Machine Learning Techniques for Fine Dead Fuel Load Estimation Using Multi-Source Remote Sensing Data

Abstract

Fine dead fuel load is one of the most significant components of wildfires without which ignition would fail. Several studies have previously investigated 1-h fuel load using standard fuel parameters or site-specific fuel parameters estimated ad hoc for the landscape. On the one hand, these methods have a large margin of error, while on the other their production times and costs are high. In response to this gap, a set of models was developed combining multi-source remote sensing data, field data and machine learning techniques to quantitatively estimate fine dead fuel load and understand its determining factors. Therefore, the objectives of the study were to: (1) estimate 1-h fuel loads using remote sensing predictors and machine learning techniques; (2) evaluate the performance of each machine learning technique compared to traditional linear regression models; (3) assess the importance of each remote sensing predictor; and (4) map the 1-h fuel load in a pilot area of the Apulia region (southern Italy). In pursuit of the above, fine dead fuel load estimation was performed by the integration of field inventory data (251 plots), Synthetic Aperture Radar (SAR, Sentinel-1), optical (Sentinel-2), and Light Detection and Ranging (LIDAR) data applying three different algorithms: Multiple Linear regression (MLR), Random Forest (RF), and Support Vector Machine (SVM). Model performances were evaluated using Root Mean Squared Error (RMSE), Mean Squared Error (MSE), the coefficient of determination (R2) and Pearson’s correlation coefficient (r). The results showed that RF (RMSE: 0.09; MSE: 0.01; r: 0.71; R2: 0.50) had more predictive power compared to the other models, while SVM (RMSE: 0.10; MSE: 0.01; r: 0.63; R2: 0.39) and MLR (RMSE: 0.11; MSE: 0.01; r: 0.63; R2: 0.40) showed similar performances. LIDAR variables (Canopy Height Model and Canopy cover) were more important in fuel estimation than optical and radar variables. In fact, the results highlighted a positive relationship between 1-h fuel load and the presence of the tree component. Conversely, the geomorphological variables appeared to have lower predictive power. Overall, the 1-h fuel load map developed by the RF model can be a valuable tool to support decision making and can be used in regional wildfire risk management.

Is Experience the Best Teacher? Knowledge, Perceptions, and Awareness of Wildfire Risk

“…Such unbalanced stratification of the sample in favor of the most educated and youngest could leave aside important considerations on wildfire impact and management in at-risk communities, such as those with poor education, low income, and the elderly [44][45][46]. Future research aimed at evaluating the interplay between the characteristics of vulnerable groups and fire-related environmental variables, such as climate variability, forest fuel distribution, and topographic features is warranted [47,48]. Lastly, we acknowledge the need to use validated scales for assessing wildfire risk perception.…”

Prediction of Outcomes in Mini-Basketball Training Program for Preschool Children with Autism Using Machine Learning Models

“…Mean square error (MSE), root mean square error (RMSE) and coefficient of determination (R 2 ) [40][41][42] were used to evaluate the performance of the models. MSE was expressed as:…”

Integration of a deep-learning-based fire model into a global land surface model

“…To address the spatiotemporal context for wildfire danger, (Kondylatos et al, 2022) applied a convolutional-LSTM network (Shi et al, 2015) integrating meteorological, environmental, and anthropogenic drivers. Other studies leveraged ML/DL methods to characterize various aspects of fire occurrence, such as fire weather (Son et al, 2022), lightning ignition (Coughlan et al, 2021), fire susceptibility (Zhang et al, 2021) and fuel availability (D'Este et al, 2021).…”