Combining sub-seasonal time scale precipitation forecast and vegetation health to predict fire anomalies in the Amazon biome

Fires in the Amazon result from human activities in combination with natural processes, among them the prevailing atmospheric conditions at various time scales and the state of vegetation health. Current forecast systems provide reliable deterministic forecasts at the scale of weather (1 to 7 days) and probabilistic outcomes at the scale of seasons (1 to 9 months ahead), leaving a forecast gap of 2 to 4 weeks. Only in recent years the potential for successful numerical prediction at this timescale, known as sub-seasonal, has been established by research-to-operation efforts, such as the Subseasonal Experiment (SubX) and the SubSeasonal to Seasonal (S2S) Project. Using a combination of SubX models to create a multi-model ensemble mean (MME) precipitation reforecasts for days 8 to 15 (equivalent to week 2 forecast) over the period 2000-2016, we find that it can be used to predict spikes in fire occurrence during the Amazon's dry season. The dry season is defined for each grid cell in tropical South America and consists of the trimester starting with the climatological driest month. MME precipitation is used as predictor in logistic regression models and result in skillful (measured as area under the ROC curve) probabilistic prediction of positive fire anomalies. Fire anomalies are obtained from the MODIS 8-day active fire product (MOD14A2) and period 2000-2019. Skillfully predicting the probability of above average fire occurrence several days in advance is also obtained from satellite estimates of vegetation health index (VHI) either as a single predictor or in combination with SubX precipitation reforecasts. This illustrates the potential for using recent advances in dynamical model forecasts at the sub-seasonal time scale in combination with remote sensing of vegetation to improve the skill and increase the lead time of fire probabilistic forecast in the Amazon. The operationalization of the methods presented in this study could allow for better preparedness and fire risk reduction in the Amazon with lead time of 1 to 2 weeks.