Methodology
Fire Watch Dashboard
The Fire Watch Dashboard uses satellite observations to provide a consistent view of agricultural burning across Punjab, Haryana, Uttar Pradesh and Madhya Pradesh. It brings together three complementary indicators:
| Indicator | What it shows | Unit |
|---|---|---|
| Active fire count | Where and how frequently thermal anomalies associated with burning in marked croplands were detected | Number of detections |
| Burnt area | How much cropland shows a spectral signature consistent with burning | km² and % of cropland |
| Daily fire timing | When during the day, burning activity was detected | % of daily detections by 1 hour interval |
The indicators answer different questions and should be interpreted together. Active-fire detections indicate observed fire activity; burnt area captures the spatial extent of affected cropland, and daily fire timing shows when detectable activity occurs.
The dashboard covers both Kharif and Rabi burning seasons. A common analysis window is used across the four states to maintain consistency in reporting. Results are primarily presented at the district level, with state-level aggregation also available.
The dashboard combines data from different satellite systems, with each source selected for a specific analytical purpose.
| Indicator | Orbit | Sensor | Satellite | Nominal overpass time (India)* | Spatial resolution |
|---|---|---|---|---|---|
| Active fire count | Polar | VIIRS | NOAA-20 | 1:30 PM 1:30 AM |
375 m |
| MODIS | Terra | 10:30 AM | 1 km | ||
| Aqua | 1:30 PM | ||||
| Burnt area | MSI | Sentinel-2 | 10:30 AM | 20 m | |
| Daily fire timing | Geostationary | SEVIRI-MSG | Meteosat 8 & Meteosat 9 |
Every 15 minutes | ~4 km × 4 km to 6 km × 6 km |
*Nominal local overpass time refers to the approximate equator-crossing or scheduled acquisition time documented for the satellite mission. Actual acquisition times over India may vary due to orbital drift and changes in the satellite ground track and in Geostationary case its observation interval which means how frequently it is giving data.
Active fire observations come from polar-orbiting satellites, providing observations at different times of day and spatial resolutions of 375 m (VIIRS) and 1 km (MODIS). Sentinel-2 provides higher-resolution imagery at 20 m for mapping burnt areas, with a nominal revisit of five days. SEVIRI, on geostationary Meteosat satellites, provides observations every 15 minutes, enabling the dashboard to examine the daily timing of detected burning activity.
This combination allows Fire Watch to examine three complementary dimensions of agricultural burning: activity, extent and timing.
Active-fire data identify thermal anomalies observed by satellites. They do not represent a direct count of individual fires.
The processing involves four main steps:
- Quality filtering: Low-confidence detections are removed. For NOAA-20 nominal- and high-confidence detections are retained; for MODIS, detections with confidence of at least 30% are retained.
- Cropland filtering: Detections are retained where the location of the thermal anomaly is on the cropland mask derived from ESA WorldCover 2021 (version 200).
- Spatial and temporal assignment: Each detection is assigned to a district and state and converted to Indian Standard Time (IST).
- Aggregation: Valid detections are aggregated by district, season and burn-day to produce the active-fire count.
Note: The dashboard reports fire detections, not the number of fires. The same burning event can be detected more than once, while small, short-lived or cloud-obscured fires may not be detected.
Burnt area is estimated to be independent of active-fire detection using Sentinel-2 imagery. It measures the area of cropland that shows a surface change consistent with burning.
The method compares pre- and post-burning imagery using the Normalized Burn Ratio (NBR). The difference between the two observations, known as dNBR, is used to identify areas where vegetation and surface characteristics have changed following burning. Higher dNBR values generally indicate a stronger burn-related change.
The processing involves the following steps:
- Sentinel-2 imagery is divided into 5-to-10-day intervals.
- The least-cloudy valid observations are selected for pre- and post-burning comparison.
- Clouds, shadows, cirrus and saturated pixels are removed.
- dNBR is calculated at 20 m resolution.
- A state- and season-specific threshold is applied to classify pixels as burnt or unburnt.
- The result is restricted to cropland, and small isolated pixels are removed.
- A pixel is counted only once per season, preventing the same persistent burn scar from being counted repeatedly.
| Measure | Meaning |
|---|---|
| Burnt area (km²) | Total cropland area classified as burnt |
| % cropland burnt | Burnt area as a share of the mapped cropland area |
Because cloud cover or missing imagery can limit observations, interpretation should also consider the proportion of cropland that had valid imagery during the season.
Note: Burnt area is a seasonal measure, not a live count of burning activity. It is derived from observed changes in satellite imagery and is therefore different from the active-fire indicator.
Daily fire timing characterizes the intra-day distribution of satellite-detected thermal activity, indicating when detectable burning activity occurs during the day. Unlike polar-orbiting satellites, which provide observations of a location only at specific overpass times, the SEVIRI sensor aboard the Meteosat geostationary satellites provides observations at approximately 15-minute intervals throughout the day. This high temporal frequency enables the diurnal pattern of detectable fire activity to be examined.
The methodology involves the following steps:
- Time-zone conversion: SEVIRI observations are originally recorded in Coordinated Universal Time (UTC) and are converted to Indian Standard Time (IST, UTC+5:30) to represent the local timing of fire activity in India.
- Temporal aggregation: Individual SEVIRI observations are aggregated into hourly time slots based on their IST timestamp. This converts the high-frequency observations into a consistent hourly profile that is easier to interpret and compare across locations and periods.
- Daily activity calculation: The number of valid thermal detections recorded within each hourly interval is calculated for each day.
- Normalisation: Hourly detection counts are normalised by the total number of valid detections for the corresponding day to calculate the percentage of daily detected thermal activity occurring in each hour.
The resulting profile represents the relative distribution of detectable thermal activity throughout the day. It is intended to identify periods of increased or decreased fire-detection activity rather than to estimate the exact number of fires occurring at a particular time.
Because SEVIRI has a substantially coarser spatial footprint than NOAA-20/VIIRS and Sentinel-2, individual observations may contain multiple land-cover types within the same satellite pixel.
The temporal profile is also more likely to represent larger, hotter or longer-lasting fires that generate a detectable thermal signal at SEVIRI's spatial resolution. Smaller, short-duration or lower-intensity agricultural fires may not produce a sufficiently strong or persistent signal to be detected consistently.
Therefore, the Daily Fire Timing indicator is used primarily to understand the diurnal pattern of observed burning activity and should be interpreted alongside the higher-spatial-resolution active-fire and burnt-area indicators.
Satellite observations provide a consistent way to monitor agricultural burning at scale, but they have inherent limitations. The three indicators should therefore be interpreted as complementary measures rather than interchangeable measures of fire activity.
| Consideration | Implication for interpretation |
|---|---|
| Not all fires are detected | Small, short-duration or low-intensity fires may be missed by satellite observations. |
| Cloud and atmospheric conditions can obscure observations | Clouds and other atmospheric conditions can interfere with thermal observations and may reduce the number of detectable fire signals during particular periods. |
| Fire count ≠ number of fires | Multiple satellite observations can correspond to the same burning event. |
| Burnt area is not a direct fire detection | Other land-surface changes, such as harvesting or tillage, can produce spectral changes similar to burning. |
| Daily fire timing has higher sensitivity to larger or hotter fires | Smaller, cooler or short-duration agricultural fires may not generate a sufficiently strong thermal signal for reliable detection. The daily fire timing may therefore be more representative of larger or more intense burning events. |
| Cropland mask affects results | Reported percentages depend on the mapped extent of cropland used as the denominator. |
| State and district comparisons require context | Differences in cropland extent, field size, crop types and burning practices can influence satellite observations. |
| Absence of detection does not mean absence of burning | A low satellite-derived count may reflect limited detection as well as genuinely lower activity. |
The dashboard should therefore be used to understand patterns and spatial-temporal differences in observed agricultural burning, rather than as a farm-level compliance or enforcement tool.