The Weather Generator project is a collaborative research project together with SINTEF Energy, NTNU, Irstea, met.no, Statkraft, Vattenfall, E-co and GLB.
The prototype will be implemented within either ENKI or R, and made available under the GNU GPL Open Source license. R is available under GNU GPL; ENKI is available under GNU LGPL.
met.no will participate in tasks 1,2,6 and 7 and contribute 2,3,5 and 6.
Task 1: Implement weather type classifications based on a review of earlier work This task will mainly be carried out at met.no, based on results from the EU Cost Action 733 working group. Within the first month of the project, a decision will be made with regards to which weather classification system that will be used in the pilot model. The analysis of classification systems and their explanatory power for different variables will continue through the pilot.
Task 2: Analyse the dependency of wind and precipitation on weather type Based on reanalysis data or measured time series, the statistical properties of wind and rainfall fields given season and weather type will be described. These properties include the spatial and temporal covariances within the time span of a weather situation.
Task 3: Formulate the space-time wind/precipitation model based on SPDE This task directly addresses the main objective and deliverable for this study. A pilot model will be built for precipitation and wind, confining the wind field to a spatially homogeneous advection vector. These limitations will be relaxed in a full scale project. The recently developed SPDE methodology (Stochastic Partial Differential Equations; Lindgren et al, 2008) will be used as simulation engine. An important challenge is to identify the assumptions which avoid excessively long computing times, while still capturing the statistical properties of the data.
Task 4: Implement sub-model for wind field-based orographic enhancement This task will be based on a linear orography model developed by Smith and Barstad (2004). Using an advection field and temperature/humidity information, the terrain-driven uplift and its associated precipitation enhancement is simulated. It is anticipated that this sub-model will explain some of the spatial variability given a frontal situation and a wind field.
Task 5: Estimate stochastic field parameters for different weather types using INLA Within this task, the parametric model will be selected, and its parameters for different weather types estimated using the Integrated Nested Laplace Approximation technique developed at NTNU (Rue et al).
Task 6: Analyse advection and temporal covariance on hourly time resolution This task will analyse the wind field temporal covariance over consecutive time steps; evaluate the spatially-stationary wind field assumption stated for this pilot’s model development, and point out improvements for a full version.
Task 7: Analyse the wind-temperature-humidity-precipitation covariance This task will evaluate cross-variable dependencies, including also variables which are omitted from the pilot model. In particular, the precipitation’s dependency of the other three will be analysed to find out if these allow an easier solution to the space-time model. This may be the case because the distribution of daily or hourly precipitation cannot be transformed into Gaussian, due to the high probability of the single value 0. Other results from this analysis will also provide suggestions for the full weather generator model, which is not confined to precipitation and wind.