Vegetation Fire Fuels Mapping in the San Diego City Canyons – a Method Comparison

Fire risk is a major threat to life, property and natural resources in southern California. Recent fire disasters occurred in autumn 2003 and 2007. Fire risk management deals with these hazards, input data are collected, analyzed and evaluated. One of the most important input data is the vegetation density in the endangered areas. Here we describe methods to map vegetation density forming five hazard classes. The main objective of this study is to explore the benefits of using remote sensed data for the accurate classification of vegetation in San Diego city canyons. Three very high resolution remote sensing data sets (< 1 m) were used in comparison: scanned color infrared film (CIR) airborne, digital multi-spectral airborne (ADS40) and digital multi-spectral satellite imagery (QuickBird). Different classification approaches (e.g. pixel-based, segment-based and knowledge-based) were tested and analyzed to separate the vegetation into five hazard classes. Accuracy assessment indicated low overall accuracies of 58 % on average. With regard to an optimized classification result in particular unsupervised and segment-based classification can be recommended. The overall accuracy for these two methods reached around 62 %. The use of specially selected reference areas for validation helped to increase the accuracies up to 81 %. Also a separating between three instead of five different hazard classes resulted in accuracies around 80 %. Furthermore it could be shown that all three data sets can be used for successful classification procedures. The resulting fire risk maps can help to reduce or prevent fire hazards. The maps are a basis for the brush management of the Fire Department to manage sites of high risk next to residential areas (e.g. establish a 30 m fire break around properties concerned). * Corresponding author. 1. MOTIVATION AND AIM OF THE STUDY Disastrous wildfires are a major threat to life, property and natural resources in southern California (USA) and represent a risk not to be underestimated. Especially in the region of the San Diego County fire disasters occur periodically – most recently in autumn 2003 and 2007. Additionally, due to the climate change the area is subject to increased warming and aridity. These facts are indicated in the maps of the fourth IPCC report (IPCC, 2007). In consequence the fire risk will rise in the future. The City of San Diego Fire-Rescue Department (SDFRD) has developed a concept to minimize the risk of fire damage, the socalled brush management. Thus, residents are required to clear or reduce vegetation within a 30 m safety zone between endangered properties and the vegetated areas to reduce the risk. To assess high risk areas detailed information of the vegetation fire fuels that are close to residential areas, the SDFRD seeks to map vegetation density representing different fire hazard classes. Furthermore, a continuous monitoring for the fire risk management is of interest. Until now the vegetation density mapping is based on coarse air-photo interpretation (1:24,000 scale) and outdated (1995) vegetation maps. This study aims to support the efforts of the SDFRD to use high spatial resolution, remotely sensed imagery to generate detailed maps of the vegetation fire fuels within the City canyons and other urban-wildland interfaces. Remote sensing data are available today differing in the platform used (air-borne/space-borne) and the data format (analogue/digital). A pilot study was completed in 2005 that evaluated multi-spectral/color-infrared airborne and satellite imagery for classifying vegetation fire fuels (Coulter et al., 2005). The suitability of such different multi-spectral remote sensing imagery for mapping vegetation density in the San Diego City Canyons is further analyzed here. Furthermore, a variety of classification approaches is available today for information extraction. Within this study, the most suitable method for mapping fire hazard classes should is sought. Additionally, the use of vegetation indices to discriminate some fire hazard classes is one objective of this research. Especially the differentiation of trees from shrub vegetation is necessary to increase risk map quality.