ML Fire Prediction Using Derived Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models for predicting fire behavior face data sparsity issues due to the lack of training data with labeled fire behavior features, limiting their ability to produce effective predictions.
Innovation Solution
The approach involves obtaining and processing data layers representing geographic regions, determining derived fire-related metrics such as speed, size, and duration, and using these metrics to train machine learning models like gradient boosted decision trees or convolutional neural networks to predict fire behavior, thereby supplementing initial training data and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional simulation techniques are used to predict fire behavior, then predictions can be generated for regions with historical fire data, but the method cannot predict fire characteristics in regions that have never experienced fire events and cannot predict far in advance
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models using derived fire behavior metrics from historical fire data before actual fire events occur. The system calculates fire behavior characteristics (spread rate, intensity, duration) from historical burn perimeter data and uses these pre-computed metrics to train models that can then predict fire behavior in advance for both historical and future scenarios, enabling predictions up to a year ahead and in regions without prior fire events.
2Productivity
If machine learning models are trained with limited labeled fire behavior data, then training can proceed with available data, but the models produce inaccurate predictions due to data sparsity
Solution Approach 1:
The system applies self-service by automatically deriving fire behavior metrics from historical fire data without requiring manual labeling. The patent computes fire spread rate, intensity, and duration by analyzing burn perimeter data over time, generating training labels autonomously from the raw data. This self-derived labeling process creates abundant training data from existing historical records, enabling effective model training without external annotation resources.
3Quantity of substance
If fire behavior metrics are not derived from historical data, then training data remains sparse, but deriving metrics increases data availability and improves model training
Solution Approach 1:
The patent introduces an intermediary processing layer that transforms raw historical fire data into derived fire behavior metrics. This intermediary step calculates intermediate values such as fire spread rate (change in burn perimeter over time), fire intensity, and duration by processing sequential burn perimeter data. These derived metrics serve as mediating features that bridge raw data and model training requirements, automatically generating abundant training labels while managing processing complexity through systematic computation.
Data Source
AI summary
Methods, systems, and apparatus for obtaining a first plurality of data elements, each data element representing a fire-related metric of a geographic region, determining, using at least a subset of the first data elements, one or more values representing one or more derived fire-related metrics, associating the one or more values with the first data elements, obtaining a second plurality of data elements, each data element representing a fire-related metric of the geographic region, and training a machine learning (ML) model using at least a subset of the first plurality of data elements, at least a subset of the second plurality of data elements, and values associated with the subset of the first plurality of data elements to provide a trained ML model.