1. Inquiry into the appropriateness of a TILE/MOSAIC approach for the representation of surface inhomogeneities B. Ritter and J. Helmert

2. Objective Concept of aggregation/disaggregation Pro&Con of TILE/MOSAIC Options of TILE and MOSAIC Implications for global and limited area NWP models Outline

3. Account for non-linear effects of sub-grid inhomegeneities at surface on the exchange of energy and moisture between atmosphere and surface (cf. Ament&Simmer, 2006) mosaic approach surface divided in N subgrid cells tile approach N dominant classes (e.g. water, snow, grass) (Figure taken from Ament&Simmer, 2006) Objective

4. Coupling of coarse atmosphere and high resolution surface E.g. Latent Heat Flux for one patch : atmospheric variables surface variables Grid box average Objective

5. Disaggregation: fluxes directed to the surface (downw. Radiation, Precipitation) Aggregation: fluxes from the surface to the atmosphere (upw. Radiation, turb. Fluxes) Concept

6. Disaggregation: Tiling shortwave radiation Gridbox value of net shortwave radiation (radiation scheme) S net Broadband (or spectral) albedo for each tile Energy conservation Concept

7. Pro & Con of TILE/MOSAIC Con: increase in computational effort & complexity additional requirements for external parameter software uncertainty with regard to suitable ‚blending height & depth‘ Pro: unsatisfactory handling of situations like snow melting phase (partial snow cover) of current approach should be alleviated simple integration of submodels (e.g. Flake, Urban) self-adaptation to model resolution

8. Options of TILE & MOSAIC MOSAIC (i.e. explicit sub-grid approach) initial selection of resolution enhancement factor, independent of heterogeneity resp. homogeneity of underlying surface sub-optimal self-adaptation to atmospheric model resolution unnecessary computational burden over homogeneous terrain (Stoll et al., 2010) TILE (i.e. weighted averaging of contributions from flexible number of surface classes) self-adaption to atmospheric model resolution and heterogeneity of surface occurs automatically computational burden adjusts to required number of surface classes

9. MOSAIC versus TILE approach preference for tile approach (Figure taken from Ament, 2006) Options of TILE & MOSAIC

10. Blending height ‚standard‘: assume homogeneity of atmosphere at lowest atmospheric level alternative: allow heterogeneity also in atmosphere near surface (e.g. downscaling/disaggregation of atmospheric variables at the lowest model level; cf. Schomburg et al., 2009) The ‚standard‘ approach creates neither technical problems nor computational overhead, but may not be justified in situations with large surface heterogeneities. A ‚downscaling‘ approach in the spirit of Schomburg et al. may alleviate this problem. Options of TILE & MOSAIC

11. Downscaling system for atmospheric variables on lowest model level (A. Schomburg) (Schomburg et al., 2009) 3 Steps: 1. Smooth field with splines 2. Downscale field “deterministically“by regression techniques (use subgrid-scale information like orography) 3. Add noise to reproduce original fine-scale variance If no rules for step 2. can be found (e.g. for precipitation) apply only step 1. and 3 Options of TILE & MOSAIC

12. Blending depth A proper tile/mosaic approach requires the simulation of soil internal processes like heat conduction for each indivual class resp.sub-cell Assuming homogeneous conditions within the soil (e.g. ECMWF IFS) leads to a major simplification and saving of computational ressources but is hardly justifiable. In particular in the framework of DWD‘s multi- layer soil model with a top layer depth of only 1 cm, it appears to be a rather crude and unrealistic assumption. implement tile approach in a consistent manner for all soil layers Options of TILE & MOSAIC

13. Implemention of tile approach requires: development and implementation of corresponding extensions in external parameter software (i.e. landuse dependend parameters for a no. of dominant classes within each atmospheric grid cell) code structure to support multiple ‚soil columns‘ within each grid cell (TERRA adaptions in COLOBOC) physics interface routine or multi-layer soil model, which controls the computation over (flexible) number of classes within each cell and performs necessary aggregation (&disaggregation) suitable diagnostics (within soil model) to allow proper validation of tile scheme a computationally efficient and flexible implementation (vectorisation?) Implications for NWP

14. AROME (SURFEX) 4 tiles: nature, town, sea, inland water Nature: ISBA 3L (Boone et al 1999) 1L snow scheme (Douville, 1995) Town Sea, inland water: constant T_s, Charnock formula UM (Jules) 9 tiles, 5 veg + 4 non-veg Broadleaf and needleleaf trees, temperate and tropical grasses, Shrubs, urban, inland water, bare soil, land ice. IFS (HTESSEL) 6 land-surface tiles High vegetation, low vegetation, interception reservoir, bare ground, snow on ground and low vegetation, Snow under high vegetation Implications for NWP