Open-Source GIS for Post-Fire Analysis

Scale burn severity: drone maps to satellite. Open-source GIS, Python for post-fire analysis.

Scaling Burn Severity Metrics: From Drones to Satellites for Post-Fire Watershed Management  

Wildfires are a growing concern, particularly in regions with vulnerable wildland-urban interfaces. These events often expose steep watersheds to significant post-fire hazards, including increased soil loss and debris flows. Accurate mapping of burn severity immediately after a fire is essential for effective post-fire management and response. This article explores how combining drone and satellite imagery using open-source GIS software and Python can standardize workflows and offer valuable information for understanding burn impacts.

Uncrewed Aerial Systems in Post-Fire Assessment  

Uncrewed Aerial Systems (UAS) offer rapid and detailed post-fire assessments by capturing very high-resolution imagery, at around 1cm per pixel. This high-level of detail helps inform emergency management and allows post-fire hazard evaluation. However, the workflows for data acquisition, processing and analysis may be complicated, often hindering the widespread adoption of UAS. Utilizing an open-source platform facilitates a transparent and repeatable method for generating these products. By combining established GIS software, such as QGIS with Python, powerful analytics can be made accessible to more practitioners in the geospatial field.

Key Software Components  

A fully integrated workflow needs various open-source software for optimal functionality.

• WebODM (OpenDroneMap): Used to generate structure-from-motion (SfM) products by processing UAS imagery.
• QGIS: Functions as the primary GIS software. QGIS performs advanced geospatial analyses and manages map layers for enhanced understanding.
• Cloud Compare: Processes the 3D point cloud data including classifications of ground points to then derive a height-above-ground model.
• Python: Serves as the foundation of the analysis by handling raster calculations and applying multi-Otsu threshold to optimize classifications.
• Jupyter Notebook: Enables documentation and code sharing and offers an opportunity for a heuristic user experience.

By working across multiple FOSS products it’s easier to develop robust and standardized workflows, fostering greater transparency and reproducibility in scientific analyses.

Building a FOSS Workflow for Burn Severity Mapping  

Let’s explore an end-to-end methodology for producing burn severity maps. The goal is to deliver detailed insights for areas at high-risk for soil erosion after wildfires by combining very high-resolution drone data with wide coverage of satellite imagery.

1. UAS Data Collection and Processing  

The workflow starts with careful image acquisition. It is very important to use proper collection methods with low-cost drones carrying RGB cameras at optimal altitudes. Once collected these images are input into WebODM for structure-from-motion (SfM) processing generating 3D point clouds. Noise in the point cloud and distortions in the orthomosaic, however, may appear in some situations. Following best-practices during acquisition such as optimizing flight paths, or accounting for scene illumination conditions may help minimize any potential issue during image acquisition. The process produces orthoimages and digital surface models at a fine-grained spatial resolution of approximately 1 cm, a ground sampling distance (GSD) essential for detail at such a fine-scale resolution.

Following 3D point cloud generation the next important step is to clean the data in Cloud Compare. Using algorithms to classify ground and above-ground point data, combined with manual outlier removal helps generate highly detailed height above ground (HAG) model. Careful inspection at every stage and proper parameter tuning specific to local environments, ensure reliable processing steps for later analysis.

2. In-Situ Data Sampling  

Field sampling is critical to ground-truth the remotely sensed data. This can be completed concurrently at the same time of drone flights for time effectiveness. A modified composite burn index (mCBI) offers valuable reference data, with multiple 1-m diameter plots to quantify different soil and vegetation conditions, such as light/dark soil and vegetation loss from the burn. These are then labeled using a handheld GPS unit for precise location to match against orthophotos. With careful note-taking it can support later validations for analysis done on remotely sensed products, enhancing accuracy. It helps ensure proper classification of the detailed image products collected through remote sensing methods, allowing users to correlate classifications with observed burn effects, specifically to quantify fractional covers for post-fire surface classes.

3. UAS Image Classification for Burn Severity Metrics  

Classification is crucial to transform drone data into meaningful layers and ultimately produces burn severity maps. In the first step to classify, Python, SciPy, Numpy, and Rasterio, play a critical role in extracting information from the orthoimages and HAG. Starting with a mask of shadows the next most critical stage is separating the green fraction from the RGB composite based on excess greenness, ExG index. Separating rocks is another important part of the workflow which uses multi-scale, object-based classification methods for greater flexibility, accounting for different sizes and properties for optimal classifications of various post-fire components.

Once the key parameters for vegetated features are masked and rocky outcrops are identified, attention is given to bare-earth. Utilizing two different reference image classifications using multi-Otsu thresholding. One classification utilized the weighted brightness (WB) model and the other uses the char index (CI). By inspecting classified results the ideal reference images and parameter tuning, provides more reliable and useful classifications by connecting local biophysical conditions to digital components. With proper attention to parameter settings you can enhance class accuracy and differentiate key soil classes, like char/dark soil from other lighter soils. This information ultimately forms a basis to upscale drone data for large landscapes to address post-fire hazards.

4. Integrating Satellite Data with Drone-Derived Metrics  

Integrating satellite imagery with UAS data offers an innovative way to scale up detailed insights. For satellite data use an open access service like the Google Earth Engine to gain wide access and reduce complexity in retrieval. By querying surface reflectance of satellite images, indices like the Normalized Burn Ratio (NBR), Normalized Difference Vegetation Index (NDVI), and Normalized Difference Water Index (NDWI), all capture different burn conditions. Sentinel-2 multispectral imagery is frequently utilized to model large watersheds because it measures areas sensitive to vegetation and soil reflectivity. Using best practice by matching the data with the same acquisition date as UAS data will help with proper georegistration by avoiding surface changes due to external factors. This approach to mapping from drones to satellites demonstrates how fine details may upscale to a landscape scale while preserving the accuracy and validity of high-resolution data products.

5. Evaluating and Modeling Burn Severity Relationships  

An accuracy assessment of classified results, allows an evaluation using two techniques, field observations against drone classified fractions and expert observer assessments of the data in terms of the UAS orthoimage. These results highlight any misclassifications associated with individual categories, which is needed to perform an optimization of processing parameter to fine-tune the workflow for optimal classification products. This procedure ensures consistency of findings across diverse sites by evaluating strengths and weaknesses for robust applications of image classification.

Ordinary least squares (OLS) linear regression establishes significant statistical relationships of UAV metrics derived through FOSS workflow and large area satellite derived products. By using a linear equation with proper variable input from statistical fitting this modeling method links key factors, char/dark soil cover and post-fire satellite products to scale up fractional cover of different surface components, such as a soil or char/dark cover across entire wildfire regions. The approach combines detailed UAS analysis and satellite modeling producing high quality landscape scale metrics for burn severity at minimal cost and with great efficiency.

Key Insights & Applications  

By leveraging the FOSS workflow approach and statistical analysis for accurate surface classification you’ll see that char index offers the most reliable method for separating dark/char, demonstrating the utility for measuring and evaluating burn conditions. Results highlight a strong relationship between drone classifications of dark/char cover and post-fire satellite NBR, proving that UAS derived detailed products may apply across wide extents with satellite platforms. Additionally including a unique rock fraction also improve regressions further enhancing how important understanding various components and features are.

Here are a few ways post-fire management and decision-making teams may benefit:

• UAS data provides very detailed (1cm resolution) maps with low-cost equipment, improving the data foundation.
• Free and open source workflow with widely available FOSS enables replicable science by standardizing workflow.
• Statistical analysis proves an important role in validating methods by directly linking UAS-derived severity classifications with broader, landscape scale products.

This FOSS method allows teams to improve hazard mitigation while also gaining crucial insight from accurate data, making efficient methods accessible for monitoring burn effects.