Validation framework for EO measurements of transitory events based on robust statistics retrieved from non-simultaneous observations: A case study applied to Fire Radiative Power (FRP) products.

The current methods to systematically validate Earth Observation (EO) products capturing transitory events such as fire activity rely mostly on the intercomparison between Near-real-time products without clearly identifying one as the reference dataset. In addition, due to the highly dynamic and ephemeral nature of such events, comparisons are restricted to near-simultaneous measurements which significantly limits the sample size of any intercomparison. In this study, we propose a new comparison framework that overcomes these limitations. This novel approach is based on a robust analysis of the frequency density (f-D) distributions of each product’s assessment of the event. We start by defining the concepts associated for distribution fitting and performance, temporal and spatial requirements, comparison metrics, and then provide an overview of the various sources of uncertainty contributing to the intercomparison exercise, and how and what uncertainties are propagated.

In this study we inter-compare eight operational remotely sensed active fire detections and fire radiative power (FRP) retrieval products: the polar-orbiter products derived from active fires detected using the Moderate Resolution Imaging Spectroradiometer data (MCD14ML), the Visible Infrared Imaging Radiometer Suite (VNP14IMGML), and the Sea and Land Surface Temperature Radiometer (SLSTR) Non-time critical product from European Space Agency (SLSTR-NTC), and the geostationary products derived from data collected by Meteosat’s Spinning Enhanced Visible and Infrared Imager (LSA-SAF FRP-PIXEL), and the three available products based on Advanced Baseline Imager (KCL/IPMA-GOES16, KCL/IPMA-GOES17, and KCL/IPMA-Himawari). We focus on annual detections and perform the analysis at 0.5° grid cell resolution, for the overlapping product’s time-series. The results are analysed for their temporal and spatial consistency, and inter-product differences are analysed in the context the product’s metadata.

The results show that an Inverse-gamma distribution can be used to characterize the fire ‘statistical signature’ and provide a reference baseline on to which all FRP products can be compared to, and their ‘representation uncertainty’ assessed. Individually, the fitting results show the degree of under representation of each sensor’s detections, namely the identification of minimum FRP detection limit, which typically precludes the detection of a proportion of the highly numerous but individually relatively small and/or low intensity fires. Furthermore, inter-comparison differences allowed for the identification, and assess the impact, of some of the key non-fire effects such as: pixel size, off-nadir pixel area growth, algorithm limitations, quality information, and the impacts of low temporal resolution of polar-orbiting sensors.

This proposed framework is a useful tool to compare EO-based FRP products and transferable to any product measuring transitory event properties that do not rely on simultaneous observations. It complements existent comparison exercises by identifying additional sources of uncertainty, the conditions under which these occur and how these translate into product inconsistencies. It is an essential tool, providing users with product-specific information on measurement limitations that, in principle, can be corrected and assimilated to higher level products and downstream applications such as GHG emission estimates from biomass burning, providing better quality information used for adaptation and mitigation policies.