1. Introduction
Dynamical downscaling is used to produce high-resolution regional or local climatological datasets based on regional climate models (RCMs) with lateral and initial boundary conditions from global climate models (GCMs) or reanalysis (Omrani et al. 2013; Lucas-Picher et al. 2016; Hu et al. 2018; Komurcu et al. 2018). Dynamical downscaling can capture microscale and mesoscale details through topographical feature and coastline representation by RCMs (Laprise 2014; Giorgi and Gutowski 2015; Komurcu et al. 2018; Liu et al. 2018). High-resolution regional climatological datasets based on dynamical downscaling can be applied in climate resource assessments, hydrological applications, large project site selection, climatic feasibility, urban planning, and so on (Xue et al. 2014).
Physical-based dynamical downscaling can potentially capture significant changes in both climate means and extremes. However, RCM results usually accumulate errors from the physical process and boundary conditions of GCMs. Long-term continuous simulation usually causes simulation drift away from large-scale driving data or outflow boundaries and increases the mismatch between RCMs and driving data (Alexandru et al. 2009). In the past two decades, some RCMs adding a forcing term in the prognostic equation (Newtonian relaxation, also called nudging), which is based on the difference between the global climate model’s realization of a prognostic field and the corresponding reference state, can avoid the divergence problem with continued model integration (Lo et al. 2008; Alexandru et al. 2009; Gula and Peltier 2012; Spero et al. 2018).
The Weather Research and Forecasting (WRF) Model has been widely used as a regional climate model to achieve downscaling analysis (Lo et al. 2008; Otte 2008; Alexandru et al. 2009; Bowden et al. 2012; Wootten et al. 2016; Spero et al. 2018). Currently, the nudging methods in the WRF Model mainly include analysis nudging (AN; e.g., grid nudging) and spectral nudging. AN relaxes the model toward the reference states that are usually obtained from coarse-resolution model analysis (Stauffer and Seaman 1990) by nudging the variables such as horizontal wind components, potential temperature, and the water vapor mixing ratio. Spectral nudging in WRF involves a Fourier decomposition of the difference between modeled and reference states and only the selected wavenumbers are kept in the nudging term, and the nudging variables include horizontal wind components, potential temperature, and geopotential as well as moisture with the extension from Spero et al. (2014).
Both AN and spectral nudging are widely used for dynamical downscaling. Spectral nudging shows a better performance for the internal divergence of surface variables than AN; however, many works have shown that AN in WRF can also improve the accuracy of the downscaled fields (Otte 2008; Bowden et al. 2012, 2013; Bullock et al. 2014). Bowden et al. (2012, 2013) showed that analysis nudging in WRF can improve the accuracy of the simulated climate and did not squelch the temperature and precipitation extremes. The results of Salathé et al. (2008) showed that the application of grid nudging could prevent large-scale drift from driving fields and allow the RCM to capture mesoscale details over fine domains. Lo et al. (2008) used AN and reinitialization in the WRF Model and found that both methods could improve the accuracy of the downscaled fields. Ma et al. (2016) performed downscaling experiments with nudging over continental China, and the results showed that the simulation of air humidity and wind speed improved with AN.
Other than AN and spectral nudging, observation nudging, which nudges model simulation toward the observations, has recently drawn some attention. Since observations are irregularly distributed, observation nudging adjusts the model variables within a given radius of influence near the observation locations and within a given time window surrounding the observations (Deng et al. 2009). Both low-density conventional surface observations and multilevel upper-air observations, such as radiosondes, wind profilers, and radiometers, have been used in recent studies. Deng et al. (2009) found that the use of both AN and observation nudging resulted in improved simulations of wind, temperature, and moisture compared to the use of only observation nudging or AN. Zhou and He (2018) used AN and observation nudging for complex terrain, and the results showed that the surface wind field was significantly improved. Recently, studies have shown that using observation nudging to assimilate surface observations in the nest grid domain and AN to assimilate GCM analysis data can reduce surface temperature and wind speed errors better than traditional dynamical downscaling methods (Jiang et al. 2018; Yi et al. 2018b).
Driven by large demands to improve severe weather monitoring and forecasting, many regions throughout the world have implemented high-density automated surface station networks. Liaoning Province, China, is one of these regions possessing such a surface network. Starting in 2005, the Liaoning Meteorological Bureau gradually installed 1552 intensive automatic weather stations (IAWSs) across 148 000 km2 of land area, in addition to the existing 61 conventional weather stations (CWSs). Most IAWSs observe precipitation and temperature, approximately 70% of IAWSs observe the four surface fields, including precipitation, temperature, wind speed, and wind direction, and approximately 10% also monitor pressure and humidity. While the high-density surface networks could potentially add value to high-resolution dynamical downscaling analysis in the regions where they are available, their impact has never been evaluated. In this study, we assess the role of using IAWS surface variables in regional climate downscaling using both AN and observation nudging with a set of experiments based on WRF. We will focus our verification on surface variables against measurements from the 61 conventional weather stations that are not used in the nudging experiments, including precipitation that was not evaluated by previous observation nudging studies. This paper is organized as follows. In section 2, the model setup, experimental designs, and evaluation methods used for the experiments are briefly described. In section 3, the model performance is quantitatively evaluated for temperature, precipitation, wind, relative humidity, and pressure. Possible reasons for large-scale and small-scale structures with the application of AN and combined analysis nudging and observation nudging with surface observations (AON) compared with the Climate Forecast System, version 2 (CFSv2), analysis datasets are also discussed in section 3. A summary and discussion of the downscaling strategy with surface observations are presented in section 4. In addition, the abbreviations utilized in this paper and their definitions are listed in Table 1.
Abbreviations and definitions in this study.
2. Model configuration and experimental setup
a. Model configuration
The WRF Model has been developed as a research and operational numerical weather prediction model, and it also showed a good ability to describe climate simulations in different regions (Lo et al. 2008; Leung and Qian 2009; Zhang et al. 2009; Soares et al. 2012; Glisan et al. 2013). In this study, WRF Model, version 3.5.1, is used for dynamic downscaling. The simulation domain is centered at 42°N and 116°E, and a two-way interactive nest with horizontal resolutions of 36 km (160 × 150 grid points), 12 km (184 × 220 grid points), and 4 km (169 × 169 grid points), which covers most of China, Mongolia, and parts of Russia, is utilized (Fig. 1). The downscaling simulation was run with 51 terrain-following eta vertical layers that extended up to 10 hPa, with approximately 15 layers in the lowest 1.5 km of the atmosphere. The initial conditions (ICs) and boundary conditions (BCs) for the large-scale atmospheric fields, sea surface temperature (SST), and initial soil parameters (soil water, moisture, and temperature) are driven by the NCEP CFSv2 6-hourly (0000, 0600, 1200, and 1800 UTC) analysis data at 0.205° × 0.204° for the surface field and 0.5° × 0.5° for the upper layers (Saha et al. 2011). The CFSv2 is the extended CFSR and incorporates a number of new physical packages for cloud–aerosol–radiation, land surface, ocean, and sea ice processes, as well as a new atmosphere–ocean–land data assimilation system (Saha et al. 2010), and the CFSv2 has a higher resolution than NCEP Final (FNL) Operational Global Analysis data and a good ability to describe the diurnal structures of large-scale circulation (Chen et al. 2014).
Model domains (d01, d02, and d03) used in the WRF Model.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
The chosen parameters are based on the work of Ma’s dynamic downscaling over China (Ma et al. 2016). The main physical options used here include the WRF single-moment 3-class microphysics scheme (WSM3; Hong et al. 2004), Rapid Radiative Transfer Model longwave radiation (RRTM; Mlawer et al. 1997), Dudhia shortwave radiation scheme (Dudhia 1989), the Yonsei University planetary boundary layer scheme (YSU; Hong et al. 2006), and the Noah land surface model (Chen and Dudhia 2001). Because the convective processes are not resolved at 1–5 km (Hong and Dudhia 2012), it is also referred to as the “gray zone,” and the Kain–Fritsch cumulus parameterization (Kain 2004) is only used in the first and second domains. Four monthly sensitivity experiments on microphysical parameterization schemes, including the WSM3 and Thompson scheme (Thompson et al. 2008) and the planetary boundary layer of YSU and Mellor–Yamada–Janjić (MYJ; Janjić 1994), were also conducted over Liaoning. The results show that the surface temperature and wind with YSU and WSM3 are better than those with other combinations, and the surface moisture of WSM3 is better than that of the Thompson scheme but not sensitive to the boundary layer schemes. Although precipitation with the Thompson scheme is better than that with WSM3, here, we not only focus on precipitation but also consider the accuracies of other surface fields; therefore, the YSU and WSM3 simulations are applied for this simulation. To provide a more accurate and dynamic surface field, data such as the SST, green vegetation fraction, and albedo from CFSv2 are updated every 6 h during the simulations.
b. Experimental design
Liaoning is located in northeastern China, and this area is affected by the East Asian monsoon climate; precipitation mainly occurs in July and August. The annual average precipitation is approximately 700 mm, and there is a large spatial difference between east and west due to the influence of topography. Three 1-yr continuous integration simulations for 2015 are carried out to investigate the downscaling strategy with analysis and observation nudging, and the CWS observations are used to evaluate the performance of nudging dense surface observations for improving the surface fields. The control experiment (referred to as CTL) is a continuous 1-yr integration without analysis or observation nudging. Sensitivity experiments using AN in the 4-km domain are first tested and show significant underestimation in the precipitation simulation (Yi et al. 2018a), which is consistent with the result of Wootten et al. (2016). AN uses analysis nudging at spatial resolutions of 36- and 12-km grids. AON is similar to AN, but the difference involves using observation nudging at a 4-km grid resolution. IAWS surface variables nudged in downscaling, and nudging variables include surface temperature, wind, and relative humidity. The three continuous integral experiments started at 0000 UTC 28 December 2014 and ended at 0000 UTC 1 January 2016, and the beginning of the 96 h of integration was regarded as the model spinup. The hourly output of 0000 UTC 1 January 2015 to 2300 UTC 31 December 2015 is used for the downscaling analysis. The output of the WRF Model every 1 h is employed for the model evaluation.
Similar to the work of Deng et al. (2009) and Li et al. (2016) for AN, horizontal wind is applied to all layers, and the potential temperature and water vapor mixing ratio are applied above the PBL. Nudging above the PBL is advantageous because it allows WRF freedom to develop and respond to mesoscale forcing while simultaneously being constrained to the large-scale atmospheric circulation in the free atmosphere. Observation nudging is applied to the surface horizontal wind, relative humidity, and air temperature. The nudging surface field can help the model increase some small-scale information at the surface layer.
Studies have shown that AN is sensitive to the strength of the nudging coefficients, and inappropriate use can also cause detrimental effects (Spero et al. 2018). With nudging that is too strong, downscaling will result in the loss of some mesoscale features and reduce intervariability (Bowden et al. 2012; Spero et al. 2018); otherwise, it will allow some phase and amplitude errors to grow. Omrani et al. (2013) suggested that the optimal nudging coefficients may also relate to the model configuration, and the nudging coefficient needs to be carefully defined (Deng et al. 2009). Otte (2008) and Spero et al. (2018) suggest that downscaling the simulation with default coefficients in WRF can make better improvements than the GCM analysis datasets. Because the nudging strategy has a greater impact than the nudging coefficients on downscaling, in this work, the effect of coefficients on downscaling is beyond the scope of this study. The default AN coefficient Gα = 3 × 10−4 s−1 with all variables in AN (Stauffer et al. 1985; Stauffer and Seaman 1990) and the observation nudging coefficient Gα = 6 × 10−4 s−1 come from the recommended values of the WRF User’s Guide (Wang and Kotamarthi 2013) and a previous study by Sommerfeld et al. (2019).
c. Datasets and evaluation methods
The surface observation stations include 1552 IAWS stations and 61 basic CWS stations for Liaoning Province of China in 2015 provided by the Liaoning Meteorological Bureau (unpublished data); the data from the two types of stations are employed for nudging and evaluation, respectively. To ensure the quality of the observations, quality control (QC) included consistency and continuity checks, so hourly temperatures from 1165 sites (Fig. 2a), hourly wind speeds and directions from 955 sites (Fig. 2b), and hourly relative humidities from 60 sites were recorded (Fig. 2c) and used for observation nudging. The hourly pressure, precipitation, temperature, relative humidity, and wind speed data from the CWSs are used to evaluate the downscaling results as independent samples from the IAWSs (Fig. 2d).
Distribution of IAWSs data, including (a) temperature at 2 m, (b) wind at 10 m, (c) relative humidity at 2 m, and (d) CWSs in Liaoning.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Dynamic downscaling is more focused on the accuracy and representativeness of the mean and extreme variables. An analysis is conducted for the 2-m temperature (mean, daily maximum, and daily minimum) and accumulated monthly, 5-day, and daily precipitation, as well as wind speed, relative humidity, and surface pressure, and only the 4 km resolution of the finest domain was evaluated with observations. The correlation coefficient (CORR), bias, root-mean-square error (RMSE), and mean absolute error (MAE) are utilized to compare the simulated results with the previously mentioned observed data:
In Eqs. (1)–(4), Pi and Oi represent the model simulation and the observed values at hourly i and N represents the total number of verification points and time.
Moreover, the threat score (TS) is used to measure the ability of the model to predict precipitation for a certain threshold. In this study, the threshold amounts used for daily, 5-day, or monthly accumulated precipitation are 0.1, 10, and 25 mm, respectively:
In Eq. (5), ak is the number of points that were forecast to occur by the threshold precipitation and did occur, bk is the number of points that were forecast to occur by the threshold precipitation but did not occur, and ck is the number of points that were forecast to not occur by the threshold precipitation but did occur; k is the different threshold.
3. Results
In this section, we focus on the evaluation of the AN and observation nudging strategies in improving the surface fields. Section 3a focuses on both the extremes and means of temperature at the surface in all three experiments. Section 3b focuses on the precipitation simulation with site observations using AN and observation nudging. Section 3c focuses on the surface wind, moisture, and pressure. Section 3d focuses on the possible reasons for large-scale and small-scale structures with AN and AON compared to CFSv2 analysis datasets.
a. Temperature verification
As a result of the influence of the East Asian monsoon, orography, and ocean and continental features, the climatic characteristics of northwest Liaoning belong to semiarid areas, and southeast Liaoning belongs to humid regions. The observational temperature of the daily maximum temperature at 2 m (T2max) (Fig. 3a), daily minimum temperature at 2 m (T2min) (Fig. 3e), and daily mean temperature at 2 m (T2mean) (Fig. 3i) has a decreasing trend from the coast to inland. All three downscaling experiments (CTL: second column, AN: third column, and AON: fourth column) can capture the spatial distributions of T2max (Figs. 3b–d), T2min (Figs. 3f–h), and T2mean (Figs. 3j–l), but there is still a large difference between them. For the extremes in the T2max and T2min temperatures, the simulation from CTL is significantly colder than the observations for both throughout the study area. With AN, the downscaling of T2max and T2min shows a great improvement over CTL but is still slightly lower than the observations in the central and coastal areas of Liaoning. Surprisingly, the downscaling of T2max and T2min from AON has a further improvement over AN, and the extremes in T2max and T2min at the center of Liaoning have a very good consistency with the observations. Similar to the extreme temperature, the daily mean temperature (T2mean) from AN and AON also showed a great improvement over CTL, and AON was closest to the observations because it nudged the surface variables.
Spatial distribution of annual mean daily (a)–(d) maximum, (e)–(h) minimum, and (h)–(k) mean temperatures at 2 m (°C) from (left) CWSs, (left center) CTL, (right center) AN, and (right) AON.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
To further analyze the bias features of T2max, T2min, and T2mean, Fig. 4 shows the bias between the observation and simulation interpolation from the downscaling result. As in Fig. 3, the downscaling of T2max, T2min, and T2mean from CTL shows cold biases of more than 3 K in most of Liaoning compared to the observations. For AN and AON, the biases of T2max, T2min, and T2mean are smaller than 2 K, and the results from AON are more consistent with the observations than AN. AON corrects the cold biases of 1–2 K for T2max, T2min, and T2mean (Figs. 4c,f,i) in central Liaoning. However, the temperature simulation from AON in western Liaoning is slightly worse than that from AN (Figs. 4b,e,h), which is possibly because complex topography causes large internal changes in surface temperature, and observation nudging also introduces too much small-scale surface information.
Spatial distribution of biases for daily (a)–(c) maximum, (d)–(f) minimum, and (g)–(i) mean temperatures (°C) from (left) CTL, (center) AN, and (right) AON.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
To analyze the temperature deviations at different time scales, Table 2 also summarizes the CORR, bias, RMSE, and MAE for the daily, 5-day, and monthly averages. The deviation in the surface variables was computed by pooling all the weather stations that can avoid error cancelation, which is associated with spatial averaging and trace localized extreme values (Soares et al. 2012). All p value for correlation coefficients in Table 2 are less than 0.01 (>95% significance). For the T2max, the observed and downscaled correlation coefficients are 0.991 for AON, 0.987 for AN, and 0.952 for CTL, and the correlation coefficient increases from the daily to monthly time scale. The T2min and T2mean have the same trend as T2max. This similarity means that the downscaling of the overall average state is better represented than the variety of temperature details. The bias of downscaled T2max from CTL is −5 K, which is reduced to −2.2 K by AN and further reduced to −0.7 K by AON. The RMSE and MAE of T2max for CTL are approximately 3.5 times those of the AON and approximately 2 times those of the AN. Similar to the findings of Zhang et al. (2009), the results also showed that the T2min correlations are smaller than those of T2max, indicating that the ability of the WRF Model to simulate T2min is weaker than its ability to simulate T2max. In most cases, the downscaling of extreme temperature and mean temperature from AON is better than that from AN, demonstrating the benefits of the high-density surface observations.
Daily to monthly maximum, minimum, and mean temperature errors. Double asterisks indicate that P < 0.01 (>95% significance). The boldface values show the best values for each indicator from the three experiments.
The RCM usually has a different ability to represent the seasonal change in temperature. The T2max, T2min, and T2mean from the three downscaling experiments show a good representation of the monthly surface temperature variety (Fig. 5). All three experiments have a cold bias of temperature throughout the year, and with the nudging of large-scale information, both the AN and AON improve the cold bias compared to the CTL, and in most months, this deviation is less than 1°C. It is also shown that the simulation of extreme and mean temperatures from the three experiments is better in summer than in winter. The simulation of T2max from AON is better than CTL and AN throughout the year; however, it is not completely applicable in all months for T2min and T2mean. The T2min (Fig. 5b) simulated in the CTL reveals a cold bias of approximately 7.2°C in March and that from AN is slightly better than the AON, and the cold bias is generally no more than 2.5°C. From the seasonal variation in the simulations of T2min (Fig. 5b) and T2mean (Fig. 5c), compared to the observation, it is shown that the simulations of T2min and T2mean from AN are better than AON in summer and worse in winter.
Mean monthly 2-m (a) maximum, (b) minimum, and (c) mean temperatures (°C) for observations, CTL, AN, and AON in 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
A summary of the seasonal error analyses of T2max, T2min, and T2mean are shown in Table 3; the correlations are all significant at the 95% level. The correlation of T2max between the simulation and observation from AON is higher than the CTL and AN in all four seasons, and the bias of T2max from AON is smaller than that in the other two experiments. For the T2min, as in Fig. 5b, the simulation from the three experiments is colder than the observations in all seasons, and the simulation from AN has the smallest bias within 1.2°C in most seasons, but AON has the smallest RMSE in most seasons. The RMSE of T2min is generally larger than that of T2max, which also demonstrates that the model’s ability to simulate T2min is weaker than its ability to simulate T2max in Liaoning. The statistical errors (correlation, bias, and RMSE) of the daily T2mean in the AON experiment are the best in most seasons except for JJA. The CTL has larger errors in spring and autumn for T2max, T2min, and T2mean, the AON reduces the errors significantly, and in winter, the AON improves the simulations but is worse compared to other seasons, indicating that WRF with observation nudging has the worst simulation for winter and better simulations in spring and autumn.
Seasonal errors of daily maximum, minimum, and mean temperatures. Double asterisks indicate that P < 0.01 (>95% significance). The boldface values show the best values for each indicator from the three experiments.
b. Precipitation evaluation
Precipitation is another variable that is considered in dynamic downscaling (Hu et al. 2018). Figure 6 shows the annual precipitation distributions of CTL, AN, AON, and observations in 2015. The observed spatial pattern of precipitation has an east–west gradient with the Liaoning topography and increases from west to east from 300 to 1050 mm, respectively. The CTL seriously overestimates precipitation, and both of the AN and AON nudging experiments well capture the annual pattern of precipitation. Additionally, the precipitation from AON is higher than that from AN in the north and center of Liaoning.
Spatial distribution of annual precipitation (mm) from (a) CWSs, (b) CTL, (c) AN, and (d) AON in 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Figure 7 also shows the spatial distribution of the bias of precipitation from CTL, AN, and AON. The precipitation downscaling from CTL is obviously overestimated by 0.3–1.58 mm day−1 for the entire area. The AN simulation is also better than the CTL, with 49.2% of stations having biases in the interval range of −0.2–0.2 mm day−1 and 47.5% of stations with biases in the interval range of 0.2 ~ 0.6 mm day−1. Nudging with AON surface observations shows a better performance in western Liaoning than that in AN but overestimates the precipitation in central and southern Liaoning.
Spatial distribution of annual mean daily precipitation (mm day−1) biases from (a) CTL, (b) AN, and (c) AON.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Similar to Table 2, Table 4 shows the CORR, bias, RMSE, and MAE of the precipitation at different time scales for the daily, 5-day, and monthly accumulations. The correlations of the simulations from three downscaling experiments with observations rise with an increase in time scale and greatly improves with the two nudging downscale experiments, where all correlations are significant at the 95% level. The correlation of AN is slightly better than that of AON. A wet bias exists in the three simulations, where AN has the smallest wet bias and is slightly better than that of AON. The bias, RMSE, and MAE also show that the simulations of precipitation from AN and AON are greatly improved at different time scales. This result also means that precipitation is a relatively independent variable, and even if the temperature is significantly improved, the change in cloud microphysical processes in the RCM model cannot guarantee improved precipitation simulation.
Daily to monthly precipitation statistics and TS at different thresholds. Double asterisks indicate that P < 0.01 (>95% significance). The boldface values show the best values for each indicator from the three experiments.
The TS scores are usually used to evaluate the ability of models to simulate precipitation for a certain threshold. Table 4 also gives the TS scores at 0.1, 10, and 25 mm with daily, 5-day, and monthly time scales from three downscaling experiments. The TS score decreases with increasing precipitation intensity and increases with time scale. For bias, RMSE, and MAE, the TS score of AN is increased by approximately 20% for different thresholds of precipitation compared with CTL, and AON has a close accuracy that is comparable to AN.
The monthly mean precipitation from the three downscaling experiments and observations is shown in Fig. 8. The CTL overestimates precipitation throughout the year, except in January, April, and June. Both AN and AON perform better than CTL, and AON performs slightly better than AN in most months except August and September. Although AN was found to be better than AON in terms of annual precipitation, it is not always optimal.
Monthly mean daily precipitation (mm day−1) for the observations, CTL, AN, and AON in 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Some dynamical downscaling research confirms that WRF usually overestimates precipitation and generates too many precipitation days (Zhang et al. 2009; Heikkilä et al. 2011), which is consistent with the three downscaling experimental results in this work. Figure 9 shows the average number of days per month with precipitation greater than 0.1 mm day−1. CTL has too many rainy days per month, and this deficiency appears to improve with AON and AN from March to October, while the number of rainy days predicted by CTL matches the observations better than AN and AON in winter. The curves for AN and AON track each other very well, and AN is slightly closer to the observations than AON.
Monthly precipitation frequency at 0.1 mm.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Table 5 shows the seasonal CORR, bias, RMSE, and TS of daily precipitation at 10 mm from CTL, AN, and AON. The CORR and TS show that the precipitation simulations from AN and AON are significantly improved in all four seasons. The simulation of precipitation from AN has the smallest bias and highest CORR and TS in most seasons, which is slightly better than AON. Most precipitation and convective processes occurred in summer in Liaoning, and although the precipitation bias is close to 0 for AN, the RMSE and CORR are worse than those in other seasons, which has also been found by other studies (e.g., Caldwell et al. 2009; Rauscher et al. 2010).
Seasonal errors of daily precipitation. Double asterisks indicate that P < 0.01 (>95% significance). The boldface values show the best values for each indicator from the three experiments.
c. Surface wind, moisture, and pressure
In this section, the surface wind speed at 10 m (WSP10), relative humidity (RH2) at 2 m, and surface pressure (PS) from three downscaling experiments are discussed. Figures 10–12 show the spatial distributions of WSP10, RH2, and PS. In Fig. 10, by nudging the IAWS observations, the surface wind speed from the AON shows a better preformation than AN and CTL. CTL and AN overestimate the WSP10 in central and western Liaoning relative to the observations. As shown in Table 6, the bias of WSP10 from AON is 0.1 m s−1, which is better than 1.1 m s−1 from CTL and 0.8 m s−1 from AN. The correlation coefficient from AON is also highest among the three downscaling experiments at different time scales.
Spatial distribution of annual mean wind speed at 10 m (m s−1) from (a) CWSs, (b) CTL, (c) AN, and (d) AON.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
As in Fig. 10, but for relative humidity at 2 m (%).
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
As in Fig. 10, but for surface pressure (hPa).
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Daily to monthly wind speeds at 10 m, relative humidity at 2 m, and surface pressure statistics. Double asterisks indicate that P < 0.01 (>95% significance). The boldface values show the best values for each indicator from the three experiments.
Only 60 of the IAWS site observation variables contained humidity and passed the QC, as shown in Fig. 2c; however, with observation nudging, AON showed a better performance in the simulation of surface humidity than CTL and AN in central and southern Liaoning. From Table 6, the RMSE of RH2 from AON is smallest, and the RMSE was reduced by nearly 48% relative to CTL and nearly 39.7% relative to AN at different time scales.
The PS spatial distributions from the three downscaling experiments are not much different from each other, and all three experiments can capture the basic features of surface pressure and overestimate the surface pressure in eastern and western Liaoning. The CORR, bias, and RMSE of PS for AN and AON are significantly better than CTL, as shown in Table 6.
Figure 13 also shows the comparison of monthly WSP10, RH2, and PS from the three experiments. The results are consistent with those shown in Fig. 10; the wind speed improved dramatically with AON, and AN was better than CTL. Different from the wind speed, the humidity simulated from AON shows the best performance from May to September, while it is overestimated in other months. Both AN and AON have good abilities to capture variations in surface pressure.
Mean monthly (a) 10-m wind speed (m s−1), (b) 2-m relative humidity (%), and (c) surface pressure (hPa) for observations, CTL, AN, and AON in 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
d. Large-scale atmospheric circulation compared to the driving data
In the following, we test the impact of AN and observation nudging on large-scale circulation compared to its driving analysis datasets of CFSv2. Figure 14 shows the bias of the 500-hPa geopotential height and 925-hPa relative humidity of the three downscaling experiments in the first domain at a 36 km resolution with CFSv2. Figure 14a shows a significant bias of 500-hPa geopotential height for CTL at a 36 km resolution in most regions, especially the bias that can reach 40 gpm in northeastern China. This result means that the model is drifted away from the large-scale driving data in the continuous simulation without nudging. In Figs. 14b and 14c, both the AN and AON experiments reduce the biases, and with nudging surface observations, the large-scale circulation remains the same as that in the AN experiment. This is an important result because it suggests that the additional constraint from the dense surface observations not only improves the surface simulations but also keeps the large-scale circulation intact. The mean biases of the 500-hPa geopotential height in the first domain for AN and AON are the same, both 7.3 gpm, and the correlation coefficient and RMSE are the same, 0.99 and 7.9 gpm, respectively. The 925-hPa relative humidity in CTL has a large bias with the CFSv2 in northeastern China, southern China, and the eastern Pacific (Fig. 14d). Similar to the 500-hPa geopotential height, AN and AON can significantly reduce the bias of the 925-hPa relative humidity with the large-scale driving field, and there is no significant difference in the spatial distribution between AN and AON (Figs. 14e,f). The mean biases of the 925-hPa relative humidity in AN and AON are the same, both are 6.4%, and both the correlation coefficient and RMSE of the two nudging experiments are the same at 0.93 and 6.5%, respectively.
The bias spatial distribution of the (a)–(c) 500-hPa geopotential height (gpm) and (d)–(f) 925-hPa relative humidity (%) between the downscaling results and CFSv2 in the first domain in 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
To further discuss the effects of AN and observation nudging on the finest domain of 4 km, the monthly variation in the related statistical verification techniques for the 500-hPa geopotential height and temperature, 850-hPa wind speed and 925-hPa relative humidity of three dynamical downscaling results compared with CFSv2 in the third domain are shown in Fig. 15. CTL shows the smallest correlation coefficient and largest biases and RMSEs for all the above mentioned variables. Both AN and AON show significant improvements. The error statistics of the 500-hPa geopotential height and temperature simulated by AN are similar to those in AON, and monthly changes are relatively steady. However, the error statistics of 850-hPa wind speed and 925-hPa relative humidity are different and have significant monthly changes, where AN is closer to CFSv2 than AON.
Monthly correlation coefficient, bias, and RMSE between the downscaling results and CFSv2 for (a)–(c) 500-hPa geopotential height, (d)–(f) temperature, (g)–(i) 850-hPa wind speed, and (j)–(l) 925-hPa relative humidity during January–December 2015.
Citation: Journal of Applied Meteorology and Climatology 59, 10; 10.1175/JAMC-D-20-0071.1
Although the model uses a two-way interactive nest, AON with nudging surface observations in the third domain does not have a significant impact on upper large-scale circulation compared with AN. However, for small-scale information, the overall impact of applying observation nudging is stronger in the lower atmosphere than in the upper layers. For the observation nudging scheme, the surface observation innovations at the lowest model level are extended vertically into upper layers through the mixing layer, with weights gradually reduced toward the PBL top (Liu et al. 2005; Deng et al. 2009).
4. Discussion and conclusions
In this study, the dynamic downscaling strategy of analysis nudging (AN) and observation nudging based on WRF is applied to produce high-resolution regional climate datasets. Three 1-yr simulation experiments without nudging (CTL), with AN, and with AN and observation nudging (AON) are carried out to downscale the CFSv2 analysis data to 4 km in Liaoning in northeastern China. The surface variables from 1552 IAWS surface temperature, wind, and humidity observations are nudged with observation nudging, and independent samples of CWS surface fields are used to evaluate the downscaled results.
The dynamic downscaling results show that the WRF Model has a cold bias in the simulation of the mean and extremes in the surface temperature in Liaoning, northeast China, and the simulation of daily maximum temperature is better than that of daily minimum temperature. The downscaling strategy of combining the AN and observation nudging reduces the bias of maximum temperature (T2max) from −2.2° to −0.7°C when compared with only using AN, and this bias is even as high as −5.0°C for the control experiment. The representation of the variability in extreme temperature and mean temperature at different time scales from AON are also better than those of AN and CTL. Moreover, the results show that both nudging experiments decrease the 10-m wind speed, 2-m relative humidity, and surface pressure errors compared with the CTL; in particular, using observation nudging, the AON further improved the simulations.
Although the surface fields show great improvement, with nudging of the high-density surface observations, the WRF Model cannot ensure further improvements to the performance of the precipitation simulation with nudging surface fields. All three experiments can capture the precipitation patterns, and the CTL shows an obvious overestimate of the annual precipitation. The precipitation errors appear in the central and coastal areas of Liaoning and do not coincide with the largest area of precipitation. Both AN and AON reduce the wet bias at different time scales, and AN performs slightly better for most statistical errors; however, annual cycles of precipitation are better described by AON in most months. In conclusion, combining the AN and observation nudging further improves the simulation ability of temperature, wind speed, relative humidity, and surface pressure, while the impact on precipitation and high-altitude factors is limited compared with only AN.
Some studies also show that nudging coefficients can influence the downscaling results (e.g., Bullock et al. 2014; Mai et al. 2017). However, this study is related to only a single choice of analysis and observation nudging coefficients. Further studies are needed to optimize the observation nudging strategy to reduce errors near the surface, and the combination of spectral nudging and observation nudging will be studied in future work. Additionally, static surface datasets can also affect the downscaled results, and we will update the surface static datasets and further evaluate the influence of the surface information in future downscaling work.
Acknowledgments
The authors were supported by the National Key Research and Development Program of China (2018YFC1506803 and 2018YFC1507302), National Natural Science Foundation of China (41675098), Climatic Change Research Item of the China Meteorological Administration (CCSF201910), and Guiding Plan for Agricultural Research and Industrialization of Liaoning Provincial Department of Science and Technology (2019JH8/10200023). The authors are grateful to the Liaoning Meteorological Observation Data Center for the collection of observation data. The authors also thank the insightful suggestions of Juanzhen Sun from the National Center for Atmospheric Research.
REFERENCES
Alexandru, A., R. De Elia, R. Laprise, L. Separovic, and S. Biner, 2009: Sensitivity study of regional climate model simulations to large-scale nudging parameters. Mon. Wea. Rev., 137, 1666–1686, https://doi.org/10.1175/2008MWR2620.1.
Bowden, J. H., T. L. Otte, C. G. Nolte, and M. J. Otte, 2012: Examining interior grid nudging techniques using two-way nesting in the WRF Model for regional climate modeling. J. Climate, 25, 2805–2823, https://doi.org/10.1175/JCLI-D-11-00167.1.
Bowden, J. H., C. G. Nolte, and T. L. Otte, 2013: Simulating the impact of the large-scale circulation on the 2-m temperature and precipitation climatology. Climate Dyn., 40, 1903–1920, https://doi.org/10.1007/s00382-012-1440-y.
Bullock, O. R., K. Alapaty, J. A. Herwehe, M. S. Mallard, T. L. Otte, R. C. Gilliam, and C. G. Nolte, 2014: An observation-based investigation of nudging in WRF for downscaling surface climate information to 12-km grid spacing. J. Appl. Meteor. Climatol., 53, 20–33, https://doi.org/10.1175/JAMC-D-13-030.1.
Caldwell, P., H.-N. S. Chin, D. C. Bader, and G. Bala, 2009: Evaluation of a WRF dynamical downscaling simulation over California. Climatic Change, 95, 499–521, https://doi.org/10.1007/s10584-009-9583-5.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Chen, G., T. Iwasaki, H. Qin, and W. Sha, 2014: Evaluation of the warm-season diurnal variability over East Asia in recent reanalyses JRA-55, ERA-Interim, NCEP CFSR, and NASA MERRA. J. Climate, 27, 5517–5537, https://doi.org/10.1175/JCLI-D-14-00005.1.
Deng, A., and Coauthors, 2009: Update on WRF-ARW end-to-end multi-scale FDDA system. 10th Annual WRF Users’ Workshop, Boulder, CO, NCAR, 1.9, http://www2.mmm.ucar.edu/wrf/users/workshops/WS2009/abstracts/1-09.pdf.
Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107, https://doi.org/10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.
Giorgi, F., and W. J. Gutowski, 2015: Regional dynamical downscaling and the CORDEX initiative. Annu. Rev. Environ. Resour., 40, 467–490, https://doi.org/10.1146/annurev-environ-102014-021217.
Glisan, J. M., W. J. Gutowski, J. J. Cassano, and M. E. Higgins, 2013: Effects of spectral nudging in WRF on Arctic temperature and precipitation simulations. J. Climate, 26, 3985–3999, https://doi.org/10.1175/JCLI-D-12-00318.1.
Gula, J., and W. R. Peltier, 2012: Dynamical downscaling over the great lakes basin of North America using the WRF regional climate model: The impact of the great lakes system on regional greenhouse warming. J. Climate, 25, 7723–7742, https://doi.org/10.1175/JCLI-D-11-00388.1.
Heikkilä, U., A. Sandvik, and A. Sorteberg, 2011: Dynamical downscaling of ERA-40 in complex terrain using the WRF regional climate model. Climate Dyn., 37, 1551–1564, https://doi.org/10.1007/s00382-010-0928-6.
Hong, S., and J. Dudhia, 2012: Next-generation numerical weather prediction: Bridging parameterization, explicit clouds, and large eddies. Bull. Amer. Meteor. Soc., 93, ES6–ES9, https://doi.org/10.1175/2011BAMS3224.1.
Hong, S., J. Dudhia, and S. Chen, 2004: A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132, 103–120, https://doi.org/10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.
Hong, S., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
Hu, X.-M., M. Xue, R. A. McPherson, E. Martin, D. H. Rosendahl, and L. Qiao, 2018: Precipitation dynamical downscaling over the Great Plains. J. Adv. Model. Earth Syst., 10, 421–447, https://doi.org/10.1002/2017MS001154.
Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927–945, https://doi.org/10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2.
Jiang, Z., Y. Zhu, H. Ma, and X. Qiu, 2018: Simulation study of establishment of high resolution temperature field by assimilating automatic station data in the three gorges area in January. Daqi Kexue Xuebao, 41, 289–297.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181, https://doi.org/10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Komurcu, M., K. A. Emanuel, M. Huber, and R. P. Acosta, 2018: High-resolution climate projections for the northeastern United States using dynamical downscaling at convection-permitting scales. Earth Space Sci., 5, 801–826, https://doi.org/10.1029/2018EA000426.
Laprise, R., 2014: Comment on “The added value to global model projections of climate change by dynamical downscaling: A case study over the continental U.S. using the GISS-ModelE2 and WRF models” by Racherla et al. J. Geophys. Res. Atmos., 119, 3877–3881, https://doi.org/10.1002/2013JD019945.
Leung, L. R., and Y. Qian, 2009: Atmospheric rivers induced heavy precipitation and flooding in the western U.S. simulated by the WRF regional climate model. Geophys. Res. Lett., 36, L03820, https://doi.org/10.1029/2008GL036445.
Li, X., Y. Choi, B. Czader, H. Kim, B. Lefer, and S. Pan, 2016: The impact of observation nudging on simulated meteorology and ozone concentrations during DISCOVER-AQ 2013 Texas campaign. Atmos. Chem. Phys., 15, 27 357–27 404, https://doi.org/10.5194/acpd-15-27357-2015.
Liu, P., X. Qiu, Y. Yang, Y. Ma, and S. Jin, 2018: Assessment of the performance of three dynamical climate downscaling methods using different land surface information over China. Atmosphere, 9, 101, https://doi.org/10.3390/atmos9030101.
Liu, Y., A. Bourgeois, T. Warner, S. Swerdlin, and J. Hacker, 2005: Implementation of observation-nudging based FDDA into WRF for supporting ATEC test operations. Sixth Annual WRF and 15th Annual MM5 Users’ Workshop, Boulder, CO, NCAR, 27–30.
Lo, J. C.-F., Z.-L. Yang, and R. A. Pielke, 2008: Assessment of three dynamical climate downscaling methods using the Weather Research and Forecasting (WRF) model. J. Geophys. Res., 113, D09112, https://doi.org/10.1029/2007JD009216.
Lucas-Picher, P., J. Cattiaux, A. Bougie, and R. Laprise, 2016: How does large-scale nudging in a regional climate model contribute to improving the simulation of weather regimes and seasonal extremes over North America? Climate Dyn., 46, 929–948, https://doi.org/10.1007/s00382-015-2623-0.
Ma, Y., Y. Yang, X. Mai, C. Qiu, X. Long, and C. Wang, 2016: Comparison of analysis and spectral nudging techniques for dynamical downscaling with the WRF model over China. Adv. Meteor., 2016, 4761513, https://doi.org/10.1155/2016/4761513.
Mai, X., Y. Ma, Y. Yang, D. Li, and X. Qiu, 2017: Impact of grid nudging parameters on dynamical downscaling during summer over mainland China. Atmosphere, 8, 184, https://doi.org/10.3390/atmos8100184.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 663–16 682, https://doi.org/10.1029/97JD00237.
Omrani, H., P. Drobinski, and T. Dubos, 2013: Optimal nudging strategies in regional climate modelling: Investigation in a big-brother experiment over the European and Mediterranean regions. Climate Dyn., 41, 2451–2470, https://doi.org/10.1007/s00382-012-1615-6.
Otte, T. L., 2008: The impact of nudging in the meteorological model for retrospective air quality simulations. Part I: Evaluation against national observation networks. J. Appl. Meteor. Climatol., 47, 1853–1867, https://doi.org/10.1175/2007JAMC1790.1.
Rauscher, S. A., E. Coppola, C. Piani, and F. Giorgi, 2010: Resolution effects on regional climate model simulations of seasonal precipitation over Europe. Climate Dyn., 35, 685–711, https://doi.org/10.1007/s00382-009-0607-7.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1058, https://doi.org/10.1175/2010BAMS3001.1.
Saha, S., and Coauthors, 2011: NCEP Climate Forecast System Version 2 (CFSv2) 6-hourly products (updated daily). NCAR Computational and Information Systems Laboratory Research Data Archive, accessed 26 February 2016, https://doi.org/10.5065/D61C1TXF.
Salathé, E. P., R. Steed, C. F. Mass, and P. H. Zahn, 2008: A high-resolution climate model for the U.S. Pacific Northwest: Mesoscale feedbacks and local responses to climate change. J. Climate, 21, 5708–5726, https://doi.org/10.1175/2008JCLI2090.1.
Soares, P. M. M., R. M. Cardoso, P. M. A. Miranda, J. de Medeiros, M. Belo-Pereira, and F. Espirito-Santo, 2012: WRF high resolution dynamical downscaling of ERA-Interim for Portugal. Climate Dyn., 39, 2497–2522, https://doi.org/10.1007/s00382-012-1315-2.
Sommerfeld, M., M. Dörenkämper, G. Steinfeld, and C. Crawford, 2019: Improving mesoscale wind speed forecasts using lidar-based observation nudging for airborne wind energy systems. Wind Energy Sci., 4, 563–580, https://doi.org/10.5194/wes-4-563-2019.
Spero, T. L., M. J. Otte, J. H. Bowden, and C. G. Nolte, 2014: Improving the representation of clouds, radiation, and precipitation using spectral nudging in the weather research and forecasting model. J. Geophys. Res. Atmos., 119, 11 682–11 694, https://doi.org/10.1002/2014JD022173.
Spero, T. L., C. G. Nolte, M. S. Mallard, and J. H. Bowden, 2018: A maieutic exploration of nudging strategies for regional climate applications using the WRF Model. J. Appl. Meteor. Climatol., 57, 1883–1906, https://doi.org/10.1175/JAMC-D-17-0360.1.
Stauffer, D. R., and N. L. Seaman, 1990: Use of four-dimensional data assimilation in a limited-area mesoscale model. Part I: Experiments with synoptic-scale data. Mon. Wea. Rev., 118, 1250–1277, https://doi.org/10.1175/1520-0493(1990)118<1250:UOFDDA>2.0.CO;2.
Stauffer, D. R., T. T. Warner, and N. L. Seaman, 1985: A Newtonian ‘nudging’ approach to four-dimensional data assimilation—Use of SESAME-IV data in a mesoscale model. Preprints, Seventh Conf. on Numerical Weather Prediction, Montreal, QC, Canada, Amer. Meteor. Soc., 77–82.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 5095–5115, https://doi.org/10.1175/2008MWR2387.1.
Wang, J., and V. R. Kotamarthi, 2013: Assessment of dynamical downscaling in near-surface fields with different spectral nudging approaches using the Nested Regional Climate Model (NRCM). J. Appl. Meteor. Climatol., 52, 1576–1591, https://doi.org/10.1175/JAMC-D-12-0302.1.
Wootten, A., J. H. Bowden, R. Boyles, and A. Terando, 2016: The sensitivity of WRF downscaled precipitation in Puerto Rico to cumulus parameterization and interior grid nudging. J. Appl. Meteor. Climatol., 55, 2263–2281, https://doi.org/10.1175/JAMC-D-16-0121.1.
Xue, Y., Z. Janjić, J. Dudhia, R. Vasic, and F. De Sales, 2014: A review on regional dynamical downscaling in intraseasonal to seasonal simulation/prediction and major factors that affect downscaling ability. Atmos. Res., 147–148, 68–85, https://doi.org/10.1016/j.atmosres.2014.05.001.
Yi, X., D. Li, C. Zhao, L. Shen, X. Ao, and M. Liu, 2018a: Assessment of dynamical climate downscaling methods using analysis nudging for Liaoning area (in Chinese). Adv. Earth Sci., 33, 517–531, https://doi.org/10.11867/j.issn.1001-8166.2018.05.0517.
Yi, X., D. Li, C. Zhao, X. Zhou, Y. Cui, and Y. Hou, 2018b: Dynamical climate downscaling over Liaoning area using nudging methods based on WRF Model (in Chinese). J. Meteor. Environ., 34, 1–10, https://doi.org/10.3969/j.issn.1673-503X.2018.02.001.
Zhang, Y., V. Dulière, P. W. Mote, and E. P. Salathé, 2009: Evaluation of WRF and HadRM mesoscale climate simulations over the U.S. Pacific Northwest. J. Climate, 22, 5511–5526, https://doi.org/10.1175/2009JCLI2875.1.
Zhou, R., and X. He, 2018: Numerical simulation and character analysis of wind field in complex terrain in Hami Xinjiang. Plateau Meteor., 37, 1413–1427, https://doi.org/10.7522/j.issn.1000-0534.2018.00021.
