Embarrassingly Distributed Computing for Symbiotic Weather Forecasts

The spatial resolution of publicly available numerical weather forecasts is largely limited by the computing resources of the originating weather services. Only selected parameters are available to the public from many weather services. We present a scalable system for distributed computation of high resolution symbiotic numerical weather forecasts. Each forecast is computed on the user’s desktop computer, independent of other forecasts. Forecasts from the neighborhood may be exchanged between forecast producing peers, and amalgamated into a Symbiotic Forecast. The combination of several forecasts yields additional insight into the uncertainty introduced by the interaction of the chosen grid placement of the numerical model with geographical features with steep gradients, such as steep terrain. We let forecast producers share forecasts with each other. We assume that a user frequently is interested in the forecast for the area where the user lives and that the produced forecasts are stored at a computer located where the user is. Based on this, we limit the sharing of forecasts to computers within the same geographical area. To find computers with relevant forecasts, we will find most of them at computers located within a limited geographical region. These computers can be located quickly by scanning for them every time forecasts are needed. This approach scales with respect to the number of forecasters by limiting interaction to geographical regions. It also removes the need to maintain a list of computers with relevant forecasts.