GC33E-0570:
Using NPMR to forecast fire season severity for the state of Oregon
Wednesday, 17 December 2014
Heather E Lintz, Oregon State University, College of Earth, Ocean, and Atmospheric Sciences, Corvallis, OR, United States, Karen Short, US Forest Service Missoula, Missoula, MT, United States, John Francis Saltenberger, Northwest Interagency Coordination Center, Portland, OR, United States, Andrew Yost, Oregon Department of Forestry, Salem, United States, Philip Mote, Oregon State University, Corvallis, OR, United States and Andrew W Wood, National Center for Atmospheric Research, Boulder, CO, United States
Abstract:
We demonstrate that objective prediction of fire season severity in Oregon is possible and promising. The causes of fire season severity are various but seasonal climate plays a dominant role in the Pacific Northwest. Seasonal climate prediction has been gaining momentum in recent decades, and skillful forecasting occurs when perturbed boundary conditions alter weather regimes regionally. Regimes most conducive to fire season severity in Oregon are anomalously warm, dry seasons with pronounced dry lightning activity. Here, we used Non-Parametric Multiplicative Regression, a forecasting algorithm well suited to automatically accommodating complex interactions, to identify meaningful interactions and predict fire season severity as a function of sea surface temperature anomalies and atmospheric modes. We find the severity of a fire season can be forecast several months ahead of time in March with a cross-validated R-squared of 0.53. A cross-validated R-squared is more conservative than an R-squared as it is a measure of model fit to previously unseen data. The model we developed predicts fire season severity as the result of a complex interaction between specific modes the Pacific Decadal Oscillation, Pacific North American Pattern, and the Madden Julian Oscillation. Modes were derived using Variational Mode Decomposition (VMD). VMD is a new algorithm in signal processing that improves upon the limitations of Empirical Mode Decomposition (EMD) to decompose a signal into different modes of unknown but separate spectral bands. Prediction of fire season severity from atmosheric and boundary conditions can help budgetary planning and reduce the number of stand replacing fires and related economic losses.
