The study, published in Atmospheric Chemistry and Physics, introduced a new inverse modelling framework called CarbonTracker Europe Long-Window/Short-Window, or CTE-LW/SW. The approach works in two stages. First, a long-window inversion resolves monthly to inter-annual variability in the global carbon monoxide budget using flask measurements from the NOAA Global Monitoring Laboratory network, optimising background sources such as methane and non-methane volatile organic compound oxidation, anthropogenic emissions, and chemical loss by the hydroxyl radical. Then, a short-window inversion hones in on the fast-moving signal of fire emissions, adjusting three-daily scaling factors for more than a thousand one-degree grid cells across South America using satellite observations of column-averaged carbon monoxide, denoted X(CO). This decoupling is technically elegant: each stage uses observations best suited to the spatio-temporal scales of the processes it targets, and the transport model TM5-MP, driven by ERA5 meteorology at one-degree resolution, acts as the observation operator linking surface fluxes to atmospheric columns.

Two satellite instruments supplied the crucial observations. The TROPOspheric Monitoring Instrument, TROPOMI, flies aboard ESA’s Copernicus Sentinel-5 Precursor satellite and covers Amazonia daily with footprints of roughly 5.5 by 7 kilometres, retrieving carbon monoxide in the shortwave infrared with near-uniform sensitivity through the troposphere. The older MOPITT instrument on NASA’s Terra satellite, operational since 1999, offers a thermal-infrared retrieval peaking in sensitivity in the mid-troposphere. Because these instruments differ in vertical sensitivity, overpass time, and retrieval physics, their agreement provides a powerful internal check. Remarkably, the inversions converged to nearly identical answers regardless of which satellite was assimilated and regardless of which fire emission inventory, GFED5.1 or GFAS v1.2, served as the starting point. Posterior totals ranged only from 46 to 48 teragrams of carbon monoxide, with a structural uncertainty of about 15 teragrams reflecting gaps in knowledge of carbon monoxide production and loss chemistry.

That convergence matters because the two prior inventories disagreed wildly. GFED5.1, the newest version of the Global Fire Emissions Database, estimated 41 teragrams of carbon monoxide for the Brazilian fire season, while GFAS v1.2, the Global Fire Assimilation System operated by ECMWF, put the figure at just 25 teragrams, a difference of roughly a factor of 1.5 in total carbon terms and up to a factor of two at the biome level. The satellite-constrained posterior landed within 20 percent of GFED5.1 but required an 81 percent upward revision of GFAS v1.2, well above the 50 percent average adjustment that earlier work had found for the 2003 to 2018 fire seasons. The close agreement with GFED5.1 is encouraging news for fire emission modellers, suggesting that recent improvements, particularly the inclusion of small-fire burned area data derived from Sentinel-2 and Landsat, which raised Brazilian burned area estimates by 61 percent for 2019, are capturing real fire activity that earlier products missed.

The temporal evolution of the fire season tells a story of shifting fire regimes. In mid-August, the season opened abnormally early with deforestation and forest fires in the Amazon biome, dominated by smouldering combustion of coarse woody fuels, a phase in which the two priors disagreed most sharply. By mid-September, fire activity pivoted decisively toward the Cerrado, Brazil’s vast savanna, where fast-burning flaming fires with low carbon monoxide emission factors took over. The late season, from October into November, brought a heterogeneous mix: savanna fires persisted in both the Amazon and Cerrado, and new activity emerged in the Pantanal wetlands, which require prolonged drying before they can sustain burning. Savanna fires, which accounted for 45 percent of Brazil’s carbon monoxide emissions in the priors during the final months, rose to roughly 65 percent in the posterior, underscoring how much of the season’s carbon release came not from headline-grabbing rainforest clearing but from recurring savanna burning.

Here lies the study’s most provocative finding. At the biome level, the inversions revealed that emissions from the Cerrado and the Caatinga, the semi-arid shrubland of northeastern Brazil, were systematically about twice as large as either inventory suggested, a gap of 7 to 8 teragrams of carbon monoxide. This was unexpected, because savanna fire dynamics are generally considered well constrained, with less cloud cover, lower canopy density, and abundant ground measurements. The researchers systematically ruled out alternative explanations. Transport model errors seem unlikely given the consistency between TROPOMI- and MOPITT-based inversions despite their different vertical sensitivities. Long-range transport of African smoke, which can contribute up to a quarter of tropospheric carbon monoxide over eastern Brazil, was well captured by the model over inflow regions. Weak correlations between posterior Cerrado emissions and those of neighbouring biomes indicate the increments are genuine local signals rather than compensating errors. And closing the gap through chemistry would require implausible two- to fourfold increases in local carbon monoxide production from volatile organic compound oxidation, or systematic hydroxyl radical biases that state-of-the-art reanalyses do not support.

That leaves the inventories themselves as the likely culprits, and the authors point to two main suspects: underestimated fuel loads and uncertain emission factors. Emission factor uncertainties for savanna fires are around 20 to 30 percent, but even a generous 20 percent adjustment would close only a fifth of the gap; to close it entirely, savanna emission factors would need to exceed those of tropical forest fires, which is physically unrealistic. More compelling is the fuel load hypothesis. Recent work using GEDI spaceborne lidar has shown that woody debris and litter estimates in the GFED fuel map are low for the Cerrado compared with field measurements, and these fuel pools constitute over 90 percent of the fuel consumed in savanna fires. Surface and below-canopy fuels are notoriously difficult to observe from orbit, so models rely on empirical tuning that may not capture the heterogeneity of a region where nearly half the native vegetation has been converted to other land uses over three decades. Landscape fragmentation disrupts the natural cycle of frequent, low-intensity fires, promoting woody encroachment, fuel accumulation, and ultimately larger, more intense fires with higher emission factors that static, decades-old field measurements cannot represent.

The carbon implications are considerable. Converting carbon monoxide emissions to total carbon using assumed carbon dioxide to carbon monoxide ratios, the posterior estimates imply 270 to 278 teragrams of carbon released by the 2019 Brazilian fires, roughly 100 teragrams more than previous estimates, with 50 to 70 percent of the increase attributable to the Cerrado and Caatinga. Because savanna fires have relatively low carbon monoxide to carbon dioxide ratios, meaning they release proportionally more carbon dioxide per unit of carbon monoxide, the choice of conversion ratio matters enormously; perturbing the ratios by 20 percent widens the carbon estimate to a range of 216 to 334 teragrams. This highlights a fundamental limitation of carbon monoxide-only inversions and motivates future joint inversions of carbon monoxide, carbon dioxide, nitrogen dioxide, methane, and formaldehyde, which could simultaneously constrain emission ratios, partition emissions by fire type and combustion phase, and separate source signals from background.

The broader lesson is that atmospheric inversions driven by satellite column retrievals have matured into an independent, complementary pillar of fire emission monitoring, capable of auditing the bottom-up inventories that feed climate models and carbon budgets. For Brazil, where fire-adapted savannas and fire-sensitive rainforests lie in close proximity and fire activity has intensified in recent years, such capability is especially valuable. The new CTE-LW/SW framework does not resolve individual fire plumes, and its current design cannot allocate emissions where priors contain none, but it robustly constrains sub-continental budgets and cross-validates the satellite records themselves. As climate projections point toward hotter, longer droughts and more frequent fire-conducive weather across Amazonia, even if deforestation rates decline, the ability to independently verify what fires actually release, biome by biome, will be essential for mitigation strategies, policy assessments, and the fragile accounting of one of the planet’s most consequential carbon reservoirs.

Subject of Research: Satellite-based inverse modelling of carbon monoxide to constrain fire carbon emissions during the 2019 Brazilian burning season

Article Title: Fire carbon emission constraints from space-based carbon monoxide retrievals during the 2019 intense burning season in Brazil

Article References: van den Berg, A.-W., Hooghiem, J. J. D., van der Woude, A. M., Rijsdijk, P., Vernooij, R., Botía, S., van der Werf, G. R., Miller, J. B., Luijkx, I. T., Krol, M. C., & Peters, W. (2026). Fire carbon emission constraints from space-based carbon monoxide retrievals during the 2019 intense burning season in Brazil. Atmospheric Chemistry and Physics, 26(19), 13929-13958. https://doi.org/10.5194/acp-26-13929-2026

Image Credits: AI Generated

DOI: 10.5194/acp-26-13929-2026

Keywords: Brazil fires, carbon monoxide, satellite retrievals, TROPOMI, MOPITT, atmospheric inversion, GFED5.1, Cerrado, Amazon, fire emissions, carbon cycle, biomass burning

The stakes of this measurement problem are enormous. Terrestrial ecosystems absorb roughly one quarter of the carbon dioxide released by human activities, a service that significantly slows the rise of atmospheric CO2 and the pace of global warming. NEE, defined as the difference between the CO2 released by ecosystem respiration and the CO2 sequestered by photosynthesis, is the quantity that tells us whether a landscape is a net carbon sink or source. Yet NEE can only be measured directly at the plot scale, using eddy covariance towers that monitor the vertical exchange of CO2 between vegetation and air. Everything beyond the tower footprint must be estimated, and that is where the disagreements begin.

Scientists have traditionally taken two routes to continental-scale carbon budgets. Top-down atmospheric inversions combine measurements of atmospheric CO2 concentrations with transport models to infer surface fluxes, offering an independent perspective but at coarse spatial resolutions limited by sparse sampling. Bottom-up approaches use process-based terrestrial biosphere models or data-driven machine learning to extrapolate tower observations across landscapes. The trouble is that bottom-up estimates have often diverged dramatically from inversions, partly because conventional machine learning treats each time step independently and fails to capture the complex, nonlinear ways ecosystems respond to disturbances and climate extremes. Ecosystems remember: a drought can suppress carbon uptake for months or years after the rains return, and most models simply do not carry that memory.

The LSTM architecture is what sets MemoryFlux apart. Developed originally to solve long-term dependency problems in sequence modelling, the network uses a six-month look-back window of historical climate and vegetation data to predict current carbon exchange, explicitly encoding the legacy effects of antecedent conditions. The team trained separate ecosystem-specific models for ten major plant functional types, from evergreen needleleaf forests to permanent wetlands, using tower-measured meteorology combined with a rich suite of satellite observations. Those predictors included the normalized difference vegetation index, leaf area index, the fraction of absorbed photosynthetically active radiation, and solar-induced chlorophyll fluorescence, a direct signal of photosynthetic activity, alongside ERA5-Land climate reanalysis variables such as temperature, radiation, precipitation, soil water content, and vapour pressure deficit.

The validation results are striking. MemoryFlux’s seasonal cycle of NEE correlated at r=0.96 with the ensemble mean of the OCO-2 v10 Model Intercomparison Project inversion and at r=0.97 with CarbonTracker2022, both top-down estimates derived from fundamentally different methodology. Perhaps more importantly, the mean annual carbon sink of −1.27 ± 0.12 petagrams of carbon per year sits far closer to the inversion range of −0.83 to −0.70 petagrams than existing machine-learning upscaling products, which have suggested sinks as large as −3.30 petagrams per year. That gap between bottom-up and top-down estimates has long been one of the most stubborn discrepancies in regional carbon cycle science, and narrowing it builds confidence in both approaches.

One of the dataset’s most celebrated achievements is its reproduction of the Midwest Corn Belt phenomenon. Atmospheric inversions have consistently shown that the croplands of the Upper Midwest pull in enormous amounts of carbon during the peak growing season in July and August, a continental-scale signal recognized as a model benchmark since the early days of the CarbonTracker project. Yet most existing global flux upscaling datasets fail to capture it, instead placing their strongest seasonal uptake in the southeastern United States. MemoryFlux correctly identifies the Corn Belt as the dominant carbon-uptake region during the peak growing season, matching the empirical evidence and the inversion picture, even though the uptake is transitory because most of the absorbed carbon ends up in harvested crops consumed by humans and livestock.

The spatial patterns across the continent also align with ecological understanding. The strongest annual carbon uptake appears in the humid southeastern United States, followed by the northern Pacific Coast and tropical regions, while major carbon sources concentrate in the tundra of northern and central Canada and the arid central United States, where warming could release soil carbon and accelerate climate change. Forests and wooded ecosystems emerge as the continent’s dominant carbon sink, contributing a mean uptake of −0.76 petagrams of carbon per year, with MemoryFlux attributing more than 85 percent of the total sink to forests, a share higher than most other products suggest.

To test whether the memory mechanism actually matters, the researchers ran a controlled experiment. They built a parallel dataset, dubbed nonMemoryFlux, using the same predictors, architecture, and validation scheme but feeding the model only contemporaneous variables, with no look-back window. The differences were substantial. At the plant functional type level, the coefficient of determination rose from 0.63 to 0.79 when historical information was included, and error metrics dropped accordingly. The memory-free version also estimated stronger carbon uptake in recent years, precisely the period when the 2015/16 El Niño and repeated droughts and heatwaves struck North America, suggesting it failed to capture the lingering legacy of those extremes. The gap between the two versions reached 186 teragrams of carbon in a single year, about 14.6 percent of the mean continental uptake.

MemoryFlux also proved its worth during climate disasters. The researchers examined six major events, including the 2011 drought in Mexico and the southern United States, the 2012 North American drought, the 2013 California drought, the 2017 flash drought across the northern plains, the 2019 Midwest floods, and the southwestern drought and wildfires of 2020 to 2021. In each case, the dataset’s spatial anomaly patterns coincided with negative soil moisture and fluorescence anomalies and with the documented extent of the events, showing anomalous carbon release where ecosystems were stressed. The memory-equipped version represented these regional anomaly patterns more clearly than its memory-free counterpart, particularly for the 2011 and 2012 droughts and the 2019 flood.

The authors are careful about limitations. Flux tower coverage remains uneven, with some ecosystem types represented by only two sites, and the dataset lacks a fully spatially explicit uncertainty framework. A mismatch between tower-scale training data and gridded reanalysis predictors introduces additional uncertainty, and the static land-cover map cannot track real transitions such as cropland expansion or fire-induced vegetation change. Differences with inversions persist in the managed Midwest croplands, hinting that human management information could further improve the models. Still, with the dataset and code openly available, MemoryFlux offers climate scientists a powerful new lens on how a continent’s ecosystems respond to a rapidly changing climate, and a template for bringing memory into the next generation of carbon cycle models worldwide.

Subject of Research: LSTM-based upscaling of net ecosystem CO2 exchange across North America from flux tower and satellite observations

Article Title: Improved estimation of net ecosystem CO2 exchange over North America using LSTM-based flux upscaling (2001–2021)

Article References: Improved estimation of net ecosystem CO2 exchange over North America using LSTM-based flux upscaling (2001–2021). (n.d.). https://doi.org/10.5194/essd-18-7367-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7367-2026

Keywords: carbon cycle, net ecosystem exchange, LSTM, deep learning, MemoryFlux, eddy covariance, atmospheric inversion, drought, Corn Belt, remote sensing, North America, carbon sink