Methods & Evidence
How Our Satellite Products Are Made
Robb Consulting Services Inc. combines satellite imagery, field-informed models, and GIS analysis to provide information for wildfire monitoring and planning. This page explains what each product shows, how it is produced, and the evidence supporting its development.
The application includes regional Snow Cover, Grass Curing, and Deciduous Leaf Out examples for the Calgary Forest Area, together with historical Burned Area Analysis products for Alberta wildfires. The delivery standards below describe the proposed provincial services; individual examples retain their own dates, coverage, and processing information.
Services at a Glance
| Service | What It Shows | Available on This Site |
|---|---|---|
| Snow Cover | Estimated visible snow percentage on a 375-metre grid | Regional maps and dated releases |
| Grass Curing | Estimated grass curing percentage on the 100-metre provincial fuel grid | Regional maps and dated releases |
| Deciduous Leaf Out | Modelled leaf-on fraction on the 100-metre provincial fuel grid | Regional maps and dated releases |
| Burned Area Analysis | Burned and unburned areas mapped at 30 metres | Historical fire analyses, imagery, and downloads |
Our Shared Method
Each service follows four steps, with the imagery and analysis tailored to the product.
- Define the area and date. Establish the requested fire, Forest Area, or provincial coverage and the period represented by the result.
- Select suitable observations. Assess available imagery and its quality information, retaining the acquisition dates and identifying gaps.
- Produce and review the result. Apply the relevant model or classification method and assess the output against the available imagery and reference information.
- Prepare a dated product. Package the map, GIS data, source records, and supporting metadata so users can understand and reuse the result.
Snow Cover
What It Shows
Snow Cover estimates the percentage of each mapped cell covered by snow visible to the satellite. It supports assessment of seasonal snow conditions and early-season fire-danger modelling. Water, cloud, unsuitable observations, and missing data are identified separately from valid snow percentages.
How It Works
The method uses NASA-distributed VIIRS observations and their quality information. For each location, it selects the most recent suitable observation within the preceding three days. A newer cloudy image does not replace an older clear observation that remains within this window. Satellite measurements that distinguish snow from other surfaces are converted into an estimated snow percentage.
Outputs and Evidence
The primary product is a 375-metre GeoTIFF, accompanied by observation dates, imagery age, source-satellite and quality information, and product metadata. Regional releases and a retrospective spring example demonstrate the mapping and source-tracking workflow. Further Alberta evaluation will compare estimates with suitable independent reference mapping. Because optical imagery sees snow visible to the sensor, assessing snow beneath forest canopy is a separate part of the proposed refinement work.
Processing Details
These notes describe the current regional implementation. Each release's metadata records the settings actually used.
- Contributing satellites are recorded for each release and can include NOAA-20, NOAA-21, and Suomi-NPP.
- Snow identification uses the normalized difference snow index (NDSI), which contrasts visible and shortwave-infrared measurements. A documented empirical relationship converts eligible index values into an estimated visible snow fraction from 0 to 100%.
- Source quality information is used to assess cloud, illumination, and radiometric suitability. Separate categories identify water and ice, cloud, rejected observations, low illumination, and locations without an eligible observation.
- Observations are mapped onto a fixed 375-metre grid in EPSG:3400 for consistent alignment between releases. The output cell size describes the mapping grid; the satellite's observation footprint varies across its viewing swath.
Grass Curing
What It Shows
Grass Curing estimates the percentage of grass that has cured within mapped grass-fuel areas. The product helps users examine vegetation condition across the landscape alongside field observations and other wildfire information.
How It Works
The method uses 30-metre Harmonized Landsat and Sentinel-2 imagery, known as HLS. A model trained against field-assessed curing relates satellite measurements of vegetation and moisture to curing percentage. The provincial fuel grid identifies O-1 grass-fuel areas, and additional land-cover information identifies the grassland contributing to each estimate. Suitable predictions are combined within aligned 100-metre output cells; cells without sufficient observations remain unestimated.
Outputs and Evidence
Downloads include curing percentages, image dates and ages, grass fraction and quality information, source-image records, metadata, and GIS display styles. Average Percentage Cured summarizes valid mapped curing estimates; it is separate from the amount of land successfully observed.
Model development used 566 Alberta field observations from 2024 and 2025. Testing excluded whole field sites from model fitting and reported an average absolute error of approximately 6.5 percentage points. Detailed evaluation also reported a root-mean-square error of approximately 9.1 percentage points. These findings describe the evaluated model and sampled conditions; further provincial assessment will address additional regions and seasonal conditions.
Processing Details
These notes describe the current regional implementation, which is distinct from the proposed provincial standard. Each release's metadata records the model and settings actually used.
- The service builds on grass-curing research and software developed by Robb Consulting Services Inc. for Alberta Wildfire, including field-method development and observations involving FPInnovations.
- Visible, near-infrared, and shortwave-infrared HLS measurements are used to calculate vegetation and moisture indicators. A Cubist regression model relates them to field-assessed curing for each eligible grassland observation.
- ESA WorldCover identifies contributing grassland pixels within O-1 cells, reducing the influence of trees, water, and other land cover.
- A cell is estimated when it has at least four suitable grassland pixels and suitable observations for at least half of its grassland pixels. A valid 0% curing value remains distinct from a missing value.
- Observations are selected separately across the landscape, so one map can contain imagery from several acquisition dates. Some regional examples allow observations up to 14 days old; the proposed provincial standard is seven days.
- Root-mean-square error gives greater weight to larger errors than the average absolute error.
Deciduous Leaf Out
What It Shows
Deciduous Leaf Out maps the modelled leaf-on fraction within deciduous and mixedwood fuel areas. It helps users follow seasonal changes in vegetation relevant to wildfire conditions.
How It Works
Suitable 30-metre HLS observations are assessed by several trained models to identify leaf-on and leaf-off conditions. These classifications are summarized within 100-metre provincial fuel-grid cells containing D-1/D-2 deciduous or M-1/M-2 mixedwood fuels. Other fuel types are excluded, and cells without sufficient observations remain unestimated.
The current percentage is the share of contributing classifications identified as leaf-on. It is not a direct measurement of how far each individual leaf or bud has developed. Evaluating partial seasonal development and the influence of conifers in mixedwood stands is part of the proposed refinement.
Outputs and Evidence
Downloads include leaf-on fraction, image-date and age information, source-image records, metadata, and GIS display styles. Field assessment identified leaf presence at all 77 evaluated visits across 21 Edmonton-region deciduous plots in 2025. Separate winter assessments identified leaves-absent conditions on 80–94% of evaluated snow-free observations. These results apply to the assessed conditions and do not establish universal accuracy for percentage leaf development.
Processing Details
These notes describe the current regional implementation, which is distinct from the proposed provincial standard. Each release's metadata records the model configuration and settings actually used.
- An ensemble of random-forest classifiers combines predictions from several trained models to classify each eligible observation as leaf-on or leaf-off.
- A cell is estimated when suitable observations cover at least half of its 30-metre source pixels.
- Average Leaf Out summarizes valid mapped cell values. A mixedwood cell can contain signals from both deciduous and coniferous vegetation, so stand composition is relevant to interpretation.
- Some regional examples allow observations up to 14 days old; the proposed provincial standard is seven days.
Burned Area Analysis
What It Shows
Burned Area Analysis maps burned and unburned areas around a requested wildfire to support wildfire records, forest operations, fuel inventories, and fire-danger modelling.
How It Works
The proposed service accepts a draft wildfire perimeter or a point location. Review includes the area up to 100 metres outside a supplied draft perimeter to identify additional burning; where only a point is supplied, the surrounding landscape is examined to identify the fire footprint.
The method compares suitable before- and after-fire satellite observations to identify changes consistent with burning. HLS is the primary imagery source, with direct Sentinel-2 or Landsat observations also proposed where their timing or quality better supports the request. Screening addresses cloud, smoke, snow, shadow, and other unsuitable conditions. Analyst review checks the relationship to the requested fire, classification errors, and boundary completeness before final contract delivery. Areas without sufficient suitable imagery remain identified as not observed.
Outputs and Evidence
The proposed delivery package includes a classified GeoTIFF at 30 metres or better, a wildfire-perimeter shapefile, and separate polygons for associated spot fires larger than one hectare. The outer perimeter generally ignores internal islands; the raster retains the detail of burned and unburned areas. Image-contribution maps and metadata document the sources, dates, processing, accuracy information, appropriate uses, and limitations. Source imagery can also be provided for viewing or download when requested.
Historical Alberta fire analyses are available in the application. Development comparisons with reference mapping assess missed burning, additional mapped areas, and boundary agreement. Performance varies with fire size, season, and observation conditions; small fires, spring fires, and neighbouring incidents require particular review. Detailed comparisons remain available under Evaluation Details.
Evaluation Details
Development results compare the mapped result with reference products. Agreement depends on the fires assessed and on differences in observation dates, mapping methods, and what the reference boundaries include. Spatial overlap scores range from 0 to 1, with 1 indicating complete agreement for the compared areas. These scores describe reference agreement, not a universal accuracy percentage or operational service acceptance.
| Year | Reference Set | Fires | Spatial Overlap (Dice) |
|---|---|---|---|
| 2024 | Alberta perimeters mapped from GPS, aerial photography, or infrared scans | 34 | 0.86 |
| 2024 | All assessed Alberta perimeters, including hand sketches | 43 | 0.85 |
| 2025 | Assessed Alberta perimeters; mapping source not recorded | 27 | 0.72 |
Additional Measures
| Reference Set | Commission | Omission | Per-Fire Median Dice | Perimeter IoU | NBAC vs Alberta Dice |
|---|---|---|---|---|---|
| 2024, 34 fires | 8.8% | 19.1% | 0.86 | 0.80 | 0.89 |
| 2024, 43 fires | 8.0% | 20.4% | 0.86 | 0.80 | 0.88 |
| 2025, 27 fires | 16.1% | 36.7% | 0.78 | 0.61 | 0.86 |
Terms. Commission concerns additional mapped burning relative to the reference. Omission concerns reference burning missed by the product. Perimeter IoU (intersection over union) measures overlap relative to the combined area. The final column compares the two reference products with each other and is shown as context only.
Reference products differ in how they represent boundaries, internal unburned areas, and fire margins. NBAC comparisons can also share satellite inputs or methodological features with this workflow. Reference agreement should therefore be interpreted in the context of the particular datasets and comparison used.
Development Findings and Limitations
- Spring fires. Fires that ignite before green-up are under-mapped: omission was about 30–50% in the samples above, because a dormant-season burn changes the NBR signal little. The spring thresholds were tuned on 14 fires across 2019, 2021 and 2024. On 20 held-out spring fires of 2025 they gave Dice 0.76 against NBAC, the same as in tuning.
- Small fires (under 200 ha). Agreement is lower and commission higher, because nearby unrelated change near a few detections is attributed to the fire. In 2025 the 10–200 ha fires reached Dice 0.47 against the Alberta perimeters, with 65% commission. Every polygon of a small fire needs analyst review.
- Neighbouring fires and re-flares. Fires that share one hotspot event are split by nearest request point, and fires that go quiet for more than 30 days are separate events. Both rules are proximity and timing rules, not proof of association. Large 2025 complexes show the resulting commission.
- Perimeter requests are bounded by the supplied perimeter. A request made with a perimeter maps burn only within that perimeter plus 500 m. Requested with each fire's first agency perimeter (a median 20% of the final area), it mapped most of the fire as missing (Dice 0.24 against the final perimeter), although agreement inside the requested area was high (Dice 0.85). Requests should carry the latest available perimeter.
- Scope. One province, four seasons (2019, 2021, 2024, 2025, plus 2023 in progress), Alberta forest and parkland fires of 10–50,000 ha. Fires above 50,000 ha were not evaluated.
- Freshness. HLS v2 is published about 2–4 days after acquisition. The 24-hour in-season freshness clause therefore cannot be met from HLS alone, and the 7-day post-season clause depends on cloud cover. Every deliverable reports the age of its newest suitable image.
These figures and findings are reproduced from the version 2.4 development evaluation record without recalculation; they describe that development version, not independent verification or the proposed service configuration. Record dated 2026-10-20.
Processing Notes
The normalized burn ratio (NBR) uses near-infrared and shortwave-infrared measurements to characterize burn-related change; comparing before- and after-fire values produces a differenced burn ratio that helps identify candidate burned areas. Earlier and current development versions use different detailed compositing and threshold rules, and each historical analysis records the version and settings used.
Proposed Delivery Standards
The following standards apply to requested provincial services. They are separate from the geographic coverage and processing history of individual website examples.
| Service | Imagery Requirement | Delivery Target |
|---|---|---|
| Snow Cover | Suitable observations within the preceding three days | Daily by 08:00 Alberta time when requested |
| Grass Curing | Suitable observations within the preceding seven days | Daily by 08:00 Alberta time when requested |
| Deciduous Leaf Out | Suitable observations within the preceding seven days | Daily by 08:00 Alberta time when requested |
| Burned Area Analysis: individual in-season request | Most recent suitable imagery less than 24 hours old, unless suitable imagery is unavailable | Within 24 hours of the request |
| Burned Area Analysis: post-season batch | Most recent suitable imagery less than seven days old, unless suitable imagery is unavailable | Within one week of the request |
The proposed snow and vegetation composites will meet the specified provincial cloud-free coverage requirement, with coverage measured and reported, and imagery as smoke-free as possible. The operational coverage requirement is at least 75%; provincial cloud-free coverage and valid estimates within the target fuel areas are reported separately where relevant. Imagery availability and coverage shortfalls will be communicated to Alberta Wildfire. Any alternative arrangements will require provincial approval and documented limitations.
GIS Products and Delivery Options
Proposed spatial deliverables use NAD83 / 10TM Forest (EPSG:3400) and support ArcMap 10.7/10.8 and ArcGIS Pro 3.3 or newer. Snow and vegetation percentage rasters are supplied as single-band, 8-bit GeoTIFFs. Required downloadable web services, polygon shapefiles, GIS styles, and source-imagery access form part of the applicable delivery package. Grass and leaf-out rasters align with the 100-metre provincial fuel grid.
Metadata records acquisition dates and times, processing times, satellites, input and output resolution, contributing image dates, methods, quality information, accuracy findings, appropriate uses, and restrictions. Where images are combined, source-footprint or seamline information identifies their contribution across the map; grass and leaf-out delivery includes the required downloadable seamline shapefiles and web services.
For Snow Cover, Grass Curing, and Deciduous Leaf Out, the proposal includes development support within Alberta Wildfire's environment, contractor-hosted delivery, or both as selected by the Province. Development support includes updating the snow workflow or converting the grass and leaf-out R workflows to Python, together with testing, documentation, deployment support, and training.
All displayed times use Alberta local time, with the applicable UTC offset. The underlying records retain the acquisition and processing timestamps needed to interpret each release.
Reading the Maps
- Observation date and release date have different meanings. The observation date records when an image was acquired. Processing and publication dates record when the product was made. Historical examples may be processed much later than the period they depict.
- Image age is assessed against the product's stated cutoff. Reusing an observation does not make it newer. Some regional vegetation examples use a longer window, up to 14 days; the proposed standard provincial grass and leaf-out products use seven days. Each release records the settings actually used.
- Coverage and condition are different measures. Coverage describes how much land has suitable observations or estimates. Average curing and leaf-out values describe the valid mapped estimates. Missing data is not a measured zero.
- The basemap supplies context. Its imagery can have a different date from the analytical product. Use the selected product's observation records to understand the evidence behind the result.
Quality Review
Quality assessment addresses three things: whether files and spatial data are correct, how well the method represents the landscape, and whether the proposed service can deliver reliably.
Checks cover spatial reference, grid alignment, valid values, missing-data handling, imagery age, coverage, source records, and file readability. Field observations and suitable reference mapping support evaluation of model and classification performance. Scheduling, publication, error handling, and GIS compatibility are assessed during implementation and commissioning.
Evidence is reported with the relevant product, method version, observations, and assessment conditions. A retrospective example, a model evaluation, and acceptance of an operational service answer different questions and are described separately.
File Integrity and Verification
Each release keeps the pipeline's checksum file, and the publishing step records the sha256 of every file. To check a download, compare its sha256 (certutil -hashfile file.tif SHA256 on Windows, or sha256sum file.tif) with the value under Product information.
Data Sources and Acknowledgements
The products draw on NASA-distributed VIIRS and HLS imagery, Sentinel-2 and Landsat where applicable, ESA WorldCover, and provincial fuel and geographic information. Release records identify the sources actually used. Dataset acknowledgements and applicable terms remain with the product information; basemap, software, and typeface credits are provided separately below.
Sources, Licences and Credits
| Source | Used for | Terms |
|---|---|---|
| NASA VIIRS snow (VJ110_NRT, VJ210_NRT, VNP10_NRT) | Snow Cover | NASA Earth Science data policy; DOIs in each release |
| Harmonized Landsat and Sentinel-2 (HLSL30, HLSS30 v2.0) | Grass Curing, Deciduous Leaf Out, Burned Area Analysis | NASA open data; contains modified Copernicus Sentinel data and USGS Landsat data |
| ESA WorldCover | Selects grass pixels for curing | CC BY 4.0, © ESA WorldCover project |
| Provincial fuel grid (Government of Alberta) | 100 m grid and fuel-type masks | Used in processing; not redistributed here |
| Calgary Forest Area outline | Map outline | From the project's own area-of-interest file, geometry only |
| OpenStreetMap | Basemap (default) | © OpenStreetMap contributors, ODbL; tiles under the OpenStreetMap Foundation tile usage policy |
| NASA GIBS: Landsat WELD annual true-colour composite (c. 2000) | Optional satellite basemap | NASA Earth Science open data policy (no restriction on use, no key); Landsat data courtesy of the U.S. Geological Survey; acknowledgement: imagery provided by NASA's Global Imagery Browse Services (GIBS), part of NASA's Earth Science Data and Information System (ESDIS) |
| Inter typeface | Interface text | SIL Open Font License, served by Google Fonts |
| Leaflet 1.9.4, geotiff.js 2.1.3, proj4js 2.11.0 | Map software in the page | BSD-2-Clause, MIT, MIT; served from this site |
Site and privacy notes. The site sets no cookies and has no analytics or tracking code; map tiles come from OpenStreetMap (or, when the satellite basemap is selected, NASA GIBS) and the typeface from Google Fonts. The basemap choice is remembered only in your browser.
Contact
For company information or enquiries, visit our Contact page.