Limitations and utilisation of Monitoring Trends in Burn Severity products for assessing wildfire severity in the USA Crystal A. Kolden A,B , Alistair M. S. Smith A and John T. Abatzoglou A A University of Idaho, 875 Perimeter Drive, MS3021, Moscow, ID 83844-3021, USA. B Corresponding author. Email: ckolden@uidaho.edu Abstract. The Monitoring Trends in Burn Severity project is a comprehensive fire atlas for the United States that includes perimeters and severity data for all fires greater than a particular size (,400 ha in the western US, and ,200 ha in the eastern US). Although the database was derived for management purposes, the scientific community has expressed interest in its research capacity. As with any derived data, it is critical to understand inherent limitations to maximise the utility of the dataset without compromising the inferences. The classified severity product in particular is of limited use to research due to a lack of both consistency in developing class thresholds and empirical relationships with ecological metrics. Here we review the products available and their development process, and characterise and quantify the limitations of the classified burn severity data product based on the use of highly variable and subjective classification thresholds. We suggest a framework for overcoming these limitations by developing a more robust classified product that will support ecological management and applications. This framework utilises field data to develop consistent, ecologically based thresholds that incorporate existing ecoregion classifications from LANDFIRE or other fire management frameworks already widely integrated into planning efforts. Additional keywords: dNBR, Landsat, MTBS, RdNBR. Received 9 April 2015, accepted 28 July 2015, published online 14 September 2015 Introduction Wildfire burn severity maps reveal immediate fire effects and long-term ecosystem changes applicable to many management needs, including fire planning and mitigating post-fire watershed effects (Robichaud et al. 2007). Focusing on fire planning, the Monitoring Trends in Burn Severity project (MTBS, www.mtbs. gov) evolved from the joint National Park Service (NPS) and United States Geological Survey (USGS) Burn Severity Mapping project developed to map and monitor wildfire effects across national parks from remotely sensed data (Eidenshink et al. 2007). Two primary products of the initial effort were the Composite Burn Index (CBI), a field methodology for quantify- ing total fire effects at the spatial resolution of Landsat reflectance data, and the Normalised Burn Ratio (NBR), a spectral index that differentiates between healthy green, dry senesced and charred vegetation (Key and Benson 2006). Subsequent research pro- duced a differenced NBR (dNBR) from pre- and post-fire scenes more strongly correlated with 1-year post-fire field measure- ments of burn severity than the single, post-fire NBR scene for some ecosystems; although considerable uncertainty remains over the nature of the biophysical responses detected by dNBR (Lentile et al. 2009). A relative version of dNBR (RdNBR) accounting for pre-fire fuel heterogeneity was more accurate in delineating higher degrees of fire effects in ecosystems with lower fuel loads (Miller and Thode 2007; Cansler and McKenzie 2012). MTBS retrospectively mapped NBR, dNBR and RdNBR for wildfires in the United States dating back to 1984 and classified dNBR into thematic severity maps. Additionally, some prescribed and agricultural fires were mapped because the burn scars were visible on the Landsat scenes. There is a ,2-year lag for inclusion of new fires into the database. The research community has increased its focus on larger- scale assessment of burn severity in recent years, including analysing trends and patterns (e.g. Miller et al. 2009; Lannom et al. 2014; Cansler and McKenzie 2014) and relating burn severity to biophysical phenomena and management practices (e.g. Smith et al. 2007; Wimberly et al. 2009; Kolden and Abatzoglou 2012; Hicke et al. 2013; Morrison and Kolden 2015). Robustness of such studies is dependent upon MTBS product accuracy; because MTBS was developed specifically for management needs and has not undergone systematic, field- based evaluation to quantify individual fire accuracy for scientific purposes (e.g. including error bars in analyses), most of these studies used MTBS protocols to develop their own equivalent data products with field-based accuracy assessments. However, given the widespread availability of the MTBS dataset, reprodu- cing its methodology on a case-by-case basis seems counter- productive to the development of a broadly transferable tool. Here, we review sources of error in the MTBS products that affect research accuracy and robustness of results, including:  Areas of no detectable change are included in MTBS fire perimeters.  The phenology offset is not applied to continuous spectral indices (i.e. dNBR and RdNBR). CSIRO PUBLISHING International Journal of Wildland Fire http://dx.doi.org/10.1071/WF15082 Journal compilation Ó IAWF 2015 www.publish.csiro.au/journals/ijwf Communication