Printer Friendly

Mesoscale climate modeling procedure development and performance evaluation.


Current architectural and engineering practice involves careful consideration of local meteorology as a key factor in many design projects. Parameters such as wind speed, temperature, humidity, precipitation, and incoming radiation can all influence a building's design and ongoing performance. ASHRAE members typically use climate data derived from multiyear (minimum 8 years, typical 30 years) measured data, provided by ASHRAE for locations worldwide (ASHRAE 2013). However, these data are often taken from meteorological stations, which may be remote from a study site by tens to hundreds of kilometers or perhaps in completely different terrain (urban versus rural, mountain versus valley), which may not be representative. Factors such as urban heat island effects, sea/land breezes, and varying terrain can all significantly alter meteorological conditions between a weather station (often at an airport outside of an urban center) and an actual project site. Alternatively, there may be many stations located in close proximity to a design site and the user must select the most appropriate data source. For instance, the three major airports serving New York City (JFK, LaGuardia, and Newark) are all located within a 15 km (9.3 mi) radius of the city center (Manhattan) and have a 2.4[degrees]C (4.3[degrees]F) difference in 0.4% cooling dry-bulb temperature, a difference of 265 cooling degree days (CDD65) and a difference of 225 heating degree days (HDD65). While this example may account for 10% differences in degree days, considering the number of HVAC systems in the city designed to meet the conditions, the accuracy of climate data becomes very important from an energy conservation and efficiency perspective. The challenging question is which airport data shouldbe treated as representative of a site on Manhattan Island.

There are a number of techniques available to derive the meteorology of a given a site in the absence of a suitable observational station. These techniques often consist of some form of interpolation or more complicated aerodynamic physical process/modeling to downscale data to a finer resolution. The interpolation approach can be as simple as spatial linear interpolation of neighboring observational stations or more complex, such as forming linear regression models based on predictors accounting for elevation or proximity to a coast (e.g., the PRISM model [Daly et al. 2008]). At the far end of complexity, mesoscale weather models, such as the Weather Research Forecast (WRF) model, consider the conservation of mass, momentum, energy, and moisture, coupled with representations of atmospheric radiation and cloud formation. These mesoscale models represent the meteorology on grids approaching the microscale (i.e., 1 km [0.6 mi] in horizontal extent), commonly used by meteorologists and climatologists for high-resolution weather forecasting or local climate extreme analysis.

Although these mesoscale models can be used to refine historical climate data at a very high spatial resolution, inherent with mesoscale model's complexity and power is a barrier to entry for first-time users; it can be difficult to execute the model without fully understanding the physics schemes and dynamic mechanisms that are being used for various weather conditions. Additionally, significant computational resources are required to execute these models. In an effort to provide engineers and designers with more accurate meteorological data, a methodology has been developed for modeling site-specific meteorological data with next-generation meteorological models, potentially on desktop computers.

ASHRAE RP-1561, Procedure to Adjust Observed Climatic Data for Regional or Mesoscale Climatic Variations, was created with two main goals in mind:

* To develop a methodology for ASHRAE members to harness the power of modern mesoscale modeling techniques in order to derive meteorological conditions specific to a study area

* To evaluate this methodology against available meteorological and climatic observational data in a variety of geographic categories, including coastal, mountain valley, mountain plateau, and major city

Here, an overview of the modeling methodology is provided and evaluated based on performance in estimating both hourly meteorology (i.e., the weather) and design elements such as the 99.6% heating dry-bulb temperature (i.e., the climate). A discussion and evaluation of a simplified methodology designed for ASHRAE members with intermediate computer skill levels is also provided. This paper provides a condensed version of the project final report (Qiu et al. 2015).


There are several numerical meteorological models, commonly used within the meteorological community for high-resolution weather forecasting, which are capable of producing high-quality gridded hourly climate data. The WRF model was used in this study due to its ability to generate high-resolution and reliable climate data at any location within its modeling domain. The WRF model is a next-generation mesoscale numerical weather prediction system designed to serve both operational forecasting and atmospheric research needs. The model is suitable for a broad spectrum of applications across scales ranging from meters to thousands of kilometers (yards to thousands of miles). The WRF model is developed and maintained as part of a collaborative effort principally among U.S. government agencies, universities, national laboratories, and international communities (Skamarock et al. 2008).

Domain Configuration

The modeling procedure begins with the establishment of the model domain. Mesoscale modeling is three-dimensional; the methodology divides the atmosphere from the ground surface to the top of the troposphere, around 100 hPa (1.45 psi), into 35 vertical layers and horizontally by grid cells in kilometers (miles) covering an entire domain. Domains defined in this paper are very large based on the requirements of the scope of work in RP-1561. The final recommended domain sizes for ASHRAE members can practically be much smaller than those in this study.

The WRF model takes a nested approach where a large coarse-resolution domain feeds into a small but fine-resolution domain. For example, the parent or first domain spans approximately 1800 * 1800 km (1120 x 1120 mi) with 36 x 36 km (22.5 mi x 22.5 mi) cells. Nested in this domain is a medium-resolution child domain, which is 730 x 730 km (450 x 450 mi). Nested in this domain is an even finer-resolution child domain, which is 280 x 280 km (170 x 170 mi). The final grid resolution in the smallest domain is approximately 4 km (2.5 mi).

Input Data

Model topography was interpolated from the 30 arcsec (about 1 km [0.6 km]) USGS geophysical data center global data coverage, which are included with the WRF preprocessing system. Vegetation type and land use data at 1 km (0.6 mi) horizontal resolution were also interpolated in to the WRF grids.

The WRF model must be coupled with the output of a larger three-dimensional reanalysis data set, one with greater area coverage:

* The larger model is interpolated in to the smaller regional model and serves as the initial state of the atmosphere, (i.e., the initial condition).

* The larger model also provides the air masses that enter the domain along the lateral boundaries as the simulation evolves (i.e., the boundary conditions).

* Finally, the large model also constrains the coarsest domain such that the model simulation does not drift too far from reality; this is termed nudging.

Here, the reanalysis data set used in this study was the 32 km (22.5 mi) North American Regional Reanalysis (NARR) data set (Mesinger et al. 2006). While NARR was used throughout this project as a WRF model input, readily available alternatives include Climate Forecast System Reanalysis (CFSR) (Saha et al. 2010) and Modern Era Retrospective-Analysis for Research and Applications (MERRA) (Rienecker et al. 2011), two data sets with global reach.

The WRF model is also able to assimilate available upper-air and surface-station observational data to provide increased accuracy and solution stability. While this observational nudging was used throughout the evaluation, it is not used in the simplified procedure discussed later in "Simplified Methodology" section.

Model Physics

The physical modules applied in the model were selected based on a series of sensitivity simulations to provide a solution that adequately represents the regions studied:

* Cumulus parameterization: The Kain-Fritsch scheme (Kain and Fritsch 1993) was applied to coarse WRF domains (36 and 12 km [20 and 7.5 mi]) to parameterize or represent cloud processes not able to be explicitly represented at those scales. However, this parameterization was turned off for the fine 4 km (2.5 mi) domains as, at this scale, clouds are partially resolved explicitly.

* Planetary boundary layer scheme: Planetary boundary layer (PBL) physics models estimate low-level wind, temperature, turbulence, cloud cover, and radiation. The Yonsei University (YSU) scheme (Hong et al. 2006) was selected.

* Explicit moisture scheme: The explicit moisture scheme is used to resolve cloud formation and precipitation within a grid cell for the fine grids (4 km [2.5 mi]). Here, we used the single-moment WSM6 scheme (Hong and Lim 2006).

* Radiation schemes: For longwave radiation, the rapid radiative transfer model (RRTM) (Mlawer et al. 1997) was used. For shortwave radiation, the Dudhia scheme (Dudhia 1989) was used.

* Land surface scheme: The commonly used unified Noah land surface model (Chen and Dudhia 2001)--a five-layer thermal diffusion scheme--was used to simulate surface skin temperature.

It should be noted that for a given region, it is possible that a different selection would better represent the local climate conditions. For example, the solar radiation schemes recommended here do not include the regional impact of aerosol and ozone. If solar radiation is of great importance, much more complex schemes are available. In addition, the explicit moisture scheme was tested and selected differently in the Orlando, Florida, study area to appropriately address precipitation simulation.

Model Simulation

First, a multiyear run is split into a number of three-day periods (i.e., 72-hour chunks). Each period is preceded by a 12-hour spin-up period. For example, a single period may be set up to run from January 1 to January 3, 2014. The spin-up added on the front of this simulation is a 12-hour portion (12:00 to 24:00 in December 31) that allows the model to reach a balanced state with the boundary conditions. Thus, the total run period becomes 84 hours. Beyond this spin-up time, a realistic flow field develops from which a usable portion of the simulation is generated. Splitting the simulation into fully contained, independent periods allows minimizing model cumulative (systematic) errors and running different time periods in parallel, if resources are available. That is, the simulation does not have to be run in sequence, saving a significant amount of time.


The output of the WRF simulations are a massive set of files (~1 TB/year/study area) consisting of hourly three-dimensional fields. As the ultimate goal of the methodology is to produce a time series of the relevant meteorological quantities at the surface, these three-dimensional files need to be mined. Derived products such as typical meteorological year (TMY) files or climatic data sheets following ASHRAE Handbook--Fundamentals criteria (ASHRAE 2013) can be generated from these data sets.


One of the main objectives of this project was to determine if the WRF model and proposed methodology sufficiently approximate meteorological data to ASHRAE's standards for distribution. Furthermore, this evaluation sought to provide reasonable estimates of the model's biases and errors. To address these objectives, it was necessary to complete both an operational evaluation (i.e., quantitative, statistical, and graphical comparisons) as well as a more phenomenological assessment (i.e., qualitative comparisons of observed features versus their depictions in the model).

In this section, annual simulations (2008) from multiple domains were evaluated to determine how well the proposed methodology correlates against the hourly observed meteorological data (i.e., the weather). In the "Climate Evaluation" section, we discuss using a series of eight-year simulations to assess how well the methodology correlates against the observed statistical data (i.e., the climate).

Domain Configuration

The WRF model was evaluated over eight separate domains with 4 km (2.5 mi) grid resolution within North America (see Figure 1). The eight domains shared five large 36 km (22.5 mi) grid resolution domains and five middle domains consisting of 12 km (7.5 mi) grid resolution (not shown). The smallest domains that were used in the WRF model evaluation roughly centered on New York, Orlando, New Orleans, Los Angeles, San Francisco, Denver, and Seattle have approximately 150 x 160 grid cells (600 x 640 km [360 x 394 mi]), each with 4 by 4 km (2.5 by 2.5 mi) grid resolution.

Performance Criteria

The WRF evaluation focuses on comparisons of specific modeled hourly meteorological parameters to observed hourly data. Performance criteria and benchmarks for the WRF meteorological model results were based on the evaluation protocols of Zhang et al. (2006) and Wu et al. (2008), which are recommended for the evaluation of mesoscale meteorological models such as WRF. The suggested criteria for temperature, humidity ratio, wind speed, and wind direction for both regular and complex terrains are given in Table 1. The metrics used in this study are (a) correlation coefficient (R--a measure of the linear correlation between two variables), (b) mean bias (MB--the average difference between two data sets), (c) mean gross error (MGE, a measure of the absolute difference between two data sets), and (d) root-mean-square error (RMSE).

Observational Mesonet Data

The Meteorological Assimilation Data Ingest System (MADIS) mesonet (NOAA 2014) network of meteorological stations was chosen as the observational data set given its high spatial resolution and data availability. The MADIS mesonet network has stations all over the globe that are used to monitor mesoscale severe weather events such as thunderstorms, tornadoes, and winter storms. Most importantly, as the MADIS mesonet station data were not used as an input to the WRF modeling, an unbiased evaluation of the modeling results can be performed.

The WRF model was assessed against available MADIS mesonet station data in four different geographic categories (coastal, mountain valley, mountain plateau, and major city), as presented in Table 2. The quality of mesonet data varies significantly from station to station; only 148 mesonet stations were found to contain sufficient data (>80% complete) and were located within the 4 km (2.5 mi) modeling domains. Also note that, as some stations fall into multiple geographic categories (e.g., coastal and city), there is some double counting of stations.


The WRF-predicted near-surface dry-bulb temperature at 2 m (6.5 ft) above grade was extracted and compared with mesonet observational data. Figure 2 shows box-and-whisker plots of modeled temperature performance compared against MADIS mesonet data across the four geographic categories. The boxes indicate the 25% to 75% range of the relevant statistic while the whiskers indicate the minimum and maximum values seen. The greyed regions indicate the performance criteria from Table 1.

Comparison of the model results to measured (MADIS mesonet) data is also summarized in Table 3. The model temperatures track the observations with a median correlation coefficient (R) of 0.95; the variability of the model temperatures matches the observed values well within performance criteria (average MB = -0.35[degrees]C [0.63[degrees]F]), as shown in Figure 2.

The WRF model performance (Table 3) varies between stations within the modeling domains, with the average MB ranging from -1.2[degrees]C to 0[degrees]C (-2.1[degrees]F to 0[degrees]F). The overall range of MB over all 148 measurement locations was -4.1[degrees]C to 1.9[degrees]C (-7.4[degrees]F to 3.5[degrees]F), and the overall range of mean gross error (MGE) was 0.7[degrees]C to 4.2[degrees]C (33[degrees]F to 40[degrees]F). Temperature was predicted best in simple terrain. In complex terrain, the model performed within the allowable range for MGE; for mountain plateaus, the model was within the allowable range for MB; the model on average was outside the allowable range for MB for mountain valleys (-1.2[degrees]C versus -0.2[degrees]C [-2.1[degrees]F versus -0.4[degrees]F]).

Figure 3 shows the probability distribution function for the modeled and observed data across the four geographic categories. This comparison shows that, with the exception of the mountain valley geographic category, the modeled temperature distribution follows measured temperature very well. In the mountain valley geographic category, the model has a slight cool bias. Note that due to measurement and unit conversions, observed temperatures may snap to the nearest [degrees]C or [degrees]F. Subsequent additional unit conversions can lead to temperatures preferentially spilling out of some histogram bins into adjacent bins, resulting in gaps in the histograms.


Specific humidity is defined as the mass of water vapor per unit mass of moist air in units of gram per kilogram (grain per pound). Table 4 shows the summary of WRF-predicted specific humidity performance. The meanMGEs forthe WRF 2 m (6.5 ft) specific humidity were within the 2 g/kg (14 gr/lb) threshold for every geographic categories, and three of the four geographic categories had maximum MGEs below the threshold. Only the major city geographic category had any values outside the allowable range. It can be seen that WRF produced a consistently negative (dry) bias, although the median MBs for every geographic category are within the allowable range ([+ or -] 0.75 g/kg[[+ or -] 5.25 gr/lb] for simple terrain, [+ or -] 1 g/kg [[+ or -] 7 gr/lb] for complex terrain). Additionally, the range of correlation coefficients seen (0.83-0.93) indicates that the WRF model is matching measured data well.

Figure 4 shows humidity ratio across the four modeling zones (geographic categories). The variability of the model humidity ratio matches the observed values well within performance criteria (the greyed area). The cumulative distribution function and probability distribution function for the modeled and measured humidity data across the four geographic categories can be found in ASHRAE Technical Report RP-1561 (Qiu et. al. 2015).


Table 5 shows the benchmarks for WRF-predicted wind speed at 10 m (33 ft) performance. Figure 5 illustrates WRF-modeled wind speed performance by the four geographic categories. Low wind speeds (below 1 m/s [2.2 mph]) were discarded from this comparison as weather stations return unreliable measurements at low wind speeds. The MBs for the WRF 10 m (33 ft) wind speed were slightly outside the allowable range for every geographic category (0.5 m/s [1.1 mph] for simple terrain, 1 m/s [2.2 mph] for complex). The RMSEs were all between 2.2 and 3.3 m/s (7.2 and 10.8 ft/s); every geographic category except for mountain plateaus was slightly outside the acceptable range (2 m/s [4.9 mph] for simple terrain, 2.5 m/s [5.6 mph] for complex terrain). Biases are mostly within or very close to the acceptable criteria range. The cumulative distribution function and probability distribution function for the modeled and measured wind speed data across the four geographic categories can be found in ASHRAE Technical Report RP-1561 (Qiu et al. 2015).

Wind direction is defined as the direction the wind blows from, in degrees (0 to 360) clockwise from North. Table 6 shows the allowable ranges for WRF-predicted wind direction performance. Table 6 shows that the modeled MB was within the allowable range ([+ or -] 10[degrees]) for every geographic category. MGE was within the allowable range for complex terrains (<50[degrees]), while the coastal and major city geographic categories were outside the allowable range (<30[degrees]). Biases are mostly within the acceptable criteria range.

It is not surprising that microscale wind variations (observational speed and direction) do not agree with the mesoscale data as well as temperature valves do. Many of the mesonet stations are poorly sited, being in close proximity to large structures (buildings) or vegetation (tree canopies), not located at airports or other well-situated sites. Thus, individual observational error can be large, resulting in large MGEs and RMSEs. However, when combining all of the stations in a given category or region, the variations even out, providing a satisfactory MB.

In summary, the modeling procedure meets performance criteria very well for temperature, humidity, and wind. Temperature performs the best across all four geographic categories and humidity also performs well within the criteria. Wind speed and wind direction are acceptable given the fact that 4 km (2.5 mi) WRF resolution may not be able to resolve all local wind conditions where the mesonet stations are located.


Whereas the weather evaluation concentrated on how well the methodology matched the observations on an hourby-hour basis, the focus here is how accurately the methodology can represent averages (e.g., average daily temperature in July), percentiles (e.g., 1% annual wind speed), and integrated values (e.g., average incident daily solar radiation received at the ground in January).

In order to investigate long-term climatic trends and statistics, modeling was conducted for an eight-year period, from 2005 to 2012. Eight years was chosen as this time period has been shown to provide reliable design conditions (e.g., [+ or -] 1[degrees]C [[+ or -] 1.8[degrees]F]) for most stations (Hubbard et al. 2004). The aim of the modeling is to generate data sufficient to estimate the current and future (1) parameters in ASHRAE Handbook-Fundamentals tables, with the exception of the clear-sky radiation coefficients, which require a separate modeling procedure.

We start with a similar modeling methodology as the meteorological evaluation, though here we limited the domains to four 4 km (2.5 mi) regions surrounding the cities of New York, Denver, San Francisco, and Orlando (see Figure 1). These four regions represent challenging modeling environments for differing reasons:

* New York is close to the moderating influence of the Atlantic Ocean but also sees a large seasonal contrast, including snowfall.

* Denver is immediately downstream of the Rocky Mountains and subject to rapid changes in weather.

* San Francisco is subject to westerly winds blowing over relatively cold Pacific Ocean waters, resulting in a complex meteorology (e.g., fogs).

* Orlando has many of the trademarks of a tropical climate and, due to the adjacent Gulf of Mexico to the west and the Atlantic Ocean to the east, sees robust moisture convergence and thus intense rainfall and high humidity.

Data Sets

A number of gridded data sets and observational meteorological stations were used for comparative purposes:

* NARR (2005 to 2012): Though this data set was used as both initial and boundary conditions for temperature, humidity, and wind, the precipitation and solar radiation derived in WRF is completely independent of NARR (Mesinger et al. 2006). In addition, the NARR data set was used to evaluate the consistency of the modeling effort.

* CFSR (2003 to 2010): The Climate Forecast System Reanalysis (Saha et al. 2010), Version 1, provides historical data over the period of 1979 to 2010. The more recent Version 2 provides historical data from 2011 to the present and at a higher resolution, but it was not used in this project.

* PRISM (2005 to 2012): PRISM (Daly et al. 2008) is a downscaled, gridded data set for average temperature and precipitation statistics only derived by interpolating observational stations and correcting for elevation, proximity to coastlines, and other parameters. The resolution is approximately 4 km (2.5 mi) over the continental United States.

* NREL (1998 to 2009): NREL provides gridded solar data at approximately a 10 km (6.2 mi) resolution based on the Perez model (Perez et al. 2002).

* ASHRAE Handbook-Fundamentals (1986 to 2010): The climatic data found in ASHRAE Handbook-Fundamentals (ASHRAE 2013) provides a long-term comparison.

* NSRDB (1991 to 2010): The National Solar Radiation Database (NSRDB) (Wilcox 2012) database is derived from largely the same locations and data as ASHRAE Handbook-Fundamentals yet also contains some quantities not currently found in the Handbook, including solar radiation and average wind speed.

* MADIS mesonet (2005 to 2012): The MADIS mesonet (NOAA 2014) provides a further, independent source of all variables. While there are on the order of hundreds of stations operating in each region, only approximately 10% of these stations had a sufficiently complete record over the eight years required. However, the few stations that were sufficient were of excellent quality and covered the same period as simulated.

Qualitative and Quantitative Comparison

To evaluate the modeling effort against all statistics currently found in ASHRAE Handbook-Fundamentals would be a monumental undertaking. Therefore, the comparison focused on select statistics for each of the major variables of temperature, humidity, wind, precipitation, and solar radiation. For simplicity, the closest grid point to each observational station was used without any sophisticated interpolation horizontally or any corrections based on elevation (e.g., lapse rate).

Here we focus on just a few of the results found in the full RP-1561 report (Qui et al. 2015). Figures 7 and 8 contain the predicted design cooling (1%) and heating (99%) dry-bulb temperatures for the four regions.

Evidently, there are significant variations, typically due to proximity to the coasts or urban development or due to changes in elevation. San Francisco in particular exhibits all of these climate drivers.

Comparing the WRF results for San Francisco against the available gridded and point data sets for the average daily temperature in July (see Figure 8) shows general qualitative agreement.

Figure 8 indicates that WRF modelling output, the reanalysis (CFSR and NARR), and the PRISM analysis all demonstrate similar patterns of temperature spatial distribution along the coastal area, which has lower temperatures over water (in this case, the Pacific Ocean) and higher temperatures further inland. Comparing plots (e) and (f) of Figure 8, WRF output (a) matches well with observation data. It is important to note that WRF provides the finest resolution grid coverage for the study area while ASHRAE Handbook-Fundamentals, NSRDB, and MADIS observation data can only provide climate information in limited locations (as shown in [e] and [f] of Figure 8), while NARR's and CFSR's resolutions are too coarse to be applicable for design purpose.

Quantitatively, we can calculate similar statistics as for the climatic evaluation. That is, for each observational station we calculate the summary statistic (e.g., 1% cooling dry-bulb temperature as shown in Figure 6) and pair that with the summary statistic calculated for the closest available modeled point from WRF. Then for all available pairs in the region, we calculate the mean-bias error (MB), the root-mean-square error (RMSE), and the correlation coefficient (R). To provide an estimate of the robustness of these error measures, we also provide a bootstrapped estimate of the 90% confidence intervals for these quantities. As an example of the full results we provide the error measures for the 1% and 99% dry-bulb temperatures in Figures 9 and 10, respectively. In all regions, errors are typically within 1[degrees]C (1.8[degrees]F) and well correlated. There is more uncertainty in the estimates in Denver, especially for the ASHRAE Handbook-Fundamentals observations, due to a lack of stations.

Similar results can be calculated and generated for all the elements in the ASHRAE Handbook--Fundamentals tables.

Many more are included in the RP-1561 final report (Qiu et al. 2015). A general conclusion of the results is as follows:

* Temperature values are very well validated with measurement data for all regions, and errors are typically within 1[degrees]C (1.8[degrees]F).

* Extreme values of humidity are well within performance bias criteria of approximately [+ or -] 0.5 g/kg (3.5 gr/lb) but poorly correlated with station results. Due to its proximity to its continental dry air, Denver is particularly poorly represented.

* Average and extreme (1%) values of wind speed are mostly within performance bias criteria for RMSE (2 to 2.5 m/s [6.6 to 8.2 ft/s]) but typically only semi-correlated to the WRF projections. The MADIS mesonet results were particularly poor due to their often inadequate wind siting (e.g., located in building wakes).

* Precipitation shows correlations that are consistently poor and biases of approximately [+ or -]50%. Region-wide, precipitation is typically of the right magnitude, but on a grid-by-grid basis there is poor spatial accuracy (i.e., poor forecasting skill).

* Solar radiation is typically poorly to semi-correlated and consistently overestimated by approximately 5%-15%.


The preceding sections described and evaluated a WRF usage that is unfortunately difficult to set up without advanced meteorological knowledge and significant computational resources: downloading and assimilating multiple meteorological data sets, configuring the many model options and domains, running the executables on a computer cluster, and finally collecting and postprocessing the results.

To make the modeling available to a wider audience, we are recommending the use of a precompiled, simplified modeling package called WRF Environmental Modeling System (EMS) (NOAA/NWS SOO 2014). Provided by the NOAA/NWS Science and Training Resource Center (STRC), WRF EMS provides all the functionality of WRF but with simplified installation, configuration, and execution. For example, WRF EMS will download whatever meteorological data as required and only for the region of interest.

Further, to meet the specific needs of ASHRAE members, the generation of a multiyear time surface series, the project team for RP-1561 has developed a set of scripts and a step-by-step guide available online (Klimaat 2015). The guide describes how to set up a modeling domain through a graphical interface (or through a simple script) simply from knowledge of a site's latitude and longitude. A script is available to download data as required, make the necessary changes to the physics options, split the simulation into chunks, run each chunk in sequence, and finally postprocesses the results to generate a time series at a desired location (not necessarily the original central latitude/longitude). As the scripts are simply text files written in a high-level language (Python), the user is free to modify to suit (e.g., include additional variables). Additionally, the online procedure will be maintained in the near future to respond to WRF EMS developments and provide additional functionality.

However, this simplified modeling procedures has a few limitations:

* WRF EMS lacks the observational nudging discussed previously that has been found to reduce the bias found in some of the reanalysis data sets.

* WRF EMS is best run on local workstations or small clusters, as the more complicated cluster setups typically need targeted compiling.

To evaluate the limitations of the simplified method, WRF EMS modeling was performed for New York City for 2005, focusing on the three surrounding airports (JFK, LaGuardia, and Newark). Figure 11 shows the results of this comparison for temperature, demonstrating a comforting similarity in results. Full results, including other variables (e.g., wind), can be found in the RP-1561 final report (Qiu et al. 2015).


We have provided an overview of a mesoscale modeling methodology and evaluated its performance against mesonet observed data in estimating hourly meteorology (i.e., the weather) within geographic categories. Modeling results indicate that temperature predictions are excellent for most of the geographic categories. Humidity is also predicted well within the evaluation criterion. The model performs reasonably well in wind speed and direction but with larger bias than temperature.

We also performed an eight-year modeling evaluation to address common design elements such as the 99% heating dry-bulb temperature (i.e., the climate). Detailed modeling performance evaluations from those locations (as indicated in Table 2) suggested that The WRF model is able to generate gridded localized climate data with varying accuracy. In particular, temperature, and to a lesser extent humidity and wind, demonstrate great predictive skill. However, the methodology struggles to predict precipitation or solar radiation with any added skill given the challenge meteorological models face in predicting cloud cover. It is anticipated that the more sophisticated radiation models under development will improve upon this deficiency.

Lastly, we have presented a simplified WRF modeling method based on the WRF EMS. Limited model evaluations suggested this method can generate data with comparable quality to the full WRF model. A step-by-step guide for a simplified version of the methodology has been made available online (Klimaat 2015).

There are numerous potential applications for these modeling results. For example, while the focus of the procedure is the generation of a time series of data for a single location, it is often useful to examine the variability of a given climate parameter within a region (e.g., climate maps of the 99% heating dry-bulb temperature for San Francisco, Figure 7). Generally, this level of detail cannot be determined from either observational data sets or reanalysis data sets at the resolutions currently available or planned.

In summary, this study has developed a mesoscale modeling approach to generate high-resolution climate data (onsite), particularly for where ASHRAE Handbook-Fundamentals climate data are unavailable. Looking forward, it is anticipated that with further developments to model physics (such as the radiation scheme), WRF EMS and the online guide (Klimaat 2015) will produce even higher-quality data.


The authors thank all members of the RP-1561 Project Monitoring Subcommittee for their guidance during the course of the project. The authors also acknowledge Jason Slusarczyk of Novus Environmental for his effort on project technical support and review.


ASHRAE. 2013. Chapter 14, Climatic Design Information. ASHRAE Handbook-Fundamentals. Atlanta: ASHRAE.

Chen, F., and J. Dudhia. 2001. Coupling an advanced land-surface/hydrology model with the Penn State/NCAR MM5 modeling system. Part I: Model description and implementation. Monthly Weather Review 129:569-85.

Daly, C., M. Halbleib, J.I. Smith, W.P. Gibson, M.K. Doggett, G.H. Taylor, J. Curtis, and P.P. Pasteris. 2008. Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States. International Journal of Climatology 28:2031-64.

Dudhia, J. 1989. Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model. Journal of the Atmospheric Sciences 46:3077-107.

Hong, S.Y., and J.O.J. Lim. 2006. The WRF single-moment 6-class microphysics scheme (WSM6). Journal of the Korean Meteorological Society 42:129-51.

Hong, S.Y., Y. Noh, and J. Dudhia. 2006. A new vertical diffusion package with explicit treatment of entrainment processes. Monthly Weather Review 134:2318-41.

Hubbard, K., K. Kunkel, A. DeGaetano, and K. Redmond. 2004. Sources of uncertainty in the calculation of the design weather conditions in ASHRAE Handbook-Fundamentals. Research Report RP-1171. Atlanta: ASHRAE.

Kain, J.S., and J.M. Fritsch. 1993. Convective parameterization for mesoscale models: The Kain-Fritsch scheme. The representation of cumulus convection in numerical models. Boston: American Meteorological Society, pp. 165-70.

Klimaat. 2015. Generating site-specific meteorological data using mesoscale modeling.

Mesinger, F., G. DiMego, E. Kalnay, K. Mitchell, P.C. Shafran, W. Ebisuzaki, D. Jovic, Jack Woollen, E. Rogers, E.H. Berbery, M.B. Ek, Y. Fan, R. Grumbine, W. Higgins, H. Li, Y. Lin, G. Manikin, D. Parrish, and W. Shi. 2006. North American regional reanalysis. Bulletin of the American Meteorological Society 87(3):343-60.

Mlawer, E.J., S.J. Taubman, P.D. Brown, M.J. Iacono, and S.A. Clough. 1997. Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. Journal of Geophysical Research 102(D14):16663-82.

NOAA. 2014. NCEP Central Operations. Meteorological Assimilation Data Ingest System. https://madis

NOAA/NWS SOO. 2014. NWS SOO, Science & Training Resource Center. SOO STRC. http://

Perez, R. J. Schlemmer, K. Hemker, S. Kivalov, A. Kankiewicz, and C. Gueymard. 2002. A new operational satellite-to-irradiance model. Solar Energy 73(5):307-17.

Qiu, X., M. Roth, F. Yang, and H. Hains. 2015. Procedure to adjust observed climatic data for regional or mesoscale climatic variations. Research Report RP-1561. Atlanta: ASHRAE.

Rienecker, M.M., M.J. Suarez, R. Gelaro, R. Todling, J. Bacmeister, E. Liu, M.G. Bosilovich, S.D. Schubert, L. Takacs, G.K. Kim, S. Bloom, J. Chen, D. Collins, A. Conaty, A.D. Silva, W. Gu, J. Joiner, R.D. Koster, R. Lucchesi, A. Molod, T. Owens, S. Pawson, P. Pegion, C. R. Redder, R. Reichle, F.R. Robertson, A.G. Ruddick, M. Sienkiewicz, and J. Woollen. 2011. MERRA: NASA's Modern-era retrospective analysis for research and applications. Journal of Climate 24:3624-48.

Saha, S., S. Moorthi, H.L. Pan, X. Wu, J. Wang, S. Nadiga, P. Tripp, R. Kistler, J. Woollen, D. Behringer, H. Liu, D. Stokes, R. Grumbine, G. Gayno, J. Wang, Y.T. Hou, H.Y. Chuang, H.M.H. Juang, J. Sela, M. Iredell, R. Treadon, D. Kleist, P.V. Delst, D. Keyser, J. Derber, M. Ek, J. Meng, H. Wei, R. Yang, S. Lord, H.V.D. Dool, A. Kumar, W. Wang, C. Long, M. Chelliah, Y. Xue, B. Huang, J.K. Schemm, W. Ebisuzaki, R. Lin, P. Xie, M. Chen, S. Zhou, W. Higgins, C.Z. Zou, Q. Liu, Y. Chen, Y. Han, L. Cucurull, R.W. Reynolds, G. Rutledge, and M. Goldberg. 2010. The NCEP climate forecast system reanalysis. Bulletin of the American Meteorological Society 91(8):1015-57.

Skamarock, W.C., J.B. Klemp, J. Dudhia, D.O. Gill, D.M. Barker, M.G. Duda, X.Y. Huang, W. Wang, and J.G. Powers. 2008. A description of the advanced research WRF Version 3. NCAR/TN-475+STR, NCAR Technical Note. Boulder, CO: NCAR.

Wilcox, S. 2012. National Solar Radiation Database 19912010 update: User's manual. NREL Technical Report NREL/TP-5500-54824. Golden, CO: National Renewable Energy Laboratory.

Wu, S.Y., S. Krishnan, Y. Zhang, and V. Aneja. 2008. Modeling atmospheric transport and fate of ammonia in North Carolina, Part I. Evaluation of meteorological and chemical predictions. Atmospheric Environment 42:3419-36.

Zhang, Y., P. Liu, B. Pun, and C. Seigneur. 2006. A comprehensive performance evaluation of MM5eCMAQ for summer 1999 southern oxidants study episode, Part I. Evaluation protocols, databases and meteorological predictions. Atmospheric Environment 40:4825-38.


Michael Case, Program Manager, U.S. Army ERDC CERL, Champaign, IL: 1) Were TMY3 files generated for the WRF data presented? 2) Were future climate change scenarios studied?

Xin Qiu: 1) Yes. WRF data can be used to generate site-specific TMY3 data, as long as the study site is in the modeling domain. We suggest using minimum 8-year WRF model data, and longer is better. 2) No, this project only dealt with historical climate. However, the model has capability to simulate future climate scenarios.

Xin Qiu, PhD


Hamish Corbett-Hains, PEng

Michael Roth, PhD, PEng


Fuquan Yang, PhD

(1) Monthly average daily all-sky solar radiation (i.e., including clouds) is a planned addition to the next edition of the ASHRAE Handbook.

Xin Qiu is aprincipal, Hamish Corbett-Hains is an air quality engineer, and Fuquan Yang is an air quality and meteorology modeler at Novus Environmental Inc., Guelph, Ontario, Canada. Michael Roth is a director at Klimaat Consulting & Innovation Inc., Guelph, Ontario, Canada.

Table 1. Performance Criteria for WRF Meteorological Model

Terrain               Wind Speed          Wind Direction

                        RMSE:                  MGE:
Regular terrain    <2 m/s (4.5 mph)        <30[degrees]

  (coastal,              MB:                    MB:
  major city)     <[+ or -] 0.5 m/s    <[+ or -] 10[degrees]
                       (1.1 mph)
                        RMSE:                  MGE:
Complex terrain   <2.5 m/s (5.6 mph)       <50[degrees]
  mountain               MB:                    MB:
  plateau)        <[+ or -] 1.0 m/s    <[+ or -] 10[degrees]
                      (2.2 mph)

Terrain                 Temperature                Humidity

                            MGE:                     MGE:
Regular terrain        <2[degrees]C           <2 g/kg (14 gr/lb)
  (coastal,                 MB:                      MB:
  major city)     < [+ or -] 0.5[degrees]C   < [+ or -] 0.75 g/kg
                      (0.5[degrees]F)            (5.25 gr/lb)
                            MGE:                     MGE:
Complex terrain       <3.5[degrees]C          <2 g/kg (14 gr/lb)
  (mountain           (6.3[degrees]F)
  mountain                  MB:                      MB:
  plateau)        <[+ or -] 1.0[degrees]C     <[+ or -] 1 g/kg
                      (1.8[degrees]F)             (7 gr/lb)

MB: mean bias, MGE: mean gross error, RMSE: root-mean-square error

Table 2. Categories of Cites for Mesonet/WRF Comparison

City           Number     Coastal   Mountain   Mountain   Major
                 of                  Valley    Plateau    City

Boston            13         X                              X
Colorado          11                              X
Denver            3                               X
Harrisburg        12                   X
Los Angeles       15         X                              X
New Orleans       13         X                              X
New York          13         X                              X
Orlando           5                                         X
Palm Springs      10                   X
Sacramento        5                               X
San               9                    X
San               8          X                              X
Seattle           2          X                              X
Toronto           14         X                              X
Vancouver         15         X         X                    X

Table 3. WRF Temperature Performance Versus Mesonet across
Four Geographic Categories

Geographic         MB, [degrees]C   MGE, [degrees]C    R
Categories          ([degrees]F)     ([degrees]F)

Coastal              0.0 (0.0)         1.5 (2.6)      0.95
Mountain valley     -1.2 (-2.1)        2.1 (3.8)      0.96
Mountain plateau    -0.2 (-0.4)        2.0 (3.6)      0.97
Major city           0.0 (0.0)         1.5 (2.7)      0.95

Table 4. WRF Humidity Ratio Performance Versus
Mesonet across Four Geographic Categories

Geographic       MB,            MGE,        R
Categories   g/kg (gr/lb)   g/kg (gr/lb)

Coastal      -0.2 (-1.7)     0.9 (6.1)     0.92
Mountain     -0.5 (-3.4)     1.0 (6.8)     0.87
Mountain     -0.6 (-4.5)     1.1 (7.6)     0.83
Major city   -0.3 (-2.1)     1.0 (6.8)     0.93

Table 5. Summary of WRF Wind Speed Performance

Climate               MB,        RMSE       R
Zone               m/s (mph)   m/s (mph)

Coastal            1.0 (2.2)   2.2 (5.0)   0.41
Mountain valley    2.1 (4.7)   3.3 (7.4)   0.38
Mountain plateau   1.4 (3.0)   2.4 (5.4)   0.37
Major city         1.1 (2.4)   2.3 (5.1)   0.40

Table 6. Summary of WRF Wind Direction

Climate            MB    MGE     R

Coastal            3.8   39.9   0.40
Mountain valley    4.6   45.2   0.38
Mountain plateau   2.9   49.8   0.29
Major city         2.3   40.2   0.39
COPYRIGHT 2016 American Society of Heating, Refrigerating, and Air-Conditioning Engineers, Inc. (ASHRAE)
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Qiu, Xin; Corbett-Hains, Hamish; Roth, Michael; Yang, Fuquan
Publication:ASHRAE Transactions
Article Type:Report
Geographic Code:100NA
Date:Jul 1, 2016
Previous Article:A simulation-based study on different control strategies for variable-speed pumps in distributed ground-source heat pump systems.
Next Article:Modeling and testing multiple precooling strategies in three residential building types in the phoenix climate.

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |