Guidelines for effective evaluation and comparison of wildland fire occurrence prediction models

Abstract

Daily, fine-scale spatially explicit wildland fire occurrence prediction (FOP) models can inform fire management decisions. Many different data-driven modelling methods have been used for FOP. Several studies use multiple modelling methods to develop a set of candidate models for the same region, which are then compared against one another to choose a final model. We demonstrate that the methodologies often used for evaluating and comparing FOP models may lead to selecting a model that is ineffective for operational use. With an emphasis on spatially and temporally explicit FOP modelling for daily fire management operations, we outline and discuss several guidelines for evaluating and comparing data-driven FOP models, including choosing a testing dataset, choosing metrics for model evaluation, using temporal and spatial visualisations to assess model performance, recognising the variability in performance metrics, and collaborating with end users to ensure models meet their operational needs. A case study for human-caused FOP in a provincial fire control zone in the Lac La Biche region of Alberta, Canada, using data from 1996 to 2016 demonstrates the importance of following the suggested guidelines. Our findings indicate that many machine learning FOP models in the historical literature are not well suited for fire management operations.

Detection of burned areas with extreme gradient boosting algorithm in a multi-temporal setting (MBAGB)

“…The classification accuracy may be measured with different metrics of the root mean squared method such as binary classification error rate, negative log-likelihood function, mean absolute error, mean absolute percentage error or area under the ROC curve (AUC). All of them provide similar results, but it is acknowledged that the AUC criterion is not recommended for very imbalanced data sets [57]. XGBoost model exhibits higher accuracy than single decision trees, but the interpretability is also more difficult.…”

Comparing calibrated statistical and machine learning methods for wildland fire occurrence prediction: a case study of human-caused fires in Lac La Biche, Alberta, Canada

“…Recent work has shown that the machine learning models used for FOP in past studies systematically overpredict the probability of wildland fires, and thus are unsuitable for operational use (Phelps and Woolford 2021). We presented methods for fitting properly calibrated FOP models using both statistical and machine learning approaches, and then compared a set of wellcalibrated models for human-caused FOP in Lac La Biche, Alberta.…”

“…As suggested by Phelps and Woolford (2021), we used area under the precision-recall curve (AUC-PR), negative logarithmic score (NLS) and temporal and spatial visualisations to evaluate and compare the models. AUC-PR was computed using the integration approach (Boyd et al 2013;Keilwagen et al 2014) from the PRROC package (Grau et al 2015), while the temporal plots with corresponding root-mean-squared error (RMSE) values were computed using aggregated daily totals across the entire study region.…”

“…4) suggest that such a large difference may be misleading. For such heavily imbalanced data, a model's AUC-PR is very dependent on the outcomes of the observations deemed most likely to be a fire occurrence; see Phelps and Woolford (2021) for more details. The two largest predictions of the logistic GAM were both fire occurrences, contributing to the spike seen in its precision-recall curve.…”

“…Recently, Phelps and Woolford (2021) demonstrated that the machine learning methods developed thus far for FOP are not well suited for operational use. This was done through a case study that included the comparison of uncalibrated models using the same data and study region we consider herein.…”