Figure 2‑37 shows a visual comparison between the spatial patterns of daytime active fire pixel count and FRP contained within the S3 Level-3 monthly daytime FRP products retrieved from one year’s Sentinel-3A and -3B data, and from Terra MODIS’ MOD14 Collection 6 product. Both S3 and MODIS products cover the period March 2023 to February 2024 and show very similar spatial patterns, indicating a broad degree of agreement even though the MODIS data includes all daytime observations and not just those collected near-simultaneously with those of SLSTR.
The daytime AF pixel count data shown in Figure 2‑37 (a, c, e) indicate that both the Sentinel-3A and Sentinel-3B SLSTR Level 3 products show a very similar AF pixel detection spatial pattern, and also one that is very similar to that provided by MODIS. However, as the grid-cell FRP totals shown in Figure 2‑37 (b, d, f) indicate, the detected MODIS active fire pixels taken together constitute a higher FRP total (47,648,138 MW) than that from either SLSTR from Sentinel-3A or Sentinel-3B – though the value is quite close to the total of the S3A (30,486,462 MW) and S3B (20,035,027 MW) records combined. This results in part from the fact that data from the dual S3A and S3B SLSTR sensors are required to be used together to provide global active fire data coverage every day, whereas only the single MODIS Terra sensor is needed due to its far larger swath width (2330 km vs. 1400 km). These findings mirror those reported in Xu et al. (2023) during initial evaluation of the daytime Level 2 SLSTR Active Fire Detection and FRP product. Furthermore, as Xu et al. (2023) discuss for the Level 2 SLSTR daytime products, whilst there are some Level 2 AF pixels that SLSTR but not Terra detects, these have typically lower FRP values and are often weakly radiant AF pixels located at the edges of fire clusters that do not add much to the cluster FRP total. There are also some fires that Terra detects but not SLSTR, typically containing only a few or a single AF pixel and which are not detected by the SLSTR daytime algorithm due to its raised thresholds compared to the night-time version (Xu et al., 2023). These Level 2 features all feed into the performance of the C3S Level 3 daytime products, and the evaluation results of those products reflected herein.
Figure 2‑37: Daytime total active fire (AF) pixel count and total FRP of actively burning fires detected from space from March 2023 to Feb 2024 using, respectively, Sentinel-3A SLSTR, Sentinel-3B SLSTR and MODIS Terra, displayed in 0.25° grid cells. Data used are from all observations, not just those observed near simultaneously between each sensor. Note that the MODIS Terra FRP has a scale that is approx. twice that of S3A and S3B, due to the generally higher maximum grid-cell FRP totals from Terra, though once the S3A and S3B data are added together to approximate daily global coverage, the total is quite similar to that from MODIS Terra which also provides that coverage.
Figure 2‑38 provides a visual comparison between the spatial patterns of active fire (AF) pixel count and FRP seen within the Level 3 Monthly Summary AF & FRP Night-time Products, these being derived from the SLSTR thermal infrared observations. The product outputs shown in Figure 2‑38 are derived from the data collected by different instruments - SLSTR onboard Sentinel-3A (panels a and b) and -3B (panels c and d) and from MODIS on Terra (panels e and f) – and encompass a one-year period (March 2023 to February 2024). The Sentinel-3 SLSTR and MODIS Terra products covering this period show very similar spatial patterns and a broad degree of agreement, especially considering that the MODIS data including all night-time observations and not just those made near-simultaneously with SLSTR.
Figure 2‑38: Total active fire (AF) pixel count (left panels) and total FRP of fires detected (right panels) within 0.25° grid cells, as calculated for the March 2023 to February 2024 period using all Sentinel-3A SLSTR, Sentinel-3B SLSTR and Terra MODIS data. Note that the Terra MODIS active fire pixel count map has a quantitative scale two times smaller than that of S3A and S3B due to its typically significantly lower detection count of AF pixels.
The AF pixel count data show that the SLSTR products include many more AF pixel detections than the MODIS products, but the grid-cell FRP totals shown in Figure 2‑38 are similar between the two records. This is because the additional AF pixels that SLSTR detects over and above those detected by MODIS in many of the grid cells have mostly individually low FRP values. These findings mirror those reported in Xu et al. (2020), and the results from Section 2.3.1 – in that at the grid-cell level the recorded FRP is similar between the compatible Sentinel-3 and MODIS products even though the SLSTR product often includes many more AF pixels detected in the cell. This is due to the fact that at night SLSTR can detect lower FRP fires than MODIS due to its AF detection algorithms sensitivity and the smaller SLSTR pixel area across the swath compared to MODIS.
The AF detection process is inherently a trade-off between attempting to detect the lowest FRP fires (which are typically the most common type in most geographical areas), whilst also minimising false alarms (i.e., the triggering of the AF detection algorithm by non-fire phenomena). Raising the minimum FRP detection limit of the contextual AF detection algorithm used to generate the Level 2 SLSTR AF products was required by day (Xu et al., 2023) compared to at night (Xu et al., 2020) due to the fact that the noisier F1 had to be used by day more often for AF detection due to saturation of the more sensitive S7 band (Xu et al., 2023). This increase in minimum FRP detection limit very likely reduces the number of false alarms, which are common with of low FRP magnitude, but will also prevent some real fires having low FRP from being identified. Xu et al. (2023) already compared global daytime AF pixel data from MODIS Terra taken within a scan angle limit of 30° and within ±6 minutes of a Sentinel-3 SLSTR acquisition, and found that 70% of MODIS’ identified active fire pixels had a matching Sentinel-3 AF pixel detection, representing an apparent SLSTR AF detection omission error compared to MODIS of 30%. This metric assumes all MODIS AF pixel detections are indeed true active fires, whereas at the global scale Giglio et al. (2016) have determined that around 1.2% of the MODIS AF pixels are in fact false alarms – and thus a small proportion of the MODIS AF pixels that SLSTR fails to detect may actually not be true fire pixels. Conversely, of the Sentinel-3 daytime AF pixel detections present in the same daytime Level 2 AF detection dataset, 84% had a matched MODIS AF pixel detection – representing an apparent MODIS product AF pixel omission error compared to SLSTR of 16%. The analysis of daytime global S3 SLSTR and MODIS Terra AF fire data from March 2023 to February 2024 (see Figure 2‑4) showed that overall, the combined S3A and S3B datasets include around twice the number of AF pixels compared to the matching MODIS Terra data, but that the additional AF pixels have mostly low FRP and are often weakly radiating and found at the edge of an active fire cluster that contains some AF pixels having an FRP significantly higher than the daytime algorithms minimum FRP detection limit. Conversely, single pixel fires with a low FRP, or indeed fires with a few AF pixels but all of which are low FRP, that are detected by MODIS relatively close to nadir are quite often missed by SLSTR – because of the raised minimum FRP detection limit of the daytime S3 product. Indeed, this latter performance is somewhat the opposite of that shown by the S3 night-time product, where single pixel ‘low FRP’ fires appear in general more likely detected by S3 than by MODIS, even when imaged by MODIS close to nadir (Xu et al., 2020).
The C3S Level 3a Daily Gridded AF & FRP Daytime Products are intended primarily for large scale analysis of fire patterns, seasonality, anomalies and trends. Such characteristics are part of a regional ‘fire regime’, which describes the role of landscape fires within an area over time, and under this broad definition, the fires’ physical attributes such as frequency, size, intensity, type and seasonality. Fire regimes may change with changing climate and with other changes in human impact associated with, for example, land use management and land use change (Moritz 2012; Flannigan et al., 2009; Archibald et al., 2013; Hantson et al, 2015). Characterizing past and current fire regimes has historically been performed by analysing field data such as charcoal records, fire-scar networks and fire occurrence databases. However, the regular and continuous information on landscape fires that can be provided by EO satellites is increasingly being used to determine certain fire regime characteristics (e.g., Chuvieco et al., 2008; Freeborn et al., 2014). Seasonality is one of the most important component characteristics of a region fire regime, and one which products such as the C3S Level 3 gridded FRP and the MODIS Active Fire Products are well suited to determine. To evaluate the daytime C3S Level 3 Gridded FRP Products we therefore used them to derive fire regime seasonality characteristics, and compared these to the same metrics derived from the MOD14 data used to create the similar MODIS MOD14CMQ and MOD14C8Q climate modelling grid (CMG) products (Justice et al., 2002).
The fire season metrics derived from the daytime C3S Level 3a Daily Gridded AF & FRP Daytime Product files were compared to those derived from daytime MOD14 data gridded to the same 0.1° spatial resolution grid. The comparisons are made globally as well as within the geographic regions of the Global Fire Emission Database (GFED) (see Figure 1‑5). Comparisons were made in terms of AF detections but also FRP (as in Figure 2‑39). The 14 GFED regions are defined based on areas having similar biomass burning regimes rather than on country borders (Van der Werf et al., 2017). Since, as shown above, the MODIS and SLSTR FRP records are more similar than the AF pixel count records, and as shown in the daytime C3S FRP validation reports the FRP patterns are more in agreement with the seasonality patterns seen in Giglio et al. (2006), we focus on the FRP metric for fire season analysis.
As an example at the global scale, Figure 2‑39a shows daytime daily global total daytime FRP as derived from the daily gridded global AF products of Sentinel-3A, -3B SLSTR and MODIS Terra. All three products show a very similar temporal development. The cumulative percentage FRP plot shown in Figure 2‑39b shows a similar trend from each sensor, and at the global scale the fire season starts in June and ends in November according to each of S3A, S3B and MODIS – based on defining the start and end of the fire season as the point at which 25% and 75% of the total FRP is reached respectively.
Figure 2‑39: Intercomparison of the daytime daily global FRP records derived from the C3S Level 3a Daily Gridded FRP Product files for S3A and S3B SLSTR and the MOD14 daytime records from Terra MODIS. (a) Total global daily FRP from each sensor; (b) Cumulative percentage of total global daily FRP. Note that there are a few days in August 2023 when S3 SLSTR provided no input Level-2 FRP data.
Figure 2‑40 shows the fire season analysis, both globally and for the GFED regions (shown in Figure 1‑6 and defined in the caption for Figure 2-42). Results from the daytime C3S Level 3a Daily Gridded FRP Product files generated from Sentinel-3 SLSTR and also from MODIS Terra data are shown. Globally, the three satellite datasets (S3A, S3B and MODIS Terra) agree very well, all indicating a start and end of the fire season within a few days of one another as derived from these daytime records. A very similar fire duration also therefore results from each dataset. For example, the global fire season starts towards the end of June 2023 and ends between end of October to beginning of November according to these data, lasting a period of approximately six months. For the GFED regions, 10 out of the 14 regions show a very good agreement between the three satellites, with a fire season start, peak, end and duration identical to within a few days.
Whilst most of the GFED regions have a single fire season annually, driven by a combination of climate and human activity, Several South and Southeast Asian countries, including India, utilise a dual cropping system where two distinct growing seasons (e.g. one for wheat and another for rice) are each followed by agricultural burning of the crop residues. Therefore, such areas can display two distinct fire seasons rather than one, and these signatures can be exacerbated by other non-crop burning fire activity such as burning in forests and grasslands. In India for example, the first fire season annually occurs around March–May (strongest in April–May), dominated by burning of wheat stubble and also sometimes with forest fires occurring across central India, the northeast, and the Himalayan foothills. The second Indian fire season typically occurs in October–November, primarily associated with rice‑residue burning in northwestern India, notably Punjab and Haryana. In these situations, the bi-modal nature of these crop-residue burning driven fire season poses a challenge for the fire‑season indicator used herein. In the Southeast Asia (SEAS) GFED region, whilst the three satellites all agree on the April start of the fire season (Figure 2-40b), MODIS data indicates the fire‑season ends in mid‑May, whereas S3A and S3B indicate this occurs in November (Figure 2-40c). This discrepancy results from the bi-modal nature of the fire activity, and results in an inferred duration in the SEAS region that greatly differs between MODIS and S3A/B (Figure 2-40d).
Figure 2‑40: Fire season metrics, as determined from the daytime Level 3a Daily Gridded FRP Product of S3A and S3B and from daily MODIS Terra data. Fire Season (a) Start, (b) Peak), (c) End and (d) Duration. Temporal resolution of the data used in the calculations is one day. The GFED regions are shown in Figure 1‑5 and are Boreal North America (BONA); Temperate North America (TENA); Central America (CEAM); NH South America (NHSA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); NH Africa (NHAF); SH Africa (SHAF); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS); Equatorial Asia (EQAS); Australia and New Zealand (AUST). Using the different S3A, S3B and MODIS daytime data, the generally very good match between the different metrics for the majority of the GFED areas is apparent. Note in some countries in the SEAS region have two fire seasons annually, mostly driven by agricultural burning of different crop residues, and this has lead here to differing fire season end dates being calculated from MODIS data (mid‑May) as compared to S3A/S3B (November). The fire season duration is therefore also calculated as being rather different according to the data of the MODIS and Sentinel-3 SLSTR sensors.
The daytime C3S Level 3a 27-Day Gridded AF & FRP Daytime Product is simply the accumulation of twenty-seven of the daytime C3S Level 3a Daily Gridded AF & FRP Daytime Products used in Section 2.6.3, so the evaluation of the former simply focuses on verifying the correctness of the lower temporal resolution (27-day) statistical summary derived from the daily data. The daytime 27-Day products were verified thoroughly in this way, with the active fire pixel count, FRP and all other parameters being found to agree with the 27-day accumulation of the daytime daily product data. This verification indicates the correctness of the data contained within the C3S Level 3a 27-Day daytime Gridded FRP Products.
An example of this verification is shown in Figure 2‑41, where the active fire pixel count from 3 to 29 August 2023 derived from the daytime C3S Level 3a 27-Day Gridded AF & FRP Daytime Product perfectly agrees with that derived from the accumulation of the appropriate twenty seven individual daily products.
Figure 2‑41: Example verification of the daytime C3S Level-3a 27-Day gridded FRP product. (a) Total active fire pixel count from the daytime C3S Level 3a 27-Day Gridded AF & FRP Daytime Product covering 3rd to 29th Aug 2023; (b) Total active fire pixels from the equivalent twenty-seven daytime C3S Level 3a Daily Gridded AF & FRP Daytime Products covering the same period. Active fire pixel count is identical in all the grid cells in (a) and (b).
As with the verification of the daytime C3S Level 3a Daily Gridded Gas Flare Night-time Product conducted in Section 2.4.1, verification of the daytime C3S Level 3 Monthly Summary AF & FRP Daytime Product was compared to the data contained within the daytime Terra MODIS MOD14 products covering the same area and time period. As the spatial patterns of absolute AF pixel counts and FRP have been analysed already in Section 2.6.1, the focus here is on the analysis of the fire season and associated metrics defined at the global and GFED region scale. Degrees of agreement between the C3S and MODIS products are quantified using the coefficient of determination (r2) and slope of the linear best fit between them.
Figure 2‑42 and Figure 2‑43 (image has been divided into two distinct parts to enhance user accessibility and clarity) show the seasonal pattern of cumulative monthly FRP at both the global and GFED region scale, as derived from both the daytime monthly C3S FRP products and from MODIS Terra. As with the C3S daily product (Figure 2‑39 and Figure 2‑40), the seasonal pattern seen from the monthly C3S product appears very close to that derived from MODIS. Examining the global scale data, the r2 value between the S3A and Terra data is 1.0 and the slope of the ordinary least squares (OLS) linear best fit between them is also 1.0, and that between S3B and Terra is the same. For the GFED regions, the r2 value between the S3A and MODIS data ranges from 0.88 to 1.00, and with the slopes of the linear best fits lying between 0.75 and 1.02, whilst for S3B and MODIS the r2 lies between 0.93 and 1.00 and the slope 0.79 to 1.09.
Figure 2‑42: Seasonal pattern of cumulative daytime monthly FRP at the global and GFED region scales, as derived from the daytime C3S Level 3 Monthly Summary AF & FRP Daytime Product and from matching daytime MODIS Terra MOD14 data. The abbreviation of the names is as follows: Boreal North America (BONA); Temperate North America (TENA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS) – see Figure 1‑5. The statistics of the r2 values between the datasets, and of the slope of the linear best fit between them, are also shown.
Figure 2‑43: Seasonal pattern of daytime cumulative monthly FRP at the global and GFED region scales, as derived from the daytime C3S Level 3 Monthly Summary FRP Product and from matching daytime MODIS Terra MOD14 data. The abbreviation of the names is as follows: Central America (CEAM); NH South America (NHSA); NH Africa (NHAF); SH Africa (SHAF); Equatorial Asia (EQAS); Australia and New Zealand (AUST) – see Figure 1‑5. The statistics of the r2 values between the datasets, and of the slope of the linear best fit between them, are also shown.
Figure 2‑44 shows the fire season start, peak, end and duration as derived from the daytime C3S Level 3 Monthly Summary FRP Product and from the matching daytime MODIS Terra data, both globally and across the GFED regions. As before, a threshold of 25% of the total cumulative FRP was used to define the start of the fire season, and 75% to define the end of the fire season. Overall, the start, end and duration of the fire season agree very well between the data derived from S3A, S3B and MODIS Terra. At the global scale, the fire season begins and ends at similar months according to the different data sources, and therefore the duration is also very close. However, there are some discrepancies between SLSTR and MODIS in terms of the peak of the fire season. Analysis of the Sentinel-3 product reports a global fire season peak in July 2023, whilst MODIS shows a peak in October 2023. For all the GFED regions, ~ 25% of the 14 regions have 100% agreement for all the four fire season metrics analysed, whilst ~ 80% have a difference of one month or very occasionally more. Almost all the regions have a difference in the metrics of less than two months. In some countries within the SESA region, such as India, fire activity is bimodal—March–May forest fires and October–November crop‑residue burning—leading to differing season end dates between MODIS (May) and S3A/S3B (November/December). See Section 2.6.3 (Level‑3a Daily Gridded AF & FRP Daytime Product) for details.
Figure 2‑44: Fire season metrics as determined from daytime Level 3 Monthly Gridded FRP Product of S3A and S3B and from daily daytime MODIS Terra MOD14 product data. The temporal resolution of the data is one month. Analysis is conducted globally and for the GFED regions. (a) Start of the fire season; (b) Peak of the fire season; (c) End of the fire season; (d) Duration of the fire season. The abbreviation of the names is as follows: Boreal North America (BONA); Temperate North America (TENA); Central America (CEAM); NH South America (NHSA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); NH Africa (NHAF); SH Africa (SHAF); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS); Equatorial Asia (EQAS); Australia and New Zealand (AUST) – see Figure 1‑6. Note in some countries in SESA like India, fire activity is bimodal—March–May forest fires and October–November crop-residue burning—leading to differing season end dates between MODIS (May) and S3A/S3B (Dec/Nov).
The AF detection process is inherently a trade-off between attempting to detect lower FRP active fire pixels (which are typically the most common type in most geographic areas), whilst also minimising false alarms (i.e., the triggering of the AF detection algorithm by a non-fire phenomena). Raising the minimum FRP detection limit in the contextual AF detection algorithm of Xu et al. (2020) very likely reduces the number of false alarms, but will also prevent some lower FRP fires from being identified. Comparison to global data from Terra MODIS taken within a scan angle limit of 30° and within ±6 minutes of a Sentinel-3 SLSTR acquisition (see Section 2.3.1 and Xu et al., 2020) shows that > 90% of MODIS-identified active fire pixels (MOD14 Collection 6 product) had a matching Sentinel-3 AF pixel detection, representing an apparent SLSTR product omission error of less than 10% compared to MODIS. Conversely, of the Sentinel-3 AF pixel detections present in the same dataset, only around 60% had a matching MODIS AF pixel detection, and around 80% of the additional AF pixel detections made by SLSTR had an FRP of less than 5 MW. These additional AF pixel detections provided by Sentinel-3 SLSTR are mostly below the minimum MODIS FRP AF pixel detection limit, and SLSTR can provide this benefit because it has a slightly lower minimum FRP detection limit than MODIS due to the smaller mean SLSTR pixel area and the intricacies of the sensitive AF pixel detection algorithm used to generate the Setinel-3 AF products (Xu et al., 2020).
Figure 2‑45 shows a visual comparison between the spatial patterns of active fire pixel count and FRP contained within the Level-3 monthly Night-time FRP products retrieved from Sentinel-3A and -3B data, and from Terra MODIS in September 2023. The errors of omission and commission of SLSTR compared to MODIS in the input Level 2 products are expected to be similar to those of prior years, and those reported in Section 2.3.2 and indeed also in Xu et al. (2020). Very similar spatial patterns are seen in all the data, indicating a broad degree of agreement despite the MODIS data including all night-time observations and not just those made contemporaneously with SLSTR. These results mirror those obtained in prior years and indicate that the performance of the S3 product has shown no noticeable change in the current period. The SLSTR product detects more AF pixels than MODIS, but the additional pixels are dominated by low FRP detections and so the grid-cell FRP is far more similar between the two types of sensor.
Figure 2‑45: Total active fire (AF) pixel count (left panels) and total FRP of fires detected (right panels) within 0.25° grid cells for the month of September 2023 using all Sentinel-3A, -3B SLSTR and Terra MODIS data. Note the Terra MODIS active fire pixel count has a scale that is five times smaller than that of S3A and S3B due to the lower number of AF pixels the former sensor typically detects.
Similar to section 2.3.3, the fire season metrics for 2023-24 derived from the C3S Level 3a Daily Gridded AF & FRP Night-time Product files were compared to those derived from the matching MOD14 data. The comparisons are made globally as well as within the geographic regions of the Global Fire Emission Database (GFED) (e.g., see Figure 1‑6). Comparisons were made in terms of AF detections as well as FRP.
Figure 2‑46a shows daily global total FRP as derived from data collected by SLSTR onboard Sentinel-3A, -3B and by Terra MODIS and stored in the various daily gridded global products (March 23 to February 24). All three products show a very similar temporal development, and at the global scale the fire season peak occurs in August, starts in May and ends in November for each of S3A, S3B and MODIS. The cumulative percentage FRP displayed in Figure 2‑46b shows a similar trend, and using the standard global 10% and 90% of the total FRP to represent the start and end of the fire season respectively we identify June 2023 as the start of season and November 2023 as the end of the fire season using the products from each of the two Sentinel-3 satellites and Terra MODIS. These timings, as well as the duration of the global “fire season”, also agree with the prior findings of Giglio et al. (2006).
Figure 2‑46: Daily FRP timeseries comparison between SLSTR and MODIS. (a) Total global daily FRP; (b) Cumulative percentage of total global daily FRP; Note that there are a few days during the period when SLSTR has no input Level-2 FRP data and also a few when Terra MODIS delivered no FRP data.
Figure 2‑47 shows the daily peak of the fire season both globally and in the individual GFED regions defined in Figure 1‑6, both from SLSTR and from Terra MODIS. Globally the results from the three satellites (S3A, S3B and Terra MODIS) agree well, all indicating that the global fire season peak occurs in late September 2023. For the individual GFED regions, 12 out of the 14 regions show a very good agreement between the timing of the fire season peak deduced from the two sensors. Note that landscape fire is a dynamic phenomenon having a high daily variance, and further analysis is needed to study the difference for regions showing greater disparity between each dataset.
Figure 2‑47: Peak timing of the fire season as determined from Level 3a Daily Gridded FRP Product of S3A and S3B and from daily Terra MODIS data. Temporal resolution of the data is one day. The abbreviated names are as follows: Boreal North America (BONA); Temperate North America (TENA); Central America (CEAM); NH South America (NHSA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); NH Africa (NHAF); SH Africa (SHAF); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS); Equatorial Asia (EQAS); Australia and New Zealand (AUST).
The C3S Level 3a 27-Day Gridded AF & FRP Night-time Product is simply the accumulation of 27 C3S Level 3a Daily Gridded FRP Products, so its evaluation will focus on verifying the correctness of its lower temporal resolution statistical summary, as derived from the former data. The 27-Day products were verified thoroughly in this way, with the active fire pixel count, FRP and all other parameters in the Level 3a 27-Day Gridded AF & FRP Night-time Product found to agree with the accumulation of the daily product data. This indicates the correctness of the C3S 27-Day products. An example is shown in Figure 2‑15 where the active fire pixel count from 3 to 29 August 2023 derived from the 27-Day product perfectly agrees with that derived from the accumulation of the 27 individual daily products.
Figure 2‑48: Example verification of the Level-3a 27-Day gridded FRP product. (a) Total active fire pixel count from the C3S 27-Day gridded FRP product covering 3 to 29 August 2023; (b) Total active fire pixel from the 27 daily C3S daily products covering the same period. The active fire pixel count is identical in all the grid cells between the data shown in (a) and that in (b).
The C3S Level 3 Monthly Summary AF & FRP Night-time Product was compared to the same period MODIS MOD14 products. As the spatial patterns of absolute AF pixel counts and FRP have been analysed already in section 2.6.1, we focus here on the analysis of fire season and the metrics defined at the global and GFED regional scale. Degrees of agreement between the C3S and MODIS products are quantified using the coefficient of determination (r2) and slope of the linear best fit.
Figure 2‑49 and Figure 2‑50 (image has been divided into two distinct parts to enhance user accessibility and clarity) show the seasonal pattern of cumulative monthly FRP at both the global and GFED region scale, from both the 2023-24 C3S FRP products and those of Terra MODIS. As with the C3S daily product (Figure 2‑46 and Figure 2‑47), the seasonal pattern seen in the C3S products appears very close to that from MODIS, with the r2 of the cumulative FRP from S3A and Terra data being 1.00 and the slope of the OLS linear best fit, 1.07, and for S3B and Terra 1.00 and 1.06 respectively. For the GFED regions, the coefficient of determination ( r2 ) between the S3A and MODIS values range from 0.78 to 1.00, and the OLS linear best fit slopes 0.62 to 1.13, and for S3B and MODIS 0.82 to 1.00 and 0.68 to 1.13 respectively.
Figure 2‑49: Seasonal pattern of cumulative monthly FRP (%) (March 23 – February 24), both at the global and GFED region scales, as derived from the C3S Level 3 Monthly Summary FRP Product and Terra MODIS. The abbreviation of the names is as follows: Boreal North America (BONA); Temperate North America (TENA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS).
Figure 2‑50: Seasonal pattern of cumulative monthly FRP (%) at the global and GFED region scales, as derived from the C3S Level 3 Monthly Summary FRP Product and Terra MODIS (March 23 – Feb 24). The abbreviation of the names is as follows:Central America (CEAM); NH South America (NHSA); NH Africa (NHAF); SH Africa (SHAF); Equatorial Asia (EQAS); Australia and New Zealand (AUST).
Figure 2‑51shows the fire season start, peak, end and duration as derived from the C3S Level 3 Monthly Summary AF & FRP Night-time Product and from Terra MODIS, both globally and for the GFED regions. Our prior analysis shows that for the GFED regions the fire season duration and magnitude are more accurately captured by the 25% and 75% percentiles of the annual cumulative FRP curve, than by the 10% and 90% percentiles used for the global analysis. This is because the lowest and highest percentiles reflect the onset and end of the fire activity more closely, while the intermediate percentiles may include some periods of low or no fire occurrence. Overall, the start, peak, end and duration of the fire season agree very well between S3A, S3B and MODIS. At the global scale, according to these satellite data, the fire season begins and reaches its peak at the same time globally. However, there are some discrepancies between SLSTR and MODIS in terms of the peak of the fire season in the GFED regions. Around 20% of the 14 regions have 100% agreement for all the four fire season metrics analysed, and around 80% have a difference of one month. Almost all regions have a difference of less than two months.
Figure 2‑51: Fire season metrics as determined from Level 3 Monthly Gridded AF & FRP Night-time Product of S3A and S3B and from daily Terra MODIS data (March 23 to February 24). The temporal resolution of the data is one month. Analysis is conducted globally and for the GFED regions. (a) Start of the fire season; (b) Peak of the fire season; (c) End of the fire season; (d) Duration of the fire season. The abbreviation of the names is as follows: Boreal North America (BONA); Temperate North America (TENA); Central America (CEAM); NH South America (NHSA); SH South America (SHSA); Europe (EURO); Middle East (MIDE); NH Africa (NHAF); SH Africa (SHAF); Boreal Asia (BOAS); Central Asia (CEAS); SE Asia (SEAS); Equatorial Asia (EQAS); Australia and New Zealand (AUST).
Figure 2‑52 provides a visual comparison between the spatial patterns of gas flare pixel count seen from the 2023 C3S Level-3 Monthly Gas Flare products, which are derived from SLSTR shortwave infrared observations, and the gas flares identified in the SNPP VIIRS NightFire product via thresholding of the effective hotspot temperature and application of a persistence criterion requiring the flare to be present at least once in three consecutive months (Elvidge et al.,2013). These gas flare data are derived from observations made by SLSTR and VIIRS over a one-year period - SLSTR onboard Sentinel-3A and VIIRS on SNPP. The data covering the year long period show a very similar spatial pattern, which demonstrates the strong degree of agreement between them.
Figure 2‑52: Total Gas flare pixel count (left panels) from the C3S SLSTR product, and total gas flare pixel count (right panels) from VIIRS – each within 0.25° grid cells for 2023.
The C3S and VIIRS NightFire detections exhibit substantial similarities in terms of their spatial pattern, but also in the number of grid cells affected by gas flaring. In this comparison, the C3S product detected gas flares in 2,183 cells, and the VIIRS NightFire product in 2,067 grid cells. In co-located detection areas, the C3S SLSTR product registered 1,964 grid cells that had gas flares where VIIRS also recorded them, whilst VIIRS recorded 1,863 cells with gas flares where the C3S SLSTR product also recorded them. In terms of differences, the C3S product has 219 unique gas flare grid cells not captured by VIIRS, whilst the NightFire product identified 204 unique grid cells not captured by C3S. Overall, SLSTR achieved a 90.0% detection confirmation rate with VIIRS, whilst VIIRS achieved a 90.1% detection confirmation rate with C3S. The nearly identical co-location rates of approximately 90% and complementary omission rates of 10% between the two products demonstrate their high degree of consistency.
Historically, the Mid-Infrared (MIR) radiance method has been the predominant approach for calculating FRP in landscape fires and is used in the C3S FRP products and the Level 2 products from which they are derived. However, the method presents limitations when applied to gas flares, primarily due to the fact that they burn at significantly higher temperatures (1600-2200 K) than vegetation fires (650-1300 K). This difference results in a shift of the peak spectral radiant emission wavelength to shorter wavelengths, and in response a modified SWIR Radiance method was developed (Fisher & Wooster, 2018; 2019). Comparative analysis of the two SLSTR SWIR bands used within this approach revealed that whilst both the 1.6 μm and 2.2 μm bands demonstrate suitability for FRP calculations, the 2.2 μm band exhibits superior performance when taken across the entire gas flare temperature range (1600–2200 K), with a maximum error margin of ±6.3% (Fisher & Wooster, 2018; 2019). The C3S Gas Flare FRP retrieval uses the SWIR radiance method of FRP derivation, with the 2.2 μm band SLSTR data.
In contrast to the C3S products, as already stated, the VIIRS Night-Fire product employs a Planck fitting method for FRP retrieval – therein termed the Radiative Heat Flux (RH). The approach utilises spectral radiances measured in multiple spectral bands and fits a scaled Planck function to the observed spectra, thereby estimating effective hotspot temperature and sub-pixel area from which RH is calculated using the Stefan-Boltzmann Law (Elvidge et al.,2013). While this methodology provides the sub-pixel fire characteristics of effective temperature and area, it necessitates sensors with multiple precisely matched spectral bands and may exhibit reduced efficacy in detecting smaller, less radiant flares with insufficient multi-band signals.
The degree of agreement between the SWIR Radiance method of FRP derivation and that derived via the Planck fitting methodology is demonstrated in Figure 2‑53. The analysis reveals temperature-dependent correlation patterns, wherein hotspots exceeding 1600K— such as are characteristic of gas flares—exhibit strong agreement ( yellow points in figure 2-53). Conversely, thermal events below 1600K demonstrate significant methodological divergence ( purple colours in figure 2-53), and the FRP from the single band method would be better derived via the MIR radiance approach. For hotspots above 1600 K the data points predominantly align along a 1:1 line, demonstrating minimal bias (μ = 0.5 MW) and reduced scatter (σ = 1.6 MW).
Figure 2‑53: Comparison of Fire Radiative Power (FRP) estimation using the SWIR Radiance method and the VIIRS NightFire radiative heat (RH) algorithm (Fisher and Wooster, 2018).
Figure 2‑54: Total Gas flare count and FRP from the SLSTR C3S data and from VIIRS stored within 0.25° grid cells for July 2023. (a) Total Gas flare pixel count from SLSTR on board Sentinel-3A; (b) total Gas flare FRP from S3A; (c) Total Gas flare count from VIIRS on board of SNPP and stored in the NightFire product; (d) total Gas flare FRP from from VIIRS on board of SNPP and stored in the NightFire product.
Figure 2‑54 illustrates a comparative analysis of the global gas flare distribution and FRP from VIIRS and SLSTR in July 2023. The spatial distribution patterns of gas flare detections in grid cells from both sensors demonstrate strong agreement. Quantitatively, the SLSTR C3S data exhibit a high co-location rate, with approximately 92% of the grid cells (1,446 out of 1,571 total cells) corresponding to VIIRS NightFire observations. Conversely, VIIRS shows a co-location rate of approximately 84% (1,551 out of 1,850 total grid cells) with SLSTR. This substantial level of agreement between SLSTR and VIIRS provides robust evidence of the capabilities of SLSTR in detecting persistent gas flaring activities.
In terms of total gas flare FRP, the VIIRS NightFire product recorded a total of 575,375 MW, whilst the data from the two SLSTR instruments onboard S3A and S3B was used within the C3S products to derive a total of 253,935 MW and 289,044 MW respectively. The combined SLSTR FRP total from both Sentinel-3A and -3B (542,979 MW) closely aligns with that from VIIRS – being less than 6% different. This indicates very strong and consistent agreement across platforms, particularly considering that the VIIRS and SLSTR sensors view the gas flares at completely difference times of the night when cloud cover may be different.
The C3S Level 3a Daily Gridded Gas Flare Product is simply the accumulation of one day’s total gas flare AF Pixel Count and FRP, so its evaluation will focus on verifying the correctness of its summary as derived from the same day level-2 summary data. The level-3a Daily gridded products were verified thoroughly in this way, with the active fire pixel count, FRP and all other parameters being found to agree with the summary of the same day level-2 data. This indicates the correctness of the C3S daily products. An example is shown in Figure 2‑55, where the active fire pixel count and FRP from June 1st 2023 derived from the daily gridded product perfectly agrees with the accumulation of same day level-2 products.
Figure 2‑55: Daily Gas Flare Distribution and Fire Radiative Power (FRP) from S3A and S3B on 1st June 2023. The analysis is divided into four subplots. (a) Total Gas flare pixel count accumulated from the NCT Level-2 products; (b) total Gas flare FRP accumulated from the NTC Level-2 products ; (c) Total Gas flare pixel count from the C3S daily product; (d) Total Gas flare FRP from the C3S daily product.
The C3S Level 3a 27-Day Gridded Gas Flare Product is simply the accumulation of 27 C3S Level 3a Daily Gridded Gas Flare Products, so its evaluation will focus on verifying the correctness of its lower temporal resolution statistical summary as derived from the former data. The 27-Day products were verified thoroughly in this way, with the active fire pixel count, FRP and all other parameters being found to agree with the accumulation of the daily product data. This indicates the correctness of the C3S 27-Day products. An example is shown in Figure 2‑56, where the active fire pixel count and FRP from 3 to 29 August 2023 derived from the 27-Day product perfectly agrees with that derived from the accumulation of the 27 individual daily products.
Figure 2‑56: 27-Day Global Gas Flare Distribution and Fire Radiative Power (FRP) Using S3B SLSTR Data from 3-29 August 2023, derived from S3B SLSTR satellite data. The analysis is divided into four subplots. (a) Total Gas flare pixel count from SLSTR on board of Sentinel-3B accumulated from daily products; (b) total Gas flare FRP from S3B SLSTR accumulated from daily products; (c) Total Gas flare pixel count from 27-Day product; (d) Total Gas flare FRP from 27-Day product.
The C3S Level 3 Monthly Gridded Gas Flare Product is compared to the VIIRS NightFire product. Both the spatial patterns of absolute gas flare counts and FRP comparison are reported below.
Figure 2‑57: Total gas flare pixel count and FRP from the C3S Monthly Gridded product and the VIIRS NightFire product within 0.25° grid cells in July 2023. (a) total Gas flare count from SLSTR onboard Sentinel-3A; (b) total gas flare FRP from S3A SLSTR; (c) total gas flare count from the Sentinel-3B product; (d) total gas flare FRP from the S3B product; (e) total gas Flare count from the VIIRS NightFire product; (f) total gas flare FRP from the VIIRS NightFire product.
Similar to Figure 2‑54, Figure 2‑57 illustrates a comparative analysis of the global gas flare distribution from VIIRS NightFire and both SLSTR on board Sentinel-3A and -3B in July 2023, as derived from the C3S Level-3 Monthly Gridded product rather than Level-2 Monthly Summary product. As expected, the total gas flare pixel count and FRP from the Level-3 product are the same as for the Level-2 Monthly Summary product, and the spatial distribution patterns of gas flare detections from both sensors demonstrate once again strong agreement. Quantitatively, SLSTR exhibited a high co-location rate, with approximately 92% of S3A C3S product detections (1,446 out of 1,571 total grid cells) corresponding to VIIRS NightFire detections. Conversely, VIIRS NightFire showed a co-location rate of approximately 84% (1,551 out of 1,850 total grid cells) with SLSTR detections. For the Sentinel-3B C3S product, approximately 91% of S3B detections (1,427 out of 1,571 total grid cells) corresponded to VIIRS NightFire detections. Conversely, VIIRS NightFire showed a co-location rate of approximately 85% (1,567 out of 1,850 total grid cells) with SLSTR. The substantial level of agreement between these two independent satellite products provides robust evidence of their respective capabilities in consistently detecting persistent gas flaring activities. These data suggest the reliability and complementarity of both sensors and products for monitoring global gas flaring activities, despite their different orbital characteristics, instrument specifications, and product algorithms.
Regarding gas flare FRP, the VIIRS NightFire recorded a total of 575,375 MW. The Monthly C3S products from S3A and S3B 253,935 MW and 289,046 MW respectively. The combined Sentinel-3A and -3B value (542,979 MW) is very close to that of the VIIRS data, indicating very consistent measurements across platforms.
Figure 2‑58: Total Gas flare pixel count and FRP from SLSTR and VIIRS within 0.25° grid cells in December 2023. (a) total Gas flare count from SLSTR onboard Sentinel-3A and stored in the C3S Monthly Summary product; (b) total gas flare FRP from S3A SLSTR and stored in the C3S Monthly Summary product; (c) total gas flare pixel count from SLSTR onboard Sentinel-3B and stored in the C3S Monthly Summary product; (d) total gas flare FRP from S3B SLSTR and stored in the C3S Monthly Summary product; (e) total Gas flare count from VIIRS onboard SNPP and stored in the VIIRS NightFire Product; (f) total gas flare FRP from SNPP VIIRS and stored in the VIIRS NightFire Product.
While Figure 2‑57 highlights a strong agreement between the VIIRS and SLSTR products, Figure 2‑58 reveals some small differences in their gas flare counts and FRP patterns. This comparison revealed interesting patterns in gas flare detection capabilities between the VIIRS and Sentinel-3 SLSTR instruments. Out of 1,500 VIIRS NightFire gas flare grid cells, 59% (885 cells) were co-located with the S3A SLSTR Monthly Summary product dataset, while 41% (615 cells) were detected by VIIRS NightFire alone. The co-location rate of the S3A C3S product with VIIRS is remarkably high at approximately 97% (830 cells), with only about 3% (27 cells) being unique detections. 60% (893 cells) of VIIRS NightFire detections were co-located with the S3B C3S product, while 40% (607 cells) were detected by VIIRS alone. For the Sentinel-3B SLSTR C3S Monthly Summary product, the co-location rate is approximately 95% (801 cells) with the VIIRS NightFire, with only about 5% (40 cells) being unique detections. This high co-location rate suggests that the SLSTR's products gas flare detections are consistent with VIIRS observations, indicating high reliability. Overall, in terms of FRP, the VIIRS product recorded a cumulative FRP of 689,492 MW, whereas the combined measurements from the SLSTR instruments on S3A and S3B and stored in the C3S Monthly Summary products totalled 380,479 MW (with 177,464 MW from S3A and 203,015 MW from S3B). The combined C3S products accounted for 55% of total FRP included in the VIIRS product, primarily because ~ 40% of reported VIIRS hotspot detections in the NightFire record that are classified as gas flares based on their effective emitting temperature (> 1600 K) remained undetected in the NTC Level 2 SLSTR product dataset. Most of these unique VIIRS gas flares are located in northern regions, Canada and Russia, where the total gridded gas flare pixel count and FRP are relatively small overall compared to the Middle East. Whilst the latter region is largely cloud free, northern regions have high cloud cover and this may also change over the period of the night and affect the detection of flares by C3S and NightFire since both use data collected at different times. Also, as the data suggests VIIRS has a more sensitive gas flare detection capability, this may well be due to its higher spatial resolution (375 m compared to 500 m of SLSTR; and thus a pixel area of around 50%) and the different detection algorithms. The SLSTR Level-2 night-time detection algorithm on which the C3S products are based has both an absolute detection test in the SWIR and a contextual detection filter threshold typically more stringent than that from VIIRS NightFire. This may restrict its ability to detect smaller gas flares, and further investigation is needed at the level of the Level 1b and Level 2 datasets to study and possibly further improve the SLSTR detection performance
This document has been produced in the context of the Copernicus Climate Change Service (C3S). The activities leading to these results have been contracted by the European Centre for Medium-Range Weather Forecasts, operator of C3S on behalf of the European Union (Delegation Agreement signed on 11/11/2014 and Contribution Agreement signed on 22/07/2021). All information in this document is provided "as is" and no guarantee or warranty is given that the information is fit for any particular purpose. The users thereof use the information at their sole risk and liability. For the avoidance of all doubt , the European Commission and the European Centre for Medium - Range Weather Forecasts have no liability in respect of this document, which is merely representing the author's view. |