Aaron
  Boone's
  Homepage

  Home
  The Liberation of
  Chateau Gontier
  Work
  Address Book
  Sports
  Wildlife Forays
  Road Trips
  Weather

ASP: Atmosphere-Surface Prediction system


Model Overview Summary Page


Page Menu


Model Technical Description



Disclaimer

Please note that this page has nothing to do with my employer, CNRS, or where I am employed, at the National Center for Meteorological Research (Centre National de Recherches Meteorologiques: CNRM) at Meteo-France. The opinions expressed herein are my own, and are not a reflection of those where I work or of my employer.



ASP - Atmosphere Surface Prediction System

The ASP (Atmosphere-Surface forecast Prediction) model has been continuously developed since 2007. The model is It uses a horizontal resolution which is comparable to some operational NWP models (albeit slightly on the large side of this range). It has baseline physics and a solid dynamical core and thus can produce reasonable forecasts. All the code and scripts have been developed using free software and literature from the web (a special thanks goes out to Allen Press and the American Meteorological Society for making the AMS historical archive freely available on the web), scripts, plotting software, GMT, and Fortran 90/95 compilers (originally using g95, currently using gfortran from the Mandriva 2008.0 and 2009.0 releases). under Linux on my home PCs.

Description of the ASP model

ASP is a hybrid (pressure)-coordinate primitive equation hydrostatic model with equations expressed in flux form. The model uses baseline (relatively simple) physics. The 7 3-D prognostic (atmospheric) variables consist in the u and v wind components, the air temperature, T, specific humidity, q and surface pressure, ps, in addition to (3) additional microphysical variables; cloud liquid water content, qcl, cloud ice content, qci, and a passive tracer.

The three remaining prognostic variables are related to the sub-surface: the soil temperature, moisture and ice temperature for N-layers (the current default is 5). Heat transfer uses diffusion, while soil moisture transfer is modeled using Richard's equation. The time integration schemes use linearized fluxes and are implicit (soil and ice temperatures are implicitly coupled with the atmospheric temperature profiles: sea surface fluxes are computed using prescribed SSTs).

For more information on the physics, dynamics and numerical aspects, see the ASP technical description.

Atmospheric Dynamics

ASP is a hybrid-coordinate primitive equation hydrostatic model with equations expressed in flux form. The five main prognostic (atmospheric) variables consist in the u and v wind components, the air temperature, T, specific humidity, q and surface pressure, ps. Also, there are 3 additional microphysical variables; cloud liquid water content, qcl, cloud ice content, qci, and a passive tracer, for a current total of 8 prognostic atmospheric variables. The equations are solved on an Arakawa C-grid along hybrid (vertical) coordinate surfaces: a terrain following pressure based coordinate is used in the lower atmosphere while constant pressure surfaces are used in upper levels (with a gradual transition in between). The vertical grid resolution is highest near the surface.

Physics

The microphysics scheme includes 5 additional 3D prognostic variables: an optional passive tracer and 4 bulk cloud water variables (cloud liquid water and ice, and rainfall and snowfall content). The passive tracer can have one or multiple sources, and it precipitates based on a user imposed sedimentation terminal velocity. There is no feedback with the atmosphere. The microphysical prognostic variable is fully coupled with the atmosphere. The model baseline physics consist in: bulk microphysics with 5 water species variables, including evaporation, condensation, depostion, sedimentation, autoconversion, evaporation of cloud cloud water, sublimation of cloud ice, freezing/melting of cloud ice/water and rain/snow, 2 longwave radiation schemes with clouds (a simple or "fast" method based on Savijarvi, 1990, or a broadband longwave method from Chen and Cotton, 1983), 2 shortwave radiation schemes with clouds (a simple or "fast" code based on Mahrer and Pielke, 1977, and Savijarvi, 1990, or a 2-stream multiple band method based on Chen and Cotton, 1983) a mass-flux convection scheme, a vertical turbulence enhancement to represent shallow convection, vertical diffusion/turbulence with Louis-type local K closure (based on ECMWF model prior to 2004), additional non-local mixing in the daytime PBL using an EDMF approach. an implicit coupling between the atmosphere and three sub-grid surface tiles (continental surface, liquid water surfaces and sea ice). As a final note, a global borrowing scheme is used for the water vapor to correct for any negative values.

An example comparison of the 2-stream model (option) downwelling longwave flux (LWdown) at the surface predicted by ASP and the GFS model can be seen here, where the GFS LWdown is in the left hand panel. Overall features and order of magnitudes are quite similar (despite the relative simplicity of the ASP scheme).

Surface

There are 6 prognostic variables for the land surface; the soil temperature, moisture and ice content for N-layers (the current default is 5), snow density, SWE and albedo (currently a 2 layer snow scheme). The frozen water tile uses a single prognostic variables for ice temperature (2 layers). Finally, the surface water scheme (lakes and oceans/seas), uses a cool-skin prognostic temperature (with a relaxation to SSTs provided from an external source). Heat transfer uses diffusion, while soil moisture transfer is modeled using Richard's equation. The time integration schemes use linearized fluxes and are implicit (soil, water and water-ice temperatures are implicitly coupled with the atmospheric temperature profiles).

In terms of the vertical soil-grid discretization, ASP uses the GFS default 4-layer soil grid, but splits the uppermost GFS soil layer into two layers: thereby adding a 5th, thin upper layer (in theory, for better resolving diurnal near-surface soil moisture fluctuations and potentially large near-surface gradients in moisture content, temperature, and soil ice). The current layer thicknesses are (from the surface downward); 0.03, 0.07, 0.30, 0.60, and 1.00 m. The frozen water tile uses a 3-layer approach. The water surface uses a cool-skin prognostic temperature (relaxed to prescribed SSTs).

Horizontal Turbulence (Diffusion)

Options for second (physically-based turbulence Smagorinsky-type), fourth or sixth order monotonic horizontal diffusion (constant diffusivity). In addition, 4th order diffusion can be used with a spatially varying deformation-dependent diffusivity (a scaled Smagorinsky-type diffusivity). For all horizontal diffusion schemes, corrections are applied to temperature and water vapor (to quasi-horizontal surfaces). Note that when odd-ordered horizontal advection is used, then horizontal diffusion can be OFF as the odd-ordered advection schemes are dissipative.

Vertical Turbulence (Diffusion)

Vertical diffusion/turbulence is modeled using a second order approach with Louis-type local K closure above the PBL and a non-local mixing within the PBL using an EDMF approach (combined eddy-diffusivity mass flux). An implicit coupling is made between the atmosphere and three sub-grid surface tiles (continental surface, liquid water surfaces and sea ice).

Horizontal Advection

Horizontal advection can be modelled using 2nd, 4th, 5th or 6th order schemes (higher order schemes are Wicker and Skamarock-WRF type schemes). There is also a 5th order WENO scheme available for scalars. Even ordered advection schemes can be used in combination with any order horizontal diffusion scheme. The odd ordered advection scheme can be used with either i) one of the three horizontal diffusion/turbulence schemes or ii) with no horizontal diffusion (as the odd-ordered advection scheme is implicitly dissipative). Also, options exist (they are used by default) to adjust the scalar fluxes to be postitive definite (currently used for water vapor, cloud liquid water and ice, and tracer(s)). Finally, more recently an option to use 5th order WENO advection has been added: it currently can be used for scalars.

Filters

The only filtering during the forecast integration is through the use of a divergence damping term in the momentum equations. The default is a high-order scheme (the idea is to just suppress small scale noise while leaving the divergence field largely intact). Low-order damping tends to damp divergence fields in time, while high order schemes tend to be more of a filter (damping mostly the shortest wavelengths). The model is initialized either using assimilation (from a past forecast) or using a digital filter (cold-start), see section on Initialization for details.

Time integration

Physics and dynamics are split: physics (radiation, convection-precipitation, turbulence) are evaluated on the large (model) time step (vertical turbulence and land surface coupling are fully-implicit). Time integration of the dynamics uses a 3rd order Runge-Kutta scheme (an option for a 4th order R-K scheme also exist). The time stepping for each RK step is done using a relatively simple forward-backward difference (time-splitting) technique where terms involving gravity wave propagation are evaluated using a smaller sub-time step (default). Options (essentially for testing purposes only) also exist to evaluate pressure gradient using a semi-implicit (trapezoidal) time scheme (a smaller time step and more CPUs than time splitting are needed), or a fully explicit time stepping (a much smaller time step and considerably more CPSs are required, so this method is essentially not used except for testing). Also, there is an option for implicit vertical advection. The (large) model time step is, by default, currently defined (in seconds) as 4d, where d is the grid spacing in km: for an 60 km grid spacing, the time step is about 4 minutes. The model equations are discretized on an Arakawa C-grid, and in the vertical a Lorenz grid is used: the prognostic variables are centered within the vertical layers, while the vertical velocity is staggered (on the hybrid levels).

ASP Lateral Boundaries

The lateral boundary treatment amounts to a nudging data assimilation scheme. Davies-type relaxation is used along the lateral boundaries. A 9 (or optionally a 5) point sponge zone is used along with an exponential increase in the relaxation coefficients (approaching the lateral boundary). One can use i) a Raleigh or Newtonian relaxation term or a ii) a diffusion or artificial viscosity term are used, or iii) both (although the diffusive relaxation is considerably lower in this configuration). Large scale values are imposed right at the lateral boundaries for computing horizontal advection and diffusion. Lateral values of temperature, specific humidity, surface pressure and wind components are computed from linear time tendencies and nudged using the aforementioned method. The aforementioned winds are obtained either (i) using relaxation on pressure surfaces or (ii) by directly using the large scale wind components (possibly rotated to the ASP projection).

ASP Model Grid

The default vertical grid configuration uses irregularly-spaced vertical hybrid layers (an approximate exponential spacing, with the highest resolution near the ground or water surface, see an example of the current grid for a bell-shaped mountain), and a 55-km horizontal resolution (lower than some hydrostatic operational models, but tests have indicated pretty good/comparable results compared to higher vertical and horizontal resolutions). The model uses either Lambert Conformal (LAMB) or Mercator (MERC) projections (the USER selects this as a function of the domain location: Mercator is generally used for domains centered at or near the equator, while LAMB is generally used for mid-latitudes). The current default Limited Area Model (LAM) grid domains are over North America (LAMB), Western Europe (LAMB), and Western Africa (MERC). Note that map factor isolines are plotted in white. Recently, the possibility of using a global grid on a Plate Carree (PLCA) projection (with Fourier-filtering above a critical latitude) has been developed and is being tested.

ASP atmospheric model initialization

The model is initialized using near-real time NCEP Global Forecast System (GFS) model data. The procedure is outlined below. Note that OPTIONAL data and procedures are indicated for testing etc...but they are not the DEFAULT.

  1. The relative humidity, the air temperature, and the u-v wind components are extracted for all available levels (from 50 mb to 1000 mb, 50 mb intervals, except slightly more layers near the surface) from the operational GFS data files. In addition, the surface sea-level pressure and lowest level geopotential heights are also extracted.
    OPTIONAL: The absolute vorticity, and/or geopotential heights can also be extracted from the GFS files if the USER wishes to derive the wind fields from these quantities. There is also an option to compute the horizontal divergence on the GFS grid using the input wind components. The aforementioned fields are then interpolated to the model (horizontal grid).
  2. Input topography is then used with the temperature profile and sea level pressure to define the along-terrain surface pressure. The pressures along the hybrid coordinate surfaces (and therefore the pressure thickness prognostic variable) are then determined (using the user-defined values).
  3. The u and v wind components are then rotated to the ASP grid projection.
    OPTIONAL: Indirect methods can be used to derive the wind components from other input fields such as vorticity, divergence or the geopotential. See Details on Wind Component Initialization for more details.
  4. The wind components and RH are vertically interpolated (linear in pressure) to the hybrid grid, and the temperature is interpolated using (natural) log pressure weighting.
  5. The vertically integrated divergence is removed from the wind components at the initial time. This is required if using real or NWP model derived winds as input to initialize the model: if one chooses the option to use filtered winds, then this has almost no impact (except possibly in areas of steep terrain).
  6. Additional variables are then defined along hybrid coordinate surfaces: the geopotential is computed (for diagnostic purposes) from the hydrostatic equation (including moisture effects), and the specific humidity (prognostic variable) is computed from the RH and air temperatures.
  7. (i) If a past forecast is not available, digital filtering is then used to obtain the model initial state for the thermodynamic prognostic variables using short forward and backward model time integrations. (ii) If a past forecast is available, then the dynamic fields are nudged towards the analysis using a simple assimilation scheme to obtain the initial state.

Details on wind component initialization

Several methods have been used and tested to get the initial wind components, they are summarized here:

  1. Direct use of input u and v wind components (UV). Pros: i) No need to derive the wind components...direct insertion of wind components into the model ii) can be applied anywhere on globe iii) no need to extract and interpolate the vorticity (or to compute and interpolate the divergence) Cons: i) winds must be rotated to ASP projection, however this is rather minor, ii) the winds are not necessarily in balance with the initial pressure field (owing to interpolation errors), which leads to an adjustment phase and could excite some gravity wave noise. But the vertically integrated divergence is removed which seems to work quite well, and a digital filter is used so excessive initial gravity wave noise at the forecast start is not a major problem.
  2. Derive wind components from input relative vorticity (VORT). Pros: i) winds are filtered (no-divergence) so little to no gravity noise, ii) resulting vorticity fields very consistent with large scale driving model vorticity, iii) can be applied anywhere on globe iv) many studies have stated that this method gives better winds compared to using a balance equation approach Cons: i) winds are not necessarily in balance with pressure fields ii)lateral boundary condition assimilation scheme then uses these rotational components in full momentum equations: it is probably more consistent to relax these winds to ASP rotational wind components or the vorticity (both both methods lead to more complex issues for now) iii)experience has shown that wind speeds are not as accurate as those from direct insertion of wind components.
  3. Derive wind components from input geopotential using the linear balance equation (LBE) Pros: i) simple solution using input geopotentials only: economical alternative to full nonlinear balance eq. ii) many studies have stated that this method gives reasonable wind components in spite of being relatively simple. Cons: i) Can't be used near equator, so must be combined with another method (like vorticity method above, or the nonlinear balance equation for low latitudes). ii) Consistent with some studies, this method tends to over-estimate wind field maxima (experience shown that wind speeds are not as accurate as direct insertion of wind components) iii) experience showed that gravity waves tended to be excited around closed cyclonic circulation centers.
  4. Derive wind components using the nonlinear balance equation (NLBE). Pros: i) good approximation to balance between pressure and wind fields ii) can be applied anywhere on globe iii) only need input geopotential Cons: i) winds not as good in highly anti-cyclonic regions (although this does not seem to be too problematic for most cases) ii)experience has shown that wind speeds not as accurate as vorticity method or direct insertion of wind components.
  5. Derive rotational wind components from input vorticity, and divergent components from input divergence (VRTD): then combine to obtain full wind components. Pros: i)Scalar quantities (divergence and vorticity) are interpolated from the GFS grid to the ASP grid (as opposed to vector u and v components) ii)can be applied anywhere on globe iii) wind components assimilated into the model are consistent with momentum equations (both rotational and divergent components present) Cons: i) pre-processing requires interpolation of an additional field: divergence (although this does not change pre-processing time much) ii)experience has shown that wind speeds not as accurate as direct insertion of wind components (although the closest of all of the indirect methods mentioned above to the actual rotated wind components)

Currently the WIND method is used as it performs the best of the aforementioned methods (which is the DEFAULT as GFS data is being used as input). Although the VRTD method gives fairly good results, it requires more pre-processing (and disk space) than the WIND method. In addition, the WIND method seems to give better long range forecasts.

Noise in the initial fields is very effectively damped owing to (i) the removal of vertically integrated divergence in the initial winds, and (ii) by using a digital filter initialization (DFI) The DFI approach is used with short forward and backward time period integrations (1-3 hours each direction: currently 2 hours forward and backward is used, so the initialization phase totals 4 hours of integration). Sixth order horizontal advection is used in this phase. Note that there is no diffusion (vertical or horizontal), or filtering (divergence damping) activated during the initialization phase. Also, all "physics" are OFF: water vapor is transported, but there is no precipitation (and therefore no latent heating etc...). After the filtering, the specific humidity is checked for saturation, and convective adjustment is performed (moist and dry). The initial state (temperature, specific humidity, pressure thickness and wind components) is now ready for the forecast integration.

Another option to get a balanced initial state is using the assilimation option. The past forecast is nudged towards the analysis using a simple assimilation scheme. In this option, all physics are "ON", which reduces possible spin up problems as the initial state is better adapted to the model.

ASP surface model initialization

The model is initialized using near-real time NCEP Global Forecast System (GFS) model data. For now, a very simple methodology is used. The initialization of the three surface modules (tiles) is described below.

  1. Land: Soil and vegetation (monthly varying) parameters are defined using the global ECOCLIMAP data set. The advantage is that this data set is available globally at up to a 1 km spatial resolution (so parameters can readily be adapted to the ASP resolution). In terms of initializing the land surface prognostic variables, soil water content and soil temperature are interpolated to the ASP soil layers using a nearest-neighbor approach (as the surface data is at a higher resolution than the atmospheric fields and are more heterogeneous). The initial ice content is not initialized using GFS input data (the GFS unfrozen liquid water content is indeed available), but rather it initialized using the total water content and soil temperature in a method which is thermodynamically consistent with ASP (using the Gibbs Free Energy concept). Note that the GFS soil textures are not necessarily consistent with those used by ASP, but eventually the GFS textures will be used (soil moisture mapping between models is a well known and yet unresolved problem: partly owing to input data sets, and partly due to physics). Snow Depth and Snow Water Equivalent (SWE) are used to initialize the ASP SWE and snow density. Snow albedo is initialized arbitrarily for now (an average of the minimum and maximum values).
  2. Water: Sea Surface and Lake temperatures are fixed during the run using the current GFS surface temperature.
  3. Frozen Water: Where the water surface (water defined using the ASP land mask) temperature is less than 0 C, an ice tile is created. The ice surface temperature is initialized using the GFS surface temperature: deeper layers are initialized using the minimum of 0 C and an average air temperature.

ASP Future Developments and Applications

The focus of model development over the near term (not in any particular order):

  • Improve the shortwave radiation parametrization of absorption/transmission (to include more spectral bands, an explicit ozone computation...)
  • Replace local-K turbulence with a prognostic TKE scheme
  • Add tall vegetation/canopy tile to land surface routine

Tests have been done in terms of higher horizontal resolution over the same sized domain: i) Horizontal - tests moving from 80 down to 50 km resolution did not show much change out to 3 days (forecasts), except for a slight slow down and a slight general strengthening of synoptic scale systems. Some tests have been done using a "high resolution" 20-km grid with a reduced domain centered over France (and a floater-domain over the US). The obvious appearance of finer scale features over zones with significant topography (and associated precipitation and lower atmosphere winds) are found: an example of outputs can be seen for a 36 hour forecast over Europe at a 80 km resolution and centered over France at a 20 km resolution for the same date. The high-resolution tests will continue.


ASP Options and Current Defaults

The dynamics, physics and numerical schemes are summarized below.

ASP Methods for: Current Default Options
Time Integration 3rd Order Runge-Kutta 4th Order Runge-Kutta
Time Stepping (on each RK step) Forward-Backward difference (time-split) Semi-implicit Pres. Grad. terms in momentum eq.s, or Explicit
Input Data Real-time GFS NWP data from NCEP Idealized case.
Initialization Method Simple assimilation: past forecast is nudged towards the analysis Cold-Start, Using Short Forward-Backward integrations with Digital Filter method.
Initial and Boundary Winds (derived from:) Input u-v wind components Geopotential using the balance equation, vorticity (and divergence), or using input wind components. *For all non-filtered winds, the vertically integrated divergence is removed
Horizontal Diffusion 6th order monotonic with constant diffusivity 4th or 6th (both monotonic) order computational diffusion with constant (4th, 6th) or spatially variable diffusivity (4th), 2nd order physically based Smagorinski-closure. *All use quasi-horizontal corrections for air temp. and spec. humidity.
Horizontal Advection 5th order WRF & 5th order WENO for scalars 2nd, 4th, 5th or 6th order WRF, 5th WENO
Advective Flux Adjustment Positive definite flux correction (PDFC) for scalars PDFC on or off
Vertical Advection 2nd order implicit (Crank-Nicolson) for T and all other scalars, 2nd order explicit for wind components 2nd order explicit for all variables
Divergence Filter/Damping 4th order (moderate coefficient) 2nd order (low coefficient), 4th or 6th order (moderate coefficient)
Horizontal Grid Arakawa-C, Lambert Conformal or Mercator Projection (depending on domain location), or Plate Carree over the entire globe (grid spacing, geographic location, projection and dimensions are USER specified input parameters)
Vertical Grid 24 layers, Hybrid pressure coordinate, Staggered: mass and momentum variables are centered within vertical layers, vertical velocities are defined at layer interfaces. Pressure top at 50 mb (number of vertical layers and coordinate level values are USER specified input parameters)
Lateral Boundaries Davies type (Newtonian), 9 pts + Exponential fn. 6-hour updates for linear time tendencies Davies type (Newtonian, or both Newtonian and Diffusive), 9 or 5 points
Vertical Diffusion Fully implicit coupling with 3 surface-tiles, Local-K (Louis functions) above PBL with moist Richarson number in cloudy regions; EDMF within the PBL (mass flux and Non-Local-K mixed layer approach) Explicit Flux, 3 tiles (Local K)
Longwave Atmospheric Radiation Broadband longwave radiation with clouds (not every time step) Simple "fast" method with clouds clouds (each time step)
Shortwave Atmospheric Radiation Simple "fast" method with clouds (each time step) 2-stream shortwave radiation with clouds (not every time step)
Shallow Cumulus Mass-flux (Tiedtke, Zhang and McFarlane type) Geleyn type Richarson number modification to turbulence
Deep Convection Mass-flux (Tiedtke, Zhang and McFarlane type) i) Gadd and Keers Moist Convective Adjustment, ii) Kuo-type Convection, iii) Bougeault (1985)
Microphysics Bulk cloud liquid water, ice, snowfall and rainfall prognostic variables. Multiple microphysical processes are modeled. No microphysical prognostic variables: Supersaturated water and snow precipitates. Rainfall/snowfall evaporates/sublimates.
Surface: Land tile 5-layer fully implicit heat diffusion and Richard's equation for water transfer, VIC-sub-grid runoff, soil ice, sub-grid orography roughness. Single energy budget, composite snow scheme. Implicit coupling with atmosphere -
Surface: Ice tile 1-layer fully implicit heat diffusion. Implicit coupling with atmosphere -
Surface: Water tile Water surface T's prescribed from climatology/Operational NWP input Charnock surface drag -

Some Example Model Forecast Plot Comparisons with GFS

Here is a comparison plot between ASP and GFS forecasts both initialized on Jan. 10, 2009 at 12Z. The 24 and 48 hour forecast 500 mb heights and sea level (surface) pressure fields are shown for both models over Europe and US. Notable Synoptic scale features over the US consist in the exit of a system off the mid-Atlantic coast at 24 hours followed by a small but potent clipper-type system plunging into the upper Midwest by 48 hours. Notable synoptic scale features over the EU consist in anticyclonic surface conditions over central Europe and the progression of systems along an active zone across northern Europe, in addition two a cut off low over north Africa. Both the position and strength of the main features are in very good agreement at 24 and 48 hours for both domains. These plots are available for Real Time Forecasts.


ASP model updates

  • Jan. 5, 2007: Use GFS absolute vorticity to derive initial wind field as opposed to balance equation: improved wind speeds (particularly magnitude) and initial relative vorticity

  • Jan. 20, 2007: Improved initialization (pre-processor) of air temperature and geopotential on sigma surfaces: removal of unrealistically large local peaks in air temperature in mountainous terrain

  • Feb. 20, 2007: 2 items: (1) Surface layer turbulent transfer functions modified: more closely resemble those of Monin-Obukhov theory, especially for wind stress. Main result is less surface drag in rapidly deepening systems (and an improvement in corresponding forecast surface pressure drop, etc...). (2) Ocean heat and mass flux parametrization improved.

  • March 16, 2007: Test option to use second order horizontal Smagorinsky-closure turbulence scheme with no background diffusivity (as it is used with 5th order advection scheme). Simple radiative rudimentary cooling profile implemented, and simple shortave radiation heating added (to atmosphere).

  • March 26, 2007: Simple longwave cooling profile and surface downwelling flux expressions replaced by full broadband longwave flux model considering clouds. Simple shortwave algorithm replaced by more robust clear sky model, with simple empirical factor for cloud effects. Improved radiation schemes cause improved low level temperatures, especially at night over relief. Precipitation (at ground) much improved: large areas of very low level light precipitation removed, and enhancement of convective precipitation.

  • April 2, 2007: Return to using default of no explicit horizontal diffusion. Although fields slightly less smooth, relative vorticity seems to be better conserved (less damped) with no explicit diffusion.

  • April 20, 2007: Improved lateral boundary layer relaxation scheme. Added diffusive relaxation, no longer relaxing surface pressure. Overall improvement, notably in terms of surface pressure.

  • May 1, 2007: Time splitting (forward-backward time differencing) used with RK loop. A larger time step is used, with negligible impact on results.

  • May 17, 2007: 4th order horizontal diffusion used with option for spatially-dependent deformation based diffusivity. Advantage is that it only becomes significant in regions of large deformation and can effectively filter some grid-point storm type noise. Outside of regions of strong deformation, it is negligible: but the 5TH order advection is dissipative so that for these regions the implicit diffusion controls noise.

  • May 18, 2007: Improved shortwave radiation algorithm a bit: cloud impact on reflection and transmission improved. main impacts are higher downwelling fluxes at surface, and absorption by clouds (improved heating rates).

  • June 23, 2007: 2 items: (1) Improved temperature/geopotential initialization...vertically interpolate potential temperature to sigma surfaces (in order to obtain geopotentials on sigma surfaces) as opposed to directly interpolating geopotential to sigma surfaces. This all but removes sometimes anomalous temperature gradients in regions with significant topographic variability. (2) Additional trigger criteria were added to the Kuo-based convection scheme which improves spatial distribution and intensity of convection.

  • June 27, 2007: 2 items: (1) Improved Wind Initialization: initial (filtered) winds are derived from the input geopotential using the linear (mid to high latitudes) and non-linear (low latitudes) balance equations. This improved results (surface pressure, geopotentials...) and eliminated gravity waves which were sometimes excited using input vorticity to obtain the initial winds. The filtered winds are also used as lateral time varying boundary conditions. (2) A moist convective adjustment scheme can be activated now in regions where the Kuo scheme is not active (i.e. convection is not PBL rooted) and the atmosphere is conditionally unstable and the RH is sufficiently high.

  • July 26, 2007: Improved Wind Initialization: initial winds determined using a method whereby the stream function is first estimated using input relative vorticity, which is then used as the initial guess for the nonlinear balance equation.

  • Aug. 8, 2007: Have gone back to using relative vorticity to initialize winds (as they give best estimates of actual wind speeds) in DEFAULT mode (meaning when input vorticity available: if not, use NLBE to get winds). The main change or new development is an improved treatment of the lateral boundaries when determining the wind components.

  • Sept. 12, 2007: Further improved initialization of temperature and moisture fields. Slightly cooler and more humid profiles result in reduced precipitation spin-up in first 12-24 hours. Also, lateral boundary winds are determined over a region slightly larger than actual simulation domain...but winds are only extracted for initialization and lateral boundary forcing inside of actual (smaller) model domain. This is done to reduce the influence of lateral boundary conditions (when obtaining winds from the vorticity or geopotential) on the wind fields.

  • Sept. 14, 2007: Starting with 12Z forecasts, move from a 14 to a 20-layer grid configuration. This was found to remove some light spurious precipitation (improvement), and slightly smoother surface pressure values. Increased CPUs more than offset by recent code optimizations. Especially higher PBL vertical resolution.

  • Sept. 28, 2007: The transition of the model input format from grib1 to grib2 format has been completed. The global grib2 GFS files have 2x the horizontal resolution as the grib1 files (used until this date), however, this switch was found to have a fairly minor impact on the simulations (this format/resolution change should be more significant when the model horizontal resolution will be changed). Also, initial forecast data files (f00) are now used for the atmospheric initialization in place of the analysis field files (anl).

  • Oct. 1, 2007: Move back from a 20 level grid to a 16 level grid. After further tests, it was found that a 16 layer configuration could produce results very similar to the 20 layer configuration given a certain layering: The upper layers use constant sigma spacing, while lower layer thicknesses decrease exponentially approaching the surface.

  • Oct. 13, 2007: Add a simple scheme to relax average mass within domain to large scale value (remove any bias that might develop).

  • Oct. 18, 2007: 2 updates: 1) Improved initial winds: Using default option (winds from input vorticity), Neumann boundary conditions used in equator-ward latitudes, while Dirichlet used (as before) at mid to high latitudes. Result is significantly improved low latitude initial and lateral forcing wind components. 2) Also, surface stress parametrization altered to reduce surface drag a bit to improve cyclonic development: has little impact outside of rapidly deepening systems.

  • Oct. 19, 2007: Deformation-based diffusion coefficient is reduced in regions of strong topographical variations: removes some rare anomalous cooling under some circumstances.

  • Oct. 23, 2007: The fourth order divergence damping diffusivity has been increased in order to better filter gravity waves while still being low enough to have little to no impact on the synoptic scale features. This was done because the lower diffusivity using the 4th order damping effectively filtered numerical noise from the divergence fields, but gravity waves early in the simulation were only weakly damped. the second order damping was also re-tested, but the fourth order damping is more scale selective and the vertical velocity fields are less smoothed in time (than using second order damping).

  • Oct. 28, 2007: Changed from 4-day to 2 1/2 day forecasts. This was done as a trade off in preparation for moving to a higher spatial resolution (CPU run time and pre-processing times are significantly reduced). Also, model performance seems to be quite good (consistent with operational NWP models also) out to 60 or 72 hours: after that, boundary conditions and model errors (simple physics, etc...) tend to make the forecasts begin to sometimes degrade in quality (in fact, this is also seen in results on web from operational limited area models such as WRF), so for now 2 1/2 days is the new default.

  • Nov. 22, 2007: Further improvement of initial winds. Wind speeds tended to be over-estimated at high latitudes using the default (relaxation of the relative vorticity). The boundary conditions have been slightly modified at high latitudes and the improved results can be seen in terms of the development of systems (especially those at high latitudes).

  • Dec. 08, 2007: Snow (on the ground) module activated. This is a very simple composite-type scheme. An explicit snow scheme is under development and will be implemented eventually. Added SnowDepth and Runoff plot diagnostics to output.

  • Dec. 12, 2007: Go from 3 to 5 soil layers. Use the GFS default 4-layer soil grid, but split uppermost layer into two layers: thereby adding a 5th, thin upper layer (for better resolving diurnal near-surface soil moisture fluctuations).

  • Jan. 7, 2008: 3 improvements to initial conditions and lateral boundary conditions; i) Initial wind fields now have a divergent component (derived from divergence computed directly from GFS outputs and then interpolated to ASP grid). The effect is relatively small, but positive. Total column divergence is initially zero (before digital filtering). ii) lateral boundary wind computation improved, result is slightly increased and improved wind speeds. iii) pressure thickness now relaxed along with other prognostic variables: result is primarily improved mass fields (an additional pressure relaxation term has been added to prognostic equations consistent with flux form). The latter two updates have significantly improved the forecasts, especially later in the forecast period.

  • Jan. 9, 2008: Moisture effects are now considered in the hydrostatic equation. One result is a more consistent temperature initialization with respect to the GFS (there previously was a slight warm bias in moist regions in the lower atmosphere). Because of this change, GFS air temperatures are now extracted from GFS output and interpolated (as opposed to deriving the air temperatures from the geopotentials). Geopotential is still extracted for diagnostic purposes.

  • Jan. 12, 2008: Input vorticity and divergence are used to derive the winds components for the initial condition *and* the lateral time forcing (assimilation scheme). Also, the second order horizontal diffusion scheme is being used in place of the 4th order scheme (as it is more physically based)

  • Jan. 30, 2008: Cold bias in lower atmosphere (related to very cold surface temperatures) corrected: problem was with the surface snow scheme. Lower atmospheric temperatures over snow surfaces much more in line with operational NWP models now.

  • Feb. 15, 2008: Replaced simplified shortwave radiation scheme (a simple empirical adjustment to the clear-sky approach to account for clouds) with a more sophisticated 2-stream approach with explicit accounting for multiple reflections and clouds. Some slight local impacts, but overall relatively small impact on large scale synoptic fields/forecast (although in sub-tropical to tropical zones the impact is a bit more significant, but still overall relatively minor).

  • Mar. 18, 2008: Several Updates: i) Increased time step to 3.95xd, ii) use a normal distribution form exponential function for lateral BC nudging (slightly more smooth transition than using a simple exponential function), iii) turn 2nd order horizontal diffusion OFF (main effect is to dampen systems a bit, so as using odd-ordered advection, no need for numerical diffusion. When move to higher spatial res, this might be turned back ON), iv) updated PBL routine (PBL height using local Ri as opposed to Bulk), v) surface turbulent fluxes modified a bit: use a convective velocity (Beljaars, 1994). vi) Downwelling shortwave, sensible, latent heat flux diagnostics added to plots, along with PBL depth.

  • Mar. 24, 2008: Use 5th order horizontal advection with 6th order monotonic horizontal diffusion. Recent studies show that this high order diffusion is good to use for the case of weak winds with 5th order advection. It essentially acts as a noise filter, while causing relatively little damping (compared to lower order diffusion).

  • Apr. 11, 2008: Bugfix: the shallow convection parametrization was being applied below the PBL as opposed to above it, causing widespread light precipitation, especially in the tropics and sub-tropics. The correction impact is most significant on the precipitation (reaching the surface): the dynamic fields are changed very little by this fix.

  • April 18, 2008: Initial and lateral boundary u-v wind components are determined using wind components interpolated from the large scale analysis (here GFS). An improvement to the wind rotation computation results in this wind determination option (as opposed to deriving filtered winds or deriving the rotational and divergent components from vorticity and horizontal divergence) performing the best. Column integrated divergence is removed only at the initial time. This method gives better overall forecast results especially beyond 48 hours.

  • April 22, 2008: Horizontal diffusion in regions of steep topographic gradients is reduced in middle and upper atmosphere, but in lower atmosphere it is large (in contrast to previous method which consisted in a linear reduction from the surface to the top of the atmosphere). Also, post-processor improved (vertical temperature interpolation to pressure surfaces below terrain heights).

  • May 6, 2008: Forecasts extended out to 3 days (from 2 1/2). With improved lateral boundary conditions (April 18, 2008), forecast quality more consistent with NWP models for longer range forecasts. Note that it seems that operational NWP-WRF (from NCEP) tends to often diverge noticeably from GFS by about 60 hours or so, so it is of interest to see how ASP compares to these 2 models for forecasts at and beyond 60 hours or so.

  • May 7, 2008: Move from 16 to 20 vertical layers. Tests have shown slight improvements in terms of precipitation (light areas reduced), and the handling of some upper level features. The grid is exponential with the greatest vertical resolution near the surface.

  • May 16, 2008: Remove topographic reduction of horizontal diffusion: this is done owing to a new computation for computing quasi horizontal corrections to scalars (potential temperature and specific humidity). These changes allow smoothing to occur over regions with strong topographic gradients (previously such regions could be a bit "noisy" at times)

  • May 22, 2008: Improved the SST initialization along coastlines: some small fractions of water were initialized using the skin temperature and thus remained quite high during the forecast integration thereby leading to anomalous surface fluxes: this has been all but eliminated.

  • June 10, 2008: 2 changes. i) 4th order divergence damping replaced by 6th order: almost no damping of divergence, except for very small scale (noise type) features. Systems slightly stronger, and virtually no overall damping of divergence field during integration. ii) Non-local PBL mixing scheme added (in addition to existing local-K turbulence). Effect is slightly more mixed PBL.

  • June 23, 2008: Transmission of shortwave radiation modified (liquid water paths re-scaled a bit) : gives improved surface shortwave (downwelling) fluxes (larger in cloudy regions) at the surface compared to the GFS model.

  • July 1, 2008: Longwave radiation scheme slightly adjusted: slightly larger cloud-free emissivities and a simple TOA downwelling flux computed. The LWdown at the surface much more consistent with that of the GFS model.

  • July 29, 2008: Move from 80 to 60 km horizontal resolution (the time step is decreased from approximately 5 to 4 minutes). The domain geographical size is approximately the same size, thus there is about a 50% increase in the number of grid points in the horizontal mesh. To offset (somewhat) the increased CPU demand, the radiative schemes are called every 15 minutes (as opposed to each model large time step): this results in a significant CPU economy while having an essentially negligible impact on the forecasts (this is a fairly common method used by NWP/GCM models to economize CPUs).

  • Aug. 1, 2008: Slight modification to output sea level surface diagnostic computation: the result is a much better agreement of this output diagnostic in mountain regions with operational models.

  • Sep. 1, 2008: Implementation and tests of new dynamic core based on temperature (T-core) as opposed to original potential temperature (PT-core). Hydrostatic equation uses hypsometric form (and equations are still integrated in flux-form). T-core results very close to PT-core, however some differences: amplitude of buoyancy waves a bit larger in PT-core, and cyclogenesis slightly more deep/rapid in T-core (although not dramatic). T-core uses slightly more CPUs because omega-term (energy conservation) must be computed, notably for each time split (but just a few %). Overall slightly better results using T-core.

  • Sep. 5, 2008: New hurricane (simple) modifications to sea surface routine: a sub-grid tropical storm/hurricane parametrization. Modification to surface drag in presence of hurricanes: development (changes to surface drag) based on several criteria; warm SSTs, low vertical shear and deep moisture. Vortex identified based on lower atmosphere relative vorticity. Finally, surface fluxes of heat and moisture also modified. Much better agreement with NWP models at a rather low cost (CPUs) and simple coding/methodology. Tests ongoing.

  • Sep. 7, 2008: Sigma-pressure vertical coordinate replaced by a hybrid pressure coordinate. In the lower levels of the atmosphere, the model still uses a sigma coordinate, however, in the upper levels there is a smooth transition to pressure surfaces (the upper-most levels are constants pressure surfaces). The model equations are still in flux form.

  • Sep. 30, 2008: Moved from 60 to 50 km horizontal resolution for Europe and North America limited area domains. Also, the high order horizontal hyper-diffusion coefficient was reduced a bit because it was previously large enough to cause some oscillations in some fields, such as relative vorticity: this artifact has now been removed with the new lower coefficient value.

  • Nov.. 18, 2008: 2 changes: 1) Now using the GFS-provided typography and land mask (0.3125 deg horizontal resolution). The topography is nearly unchanged, however the land mask is improved in low lying coastal areas. 2) The vertical turbulence scheme was updated using increased mixing (turbulent length scale parametrization) in the PBL following Beljaars (used in the local K-portion of the ECMWF scheme). This was found to improve vertical profiles in the PBL, and the impact on the atmosphere was to slightly damp rapid cyclogenesis (which is an improvement). Also, above the PBL over land, the K values are limited...this was found to eliminate some numerical spikes which could occur associated with highly convective PBLs. ASP results are quite close to operational models out to 72 hours now (seeming to fall within the model spread).

  • Nov. 20, 2008: Additional update to turbulence scheme: a non-local K-approach (much like that used in the GFS and ECMWF models now) is used within the PBL (while the local K method is still used above the PBL). This has resulted in several improvements: i) during cold air outbreaks over oceans/seas, too much precipitation was previously produced. Now increased vertical mixing has greatly improved this. ii) improved PBL depth (diagnostic), at least compared to GFS outputs, and iii) improved wintertime mid-Atlantic off-shore cyclogenesis. This change is effective as of runs on Nov. 20, 0Z.

  • Dec. 1, 2008: Moved from 20 layers to 21 layers through the addition of a sigma level between 0.990 and 1.000, at 0.997. This level is approximately 30 m above the surface and has been added at little additional computational cost in order to improve the representation of the surface (turbulent) layer.

  • Dec. 3, 2008: Added a counter-gradient term to the vertical diffusion (turbulence) parametrization. The main impacts are to increase vertical mixing and the PBL depth. Switched from 6th to 4th order divergence damping. This has virtually no effect on synoptic scale features, but does dampen some strong local vertical velocities which appeared using the new counter-gradient method in the vertical turbulence scheme.

  • Dec. 12, 2008: The minimum dynamic PBL heights in low-latitude regions tended to be too large over ocean/sea surfaces...resulting in anomalously deep PBLs and therefore too much mixing in the vertical (in the new non-local PBL parametrization, which has a strong dependence on PBL depth). This sometimes caused problems in low level atmospheric fields over these regions (such as excessive cooling). This was corrected by simply improving the computation of the minimum PBL dynamic height to give much more realistic (lower) values for low-latitude ocean/sea surfaces. Effective for the 12z run on the 12th.

  • Dec. 16, 2008: Adjusted hybrid coordinate a bit to be a bit smoother in the vertical. Also, lowered the pressure level at which surfaces become pressure surfaces to about 350 to 400 mb (it was about 250 - 300 mb). Little change in results, except simulated jet stream at 300 mb is a bit smoother.

  • Dec. 23, 2008: Vertical interpolation for vertical advection simplified, also go back to using explicit vertical advection. This is done because for current grid resolution (vertical), implicit not really needed (and some CPUs saved by using explicit method). Also, temperature no longer a test-value in time split equations (to maintain better consistency with omega computation within time split). The aforementioned changes have virtually no impact on results, but some CPUs saved. Finally, the gfortran compiler (from the Linux Mandriva 2008.0 and 2009.0 releases) now used (i.e. replaced the g95 pre-built compiler)...a nearly 40% improvement in run time with essentially no change in results.

  • Dec. 26, 2008: After further testing, air temperature vertical advection is now evaluated using the fully implicit method, but in contrast to before, this is now done within the time split loop. This added resolution permits a bit more flexibility in terms of vertical resolution.

  • Jan. 9, 2009: Improved lateral boundary conditions (best configuration was used for N-S boundaries, but was not for E-W boundaries...this bug was fixed). Also, topographic sub-grid roughness made more general (before a reduction adjustment was made compared to baseline parametrization from the literature...this reduction has been removed), resulting in slightly higher land surface roughness lengths. These two changes caused an improvement in the simulation of synoptic scale systems, which are now quite close to GFS even out to 72h (end of current forecast window).

  • Feb. 24, 2009: Model forecasts extended from 3 to 3.5 days.

  • Feb. 26, 2009: Two changes have been implemented as of the 12Z runs. i) First, a simple entrainment parametrization has been implemented into the turbulence scheme. This has been found to significantly reduce light rainfall associated with cold air outbreaks over warm water bodies. The associated cloudiness has also been reduced, improving radiative cooling in the lower atmosphere. ii) The second change is to replace the average emissivity parametrization in the longwave radiative scheme with a maximum random overlap based approach. This has been found to improve the impact of clouds on the downwelling surface flux and to improve lower atmospheric radiative cooling. Change (ii) has been found to also have a positive impact (non-negligible) on developing synoptic waves.

  • Mar. 19, 2009: Replace the vertical diffusion of potential temperature by enthalpy, using a moisture dependent air heat capacity at constant pressure. The result is slightly increased sensible heat fluxes over relatively warm water surfaces: elsewhere the impact is generally less. The land surface routines surface energy budgets and flux computations were modified to use a more arbitrary heat variable for coupling with the atmosphere (it is now straightforward to diffuse potential temperature or enthalpy or dry static energy and maintain implicit coupling with all surface tiles).

  • Mar. 26, 2009: Go from a grid spacing of 50 to 55 km for the US and European domains (for same number of grid points, so a 10% larger domain). Very little change to results, but put's boundaries a bit further away from zones of interest. Also, speeds up the runs a bit owing to the slightly larger time step.

  • Apr. 1, 2009: 2 changes: i) The imposed maximum diffusivity in the vertical turbulence scheme and the maximum for the counter gradient terms have been relaxed (increased). The result is that they are much less often reached and diffusion can be larger, especially in the PBL. The result is slightly better (more deep) cyclogenesis (a bit more consistent with operational models). ii) Also, the convection scheme has been modified slightly such that heating from condensation is re-distributed vertically such that more heating reaches higher in the cloud layer. The result is a larger impact on the synoptic evolution owing to convection, again which is more consistent with operational models. Also, convective precipitation is increased owing to both changes.

  • Apr. 8, 2009: Modified the hybrid hydrostatic pressure coordinate. It now is defined in a very similar manner to WRF-NMM. Compared to the previous coordinate, the flattening of pressure surfaces occurs more slowly away from the surface over terrain. This change has been found to cause a noticeable change, especially late in the forecast: ASP results are now even more consistent with global operational models (this is assumed to be an improvement!).

  • Apr. 16, 2009: Raised maximum PBL sigma level from 0.6 to 0.5 based on PBL heights in the GFS model. This gives improved lower atmosphere heating over Africa for example. Bug fix: solar zenith angle computation had errors for east longitudes...this has been fixed and checked against data.

  • Apr. 22, 2009: Slight modification to Ekman-type PBL depth over water...slightly increased to give PBL depths which are more consistent with GFS over oceans. A slight improvement in surface pressure of oceanic cyclones noted. Note the actual PBL depth is the maximum value of the Ekman type depth and one based on the Richardson number.

  • Apr. 24, 2009: Slight modification to longwave radiation routine. A minimum downwelling flux at the surface is imposed. Downwelling fluxes in the entire profile are adjusted if the minimum is not respected, and energy is conserved. This correction is only rarely used for very dry (and possibly cold) conditions.

  • May 10, 2009: A modification to the turbulence scheme has been included to incorporate conditional stability (like a moist convective adjustment) to the vertical diffusion scheme. The main result is slightly amplified systems. Also, an explicit computation of the moist adiabat has been incorporated into the convection scheme to improve estimates of CAPE.

  • May 20, 2009: A slight model cold bias has been reduced substantially by reducing the atmospheric longwave radiative cooling rates. A simple factor reduction gives results much more consistent with operational models in terms of warmer lower-tropospheric air temperatures (at 850 mb for example) and accordingly higher geopotentials. Total energy is conserved when applying this factor.

  • May 27, 2009: Go from 21 layers to 22....essentially kept the same grid configuration except added another layer by extending the upper pressure (top) from 100 to 50 mb.

  • May 29, 2009: Two modifications to the vertical turbulence scheme (above the PBL). Improved modifications to moist convection in the turbulence scheme. Result is more widespread convection and reduced precipitation maxima the tropics (improvements). Also, shallow convection scheme modified to reduce vertical mixing over high altitude points, which results in lower precipitation peaks which sometimes occur over mountains.

  • June 6, 2009: Because of recent changes to increase/improve mixing, it has been found that precipitation evaporation thresholds can be lowered and are therefore more realistic and consistent with operational model values.This improves lower atmosphere temperatures notably in the sub-tropics. Also, a vertical heating profile distribution is introduced into the convection scheme...this implicitly represents vertical mixing/upward heat transport: it reduces sometimes large convection precipitation totals. Finally, the May 29, 2009, turbulence scheme modification to high altitude mixing has been undone as it has little effect and is offset somewhat by the aforementioned vertical heat redistribution method.

  • June 30, 2009: It has been found that when both the moist convective adjustment and deep convection are both active at the same grid point and time, local precipitation amounts can sometimes be quite high in the summer. A small adjustment has been made such that when there is deep convection, moist convective adjustment within the same grid cell is reduced/shut off (if deep convection is strong enough). This has been found to reduce/remove the occasionally anomalously large rainfall amounts associated with convection.

  • Sept. 15, 2009: Convection has been modified: it was found that a moist convective adjustment was being used at the same time as the deep convection scheme, sometimes resulting in very high convective rain rates. This has been improved: now, the two schemes are not used at the same time. The result is improved rain rates, and also the reduction of quasi-stationary deep convection (which was over-estimated) and a reduction of a sometimes present over-estimation of convection over relief.

  • Sept. 26, 2009: "Big Brother" 4-d assimilation scheme implemented. The assimilation method is simple: a Newtonian nudging towards the large scale model early in the forecast. For the first 6 hours, the forecast is nudged towards the large scale model forecast, then in the 6 to 12 hour forecast period, the nudging is relaxed to zero. The hypothesis is that the forecast model is very close to "reality" in the first 6 to 12 hours. Although the ASP initial fields are filtered/balanced owing to the digital filter, the assimilation can improve the forecast trajectory under some circumstances. Generally, this nudging has little to virtually no visible effect on the forecast. This is especially the case under fast flow (zonal) regimes where the influence of initial conditions is lost fairly quickly as lateral boundary forcing becomes most important (for the extended period). But under high amplitude flow regimes, sometimes this assimilation can have an effect, and it is almost always positive (a small to moderate improvement). As it is essentially cost-free (no additional pre-processing is needed and essentially no additional CPUs are needed) and almost always improves the extended forecast (under some conditions), for now it is the new default.

  • Sept. 27, 2009: A mass-flux type scheme for convection/turbulence is implemented. A vertical velocity is computed using CAPE within the convection scheme, and this is then transformed into a diffusivity within the turbulence scheme. The result is increased vertical transport of heat and moisture above the PBL within convectively active zones. This is turn results in generally larger convective precipitation amounts, and deeper weather systems (in which convection is important).

  • Oct. 6, 2009: The land surface and soil parameters are now generated from ECOCLIMAP. One major advantage is that the spatial resolution can be adapted to the ASP grid (down to 1 km). In addition, GFS volumetric water content is now transformed to SWI using GFS soil class/texture information, which is then interpolated to the ASP grid, then transformed back to volumetric water content using ECOCLIMAP texture information.

  • Oct. 18, 2009: A new, simpler longwave radiative transfer scheme has been included as an option in the code, and it is now used as the DEFAULT. A so-called "fast" method, it is much more simple than the old default 2-stream approach, but has been found to give generally improved forecasts at a much lower cost. The lower atmosphere does not cool as much in general, which is an improvement. Also, sometimes rather excessive convective precipitation totals have been reduced. A slight improvement in long term forecasting of cyclogenesis has also been noted.

  • Oct. 21, 2009: Two bugs fixed. GFS soil moisture was being over-ridden by climatological values in the initialization...this has been fixed. Also, the minimum stomatal resistance was not set, so it was zero thus evaporation was relatively high, even in dry areas. This was also fixed.

  • Nov. 5, 2009: Replaced 2-stream shortwave radiative transfer routine by a much simpler "fast" scheme. The overall synoptic scale results are nearly identical, at a much reduced cost. Also, longwave and shortwave routines are now called at each large model time step since they go so much faster (the longwave code also uses a "fast" approach, see Oct. 18, 2009 item in this listing). Note, both the more detailed (and considerably slower/more intensive in terms of CPUs) longwave and shortwave codes can be used as an option, but are no longer the defaults.

  • Nov. 17, 2009: More computationally expensive flux emissivity longwave radiative transfer algorithm reinstated since it was found it gives better long range forecasts. It is called every 20 minutes (the shortwave routine still uses the "fast" method at each time step).

  • Jan. 1, 2010: Replace mean sea level pressure computation using vertical interpolation of along topography surface temperature through mountains to a method using horizontal relaxation to obtain the mean sea level pressure field. This reduces sometimes sharp gradients in and along high altitude regions.

  • Jan. 7, 2010: Fixed a recent bug which caused soil ice to be constant.

  • Jan. 16, 2010: Added a method to filter dynamic tendencies where map factors exceed unity. This permits a larger time step to be used (for large domains). Applied each RK step. It has been found to have virtually no impact on results, but in practice for the North American and European domains, time steps 10-15% larger can be used.

  • Jan. 29, 2010: Increased moist convective mass flux (triggered more often/widespread): the result is more rapid and intense wintertime cyclogenesis along Gulf of Mexico coast and off SE US coast, and a slight acceleration of the corresponding systems (both are improvements). Results very much more in line with the NAM and the GFS for such systems. Also, use flux form of relaxation equations for lateral boundaries: this gives slightly improved surface pressures, but otherwise has little effect.

  • Apr. 1, 2010: It was found that sometimes rain rates associated with cyclogenesis tended to be a bit excessive. This resulted because several types of convection occurred at the same time (which was not very consistent). Now, the three main types of convective regimes occur separately, with a smooth but rapid transition between them. Deep convection gets priority, next convection due to conditional instability (which is implemented as a vertical diffusion augmentation), and finally if neither of the aforementioned types of convection are activated, then it is possible to have shallow convection (which is also implemented as a vertical diffusion increase). This has little to impact outside of regions of cyclogenesis.

  • Apr. 16, 2010: Modification to the PBL depth computation was made. For shallow PBLs, parcel used to compute the PBL depth is from the lowest model layer (old default). But now, when PBL is sufficiently deep, the parcel is computed at one-tenth of the PBL depth. This results in generally lower depths (especially over arid regions), which seems more realistic. Also, the temperature excess term is limited as PBL depths become very large.

  • May 13, 2010: Added a passive tracer prognostic variable to the model. Numerical treatment by dynamic core is exactly the same as water vapor scalar, with a few simple physics routines added.

  • May 22, 2010: Added a 6-hour data assimilation cycle. A very simple 4DDA method is used. The ASP forecast from 6 hours before the current forecast base time is used as the starting point, and the atmospheric prognostic variables are nudged towards the (filtered) analysis. The result is a forecast analysis which is very close to the actual analysis, but the reason to do this is threefold: i) this gives the physics a short period to spin-up, ii) this permits an initialization for certain fields not in the NWP analysis (like passive tracers), and iii) certain land surface fields in the analysis (like soil moisture) seem to better initialized using forecasts rather than the GFS analysis owing to consistency issues and doubts about the quality of the current GFS soil moisture initialization. This of course can permit drift owing to an accumulation of errors over time, so a nudging of soil moisture is being tested. The overall impact on the forecast is relatively small, but it is being used mainly owing to reasons ii and iii above. More sophisticated atmosphere and land surface assimilation methodologies may be used at some point (using observations etc..).

  • May 25, 2010: A slight modification to the shallow convection scheme. A conditional instability criteria was added over land. This was found to improve (increase) the presence of convection, improve 850 mb temperatures (by decreasing vertical mixing a bit), and to decrease sometimes very high local convective precipitation amounts. In drier regions, this has no impact.

  • May 29, 2010: 2 changes. i) An additional slight modification to the shallow convection scheme has been added. The PBL convective velocity (w*) must be positive for shallow convection, the simple idea being that it acts as a trigger mechanism. This further improves 850 mb temperatures (compared to modifications from May 25, 2010) over humid continental regions, has little impact over the ocean (where results were already reasonable), and has improved the diurnal cycle of convection a bit over the aforementioned land regions. ii) The "Big Brother" 4DDA scheme (Sept. 26, 2009) has been turned OFF for now. The reason is that it no longer seems to improve results since the May 22, 2010 assimilation cycle was added. This implies that letting the physics spin-up for 6 hours along with using land surface initialization based on the past ASP forecast state advanced to the analysis time reduce forecast errors. But tests will continue over the next year to make sure that there is not a seasonal dependence (i.e. to make sure the Big-Brother 4DDA does not improve the forecast in winter for example).

  • June 6, 2010: The "Big Brother" 4DDA scheme (Sept. 26, 2009) has been turned back ON. It seems to have a slight positive impact, although this will continue to be investigated. An additional slight modification to the shallow convection scheme has been added. A large-scale lift criteria has been added. This has been found to slightly improve the intensification of systems. For areas outside of cyclogenesis, this has little to no impact.
  • June 18, 2010: During the assimilation cycle, assimilation of the main prognostic atmospheric variables is now done as an adjustment rather than in the dynamics solver. This permits a stronger assimilation and a state at the initial time which is very close to the analysis (closer than the previous method). Also, in the deep convection scheme, CIN was used to prevent convection. CIN can now be reduced if sufficient w* and/or large scale forcing (at the LCL) is present. This allows parcels to overcome strong CIN if sufficient lift is present. Finally, an additional criteria is added for shallow convection triggering: it can be initiated if weak large scale lift if atmosphere is sufficiently unstable. This causes a bit more widespread mixing than in the previous version (which may have reduced mixing a bit too much).

  • July. 17, 2010: Three updates: i) The initial state is now based on a previous forecast integrated to the current initial time with the simple assimilation scheme which nudges the past forecast towards the analysis. This results in slight differences in the initial state. The "big-brother" assimilation scheme is still on, so the slight differences in initial state don't have much impact on the forecast...but lower atmospheric variables are much more consistent with what ASP predicts (spin up is reduced). ii) Also, along the lateral boundaries, pressure is no longer nudged (this has very little impact). iii) Finally, a simple pressure tendency filter has been implemented which has little effect on the dynamic fields except to strongly dampen noise which can sometimes develop where strong convection occurs.

  • July. 20, 2010: Big brother assimilation during first 6 hours of the forecast is now off. The use of assimilation to obtain the initial state seems to have largely negated the need to use the big brother mode. So current initialization procedure consists in i) if a past forecast is present, then this is nudged towards the analysis to obtain an initial state. The forecast then proceeeds. ii) if no previous forecast is found, then the cold-start method is used with digital filtering. Also, a constraint on maximum PBL depth is removed which seems to improve 850 mb temperatures a bit.

  • Sep. 1, 2010: The hurricane parameterization (impacts only the convection routine directly) was modified to reduce the rapid decay of systems when encountering sub-optimal conditions. The result is stronger and longer lasting tropical systems.

  • Sep. 26, 2010: Several PBL and surface related changes have been made, which result in a significant reduction (improvement) in the low level atmospheric positive temperature bias over continental regions for convective PBL conditions. Summary: It was found that surface roughness lengths in the input were a bit large relative to theoretical values, so the input is scaled accordingly. This simple fix gives 850 mb temperatures much more consistent with operational models (GFS and NAM) and was found to have the biggest impact on essentially eradicating the bias. Lowest model level air temperatures (and screen level T's) are also improved (reduced) for such conditions. Also, the surface stability functions for computing the drag coefficient were adjusted (down) slightly since drag seemed to be a bit large on low level winds. This doesn't have much effect except for developing lows, and this seems to improve their development owing to slightly reduced surface drag (this is a tuning adjustment). Finally, the PBL top computation has been adjusted slightly such that the near-surface temperature used to compute the bulk PBL depth is computed at the top of the surface layer (i.e. currently defined as one tenth of the PBL depth) rather than always at the lowest model level. This does not have much effect except that the diagnostic PBL depths are a bit lower (and a bit more realistic).Finally, the entrainment adjustment to K is no longer just right at PBL top, but uses a continuous function through and just above the PBL. But this function is nonlinear and so it doesn't have a dramatic impact, except that in some convective PBL situations it seems to further improve (reduce) the aforementioned warm bias by more efficiently evacuating heat from the PBL (than before).

  • Nov. 10, 2010: 2 changes since last update: one to physics, the other to the numerics in the dynamics. (i) A maximum PBL height increase rate has been implemented which results in slightly improved PBL heights (sometimes they seemed to be locally a bit high). (ii) Also, a simple postitive definite flux corrector has been added to the horizontal scalar advection (5th and 6th order since they are the default schemes used) computations. This does not have a very significant impact on storm development or precipitation, but an improvement in the passive tracer fields is noted.

  • Nov. 26, 2010: Bug-fix: a bug in the soil freezing routine prevented soil freezing.

  • Dec. 9, 2010: Replaced moist convection Richardson number scheme (in vertical turbulence module) by a more classic moist-Richardson number (cloud regions) scheme. The result is nearly the same in terms of cyclogenesis, except that the new scheme has a better representation of RH in mid-levels of the atmosphere (the old scheme seemed to mix too much). Also, the shallow convection scheme is now used always (if certain criteria are met: previously it was also shut off when the moist-convection Richardson number scheme was activated because excessive mixing occured when the two were both active). The new scheme is much more continuous (along with better RH fields and slightly cooler air temperatures at low levels, which is also a slight improvement).

  • Jan. 8, 2011: Two updates. First, the turbulence scheme now computes a alternate diffusion coefficient if certain convective criteria are met (large sclae lift, convective PBL, sufficient moisture in the atmosphere) when the deep convection scheme is not active. This convection is shallow to mid-level in nature. It was found to improve cycogenesis over warm waters such as along the East coast of the US in winter. The second change is the replacement of the composite snow scheme by an explicit 2-layer scheme.

  • Jan. 10, 2011: Improved explicit snow scheme: heat content became unrealistically cold during the day for certain conditions, especially for fairly thin snowpacks. This has been corrected and now results are much more realistic.

  • Feb. 05, 2011: Changed computation of sea level pressure for cold conditions. This only has an impact in mountainous regions. The only change is a new computation of the sea level temperature below relief. As air temperature warms, this gradually reverts to original method (which seems to perform better for warmer conditions). Purely a diagnostic: has no impact on model computations.

  • June. 06, 2011: Several (3) turbulence related changes. 1) Maximum convective velocity for turbulence scheme computed as a function of CAPE as opposed to MSE. Reduces mixing mainly over warm oceans: found to improve both low atmosphere temperatures (i.e. 850 mb) and near surface RH values (more moist). Shallow convection K incorporated into single bulk Richardson number (slightly offsets previous change, but generally very little impact but simpler code/equations). 2) Surface sub-grid orography effect on surface drag extended into lower atmosphere via an increased K for momentum (based on sub-grid topographic variability). Just using surface roughness increase alone indeed slows flow over mountainous areas, but tends to slow near surface winds in these areas too much. The increased PBL K coefficient increases momentum transfer towards the surface thereby offsetting the anomalously low near surface wind speeds while still increasing drag. Effects bring cyclogensis results over rough topography much more in line with operational NWP models (WRF, GFS) using a very simple scheme. 3) Final change is to slightly adjust dynamic surface roughness for heat by imposing a lower limit (to prevent too much de-coupling/too much surface heating) and allowing the ratio z0h/z0m to approach unity for the special case of dense and tall vegetation cover (forest). This mainly improves excessively hot surface temperatures over desert regions for example.

  • June. 10, 2011: Upgrade to shortwave radiation scheme. Improved cloud water computation (and scrapped a somewhat arbitrary cloud scaling method). Result is reasonable heating rates (improved in theory, based on 1D tests) and surface downwelling all-wavelength total solar flux (it is very close to GFS NWP model now). Also fixed a slight positive bias in this flux (again, relative to GFS).

  • Aug. 02, 2011: Modified convection scheme to have increased widespread convection. Essentially, the convective timescale is decreased (more rapid) if CAPE is present and the equivalent temperature is large enough. This has improved summertime cyclogenesis, and also tropical storm/hurricane formation and intensification.

  • Sep. 09, 2011: Set maximum free atmosphere (above PBL) vertical diffusion coefficient to be 100 m2ss-1. This produces atmospheric (above PBL) RHs very close to operational NWP models. The old value of 1000 (PBL value) caused RHs to mix out a bit too much. Not a very big impact on other fields. Also, removed the water vapor counter gradient term. Indeed, this also has little effect and is consistent with recent papers on this type of PBL approach (e.g. Hong et al., 2006, MWR).

  • Sep. 29, 2011: Modified the convection scheme. Removed notion of convective timescale. Also, modified convective fraction (latent heat release) to increase as a function of CAPE. This minor modification improves spatial distribution of convective rainfall and reduces zones with sometimes large CAPE (as more CAPE is used up by convection). This even improves dynamics, such as the AEJ location and intensity. Finally, a check is put in to make sure (fairly rare) excessively large convective rain rates don't occur: in such regions the convective fraction is reduced back towards the original non-CAPE modified value.

  • Nov. 1, 2011: Move from 22 to 24 layers in the vertical: only add additional vertical resolution above 300 mb. This was found to improve the smoothness of high altitude jets. Only a very minor impact on lower levels for forecasts out to 3.5 days.

  • Nov. 11, 2011: Major Update: Replaced Kuo-convective scheme by a mass-flux convection (MFC) scheme (following very closely the scheme presented by Bougeault, 1985). The only two additions to the MFC scheme relative to the original Bougeault model are (i) the introduction of a trigger function (which is a smooth transition/continuous function) based on CIN (prevents convection if the scaled CIN is too large and vertical forcing weak) and the large scale relative humidity in the cell with convection (to reduce convection if the atmosphere in the vicinity of the cloud is very dry). The other addition (ii) is the inclusion of a Kessler form evaporation of rainfall in the lower and below cloud regions (if the grid scale RH is sufficiently low in these regions). The impact of this latter change is to possibly reduce a bit the warming and drying in the lower cloud region, and possibly induce colling and moistening below the cloud (a simple downdraft representation). The impact of this latter change seems to be fairly minor. The main changes by introducing the MFC scheme noted are i) increased CAPE, ii) increased vertical velocities in convective regions, iii) increased convective warming, iv) more transient convection compared to the Kuo scheme, v) convective regions which seem to be a bit more correlated with the dynamics associated with regions of lift, and vi) increased mid-level drying. In preliminary tests it is seen, for example, that the MFC scheme produces vertical velocity fields, 850 mb temperatures and even cyclogenesis more in line with the operational GFS (and reality) than with the Kuo scheme. The only aspect which may eventually be adjusted is the occasional presence of what seem to be slightly large convective rain rate maxima, but for now no modifications for this have been done.

  • Nov. 27, 2011: Discovery of a bug: during condensation for temperatures above freezing, the latent heating was not properly accounted for in the atmosphere. For mid-latitudes in winter, this problem was not very evident as freezing levels tended to be low. But, cyclogenesis in the summer was indeed slow and weaker than observed generally, and this was directly related to this bug. It has now been fixed. However, this has lead to occasional convective- large-scale precipitation positive feedbacks. Similar effects were observed for NWP models without microphysics (like the ETA from NCEP years ago): so for now, a simple precipitation rate maximum has been imposed which alleviates this problem. Currently, a simple microphysics scheme is being developed which should remove the need for such a limit. Overall, summertime cyclogenesis over mid latitudes is greatly improved.

  • Dec. 18, 2011: Major Update: Introduction of a simple microphysics scheme. Like ECMWF (IFS 2010), a single bulk microphysical cloud water (ice and liquid phase) prognostic variable has been introduced. The ice and liquid compositions are determined as a function of temperature. Autoconversion, evaporation/sublimation of cloud water/ice and (as before) evaporation/sublimation of rainfall/snowfall are modeled. As in ECMWF, there is no rainfall/snowfall prognostic variable. Saturation specific humidities and vapor pressures are computed over ice and liquid water (using the aforementioned temperature function for the paritioning). The parametrizations for the above processes (including diagnostic cloud frazction) are taken from the IFS. There are two simplifications done compared to ECMWF: that model also uses a prognostic cloud fraction, whereas ASP uses the diagnostic fraction functions only. And, all latent heats in the microphysics routines use the latent heat of fusion as in Grabowski (1998): this ensures energy conservation and simplifies the computations considerably. The main impact of the new microphysics are i) reduced large scale or stratiform precipitation, which agrees much better with operational NWP models, and ii) also in much better agreement with NWP models, the atmosphere is considerably more humid in regions of lift and precipitation. Finally, the cloud water content is now used directly in the solar radiation routine (as opposed to the previous parametrization which was based on an empirical method tuned to agree with GFS SWdown fluxes). Note that currently no cloud water variable is imposed along the lateral boundaries, and cloud water is recycled during the assimilation/initialization phase.

  • Jan. 23, 2012: Slight modifications to deep convection scheme. Added a simple downdraft parametrization. Effect noticeable only for heavy convective precipitation events: results in formation of a cold pool near the surface (and helps reduce sometimes large local precipitation amounts when strong convection active). Also, add a simple convective parametrization to the turbulence scheme: for strong deep convection, vertical turbulence increased in the cloud region. main effect is also to reduce a bit sometimes large precipitation bulls-eyes which could occur for strong convective events.

  • Apr. 20, 2012: 3 Major modifications. (1) The Bougeault mass flux convective scheme has been extended to a bulk mass convective scheme similar to those proposed by Tiedtke and Zhang and McFarlane. For example, closure is now a function of CAPE, there is a downdraft parameterization, and upgrades to the convective microphysics. (2) Also, the resulting mass fluxes are used within the turbulence scheme to enhance mixing where convection is occuring. (3) Finally, the implicit vertical advection in the dynamic core is now ony used for temperature, and no longer for other scalers as it was more costly and essentially did not change results (but it can be reactvated at any point as it's an option). Also, the simpler vertical advection of water vapor enabled the implementation of a more detailed vertical interpolation method (for computing vertical advection fluxes at levels) which improved high-atmosphere water vapor/relative humidity profiles (occasionally numerical minima were observed at very high levels as downward fluxes could be too large owing to nonlinear change of water vapor with height). Also, a simple flux corrector was added to the explicit vertical advection. Overall warm season structures and fields are improved, including surface rainfall, monsoon dynamic features, atmospheric RH profiles, temperatures etc.

  • Apr. 29, 2012: Geleyn type (Richarson number modification) shallow convection turned off as shallow convection now modeled by the mass flux scheme (see Apr. 20, 2012 updates). This improves low level humidity, and causes an increase in CAPE.

  • May 18, 2012: A stratiform precipitation limit added to avoid high precipitation amounts sometimes occuring in convection; Has very little impact overall escept to reduce sometimes very large precipitation amounts where intense convection occurs. Also, 2m air temperature diagnostic adjusted as it was seen to be too high (too close to surface T and much larger than lowest atmopshere level T) over hot, desert regions.

  • June 9, 2012: Sea module modified to include a skin temperature diagnostic and a prognostic deeper layer near surface (currently 3 m) sea surface slab temperature. Based on the work of Fairall et al. (1996) and Zeng and Beljaars (2005). Similar to the IFS implementation described in the 2012 Physics documentation. Found to generally improve low level air temperatures in tropics and sub-tropics, and to improve precipitation.

  • June 30, 2012: Replaced relative humidity-based buoyancy effect on Richardson number (in the free atmosphere turbulence scheme) by a cloud-water (microphysical) effect (based on Smith, 1990, QJRMS). The results are generally similar (the impact is not large), however, with this new scheme, the shallow cumulus (Geleyn-type) Richarson number modification is no longer needed (it was included along with the RH effect previously).

  • Aug. 15, 2012: Modified the vertical turbulence scheme to use the so-called EDMF approach (combined Eddy-Diffusivity Mass-Flux) It is largely based on the implementation in the ECMWF model (IFS 2012). The diffusion part is unchanged, hwoever, the counter gradieint terms in the turbulence scheme are replaced by mass-fluxes computed using plume equations and simple microphysics. The main impact has been found in terms of lower atmospheric relative humidity, which tends now to be larger (and in better agreement with operational model analysis) over humid regions, especially tropical and sub-tropical oceans.

  • Aug. 23, 2012: The so-called "cool-skin" SST method was found to cause some numerical instabilities at times. So, the explicit flux computation was replaced by a locally implicit method which removed this effect. For most cases this has virtually no effect, except it has effectively prevented rare stability problems.

  • Sep. 07, 2012: 2 updates: 1) Modified convective timescale in mass-flux convection scheme: simplified it. Previously, the baseline timescale was adjsuted as a function of environmental humidity, depth of the cloud and convective vertical velocity. Now, these dependences have been removed, which has removed several model parameters and improved results (the timescale is generally lower now). The main effects of the lowered timescale are; intermittent weak convection has been increased a bit and it is a bit more widespread, especially over relief: this is a notable improvement. Also, in tropical regions, the occasional large precipitation event/amount has been reduced (which is an improvement): this is because the shorter timescale leads to a greater stabilization of the atmosphere thus reducing explicit convection (which resulted since not enough instability was being removed by the convection). 2) Also, the sea routine was modified slightly to reduce sensible heat flux slightly over warm tropical regions: the result is slightly cooler and more moist lower tropospheric temperatures, more in line with operational NWP analysis from the GFS...this has a fairly small impact on the model.

  • Dec. 24, 2012: Bug in high level cloud fraction corrected. This bug resulted in an underestimate of upper atmosphere (400-200 mb level) cloud cover fraction (diagnostic). It has been corrected. The shortwave radiation computation was adjusted slightly since the new increased cloudiness reduced incoming shortwave radiation too much: the scale factor for absorbtion/transmission was adjusted to give shortwave radiation at the ground similar to previous values (tuned to be close to GFS values at the surface). Finally, evaporation of cloud water was slightly adjusted such that cloud water can not exist in sub-saturated regions: this helps improve cloud fraction.

  • Jan. 06, 2013: A numerical problem was fixed in the PBL EDMF scheme. At some pixels, excessive drying was found in the lower atmosphere (which could even influence the soil in some cases) which was related to numerical effects: minor modifications to the CFL limit and to the simple microphysics eliminated the problem. Also, the snow cover fraction routines were slightly modified to permit a higher snow fraction for baresoil and for forested regions when LAI is low (deciduous trees in winter for example). These two changes mainly melt snow slightly less, leading to a slight improvement in snow cover and accumulation (especially for snow falling on previously snow-free areas).
  • Jan. 25, 2013: It has been noted that with the improved (mass-flux) convection and turbulence schemes (EDMF), cyclogenesis is generally now more deep than for certain operational NWP models (UKMO, ECMWF, GFS, GEM...). To remedy this (without tuning etc...), two disactivated schemes have been reactivated: physically based Smagorinski horizontal diffusion and sub-grid topographically enhanced surface roughness. Both of these schemes were disactivated in the past, in fact, because they tended to damp cyclogenesis a bit too much. But the more recent improvements in convection and turbulence now permit the aforementioned schemes to be re-activated. And, cyclogenesis (and decay of systems) is indeed more in line with operational NWP models.

  • Apr. 20, 2013: Updated precipitation evaporation computations for convection and large scale precipitation. This reduces the precipitation amounts in both cases which seems to be an improvement (compared to operational NWP models at least).

  • Apr. 25, 2013: Option to use 5th order WENO advection has been implemented. Currently, it is now only used for the scalar variables passive tracer and cloud water content. The time integration of these scalars has been moved outside of the dynamic core: thus they can use different Runge-Kutta time stepping methods (from that used for the WRF-type advection schemes currently used in the dynamic core). Also, note that this implies the wind components (horizontal and vertical and specific density/surface pressure) are constant during the RK loop and correspond to the values at the end of the time step. This doesn't impact results noticeably and makes a more modular code. The main impact of using the WENO scheme can be seen in the simulated tracer fields: the overall features after a certain integration in time are quite similar using either 5th order WRF or 5th order WENO (as one would hope!), however occasional Gibbs-type (oscillatory) phenomena (i.e. anomalous local maximums) seen using the 5th order WRF scheme are removed using the 5th order WENO scheme (which is the expected/desired effect).

  • Apr. 27, 2013: 5th order WENO advection used by default for water vapor, in addition to cloud water and the passive tracer (previous update). Overall, not much impact (convective precipitation peaks tend to be slightly lower, but overall, not much impact).

  • May 10, 2013: Mass flux convection scheme updated: now use fully coupled cloud model equations with entraining and detraining plumes to determine cloud properties (temperature, specific humidity, water content) profiles: previously more simple entraining uncoupled plume model used. The modification shifts heating profiles slightly higher (which seems to be an improvement) and atmosphere has more CAPE (which also seems to be an improvement as CAPE consumption seemed to be slightly overestimated previously).

  • May 25, 2013: Go from 24 layers to 27 in default runs. 3 new layers were added to upper troposphere to better resolve convective cloud tops. The effect is relatively minor.

  • June 12, 2013: In EDMF scheme, fluxes go to zero in a layer above PBL with a thickness based on the surface layer temperature excess. This causes CAPE to be slightly less, which is more in line with operational models (NAM and GFS). Also, in the convection scheme, assume a minimum large scale forcing timescale (1 hour) which is larger than the PBL-rooted deep minimum convective timescale. This causes slightly more intense and localized convection.

  • June 22, 2013: Updated EDMF scheme: modify detrainment computation. Detrainment now a constant in the vertical, and it is equal to the entrainment at cloud base (at the LCL). This was found to reduce the rather high convective rainfall amounts.

  • June 25, 2013: Slightly modified turbulence (above PBL) to possibly increase mixing (under certain cricumstances depending on the vertical spacific humidity and temperature profiles) : main effect is to reduce light precipitation areas over oceans.

  • July 3, 2013: Modified entrainment formulation: the old formulation was from the current IFS, whereas now it has two terms with one depending on RH (with a cubic dependence on saturation vapor pressure) and the other depending on LCL depth (with a quadric dependence on saturation vapor pressure). The result is slightly lower widespread precipitation amounts (an improvement), but more importantly, much better interactions with mid-latitude baroclinic cyclogenesis (the old method tended to either damp systems too much or result in over-estimated intensification from feedbacks: the new method works better for both cases).

  • Sep. 1, 2013: Changed to double precision coding/compilation leading to a 30% reduction in CPU time using optimization level 4 with the gfortran compiler. For now, have increased the domain size for the CONUS and Europe each by 15%.

  • Sep. 7, 2013: Initialize cloud water content from NWP input, and also impose along lateral boundaries. Only slight changes seen within the domain to precipitation and cloud water.

  • Sep. 21, 2013: Modified single cloud water prognostic variable into two variables: cloud ice and cloud liquid water (i.e. a new prognostic equation). Latent heating (sublimation and vaporization) now accounted for.

  • Sep. 29, 2013: Added bulk microphysical prognostic variables for rainfall and snowfall, thus there are now 5 microphysical variables (the 2 aforementioned variables along with cloud ice and liquid water, specific humidity).

  • Nov. 30, 2013: Activated second order physically-based deformation-based horizontal diffusion: to reduce slightly excessive cyclogenesis.

  • Jan. 10: Adjsuted microphysics parameters slightly to get cloud water content more in line with NWP models (GFS). This meant increasing auto-conversion a bit. Impact on the other fields was rather small, except for shortwave radiation at the surface which now seems to be slightly improved.

  • Feb. 10: Accretion formulation for the passive tracer was found to be too strong: a different corrected parameterization was put in place, which results in significantly less reduction in tracer concentration from precipitation.

  • March 5: Implemented a simple orographic drag scheme (based on the IFS implementation described by Beljaars et al., 2004, Q.J.R.M.S.). In fact, the scheme greatly improved the sometimes excessive cyclogensis over continental areas characterized by significant topographic variability.

  • May 17: Implemented a new entrainment forumation for the mass flux convection scheme. For dry layers, it resembles the original IFS-based formulation. For moist layers, it is somewhat similar to the Hong et al forumation in that a background entraiment dominates. Also, the LCL computation is modified and produces slightly higher (more realistic) LCL depths. The result of these modifications is that while the spatial coverage is nearly the same, the intensity is decreased (which is an improvement), mainly owing to a slightly larger entrainment. Also, the heating profile maximum is lowered a bit (the mass flux profile maximum is a bit lower), which is more in line with schemes in operational and GCM models.

  • June 12, 2014: Updated convection scheme. Parcel entrainment for finding LCL was an order of magnitude too small...small changes result. Trigger now used to weight convective and large scale forced convection, before a maximum function was used: only small changes. As in IFS c40r1 (2013), replace dilated CAPE with potential CAPE in computing CAPE-based closure. The main result is larger deep convective mass fluxes (owing to larger CAPE), which remove stability more efficiently thereby reducing occasional extreme convective precipitation events. Also modified the trigger for the Geleyn type mixing contribution: it is activated less often and not in stable regions (thus, less erosion of low straus clouds in some regions).

  • June 24, 2014: Corrected an inconsistency in the new convection update: potential CIN now computed (as opposed to diluted CIN, which was much larger and was supressing convection too much via the trigger). Main result is removal of small extreme precipitation maximums. Also modified slightly CIN based trigger (no minimum CIN specified). Finally, instead of using the post-processed CAPE as an output diagnostic, the potential CAPE used within the convection scheme is output and plotted as the diagnostic (which is more relevant to the simulated convection).

  • July 12, 2014: Convection scheme modification. Improved the heating profile by updating the detrainment scheme. A RH-dependent background value (from IFS) is used, and a linear detrainment scheme above the level of maximum vertical velocity is used (as opposed to the vertical gradient of the vertical velocity). Peak heating rates are now located around 500 mb, more in line with observational studies. Some low level peaks also removed. A minimum CIN for the trigger has been added back (a low, forecaster-based value).

  • Aug. 9, 2014: Convection scheme modification. Slight adjustment of trigger to improve some small strong precipitation amounts (slightly less sudden). Also, a more sharp transition between shallow, deep and large scale initiated convective types added. Also, large scale forcing baseflux based upon low level (below cloud base) moisture convergence rather than omega (this avoids an coupling/excessive positive feedback sometimes observed between large scale omega and the convection). This also seems to give precipitation patterns more in line with operational NWP models (NAM, GFS). Finally, owing to these changes, peak convective rain rates tend to occur between 16 and 17 LST (verses 13-15 LST in old version).

  • Sep. 4, 2014: Convection scheme modification. The scheme has been found to produce slighty exessive convective rainfall (spatially and intensity). The CIN-supression factor for deep convection has been slightly modified to improve this. This also results in better synoptic scale patterns (since too much convective heating was previously present). Also, shallow convection is triggered only based on surface parcel lift (no longer using the CIN factor as in the deep convection). The main effect is to warm up low levels of the atmosphere through more widespread shallow convection. Finally, peak rainfall is now generally around 16 to 17h local, which is more consistent with observational studies.

  • Sep. 18, 2014: Adjusted deep convective trigger slightly so that convection is not so widespread and frequent over warm tropical oceans. Convection tended to be triggered too easily over such regions, it has been reduced by modifying a bit the CIN-based trigger function.

  • Oct. 14, 2014: Changed numerical representation of convection equations from explicit mass flux to an analytical form. Code is more simple, but it was also found that lower atmospheric heating seems to be improved (at least, compared to operational NWP models and analysis) and of note, tropical cycles are much better simulated. Finally, vertical heat profiles have a more classic shape (peak heating near 500 mb for deep convection, etc.).

  • Jan. 1, 2015. Global simulations using a Plate Carree projection (along with Fourier-filtering) has been developed and is being tested.

  • Apr. 15, 2015. Convection found to give occasional very large precipitation amounts. A slight adjustment to a CIN parameter permits convection to ignite slightly more easily thus releasing energy a bit more gradually...leading to larger areas with moderate precipitation but less local areas with very large amounts.

  • Jun. 6, 2015. Convection found to still give occasional very large precipitation amounts. A slight adjustment was made to increase the assymetry of drying and heating profiles (peak in drying lower and peak of heating higher in the atmosphere). The result is generally lower peak rain rate bulls-eyes, reduced positive convective feedback, and lowered vertical velocity values in regions with strong convection. Convection also becomes slightly more widespread, but this effect is rather small.

  • Sep. 13, 2015. Modified entrainment for convection to be limited by bouyancy effects. This causes convection to be less concentrated in bands and rainfall to be lower in those same regions (improvements). Also, the lower atmosphere is dried more and CAPE is lowered, especially over warm ocean areas (in better agreement with analysis from operational models). Finally, some occasional convective feedback with the large scale (over-estimated deepening of synoptic scale systems) is reduced (also an improvement). Overall heating profile forms and time of peak convective rain rates are relatively unchanged.