9th Annual WRF Users’ Workshop, Boulder, CO, 23-27 June 2008
Using WRF for Parameterization Development: The Making of ECPP for MMF Models William I. Gustafson Jr., Richard C. Easter, Larry K. Berg, and Steven J. Ghan Atmospheric Science and Global Change Division Pacific Northwest National Laboratory, Richland, WA 1. Introduction The Weather Research and Forecasting (WRF) model (Skamarock et al. 2005) is a powerful tool for model parameterization development. The ability to easily vary grid resolution and alter the domain size makes WRF particularly attractive because one is not limited to parameterizations solely suited for mesoscale grid sizes; one can also work with coarser, global climate model-like spacing. With relatively easy modifications WRF can also be converted to work as a single column model, which allows parameterizations developed in 3-D to be tested in varying frameworks. Additionally, the chemistry features in WRF account for both cloud processing of aerosols and the effect of aerosols on cloud properties, which allows researchers to focus on parameterizations addressing some of the key uncertainties hampering climate change modeling. One example of how these features have been utilized in WRF is the development of the ExplicitCloud Parameterized-Pollutant (ECPP) parameterization (Gustafson et al. 2008) that is intended for use with Multiscale Modeling Framework (MMF) models (Grabowski 2001; Randall et al. 2003). 2. E C PP and M M F MMF is a modeling technique where most of the traditional physics parameterizations in a global climate model (GCM) are removed, e.g. clouds and radiation, and are replaced by an embedded cloud resolving model (CRM). The motivation for this is that the CRM can be run at a high resolution and, in theory, can more accurately reproduce the cloud fields within the GCM column so that the cumulus parameterization is no longer needed. This CRM can be configured in either 2-D or 3D modes depending on the available computational resources. While the MMF technique adds a significant computational cost compared to the traditional GCM framework, on the order of 200 times (Randall et al. 2003), it is still significantly less than what would be required to run the GCM at a fully cloud-resolving scale for the entire globe where the added cost would be on the order of a factor of 106. Computers today are able to handle simulations using the MMF technique, but fully global CRMs will not be feasible for climate simulations for the foreseeable future. Corresponding author:
[email protected]
FIG. 1. Schematic depiction of ECPP showing linkages between ECPP in relation to the host GCM and embedded CRMs in the MMF. Large-scale advective tendencies of tracer species (subscript a), moisture species (subscript m), and temperature ( θ ) are provided by the GCM to ECPP and the CRM. ECPP and the CRM exchange information on mass (q), number (N), cloud mass flux (Mc ), and precipitation (P), which among other things, are used to € fractions of area within the GCM column that determine are assigned to updraft, downdraft, and quiescent draft classes by level. In turn, ECPP and the CRM provide vertical advection and source–sink terms for sub-GCM scales for tracer species (ECPP) and moisture species and temperature (CRM), which are returned to the GCM.
While the computational cost of the meteorology is currently feasible for the MMF, the cost of treating chemistry and aerosols in the CRMs of the MMF is not. Depending on the complexity of the aerosol module, anywhere from a dozen to several hundred aerosol species are added to the model, with most applications limited to the lower end due to cost considerations. So, an alternative is to carry the aerosol and other chemistry species on the coarse GCM grid and to use the added cloud information from the CRMs to parameterize interactions between the aerosols and clouds. The Explicit-Cloud Parameterized Pollutant (ECPP) parameterization (Gustafson et al. 2008) is being developed to specifically address this possibility. Figure 1 schematically depicts how ECPP works for a given GCM column. The GCM portion of the model is responsible for horizontal transport of tracers and largescale, grid resolved vertical transport. (Note that the term tracer is used generically here to represent both inert and non-inert tracer species.) The large-scale tendencies for
9th Annual WRF Users’ Workshop, Boulder, CO, 23-27 June 2008
the tracers are passed to ECPP and the large-scale tendencies for the moisture and other dynamical variables are passed to the embedded CRM. The cloud mass flux and other cloud properties are passed from the CRM to ECPP, and aerosol number and mass are passed from ECPP to the CRM. With this information exchanged, ECPP calculates the sub-GCM-grid vertical transport tendency perturbations, affects of the clouds on the aerosol such as wet scavenging and aqueous chemistry, and the activation of aerosol into cloud droplets. The CRM uses the added aerosol information to modify its treatment of the clouds in order to represent the aerosol indirect effects on cloud albedo and lifetime. Finally, the grid-cell bulk results due to the sub-GCM-grid perturbations are passed back to the GCM. 3. D evelopm ent m ethodology for E C PP Since the cost of running a global MMF model is so high, development of ECPP would be prohibitive if done within the full MMF model. So, the chemistry version of WRF with interactive aerosol-cloud interactions (Grell et al. 2005; Fast et al. 2006; Gustafson et al. 2007) has been used as an alternative. WRF has been configured to represent both the GCM column and the embedded CRM within an MMF grid column. Two WRF simulations are performed. The first is a cloud-resolving domain with doubly-periodic lateral boundaries and high horizontal resolution. This simulation (WRF-CRM) serves two purposes. First, it is a control to indicate what the tracer transport characteristics should be when processed through ECPP. Second, it is used to represent the CRM in MMF by passing the relevant meteorological variables from this domain to the second WRF simulation. The second WRF setup is as a single column model (SCM) with the phyiscs parameterizations replaced by the ECPP parameterization (WRFECPP) and the large-scale forcing derived from WRFCRM. The handling of the tracers in WRF-ECPP is by ECPP and is independent of the tracers in the WRFCRM domain. So, for a perfect parameterization, the domain averaged tracer profile in WRF-CRM would be identical to the profile produced by WRF-ECPP. Currently, the methodology for handling tracer transport has been developed for ECPP (Gustafson et al. 2008) and the handling of aerosol-cloud interactions is under development. The single-column version of the Community Atmospheric Model (SCAM) (Collins et al. 2006; http://www.ccsm.ucar.edu/models/atm-cam/docs/scam)
FIG. 2. Rain rate time series for University of Washington radar-based observations (blue),WRF-CRM (black), and SCAM (dashed red). Units are mm h − 1 .
is also used for comparison with WRF-CRM and WRFECPP. SCAM has been configured to use the same largescale forcing as WRF-CRM and the same initial tracer profiles have been used. This enables comparisons between SCAM, WRF-CRM, and WRF-ECPP to see how a conventional GCM would handle tracers versus the ideal 3-D setup, and then also with a facsimile of what an MMF model with ECPP would do. 4. Tracer transport results Initial testing of ECPP has been done using a largescale forcing dataset from the Kwajalein Experiment (KWAJEX) (e.g. Yuter et al. 2005) in the Republic of the Marshall Islands. The period 17–31 August 1999 was simulated with the period up through 18 August 18 UTC neglected due to model spin-up. WRF-CRM is configured with 122x122x40 grid points, 2-km horizontal grid spacing, and approximately 0.5 km vertical grid spacing. Comparison of the precipitation rate from gauge adjusted radar observations, WRF-CRM, and SCAM are shown in Figure 2. WRF-CRM is able to capture the timing of most major precipitation events, although it tends to over estimate their intensity. The SCAM precipitation is not as good, but the precipitation rates are on par with the small to medium precipitation events. Note that in order for the WRF-CRM results to be close to accurate, it was critical for positive-definite advection to be used, both for the cloud and tracer species. Substantial additional cloud and tracer was generated when using the default non-positive definite advection algorithm. For more details, refer to Chapman et al. (2008).
9th Annual WRF Users’ Workshop, Boulder, CO, 23-27 June 2008
FIG. 3. Tracer concentration for WRF-CRM (black), WRF-ECPP (blue), and SCAM (red) for cases with both a high-level tracer and deep convection (A) and shallow convection (B); and a low-level tracer for cases with deep convection (C), and shallow convection (D). The values represent composited profiles at 12 UTC, as described in the text. The horizontal bars indicate the standard deviation between the days within the composite. Note the split horizontal axes for clarity outside of the tracer injection level. Also, a minor adjustment to the vertical levels was done to aid in differentiating overlapping standard deviation bars.
Ten inert tracer profiles were initialized in each model with each tracer set to one over a depth of 100 hPa and zero elsewhere; tracer one was 1 from the surface to 900 hPa, tracer two was 1 from 900–800 hPa, etc., up to the model top. In order to easily quantify how the tracer transport differed during different rain conditions, the tracer profiles were reinitialized each morning at 18 UTC and then allowed to freely develop over the following 24 hours. Results of the tracer mixing are presented for two tracer categories: low-level tracers based on the tracer initialized to 1 between the surface and 900 hPa, and highlevel tracers initialized to 1 between 500 and 400 hPa. Each category is composited based on the tracer profiles at 12 UTC either as a no convection, shallow convection, or deep convective case. This is done using thresholds of 0 and 0.01 ppbv of low-level tracer within the layer between 400 and 200 hPa at 12 UTC. The majority of the days in the study period have deep convection, but three days in the WRF-CRM simulations and five days in the SCAM simulation are classified as cases of shallow convection, and two days in each simulation have no convection.
Figure 3 shows these composites for the WRF-CRM, WRF-ECPP and SCAM simulations. When coupled with ECPP, the SCM version of WRF is able to better match the vertical transport of the low-level tracer with a smooth profile containing tracer mass from the surface up to the tropopause, similar to the 3-D WRF run. This is due to the extra information available to ECPP based on the range of cloud heights present in the CRM, as opposed to only the parameterized cloud height distribution available in SCAM. SCAM tends to remove the tracer from the surface and detrain too much of it near the tropopause instead of distributing it throughout the depth of the troposphere. The tendency for ECPP to distribute tracer mass throughout the depth of the profile is also evident in the high-level tracer, with the region below about 550 hPa much more accurately reproduced in WRF-ECPP than in SCAM. 5. D iscussion The experience of using WRF to develop ECPP has shown the value of having a model that is easily configurable to run in a wide range of configurations. Outside of the coding changes necessary to handle the actual ECPP
9th Annual WRF Users’ Workshop, Boulder, CO, 23-27 June 2008
parameterization, very few code modifications have been necessary in WRF for this project. For WRF-CRM the main changes involved inputting the large-scale forcing tendencies for advection (potential temperature, water vapor, and sea surface temperature) and setting up the tracers. The tracer setup mainly involved defining an appropriate package in the WRF Registry containing the extra chemistry variables and then forcing their reinitialization at a set time. Code was also added to generate time-averaged output for some variables, such as vertical velocity, between history outputs, and to output certain diagnostic domain averaged profiles for use in comparing against the WRF-ECPP output. Code changes required for WRF-ECPP are more substantial. Code is needed to read in horizontally-averaged statistics from the WRF-CRM simulation and to initialize the tracers every 24 h. The solve_em module is largely “gutted”. The dynamical (U, V, & W), state (T & P), and moisture species are now set to the WRF-CRM horizontal averages. The tendencies for chem species (currently the tracers) include vertical advection by the horizontallyaveraged vertical velocity (which is zero with the periodic boundaries) and any tendencies calculated in chem_driver. As no physics parameterizations are used in WRF-ECPP, all physics driver calls in solve_em become extraneous. Since the differences between the 3-D and 1-D versions of WRF are isolated almost entirely to the registry files and solve_em, converting the model between modes is done simply through the setting of a switch when running “configure” before compiling. Two versions of solve_em.F are maintained (solve_em.3d.F and solve_em.1d.F) and the appropriate one is symbolically linked into place. Changes in the registry are handled through a chemistry package choice and other namelist settings. The whole process works very smoothly. The only other software necessary is an offline post-processor for the WRF-CRM output to generate statistics and forcing tendencies similar to what would be available to ECPP from an MMF CRM. In conclusion, WRF is a very useful tool for developing physics parameterizations for both regional and global models. Particularly with the new global domain capabilities released in WRF v3.0 one can develop and test a parameterization on a small, cost-efficient domain and then with a simple configuration adjustment test the parameterization over large regions and over a range of grid spacings. Acknowledgments. We wish to thank Stacy Brodzik and Robert Houze of the University of Washington for their assistance with the rain rate data, and Vaughn Phillips of Princeton University for the KWAJEX large-scale forcing data. This manuscript has been authored by Bat-
telle Memorial Institute, Pacific Northwest Division, as part of the US Department of Energy Atmospheric Science Program under contract No. DE-AC0576RL01830. This work was also supported by the NASA Interdisciplinary Science Program under grant NNX07AI56G.
REFERENCES Chapman, E. G., W. I. Gustafson, R. C. Easter, J. C. Barnard, S. J. Ghan, M. S. Pekour, and J. D. Fast, 2008: Coupling aerosol-cloud-radiative processes in the WRF-Chem model: Investigating the radiative impact of elevated point sources. Atmos. Chem. Phys., submitted. Fast, J. D., W. I. Gustafson, R. C. Easter, R. A. Zaveri, J. C. Barnard, E. G. Chapman, G. A. Grell, and S. E. Peckham, 2006: Evolution of ozone, particulates, and aerosol direct radiative forcing in the vicinity of houston using a fully coupled meteorologychemistry-aerosol model. J. Geophys. Res., 111, D21305, doi:10.1029/2005JD006721. Grabowski, W. W., 2001: Coupling cloud processes with the large-scale dynamics using the cloud-resolving convection parameterization (CRCP). J. Atmos. Sci., 58, 978-997. Grell, G. A., S. E. Peckham, R. Schmitz, S. A. McKeen, G. Frost, W. C. Skamarock, and B. Eder, 2005: Fully coupled "online" chemistry within the WRF model. Atmos. Environ., 39, 6957-6975. Gustafson, W. I., E. G. Chapman, S. J. Ghan, R. C. Easter, and J. D. Fast, 2007: Impact on modeled cloud characteristics due to simplified treatment of uniform cloud condensation nuclei during neaqs 2004. Geophys. Res. Lett., 34, L19809, L19809, doi:10.1029/ 2007GL0300321. Gustafson, W. I., L. K. Berg, R. C. Easter, and S. J. Ghan, 2008: The Explicit-Cloud Parameterized-Pollutant hybrid approach for aerosol-cloud interactions in multiscale modeling framework models: Tracer transport results. Environ. Res. Lett., 3, doi:10.1088/1748-9326/3/2/025005. Randall, D., M. Khairoutdinov, A. Arakawa, and W. Grabowski, 2003: Breaking the cloud parameterization deadlock. Bull. Amer. Meteor. Soc., 84, 15471564. Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005: A description of the advanced research WRF version 2, 88 pp. Yuter, S. E., R. A. Houze, E. A. Smith, T. T. Wilheit, and E. Zipser, 2005: Physical characterization of tropical oceanic convection observed in kwajex. J. Appl. Meteor., 44, 385-415.
