A Machine Learning Methodology for Next Day Wildfire Prediction

Girtsou, Stella; Apostolakis, Alexis; Giannopoulos, Giorgos; Kontoes, Charalampos

doi:10.1109/igarss47720.2021.9554301

Cited by 9 publications

(21 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The exact fire ignition date is then derived from the active fires product, since this product records the fire event at the time of the satellite pass, whereas the burned areas are recorded in later passes. Finally, a fire inventory for the years 2010-2020 was constructed in order to be used for training, validation, and testing [13,14].…”

Section: Study Area and Evaluation Datasetmentioning

confidence: 99%

“…Dewpoint. A feature with dewpoint temperature was added in the current work, compared to [14], which measures the humidity of the air. According to ERA5 Land definition it is the temperature at which, the air at 2 meters above the surface of the Earth would have to be cooled for saturation to occur (https://cds.climate.copernicus.…”

Section: Training Featuresmentioning

confidence: 99%

“…Table 1 summarizes the aforementioned features, denoting with bold the ones that were added in the current work, as compared to our previous one [14].…”

Section: Lstmentioning

confidence: 99%

“…However, in practice, the extremely imbalanced setting of our problem requires tuning the considered evaluation measure, in favor of the extremely rare, fire class, when used during model selection. To this end, apart from the widely used measures of ROC-AUC and f-score, and based on our empirical experimentation [14], we define a set of task-specific evaluation measures that combine the two most important effectiveness indicators for the task, i.e., sensitivity and specificity, and put additional weight on sensitivity, i.e., the percentage of fires that are identified by the model:…”

Section: The Proposed Schemesmentioning

confidence: 99%

“…The presented work builds on and significantly extends our previous two works on the field [13,14], by: (i) Implementing an extended set of training features (described in Section 2.3) for capturing the most important fire driving factors within the adopted ML models. (ii) Implementing two alternative cross-validation schemes, as well as task-specific evaluation measures (described in Section 3.2.2) for model selection during validation, in order to handle the large data scale and imbalance; the second cross-validation scheme, as well as the second task-specific evaluation measure comprise new contributions compared to [14]. (iii) Presenting an extended evaluation and discussion of the proposed methodology.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Estimating Next Day’s Forest Fire Risk via a Complete Machine Learning Methodology

et al. 2022

Self Cite

View full text Add to dashboard Cite

Next day wildfire prediction is an open research problem with significant environmental, social, and economic impact since it can produce methods and tools directly exploitable by fire services, assisting, thus, in the prevention of fire occurrences or the mitigation of their effects. It consists in accurately predicting which areas of a territory are at higher risk of fire occurrence each next day, exploiting solely information obtained up until the previous day. The task’s requirements in spatial granularity and scale of predictions, as well as the extreme imbalance of the data distribution render it a rather demanding and difficult to accurately solve the problem. This is reflected in the current literature, where most existing works handle a simplified or limited version of the problem. Taking into account the above problem specificities, in this paper, we present a machine learning methodology that effectively (sensitivity > 90%, specificity > 65%) and efficiently performs next day fire prediction, in rather high spatial granularity and in the scale of a country. The key points of the proposed approach are summarized in: (a) the utilization of an extended set of fire driving factors (features), including topography-related, meteorology-related and Earth Observation (EO)-related features, as well as historical information of areas’ proneness to fire occurrence; (b) the deployment of a set of state-of-the-art classification algorithms that are properly tuned/optimized on the setting; (c) two alternative cross-validation schemes along with custom validation measures that allow the optimal and sound training of classification models, as well as the selection of different models, in relation to the desired trade-off between sensitivity (ratio of correctly identified fire areas) and specificity (ratio of correctly identified non-fire areas). In parallel, we discuss pitfalls, intuitions, best practices, and directions for further investigation derived from our analysis and experimental evaluation.

show abstract

Section: Study Area and Evaluation Datasetmentioning

confidence: 99%

Section: Training Featuresmentioning

confidence: 99%

“…Table 1 summarizes the aforementioned features, denoting with bold the ones that were added in the current work, as compared to our previous one [14].…”

Section: Lstmentioning

confidence: 99%

Section: The Proposed Schemesmentioning

confidence: 99%