Fire-climate relationships and long-lead seasonal wildfire prediction for Hawaii

Abstract

We examined statistical relationships between the seasonal Southern Oscillation Index (SOI) and total acreages burned (TAB) and the number of fires in the Hawaiian Islands. A composite of TAB during four El Niño/Southern Oscillation (ENSO) events reveals that a large total of acres burned is likely to occur from spring to summer in the year following an ENSO event. The correlation is most significant between the TAB in summer and the SOI of the antecedent winter. This relationship provides a potential for long-lead (i.e. 2 seasons in advance) prediction of wildfire activity in the Hawaiian Islands.

Logistic regression is applied to predict events of large acreages burned by wildfires. The goodness of predictions is measured by specificity, sensitivity, and correctness using a cross-validation method. A comparison of prediction skill for four major islands in Hawaii is made using the summer TAB as the response variable and the preceding winter SOI as the predictor variable. For predicting the probability of events (sensitivity), results indicate rather successful skills for the islands of Oahu and Kauai, but less so for Maui and Hawaii. It is more difficult to predict non-events (specificity), with the exception of Oahu. As a result, only Oahu has a high overall correctness rate among the four islands tested.