the estimation of heat, momentum, and moisture fluxes by the LSM. Finally, the PBL schemes determine the flux profiles within the convective boundary layer and the stable layer, and thus provide atmospheric tendencies of temperature, moisture, and momentum in the entire atmospheric column (Shrivastava et al. 2015).

The main problem in climate change research over CA is the lack of observational data, which can create some limitations for precipitation validation and optimization by other statistical and machine learning techniques. Since precipitation is the most important indicator for TWS and water resource evaluation and prediction, the numeric model coupled with the data assimilation module (WRF-DA) will be used in future studies for precipitation simulation improvement once sufficient high-quality observations or satellite products are collected in CA. Also, the intense human activities in CA are another important factor that needs to be considered. Human activities have affected the appearance of the land surface, changing the roughness, reflectivity, and water–heat balance of the land surface, causing local climate changes. With the development of human society, the breadth and depth of its influence have increased, and the influence of human activities has become increasingly important. From the perspective of water resources, the most important human activity in CA is water consumption for agricultural irrigation, which was not considered in the numerical experiments yet in this study. In the next step, the water use module and irrigation scheme in the LSM need to be activated to not only obtain the human-induced TWS results but also to quantify the impact of human activities on TWS and regional water resources. In the past 100 years, the temperature in the arid region of CA has increased significantly, with a magnitude of 1.6 °C, which is much higher than the warming in the northern hemisphere. With the intensified global warming in the future, the temperature increase in the arid region of CA will generally be higher than the global average, likely leading to the warming of the core area of Eurasia. Another important factor for regional TWS is lake water storage variation (Ma et al. 2017), which can be solved by introducing the dynamical land use/land cover change with temporal variation and activating the lake module in the land surface scheme to enhance the description of lake effects in numerical experiments. So, future work of TWS in CA needs to clarify the contributions of natural forcing, internal variation, different human activities, and land use/land cover change.

Nine sets of simulation experiments using WRF v4.0.1 at a horizontal resolution of 25 km during 2004–2008 were carried out in CA. The dominant study variables in this study are TWS and precipitation, which are evaluated based on GRACE products and ERA5 data. In comparison with the GRACE data, all the simulations basically capture the spatial features of TWS in CA but with more detailed regional information, especially in the east part of the study domain. From the regional time series of the TWS, the WRF reproduces a downward trend in CA, but the rate of decline is relatively small due to the lack of consideration of human activities such as irrigation. The box plot shows the simulated TWS has larger variability compared to the GRACE. The different physical options and combinations present less difference and small variances in the TWS simulation. This situation may be attributed to the relatively short simulation period, and thus, longer simulation periods and finer spatiotemporal resolution are required for further research. Using the TWS products provided by both GRACE and GRACE-FO missions over longer spatiotemporal scales to achieve more statistically significant and more reliable simulation results will be the key focus of the future work. Conversely, the results indicate that the TWS in CA may be insensitive to WRF physical scheme combinations, although the model can capture the regional water balance stably. In the future, we will pay more attention to the evaluation of different physical options for the water cycle components, e.g., the precipitation, the soil water evaporation, and moisture variation, and try to consider the effects of human activities in detail, lake processes, and the utilization of data assimilation modules on regional TWS investigations in CA.

This study was jointly funded by the Second Tibetan Plateau Scientific Expedition and Research Program (STEP) (No. 2019QZKK0903-02), the International Partnership Program of the Chinese Academy of Sciences (No. 060GJHZ2022057MI), the Natural Science Foundation of China (No. 42375176), and the Iran National Science Foundation (INSF, No. 4006075). We also thank for the technical support of the National Large Scientific and Technological Infrastructure “Earth System Numerical Simulation Facility” (https://cstr.cn/31134.02.EL, EarthLab, No. 2023-EL-PT-000205). The datasets used in this study are derived from European Centre for Medium-Range Weather Forecasts (https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5), National Centers for Environmental Prediction (https://rda.ucar.edu/datasets/ds083.2/), and the GRACE data centers of JPL and CSR (https://grace.jpl.nasa.gov/data/get-data/monthly-mass-grids-land/, https://grace.jpl.nasa.gov/data/get-data/jpl_global_mascons/, and http://www2.csr.utexas.edu/grace/RL06_mascons.html). In addition, we also thank the anonymous reviewers for their detailed and helpful comments.

All relevant data in this study are available from an online repository or repositories. Due to the large volume of the numerical model results, we uploaded one group of model results and all the GRACE-based TWS products. The data link is: https://pan.baidu.com/s/1v90DNZ2Znu2dO57ohCSXHQ, and the pw code is: dfxx. If the reader has further in-depth interest in the datasets mentioned in this article, please feel free to contact the corresponding author Ziyan Zheng ([email protected]) to obtain the complete datasets.

The authors declare there is no conflict.