This project focuses on predicting forest fire occurrences in Uttarakhand, India, by integrating meteorological, terrain, and land-use datasets with machine learning algorithms such as Random Forest and XGBoost.
Note:
The directories data/raw/ and data/processed/ are ignored by Git to prevent the inclusion of large and sensitive datasets.
The following datasets were utilized to train and validate the prediction models:
-
ERA5 Reanalysis Data – Provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). Citation: [1] H. Hersbach et al., “The ERA5 Global Reanalysis,” Quarterly Journal of the Royal Meteorological Society, vol. 146, no. 730, pp. 1999–2049, 2020. [Online]. Available: https://doi.org/10.1002/qj.3803
-
FIRMS (Fire Information for Resource Management System) – Active fire data from the NASA Earth Observing System Data and Information System (EOSDIS). Citation: [2] NASA FIRMS, “Fire Information for Resource Management System,” NASA EOSDIS, 2023. [Online]. Available: https://firms.modaps.eosdis.nasa.gov/
-
Land Use Land Cover (LULC) Data – Obtained from Bhuvan, the Indian geo-platform developed by ISRO’s National Remote Sensing Centre (NRSC). Citation: [3] NRSC–ISRO, “Bhuvan: Indian Geo-Platform,” National Remote Sensing Centre, Hyderabad, India. [Online]. Available: https://bhuvan.nrsc.gov.in/
-
Digital Elevation Model (DEM) Data – Acquired from Bhoonidhi, the data dissemination portal of the Indian Space Research Organisation (ISRO). Citation: [4] ISRO, “Bhoonidhi: Earth Observation Data Hub,” Indian Space Research Organisation. [Online]. Available: https://bhoonidhi.nrsc.gov.in/
The study area covers the Uttarakhand region of India, bounded by the coordinates:
-
Longitude: 77.5°E – 81.2°E
-
Latitude: 28.8°N – 31.3°N
-
Data Preprocessing:
- Extracted relevant meteorological and topographic variables.
- Aligned datasets to a uniform spatial grid.
- Performed normalization and missing value imputation.
-
Feature Engineering:
- Derived indices such as temperature anomaly, relative humidity, slope, and vegetation cover.
- Mapped fire events with environmental variables to create labeled datasets.
-
Modeling and Evaluation:
- Implemented Random Forest and XGBoost classifiers.
- Evaluated models using metrics like Precision, Recall, and F1-score.
- Produced probability-based fire risk maps.
- Incorporation of deep learning architectures such as CNNs and LSTMs for spatio-temporal prediction.
- Simulation of fire spread based on weather and terrain dynamics.