Global climate model tracking using geospatial neighborhoods

Climate models are complex systems of interacting mathematical models designed by meteorologists, geophysicists, and climate scientists, and run as computer simulations, to predict climate. These General Circulation Models (GCMs) sim-ulate processes in the atmosphere and ocean such as cloud formation, rainfall, wind, ocean currents, radiative transfer through the atmosphere etc., and their simulations are used to make climate forecasts. Due to the high variance among the predictions (projections in climate science terminology) of the approximately 20 global climate models that inform the Intergovernmental Panel on Climate Change, climate scientists are currently interested in methods to combine the predictions of this multi-model ensemble, of GCMs drawn from major laboratories around the world. Previous work provided techniques to combine the predictions of the multi-model ensemble, at various geographic scales, when considering each geospatial region as an independent problem [4, 3]. However since climate patterns can vary significantly and concurrently across the globe, this assumption is unrealistic, and could therefore limit the performance of these previous approaches. We consider a richer modeling framework in which the GCM predictions are made at higher geospatial resolutions, and our algorithm models neighborhood influences among geospatial regions. Overview The Tracking Climate Models (TCM) algorithm was introduced in [3] to dynamically combine the temperature predictions of the multi-model ensemble using Learn-α, an algorithm for online learning with experts under non-stationary observa