Long-run burned-area and MODIS-context period.
Ghana Fire Observatory
Methodology
How spatial units, time periods, satellite observations and data-quality controls are structured for transparent fire analysis
Scope
A concise account of the analytical framework
The methodology is organised around three requirements: a stable spatial framework, explicit temporal support and honest interpretation of satellite products. This page describes the analytical design.
Spatial framework
National, agro-climatic-zone and district reporting
The district framework contains 260 harmonised units. Each district is assigned to the agro-climatic zone with which it has the greatest spatial overlap, providing a stable nested structure for national, zone and district analysis.
| Zone | Code | Districts |
|---|---|---|
| Sudan Savannah | SS | 20 |
| Guinea Savannah | GS | 45 |
| Transition | TZ | 27 |
| Forest | FZ | 110 |
| Coastal | CZ | 58 |
| Total | 260 | |
- Define unitsEstablish the national district universe and zone geometries.
- Measure overlapCalculate each district’s spatial overlap with the five zones.
- Assign contextAllocate each district to the zone with the greatest overlap.
- Keep stable identifiersUse the same district–zone relationship across the monthly panel.
Temporal framework
A complete monthly spine with product-specific support
The panel spans January 2001 to December 2024, giving 288 monthly positions for every district and zone. Product values are populated only where the source and analytical design support them.
Comparison window used for MODIS–VIIRS mismatch analysis; January 2012 is excluded.
Full-calendar-year summaries for overlap-era analyses.
Structural missingness: months outside a product’s valid support period remain explicitly unavailable. They are not converted to zero, because zero means a supported observation with no detected event.
Sensor roles
Fire impact and fire activity are analysed separately
MODIS MCD64A1 burned area
Mapped burned area is used to characterise the land surface affected by fire. Monthly metrics include burned area, burned-pixel counts and normalised burned-area measures where valid denominators are available.
MODIS and VIIRS active fire
Thermal detections are used to characterise observed fire activity. Metrics include detection counts, active days and fire-radiative-power summaries within the appropriate support periods.
Agreement between products is informative, but disagreement is also scientifically meaningful. Product resolution, detection opportunity, cloud and smoke effects, overpass timing and the physical distinction between activity and impact can all contribute to mismatch.
Analysis
Methods used for long-run and overlap-era assessment
Methods are applied only where their assumptions and product support are appropriate.
Seasonality and fire calendar
The year-by-month heatmap retains supported positive values, valid zero and structural unavailability as distinct states. Absolute mode uses native monthly units. Annual-share mode normalises each supported year to 100% before monthly shares are compared, so high-total years do not dominate the seasonal shape. Monthly anomaly subtracts a labelled month-of-year climatological mean. The within-year index sets each applicable year’s supported monthly mean to 100. Burned-proportion contributions use monthly burned area because percentages are not additive. The national November–March grouping is an analysis window, not a universal climate season; no fixed zone window is used until it is scientifically approved. Custom windows are user-defined analytical groupings and may wrap across December and January.
Annual trajectories
The Observatory shows absolute annual values, values indexed to the supported selected-period mean of 100, or anomalies from that mean. A centred three-year mean is an optional descriptive overlay. Governed trend wording uses Mann–Kendall direction and Sen slope; it does not imply causation or prediction.
Active-fire detection states
For each supported annual or monthly observation, MODIS active-fire detections and VIIRS active-fire detections are classified as both zero, MODIS only positive, VIIRS only positive or both positive. These states are distinct from the governed monthly MODIS burned-area-presence–VIIRS mismatch states.
Sensor comparison
The public Sensors page compares MODIS active-fire detections and VIIRS active-fire detections as separate products. Paired monthly district, zone and Ghana comparison begins in February 2012; annual comparison uses complete years 2013–2024 because annual 2012 totals cannot remove January. Spearman rank correlation uses average ranks and measures monotonic association, not agreement, interchangeability, accuracy or causation. Difference is VIIRS minus MODIS. Ratio is VIIRS divided by MODIS and is defined only where MODIS is positive.
Burned-area–VIIRS mismatch
A separate monthly technical view classifies MODIS burned-area presence and VIIRS active-fire detections as both zero, VIIRS only, MODIS burned-area only or both positive over February 2012–December 2024. Neither product is treated as truth; disagreement is observational structure rather than an automatic error label.
Spatial concentration
Hotspot and spatio-temporal clustering methods identify concentrations that depart from the broader background pattern.
District period aggregation
Detection counts, burned area and active-fire days are summed across supported annual records. Burned percentage is recomputed as total burned area divided by total burnable area multiplied by 100; annual percentages are never summed.
Map classification
The district map can use a continuous scale, five quantile classes or five equal-interval classes. The optional 95th-percentile display cap preserves spatial contrast without changing exact values in tooltips, rankings or exports. Valid zero, structural unavailability, undefined values and out-of-scope districts remain distinct.
District ranks and percentile standing
Competition ranking assigns equal values the same rank. National and within-zone denominators include only districts with a supported value. Percentile standing is the percentage of supported districts with a lower selected-period value.
Profile benchmarks and climatology
Absolute benchmarks retain native units. Indexed benchmarks set each supported district, zone or Ghana series to its own selected-period mean of 100 and therefore compare relative trajectories rather than magnitudes. Governed district-month records support both full-supported and selected-period monthly climatology, with contributing-year counts and product support retained.
Overview summaries
The Overview separates selected-period values from annual means. Detection, burned-area and active-day indicators are summed across supported years. Period burned proportion is recomputed as summed valid burned area divided by summed valid annual burnable area multiplied by 100; annual percentages are never summed. Recent change compares the latest five complete supported years with the preceding five and requires ten consecutive supported years. National active-day totals are reported as zone-days because unique national calendar days are not available in the runtime panel.
Contextual comparison
District profiles are compared with agro-climatic-zone and national baselines to expose local departures.
Data quality
Controls applied before interpretation
Canonical unit–month spine
Every expected district–month and zone–month key is represented once, which makes missing support visible and permits complete key auditing.
Source and support flags
Source-presence, structural-missingness and overlap-support flags preserve the distinction between absent source coverage and valid measurements.
Denominator controls
Normalised measures are calculated only where the relevant land or burnable-area denominator is valid.
Audit and manifest files
Key audits, file manifests, schema records and merge summaries provide an explicit provenance trail. The Observatory quality register compares governed expected values with observed runtime values and distinguishes available, valid-zero, structurally unavailable, invalid/missing and not-applicable states.
Interpretation limits
Satellite products do not observe every fire, and district aggregation does not represent address-level hazard. The current public panel is suited to fire-regime and comparative analysis; the 30 m susceptibility framework is a separate ongoing study and its results are not included here.