A Subgrid Orography Scheme: Ready for Prime Time Steven Ghan and Tim Shippert Pacific Northwest National Lab Ghan, S. J., X. Bian, A. G. Hunt, and A. Coleman, 2002: The thermodynamic influence of subgrid orography in a global climate model, Climate Dynamics, 20,
Subgrid scheme
Implementation Applies to all column physics –clouds –radiative transfer –vertical mixing –surface physics
Elevation Classification
Progress Scheme applied to CAM and CLM –Euler dycore –Finite-volume dycore Developmental branch updated to cam Bit-for-bit agreement between SP, SMP, SPMD Restarts bit-for-bit agreement Bit-for-bit agreement with dev trunk if subgrid scheme turned off Energy conservation demonstrated to within 0.01 Wm -2 Load balancing within nodes for both dycores Load balancing between nodes for both dycores (IJHPCA, 2005) Runoff distributed according to elevation of river transport model surface elevation AMIP simulations at T42 and 2˚x2.5º resolution Detailed evaluation in eight regions (J. Climate, 2006a) IPCC A1B simulation at 2˚x2.5˚ resolution (J. Climate, 2006b)
Experiments AMIP SST –Initialize September 1977 –Run through 1989 –T42 TOPOG 11 elevation classes No TOPOG –2x2.5 TOPOG 11 elevation classes No TOPOG –1x1.25 No TOPOG IPCC A1B –2x2.5 –TOPOG 11 elevation classes –
Evaluation
temperature
March snow
March Snow
Precipitation in other regions
RMS Error Western US
North America Glaciers
River Discharge
Does the Scheme Impact the Grid Cell Mean Climate?
Does the Scheme Impact the Grid Cell Mean Climate? Precipitation at 2x2.5 resolution
Surface Air Temperature T42
Surface Air Temperature 2x2.5
Zonal and Annual Mean
Planetary Energy Balance
Global Annual Mean T42 TOPOG T42 NOTOP T42 2X2.5 TOPOG 2X2.5 NOTOP 2X2.5 FSNT FSNS FLNT FLNS RAD SHFLX LHFLX TFLX RAD+TFLX PREC QFLX
Application to IPCC A1B Intergovernmental Panel on Climate Change A1B Scenario CAM3 and CLM3 run offline with the subgrid scheme, using ocean surface conditions from CCSM3 simulation Adjust ocean surface conditions to correct for biases: Greenhouse gas and aerosol concentrations from the coupled simulation 2˚x2.5˚ horizontal resolution Simulate the period
Snow Water in Western U.S.
Snow Water in New Zealand
Computational Burden
Zonal Mean Burden
Global Burden
Balancing the Load Distribute elevation classes across physics chunks –Rank grid cells according to number of classes –Starting with cells with most classes, assign all classes in cell to the chunk with the least number of columns –Continue distribution until classes for all cells have been assigned to chunks –Save cell index and area corresponding to each chunk column Distribute chunks across nodes -Minimize dynamic-physics transpose cost by assigning chunks to node with most dynamics grid cells from chunk -fraction of columns assigned to same node as dynamics: 40% for T42 11 classes on 8 nodes
Load Balancing Performance
Code Changes Unlike any other parameterization –All column physics applied to each elevation class Mostly manifest at higher levels –phys_grid phys_grid_init create_chunks scatter_field_to_chunk gather_chunk_to_field –phys_types –dp_coupling –history
Summary The subgrid scheme provides valuable regional detail at a modest computational cost. It produces some biases, largely due to neglect of rainshadow. Its biases can be ameliorated by reducing grid size to km. It produces little impact on grid cell means of most fields, except for snow water. It can be treated as an option with minimal retuning.
Future Work Update to support CCSM and hi-res CLM: –pass elevation class fields through the coupler –optional online mapping of atmosphere fields to CLM grid Present to LMWG for approval. Submit proposal to SSC. Merge on to developmental trunk. Continue to maintain as CCSM changes.
Greater Emphasis on Snowpack 70% of runoff in Western US is from snowmelt. Mountain snowpack is heavily relied on as a natural water reservoir. Simulations of greenhouse warming by regional climate models suggest a 50% reduction in mountain snowpack in the Cascades and Sierras during this century. A similar sensitivity can be expected in any region marginally cold enough for significant snow accumulation. Greenland icepack is expected to melt, but how soon? The subgrid orography scheme can be used in CCSM to explore these issues for the whole Earth.
Land Model Options 1.Continue to use elevation classification framework –Use different (multiple) surface types for each elevation class –Treat influence of slope/aspect on solar absorption –Two-fold computational cost without slope/aspect –Inconsistent with high-resolution gridded CLM 2.Map atmosphere to a separate CLM grid –Already being done for RTM –5’ resolution required to resolve orographic signature of precipitation –Memory cost is ~nlat*nlon*maxclass*30~3 Gb per node for 5’ grid –Hundred-fold computational cost for CLM, if distributed across nodes –Could be distributed across nodes easily if no horizontal interpolation –Horizontal interpolation would require communication and would have conservation challenges