Sensitivity of physical parameterization schemes in WRF model for dynamic downscaling of climatic variables over the MRB

The Weather Research and Forecasting (WRF) model was tested through 18 different combinations of physics parameters to simulate the regional climate over the Mackenzie River Basin (MRB). The objective was to investigate the response to the physics parameters for dynamic downscaling of climatic variables. The rainfall, temperature, albedo, and surface pressure from the 18 different WRF setups were compared with the reference data and were found sensitive to land surface physics and microphysics and to the radiation physics. The combination of Noah Land Surface Physics with the WRF Single-moment 6-class microphysics and CAM shortwave and longwave schemes produced comparable results for summer 2009. This WRF setup was further tested for summers 1979–1991 and it was found that WRF could simulate air temperature more accurately than the rainfall, since the rainfall over the mountainous regions was over-simulated. Then the selected combinations of WRF parameterizations were used to downscale the CanESM2 historical temperature and rainfall for summers 1979–2005, which showed good agreement with the reference data. The suggested WRF parameters from this study could be utilized for regional climate modeling of MRB.

In this study, our objective was to find out how the choice of physics parameters could possibly affect the WRF model simulations over a large and complex river basin like the MRB and also how realistically WRF can simulate the long-term summer temperature and rainfall over the MRB. This effort using different WRF parameterizations for sensitivity analysis will enhance the knowledge of regional climate modeling performance.

Study area
Mackenzie River Basin (MRB), a high latitude continental basin with a total area of about 1.8 million square kilometers, extending from 52 to 69 N and 140 W to 102 W, which enjoys being the largest river basin in Canada.
The basin possesses a unique hydrological and environmental variability as it has continuous permafrost at the north and warm summer at the south. It can be described with three major physiographic regions: the Cordillera, the Interior Planes, and the Precambrian Canadian Shield.
Atmospheric circulation is greatly influenced by the topography: the Rockies at the western side and the lower latitude at the central part induce strong atmospheric circulations; the lakes and wetlands affect the energy and water balance of the basin (Woo et al. ). Figure 1 shows the model domain and terrain height of the study area.

Selection of physics parameters for WRF model configuration
The WRF model is being used as a regional climate model (RCM) for dynamical downscaling of global model data in different parts of the world, because WRF provides the freedom to select the options that best describe the regional climate of interest and produces finer resolution data (even for 3 km by 3 km resolution) for that geographic region (Prabha et al. ).
The Advanced Research WRF (ARW-WRF) modeling system gives a large selection of physical parameters which makes the model more specified according to our needs. The selection of physics parameters are also influenced by the geographic position and topography of the area. Microphysics, longwave and shortwave radiation, surface layer, land surface, planetary boundary layer and also cumulus parameterization play important roles for simulating the regional climate

Tests setup
The WRF model was set up to test the sensitivity of the physics parameters to simulate the climate over the MRB.  used for longwave radiation (ra_lw_physics), and Dudhia scheme and RRTMG shortwave schemes for shortwave radiation (ra_sw_physics). These selections also combine with Yonsei University scheme and Asymmetric Convection Model 2 scheme (ACM2) for its planetary boundary layer (bl_pbl_physics) and for surface layer option MM5 Similarity scheme was chosen, and Kain-Fritsch scheme was taken as the cumulus parameterization option (cu_physics). Table 1 shows the combination of the physics parameters with 5-layer Thermal Diffusion scheme; here numerical numbers correspond to the physics scheme as mentioned in the WRF-ARW user manual.
For the second category, the Unified Noah Land Surface model was chosen as the land surface scheme, where a total of 12 different combinations of physics parameters were used. Table 2 shows the combinations of the physics parameters with Unified Noah Land Surface physics (using the same numerical numbers used in WRF).
The third category is the combinations of physics options with the RUC Land Surface Model. A total of three sets were tested for this category with the Kessler scheme and WRF Single-moment 3-class microphysics scheme combined with MM5 Similarity scheme for surface layer options. The radiation physics were taken as CAM shortwave and longwave schemes, RRTMG shortwave and longwave schemes and Fu-Liou-Gu shortwave and longwave schemes, and Kain-Fritsch scheme was selected for the cumulus parameterization option for all the three tests of this category. Table 3 shows three different combinations of the physics parameters with RUC Land Surface model.

Temperature
The WRF-simulated 2 m air temperature data were averaged over the testing period (May, June, July, August of 2009) and compared with the ANUSPLIN temperature data for the same period. The ANUSPLIN data are daily observational  Temperature bias over the basin was calculated and is shown in Figure 2. For the first category of WRF test (exp. no. 1 to 3), it was observed that using the 5-layer Thermal Diffusion scheme temperature was well simulated, although for test no.1 cold bias was observed up to À5 C in the lake region; whereas exp. no. 2 and 3 show 1 to 3 degree positive bias in the lake region and slightly negative bias (up to À1 degree) in the north-western part of the basin. Figure 2 shows the 2 m air temperature bias (WRF-ANUSPLIN)  Figure 3 shows the plotting of average 2 m air temperature from the ERA-Interim data and from the 18 different WRF experiments.
From Figure 3 it can be stated that exp. no. 11 and 12 showed close agreement with the reference ERA data.
A Taylor diagram (Taylor ) was plotted using the 18 different WRF simulations and the ERA-Interim dataset to show the statistical relationships between them. Figure 4 shows the correlation, RMSD, and standard deviation for each simulation (the numbers on the plot indicate the WRF experiment numbers). From Figure 4     Depending on the basis of the 18 experiments, it was found that the summer rainfall from exp. no. 11 and 12 agreed well with the ANUSPLIN 10 km gridded data.
From the spatial distribution of rainfall pattern, it is clear that the model has a tendency to produce higher rainfall over the mountainous region while giving lower rainfall over the northern part of the basin.
To compare the WRF rainfall with the observed rain gauge data, eight climate stations located within the three major physiographic regions of the basin were selected.
These eight stations were scattered over the MRB subbasins, thus representing the different parts of the basin.
The 'Second Generation of Daily Adjusted Precipitation for Canada' was collected from Environment Canada web site to use as the reference data. Figure 6 shows the selected rainfall station location over the MRB. Figure 7(   We also compared the ERA-Interim 2 m air temperature with WRF simulation over the same period. Figure 13 shows the bias plot for WRF and ERA-Interim data. A scatter plot of these temperature data is shown in Figure 14. The correlation coefficient was 0.81, which indicates a fairly good correlation between WRF simulations versus the ERA-Interim 2 m air temperature data.
To assess WRF simulated precipitation over the MRB for the historical period using CanESM2 data we compared the outputs with some observed station rainfall data. It was found that WRF simulated historical average rainfall data matches reasonably well with some observed rainfall data     From Figure 15, it was observed that most of the mountainous area has much higher precipitation than the low lying plains of central and eastern parts of MRB. It was also noticeable that WRF oversimulated precipitation over the western part of the MRB, but it produced reasonable precipitation over other parts of the MRB for the historical period. By comparing with the ANUSPLIN data, it was clear that the bias of WRF simulation was generally modest although about 250 mm of positive bias in the western part and about 100 mm in the north-eastern part of the MRB were observed.
Based on 2 m air temperature and precipitation data from WRF simulations for the base period  using CanESM2 historical data or from the ERA-Interim reanalysis data (for 1979-1991), it was found that the chosen WRF setup (exp. no. 11) was suitable for long-term summer temperature and rainfall simulations for MRB.

CONCLUSIONS
In this study, a physically based mesoscale model called WRF was tested for 18 different physics combinations to simulate the climate of MRB. The outputs from the WRF were compared and analyzed with the reference data to investigate the sensitivity of the model parameters and significant sensitivity was found; especially to microphysics, land surface, planetary boundary layer, and longwave radiation schemes. Temperature, rainfall, albedo and surface   It was observed that the WRF simulated precipitation from experiment 11 had good agreement with the reference data although some positive bias (WRF-ANUSPLIN) was observed over the mountainous region and negative bias for the other parts of the basin. We found that the special variability of rainfall distribution over the basin was more realistically captured by exp. no. 11.
In the case of air temperature simulation, WRF showed high proficiency in simulating the summer 2009 temperature using the Noah Land Surface Model. It was observed that exp. no. 11 could realistically downscale the surface air temperature. Also, the spatial variability of temperature distribution over the MRB was well captured by WRF. We found that the temperature simulation was sensitive to land surface model, microphysics, and radiation schemes.
We also compared the WRF simulated albedo with the observed albedo data as well as with the ERA albedo data, and good results were found.
However, the mean surface pressure (PSFC) simulated by WRF showed continuous undersimulation in comparison to ERA-Interim data, which was a similar finding from others.
This study provided a guideline for the selection of model parameters to simulate the climate for a large and complex area. The proposed WRF setup from this study was utilized to simulate long-term climate over MRB for the base period. WRF simulations for the summer period from 1979 to 1991 based on ERA-Interim reanalysis data were compared with the ANSUPLIN data and the ERA data, which showed good agreement. The current WRF setup was further utilized to downscale MRB climate (air temperature and rainfall) using CanESM2 data for the base period (summer 1979-2005), and reliable results were obtained by comparing with the reference data for the same period. From this study, the recommended WRF parameterizations would be useful for regional climate modeling and for future climate change projection of that area.