Several application areas as high energy physics or bio-applications have benefited for years from Grid technologies. Applications from the Earth Science community are starting to take advantage of this technology (see e.g. www.eu-degree.eu) . Earth science applications and, in particular, a climate and meteorological models poses a great challenge to the Grid in terms of the computing and storage requirements. These models are resolved by numerical equations that are CPU intensive applications which usually require long walltimes and produce large amounts of data.
In this work we present the port to Grid environments of the Weather Research and Forecasting (WRF) model (http://www.wrf-model.org, . This model is a state-of-the-art mesoscale numerical weather prediction system designed to serve both operational forecasting and atmospheric research needs. It has a rapidly growing community of users all around the world.
In a previous work, within the EELA project, we ported the Community Atmospheric Model atmospheric model (CAM, ) to the Grid, developing a workflow to solve the most common problems encountered in the Grid . The limited area models (another family of numerical atmospheric models) pose an additional challenge for the Grid, since not only the output data is large, but also the input data required before the simulation starts can be quite large. These models can be nested into global atmospheric models, thus, they are the natural step forward after the Gridification of the CAM model. The Gridification of a limited area model can give rise to a cascade of simulations running first a global model and, then, nesting one (or several) limited area simulations.
WRF for Grid (WRF4G) is a port of the WRF Modelling System to Grid environments. Small modifications to the source code of the model allow the monitoring and output data management in a flexible way. In addition to the model, the WRF Grid Enabling Layer (WRFGEL) is an interface between the model and the Grid, allowing the model to inform about its status, get the required input data and save the output data to a Storage Element (SE) in the Grid. Finally, a set of user scripts permits a flexible design of experiments consisting of realizations which can span different physics/parameters and/or a sequence of independent hindcasts.
Currently, the heterogeneous Grid infrastructure is subject to common failures and intermittent availability of resources the numerical weather models are not prepared for. For those reasons, in this contribution we present a new execution framework providing a software wrapper for a numerical prediction model. Since multi-site parallelism cannot be used due to latency, the Grid is best suited for large amounts of independent and relatively
short simulations (ensembles). WRF4G is able to benefit from intrasite parallelism where available, though.
As an illustrative example, the EELA-2 infrastructure can be used to perform an ensemble of seasonal prediction simulations by nesting the global predictions over key regions such as El Niño sensitive areas over Latin-America. Taking advantage of the Grid, all members of the global ensemble could be nested and each could be perturbed in the physics to obtain a larger ensemble and a better estimation of the uncertainty.