DYNAMICAL DOWNSCALING OF COUPLED MODEL HISTORICAL RUNS

FINAL REPORT FOR PROJECT 1.5.4 DYNAMICAL DOWNSCALING OF COUPLED MODEL HISTORICAL RUNS PRINCIPAL INVESTIGATOR: DR. JOHN MCGREGOR, CSIRO Marine and Atmospheric Research, John.McGregor@csiro.au, Tel: 03 9239 4400, Fax: 03 9239 4444, Address: PB1 Aspendale, Vic. 3195 CO-AUTHOR: Dr. Kim Nguyen, CMAR. Date: January 2008

Abstract A 40-year simulation has been performed with a resolution of about 20 km over the GRDC region, using CCAM for downscaling from a Mk 3.0 coupled GCM simulation. The CCAM present-day climatology shows realistic detail compared to the coarser Mk 3.0 simulation. Over the GRDC region, seasonal rainfall biases are generally small. The simulated average daily maximum and minimum temperatures are found to be acceptable; the maximum temperatures are up to 3 degrees too cool in autumn and winter; the minimum temperatures are a little warm, except in Victoria in winter when they are up to 2 degrees too cool. The interannual variability of the mean annual rainfall of the simulation over the GRDC region is qualitatively similar to the observations, in regard to peak values and number of wet and dry episodes. The simulation produces an increasing trend of rainfall, whereas the observations show little trend over this period. The simulation also calculates an estimate of pan evaporation. No trend is found for pan evaporation over the GRDC region for the simulation. Date: January 2008 Project Objectives: Using CCAM, dynamically downscale the CSIRO Mk 3 GCM for a period corresponding to 1961-2000. Compare the simulated climatology, in particular rainfall and temperature, against observations taken within the study region Produce derived fields (including surface fluxes and measures of potential evaporation) and compare them against available observations and observed trends 1. Model description CCAM is formulated on the quasi-uniform conformal-cubic grid. The CCAM regional climate simulations for SEACI use a C72 global grid (6 x 72 x 72 grid points). By using the Schmidt (1977) transformation with a stretching factor of 0.15, CCAM then achieves a fine resolution of 20 km for the central panel which is located over eastern Australia. The grid is illustrated in Figure 1. The previous SEACI Project 1.3.6 described experiments using CCAM to downscale NCEP reanalyses from 1951-2000. This project follows on from that project, using similar techniques to downscale from the Mk 3.0 coupled climate model, for model years corresponding to 1961-2000. This report is laid out in a similar manner to that of Project 1.3.6. The dynamical formulation of CCAM includes a number of distinctive features. The model is hydrostatic, with two-time-level semi-implicit time differencing. It employs semi-lagrangian horizontal advection with bi-cubic horizontal interpolation (McGregor, 1993; McGregor, 1996), in conjunction with total-variation-diminishing vertical advection. The grid is unstaggered, but the winds are transformed reversibly to/from C-staggered locations before/after the gravity wave calculations, providing improved dispersion characteristics (McGregor, 2005a). Three-dimensional Cartesian representation is used during the calculation of departure points, and also for the

advection or diffusion of vector quantities. Further details of the model dynamical formulation are provided by McGregor (2005b). CCAM includes a fairly comprehensive set of physical parameterizations. The GFDL parameterization for longwave and shortwave radiation (Schwarzkopf and Fels, 1991) is employed, with interactive cloud distributions determined by the liquid and icewater scheme of Rotstayn (1997). The model employs a stability-dependent boundary layer scheme based on Monin-Obukhov similarity theory (McGregor et al., 1993), together with non-local vertical mixing (Holtslag and Boville, 1993) and also enhanced mixing of cloudy boundary layer air (Smith, 1990). A canopy scheme is included, as described by Kowalczyk et al. (1994), having six layers for soil temperatures, six layers for soil moisture (solving Richard's equation), and three layers for snow. CCAM also includes a simple parameterization to enhance sea surface temperatures under conditions of low wind speed and large downward solar radiation, affecting the calculation of surface fluxes. The cumulus convection scheme uses the mass-flux closure described by McGregor (2003), and includes both downdrafts and detrainment. Figure 1. The C72 conformal-cubic grid used for the CCAM simulations. 2. Simulation design The current project uses CCAM to downscale from the Mk 3.0/M20 simulation, run for years corresponding to 1961-2000. After 2000, the Mk 3.0 simulation branches into two simulations corresponding to the A2 and A1B scenarios. This report describes the present-day part of the simulation, corresponding to 1961-2000. It should be noted that the Mk 3.0 model does not employ flux correction; hence there

are some biases of sea-surface temperatures (SSTs), up to 2 degrees near Australia, compared to observations. These average monthly biases are shown for a 30-year period (1971-2000) for January and July, in Figure 2. The CCAM simulation uses daily SSTs from Mk 3.0, but with this monthly two-dimensional bias first subtracted. Sea-ice distributions are interpolated directly from the daily values of Mk 3.0; this simulation was submitted by CMAR to the Fourth Assessment Report of IPCC. Figure 2. SST biases in the Mk 3.0 coupled climate simulation, corresponding to 1971-2000 for January (top) and July (bottom). For the prior simulation of Project 1.3.6 downscaling from NCEP reanalyses, global nudging of winds above 500 hpa from the large-scale fields was employed, whilst outside the central high-resolution panel, gradually-increasing far-field nudging was also employed for MSL pressures and winds between 900 hpa and 500 hpa. This technique was adopted to help ensure that the north-south shifts of the jet-stream were captured. Likewise, we wish to have jet-stream locations in this CCAM simulation that are similar to those of the Mk 3.0 simulation. A new digital-filter technique is used in the present project, whereby large-scale features of MSL pressure and the winds above 500 hpa are similar to those of Mk 3.0; large-scale here is specified as

a length scale approximately the width of NSW. The digital filtering is efficiently achieved by a sequence of three one-dimensional passes, two of which are shown schematically (for a 60 km grid) in Figure 3, performed twice-daily. The model output was saved twice per day at 00 GMT and 12 GMT. These data have been post-processed onto a 0.2 degree grid covering 130E - 160E and 50S - 5S. Many prognostic and diagnostic fields have been saved. Figure 3. Schematic showing two one-dimensional passes of the Gaussian digitalfilter. 3. Seasonal averages of rainfall and temperature The 40-year (1961-2000) monthly-mean CCAM rainfall was averaged to produce seasonal averages for December, January and February (DJF), March, April and May (MAM), June, July and August (JJA) and September, October and November (SON). These averages are compared against the observed seasonal rainfall, and the maximum and minimum temperatures provided by the Bureau of Meteorology. In general, CCAM simulates well the mean seasonal rainfall patterns (Figure 4). The large rainfall along the eastern coast is well captured for all seasons. The large rainfall along the Great Dividing Range and the eastern coast is also well captured, except in winter when it is deficient. Inland of the Great Dividing Range, summer rainfall is well captured, autumn and winter are somewhat lower than observed, whereas spring rainfall is somewhat larger than observed over inland New South Wales and Queensland. Seasonal rainfall biases are shown in Figure 5 for both CCAM and the host Mk3.0 simulations. For autumn, winter and spring the biases of Mk3.0 and CCAM are both generally quite small, with CCAM biases being a little smaller near the coastline. For summer, Mk3.0 has a rather large wet bias of 1-2 mm/day over much of New South Wales and Queensland, whereas the CCAM biases are again quite small.

Figure 4. Seasonally-averaged rainfall (mm/day), with observations (top) and CCAM (downscaled from Mk3.0, bottom). Figure 5. Bias of seasonally-averaged rainfall (mm/day) for Mk3.0 (top) and CCAM (bottom).

Figure 6. Seasonally-averaged maximum temperatures (degrees C), with observations (top) and CCAM (bottom), compared to observations. The seasonally-averaged daily-maximum surface air temperatures are shown for observations and CCAM in Figure 6. The maximum temperature patterns are generally well simulated for all seasons; terrain effects are well captured. Maximum temperature biases are shown in Figure 7 for both Mk3.0 and CCAM. Over New South Wales and Queensland the CCAM biases are generally less than 1 degree, whereas Mk3.0 maximum temperature biases are up to 5 degrees too cool, especially in autumn and winter. In summer and spring, maximum temperatures over Victoria from CCAM are up to 2 degrees too warm.

Figure 7. Bias of seasonally-averaged maximum temperatures (degrees C) for Mk3.0 (top) and CCAM (bottom), compared to observations. The seasonally-averaged daily-minimum surface air temperatures are shown for observations and CCAM in Figure 8. The minimum temperature patterns are also generally well simulated for all seasons, and again terrain effects are well captured. Minimum temperature biases are shown in Figure 9 for both Mk3.0 and CCAM. Mk3.0 minimum temperature biases (typically up to 1 degree too cool) are now generally smaller than for CCAM (up to 2 or 3 degrees too warm) for most seasons, and especially summer and spring. During summer, both Mk3.0 and CCAM are too warm over the southern half of the MDB

Figure 8. Seasonally-averaged minimum temperatures (degrees C), with observations (top) and CCAM (bottom). Figure 9. Bias of seasonally-averaged minimum temperatures (degrees C) for Mk3.0 (top) and CCAM (bottom), compared to observations.

4. Interannual variability of rainfall and pan evaporation The interannual variability of the simulation is illustrated by means of time series of the annual averages for rainfall and pan evaporation. The averages are taken over land points for two regions: Region 1: the GRDC southern region, 134E - 154E and 39S - 32S Region 2: a larger MDB region, 141E - 154E and 39S - 25S. Figure 10 shows the annual rainfall for region 1. While the individual high and low rainfall years do not coincide, this is not to be expected as the model is using GCM years rather than actual years. However, the peak values are similar. Both observations and CCAM have about 6 episodes with rather low (El Nino-like) rainfall, and about 6 episodes with higher rainfall. The simulation shows an increasing trend (2% per decade) for rainfall during this period, whereas the observations show no trend. The situation is similar for Region 2 (Figure 11); this region has generally higher average rainfall than Region 1, as it includes more of the tropical part of Australia. 3 2.5 Reg1 Rain CCAM Obs 2 1.5 1 1970 1980 1990 2000 Figure 10. Time series of observed and simulated annual rainfall for region 1 (the southern GRDC region) for 1961-2000. CCAM is shown for GCM model years, whilst OBS refers to actual years. Trend lines are also shown for CCAM (increasing trend) and observed (little trend).

3 2.5 Reg2 Rain CCAM Obs 2 1.5 1 1970 1980 1990 2000 Figure 11. As for Figure 10, but showing the time series for the larger region 2. Plotted for 1961-2000 observed and simulated. CCAM is shown for GCM model years, whilst OBS refers to actual years. Trend lines are also shown for CCAM (increasing trend) and observed (little trend). In the CCAM simulation, we also explicitly calculate pan evaporation, by predicting at each time step the water temperature for a pan located at each grid point. Sensible and latent heat fluxes are calculated as part of this process. Figure 12 shows the time series for the annual pan evaporation averaged over both regions. The Figure also shows the pan evaporation for 1961-2000 from our prior simulation downscaling from NCEP reanalyses (Project 1.3.6). During this period, there is a negligible trend in pan evaporation simulated for either simulation or either region. Region 2 is on average warmer than region 1 and produces larger pan evaporation, as is to be expected. The current simulation produces somewhat (by up to 10%) larger estimates of pan evaporation than our prior simulation that downscaled from NCEP, for reasons we are yet to clarify (presumably related to one or more of low-level temperature, humidity or wind speed being different between the two simulations).

3500 3000 NCEP: Pan Evaporation Reg 1 Reg 2 2500 2000 1500 3500 3000 1970 1980 1990 2000 A2: Pan Evaporation Reg 1 Reg 2 2500 2000 1500 1970 1980 1990 2000 Figure 12. Time series of simulated pan evaporation (mm/year) averaged over regions 1 and 2 for 1961-2000. The upper plot is for the prior CCAM simulation downscaling from NCEP reanalyses. The lower plot is for the current CCAM simulation, for Mk3.0 GCM years. Trend lines are also shown. 5. Rainfall extremes Rainfall extremes have been calculated from the daily output files for the maximum daily rainfall on a 0.5 degree latitude/longitude grid. These are compared in Figure 13 against Bureau of Meteorology values plotted on a similar 0.25 degree grid. The simulated maxima agree quite well along the coastal zone (up to about 250 mm/day). The interior extreme values are somewhat low. This will be partly due to interpolation

from the model 20 km grid to sparser points at 0.5 degrees (about 50 km). We plan to investigate this behaviour further. Figure 13. Maximum daily rainfall (mm/day) at each point (on a 0.5 degree grid) for the duration of the 40-year run. CCAM is shown on the left, and observations on the right. 6. Summary This report presents some results from a 40-year present-day simulation downscaling a present-day simulation of the Mk 3.0 coupled atmosphere-ocean GCM by using the CCAM variable-resolution model at 20 km resolution over eastern Australia.. Many detailed 6-hourly output fields were produced and these can be made available to other researchers. The CCAM present-day climatology shows realistic detail compared to the coarser Mk 3.0 simulation. CCAM reproduces well the mean seasonal rainfall patterns; the only deficiencies are that the winter rainfall is somewhat low along the east coast, inland autumn and winter rainfall is somewhat low, and spring rainfall is somewhat large over inland New South Wales and Queensland. The mean daily temperature patterns are also well captured, and generally the maximum temperatures have smaller errors than Mk 3.0; terrain effects are also better captured. The largest maximum temperature errors (up to 2 degrees) are in summer and spring over Victoria. The CCAM minimum temperatures are up to 2 or 3 degrees too warm, especially for summer and spring, whereas Mk 3.0 is typically about 1 degree too cool. During summer, both CCAM and Mk3.0 are too warm over the southern half of the MDB. The interannual variability of the mean annual rainfall of the simulation over the GRDC region is qualitatively similar to the observations, in regard to peak values and number of wet and dry episodes. The simulation produces an increasing trend of rainfall, whereas the observations show little trend over this period. The simulation also calculates an estimate of pan evaporation. Over the GRDC region, the values are about 10% larger than for the prior simulation downscaling from NCEP reanalyses. No trend is found for pan evaporation over the GRDC region for either of the simulations.

Maximum daily rainfall throughout the 40-year simulation was also calculated. Near the coastline it agrees quite well with observed values. Inland, the maximum values are somewhat lower than observed. The ongoing project 2.1.5, continues the simulation on to 2100 for both the A2 and A1B scenarios. List of publication titles: McGregor, J. L., Nguyen, K. C., and M. Thatcher, 2008: Regional climate simulation at 20 km using CCAM with a scale-selective digital filter. Research Activities in Atmospheric and Oceanic Modelling Report No. 38 (ed. J. Cote), WMO/TD. Thatcher, M., and J. L. McGregor, 2008: Using a scale-selective filter for dynamical downscaling with the conformal cubic atmospheric model. Submitted to Mon. Wea. Rev. Acknowledgements Dr. Marcus Thatcher co-developed the digital filter used in this simulation. Dr. Ian Watterson kindly produced Figure 13 displaying the extreme rainfall values. This project References Holtslag, A. A. M., and B. A. Boville, 1993: Local versus non-local boundary layer diffusion in a global climate model. J. Climate, 6, 1825-1842. Kowalczyk, E. A., J. R. Garratt, and P. B. Krummel, 1994: Implementation of a soilcanopy scheme into the CSIRO GCM -regional aspects of the model response. CSIRO Div. Atmospheric Research Tech. Paper No. 32, 59 pp. McGregor, J. L., 1993: Economical determination of departure points for semi- Lagrangian models. Mon. Wea. Rev., 121, 221-230. McGregor, J. L., 1996: Semi-Lagrangian advection on conformal-cubic grids. Mon. Wea. Rev., 124, 1311-1322. McGregor, J. L., 2003: A new convection scheme using a simple closure. In "Current issues in the parameterization of convection", BMRC Research Report 93, 33-36. McGregor, J. L., 2005a: Geostrophic adjustment for reversibly staggered grids. Mon. Wea. Rev., 133, 1119-1128. McGregor, J. L., 2005b: C-CAM: Geometric aspects and dynamical formulation [electronic publication]. CSIRO Atmospheric Research Tech. Paper No. 70, 43 pp. McGregor, J. L., H. B. Gordon, I. G. Watterson, M. R. Dix, and L. D. Rotstayn, 1993: The CSIRO 9-level atmospheric general circulation model. CSIRO Div. Atmospheric Research Tech. Paper No. 26, 89 pp. Rotstayn, L. D., 1997: A physically based scheme for the treatment of stratiform clouds and precipitation in large-scale models. I: Description and evaluation of the microphysical processes. Quart. J. Roy. Meteor. Soc., 123, 1227-1282. Schmidt, F., 1977: Variable fine mesh in spectral global model. Beitr. Phys. Atmos., 50, 211-217. Schwarzkopf, M. D., and S. B. Fels, 1991: The simplified exchange method revisited: an accurate, rapid method for computation of infrared cooling rates and fluxes. J. Geophys. Res., 96, 9075-9096.

Smith, R. N. B., 1990: A scheme for predicting layer clouds and their water content in a general circulation model. Quart. J. Roy. Meteor. Soc., 116, 435-460. Project Milestone Reporting Table To be completed prior to commencing the project Milestone description 1 (brief) (up to 33% of project activity) 1. Compare alternative downscaling techniques (wind forcing from upperair versus farfield) 2. Complete the simulation from 1961-2000 at 20 km resolution Performance indicators 2 (1-3 dot points) Complete 10- year trial simulations, and select the most appropriate downscaling methodology from the Mk 3.5 GCM. Completion of full simulation Satisfactory climatology of rainfall and maximum and minimum temperatures Completio n date 3 Xx/xx/xxx x Budget 4 for Mileston e ($) Completed at each Milestone date Progress 5 (1-3 dot points) 31/3/07 33k Trial simulations were performed, comparing methods for forcing by upper winds Downscaling was compared from both Mk 3.0 and Mk 3.5 30/6/07 34k A present-day climate simulation has been carried out downscaling from Mk 3.0 for years corresponding to 1961-2000 Recommended changes to workplan 6 (1-3 dot points)

3. Comparison of model output with available observations Complete an analysis of rainfall, extremes and trends against available observations Produce an analysis of maximum and minimum temperatures Analyse measures of potential evaporation produced by the simulation Examine the model output with regard to behaviour during El Nino years. Produce 8 page report. 31/12/07 33k The various modelled fields have been compared against observations and a report produced.