The Linked Environments for Atmospheric Discovery (LEAD) project is a large-scale, interdisciplinary NSF-funded research project. The objective of LEAD is to develop a cyber-infrastructure for identifying, accessing, decoding, assimilating, analyzing, mining, and visualizing a broad array of meteorological data and model output necessary for dynamic and adaptive forecasts in response to weather events. One key element in event-driven forecasts is the identification of weather events from observations and model outputs, representing regions of interest. We propose that a difference field --defined as the absolute difference of a model output field from two consecutive runs valid at the same time-- contains weather phenomena information that can be used to determine areas of interest for an on-demand forecasting. The difference fields indicate where weather is changing rapidly or is most sensitive to variations in model initial conditions. The Phenomena Extraction Algorithm (PEA) is applied to North American Mesoscale (NAM) model difference fields to detect regions of interest in these fields. Then, a composite regions-of-interest field is constructed by weighting the features identified in individual fields. The PEA is then re-applied to the composite field to extract the final regions of interest for consideration. Using 850 hPa wind patterns, preliminary results show that cyclones and fronts are a couple of identified regions of interest. We plan to investigate the sensitivity of this approach to the phenomena that are extracted, their weighting, and further explore how the resultant regions relate to ongoing weather systems.