Quantifying the Importance of Vegetation Characteristics and Fuel Build Up for Fire Seasonality

Abstract:

Vegetation build up is of fundamental importance for global wildfires due to its impact on fuel availability. The relationship between wildfire activity and past vegetation productivity is still unclear, however, with current fire models poorly representing observed vegetation-fire relationships. Likely as a result of this mismatch, fire season length and inter-annual burned area variations are not reproduced well. Given the anticipated increase in conditions conducive to extreme fires in many regions of the world, this relationship needs to be investigated to better understand and manage such events. To achieve this, we use global satellite datasets from 2010 to 2015 and a random forest (RF) machine learning model to assess the relationships between global climatological burned area and both climatic and biophysical variables. Fuel accumulation processes are accounted for by the inclusion of lagged relationships with dry day period and productivity proxies by up to 2 years. Multiple RF models with varying sets of variables are constructed to investigate their relative importance. The inclusion of lagged effects while maintaining an equal number of variables results in an improvement of out-of-sample R2 from 0.53 to 0.64. The fraction of absorbed photosynthetically active radiation, related to vegetation abundance and productivity, and solar-induced fluorescence, a measure of photosynthetic activity, are found to be the most significant vegetation variables. The marked improvement in model performance resulting from the inclusion of lagged relationships underlines their importance and the need for their accurate representation to skilfully predict burned area. Lagged relationships are found to have a negligible importance for lags ≥1 year, suggesting that for most biomes, fuel accumulation can be discerned accurately using information from within the last year only. The relative importance of different antecedent time periods is found to vary significantly across biomes with a clear difference between fuel-limited and moisture-limited regions. These results indicate that the representation of fuel build up in fire models could be improved by the refinement of relationships with fuel moisture and vegetation productivity at different timescales.