2.7 Product Verification Results: March 2024 to Feb 2025 C3S AF and FRP Daytime & Night-time Products and C3S Gas Flare Products
With the impending retirement of the Terra satellite in late 2025 or early 2026, MODIS-based assessments are no longer feasible. As a result, the evaluation of C3S daytime and night-time AF & FRP products for March 2024 to February 2025 is now conducted through continuous verification between the C3S baseline dataset and the scientific prototype. September 2024 was selected as a representative month due to the high incidence of global fire activity during this period. Verification results from 2024–2025 demonstrate approximately 98% agreement between the two data versions. Given the persistent nature of gas flares, validation of the corresponding products relies on data from previous years. A comparison of the annual gas flare datasets for 2024 and 2023 reveals approximately 90% agreement. These consistently high levels of concordance confirm that all twelve C3S products perform in accordance with established expectations.
2.7.1 Daytime Product Verification Results
Figure 2-59 shows the analysis of daytime active fire detection performance revealed very small differences between the C3S baseline and experimental prototype algorithms across daily fire count and Fire Radiative Power quantification metrics, given the different languages and computer architecture used.
During daytime processing conditions, the C3S baseline algorithm detected 130,482 individual active fire pixels with a cumulative FRP magnitude of 3,002,623 MW across the September 2024 analysis period. In comparison, the original science prototype) code identified 131,448 fire pixels with a total FRP of 3,038,933 MW under identical processing conditions and temporal coverage. These results demonstrate a systematic 0.7% increase in daily fire count detection by the prototype compared to the operational C3S baseline. The FRP quantification showed a corresponding 1.2% increase from the prototype compared to the baseline approach. The consistent directionality of these differences across both detection count and energy quantification metrics indicates genuine algorithmic agreement.
Figure 2‑59: Verification results for the daytime (a) daily fire count and (b) daily FRP between C3S baseline and the prototype for September 2024.
Comprehensive verification of the daytime level-3 gridded products demonstrated exceptional consistency across multiple temporal aggregation scales. At the daily level, systematic algorithmic differences between the C3S baseline and science prototype algorithms ranged from 0.7% to 1.9%, with the experimental prototype consistently detecting more fire activity, while preserving excellent spatial correlation and temporal coherence. Verification of the 27-day composite product revealed near-perfect agreement (<1% difference) between the aggregated 27-day product and the corresponding sum of daily products over the September 11–October 7, 2024 analysis period. Similarly, monthly product verification achieved near-perfect agreement (<1% difference) for both fire pixel counts and FRP quantification in September 2024.
2.7.2 Night-time Product Verification Results
Night-time fire detection analysis revealed similar results to daytime between the operational C3S baseline and experimental prototype. During night-time processing, the C3S baseline algorithm detected 436,242 fire pixels with a total FRP of 2,942,775 MW, while the experimental prototype identified 444,771 fire pixels totalling 3,010,214 MW. These results represent a 1.9% increase in fire count detection and 2.3% increase in FRP quantification for the experimental algorithm during night-time conditions. Verification of the night-time Level-3 gridded products demonstrated similar consistency across multiple temporal aggregation scales as daytime. At the daily level, systematic algorithmic differences between the BC baseline and EI prototype algorithms ranged from 0.5% to 2.4%, with the experimental prototype consistently detecting higher fire activity while preserving excellent spatial correlation and temporal coherence. Verification of the 27-day composite product revealed near-perfect agreement (<2% difference) between the aggregated 27-day product and the corresponding sum of daily products over the September 11–October 7, 2024 analysis period. Similarly, monthly product verification achieved near-perfect agreement (<2% difference) for both fire pixel counts and FRP quantification in September 2024.
Figure 2-60 presents the verification results for the night-time Level-3 monthly gridded product for September 2024. The analysis reveals that both the active fire detections and the Fire Radiative Power (FRP) exhibit highly consistent global spatial distributions. This strong spatial agreement underscores the reliability of the C3S baseline product in correctly capturing fire activity patterns at a global scale during night-time observations.
Figure 2‑60: Verification results for the night‑time Level‑3 monthly gridded product, September 2024. (a) baseline fire count; (b) prototype fire count; (c) baseline FRP; (d) prototype FRP. Active fire detections and Fire Radiative Power (FRP) display highly consistent global spatial distributions, demonstrating strong spatial agreement and confirming the reliability of the C3S baseline product in capturing night‑time fire activity patterns at the global scale.
2.7.3 Gas Flare Interannual Comparison
Given the persistent nature of gas flares, validation of the corresponding products relies on data from previous years. To assess the quality of the 2024 gas flare product, a comparison was conducted between the S3B annual gas flare datasets for 2024 and 2023. This comprehensive year-to-year comparative analysis of S3B SLSTR gas flare detections demonstrates exceptional temporal stability and consistency in the monitoring capabilities of gas flare infrastructure. The analysis encompassed 402,345 gas flare detections from 2023 and 418,269 detections from 2024, representing a 3.9% increase in total gas flare pixel detection count (+15,924 detections) across the annual comparison period. The co-location analysis employed a sophisticated 3×3 pixel spatial window methodology to account for minor geolocation uncertainties and atmospheric effects, achieving exceptional agreement rates that demonstrate the inherent stability of gas flare infrastructure over annual timescales (Figure 2-61). The 2023 dataset exhibited a 90.0% co-location rate (1,965 out of 2,183 total cells), while the 2024 dataset achieved an even higher 91.5% co-location rate (1,986 out of 2,170 total cells). Cell persistence analysis revealed that 90.0% of gas flare cells detected in 2023 remained active and detectable in 2024, with only 218 cells exclusive to 2023 and 184 cells exclusive to 2024, representing operational changes rather than algorithmic inconsistencies. This high persistence rate confirms the methodology's effectiveness for long-term infrastructure monitoring applications and validates the algorithm's temporal consistency across annual periods, providing critical evidence for the stability of satellite-based gas flare monitoring capabilities in static infrastructure applications where detection consistency over multi-year periods is essential for carbon emission quantification and environmental monitoring studies.
Figure 2‑61: Annual gas flare comparison between (a) 2023 and (b) 2024 from Sentinel-3B SLSTR
