Urban Surface Energy Balance Models: Model Characteristics and Methodology for a Comparison Study - Meteorological and Air Quality Models for Urban Areas - page 104

calculation of the turbulent sensible heat flux, the methods for calculating the latent
heat flux typically involve some form of resistance scheme whereas some prescribe
a value based on areal extent of vegetation (e.g. GCTTC).
11.3 Methodology for the Model Comparison
The methodology to be adopted follows that used in PILPS, the Project for Inter-
comparison of Land-Surface Parameterization Schemes (Henderson-Sellers et al.,
1993). This involves four stages. The models are all run offline so the forcing data
are provided for the 'top' of the model and so there is no feedback to larger scale
conditions within the modelling domain. The smallest time unit of analysis will be
hourly and the spatial unit will be an area that is representative of the local scale
(equivalent to one grid point in a meso-scale model). Initially participants will be
provided with very limited information about the chosen site(s), only being given the
forcing variables. In subsequent runs, more information will be provided to ensure
that a controlled experiment is achieved. By undertaking this staged approach it
should be possible to establish the required accuracy for each of the model parame-
ters by comparing the quality of the simulation at each stage. This methodology is
endorsed by the GEWEX Global Land Atmosphere System Study (GLASS) panel,
which coordinates the PILPS experiments (A. Pitman, personal communication,
2006). More specifically the steps are:
1) Forcing data only : The models are run with no prior knowledge of the urban sur-
face (i.e. each group determines their own default values for all of their parame-
ters); only the main forcing data will be supplied, e.g. winds, temperature, solar
radiation.
2) Add urban morphology : This involves releasing morphological information to
the modellers, e.g. building density, mean building height, vegetation fraction.
These data are more readily available on a global basis.
3) Add urban fabric properties : Details of building materials are then given, such
as thermal properties, surface cover fraction and albedo. This information is spe-
cific to each city and is not generally known on a global basis. Reliance on these
types of data makes a scheme such as this difficult for global applications.
4) Add evaluation data : At the final stage, the evaluation dataset is released to allow
optimisation of model parameters for best fit to observations. At this stage, mod-
elling groups will also return information on their optimised parameters as well
as the standard outputs. The methods used by individual groups to determine
what they regard as their optimized parameter set will be also gathered.
The model evaluations will involve statistical analysis of the performance of the
models relative to the observations . The observations will be for one site and will
consist of a range of data that varies seasonally. This assessment will be conducted
flux-by-flux (radiative, turbulent sensible and latent heat fluxes) and will consist