IRI-Danger: Probability of Human Ignition (PHI) [ET] [Version 1]

The product estimates the probability that a human caused fire will be ignited. The product was developed using Random Forest algorithm based on socio-economic and climatic drivers of ignition from geospatial indicators and demographic databases. The model obtained a performance of 80% estimated by the area under the curve (AUC). The model uses all ignitions that generated burned areas > 100 ha for a total of 33,388 ignitions. In addition, different data balancing methods were tested for the training, the stratified method being the most suitable for this model. This method consists of using a descriptive layer of European ecological regions to obtain a more representative sample of absence points. The results of the climate model showed that the probability of ignition in northern Europe is masked, the human models encountered challenges with classification accuracy, suggesting that the RF algorithm struggled to make accurate predictions based solely on human variables. These findings imply that predicting large fires initiations across broad geographic extents may not be feasible with human variables alone, highlighting the necessity of including climatic variables. This limitation depends on the scale of the analysis, as it dictates the magnitude of potential environmental gradients that drive initiation patterns. Whereas the mixed model proved to be the best candidate to represent the probability at the European scale. This might be because climatic variables help the model to differentiate between northern and southern Europe and thus find patterns in the data. It is important to note that by using the mixed models, areas of high probability are seen in northern Europe, which would otherwise be masked. Because of the results, we suggest the use of mixed models for global studies. For further information about the methodology and variables used see (Ochoa et al 2024) Ochoa, C., Bar-Massada, A., & Chuvieco, E. (2024). A European-scale analysis reveals the complex roles of anthropogenic and climatic factors in driving the initiation of large wildfires. Science of the total environment, 170443. https://doi.org/10.1016/j.scitotenv.2024.170443