Interannual variability in global biomass burning emissions from 1997 to 2004

In nineteen ninety-eight, fires burning across the planet released more than three billion tonnes of carbon into the atmosphere. That's more than the entire European Union emits from burning fossil fuels in a year. Until van der Werf and colleagues published this work, no one had a reliable way to measure how much fire emissions fluctuate from year to year or why. The problem is harder than it sounds. To calculate fire emissions, you need three things at once: how much area burned, how much fuel was available to burn, and what fraction of that fuel actually combusted. Miss any one of these and your estimate falls apart. Datasets on burned area had only recently become available through satellite efforts like GBA2000 and GLOBSCAR, and even those were poorly constrained in deforestation regions where cloud cover and mechanized clearing obscure detection. Fuel loads were typically set as fixed biome averages, which misses the entire point — fuel availability changes with climate. Combustion completeness, the fraction of available fuel actually consumed, varies enormously depending on how dry the season is and what is burning. Meanwhile, atmospheric scientists using top-down inversion methods — working backward from measured carbon monoxide concentrations in the atmosphere — were estimating global fire emissions around three point four petagrams of carbon per year. Bottom-up approaches, estimating from the ground up, were coming in between one point three and two petagrams for the same period. That's not a small discrepancy. Something was missing. To close that gap, van der Werf and colleagues built a framework that combined new satellite observations with a dynamic biogeochemical model. For burned area, they used the Moderate Resolution Imaging Spectroradiometer, or MODIS, satellite's five hundred meter resolution burned-area product covering fifty-two globally distributed tiles from two thousand one through two thousand four. To push the record back to nineteen ninety-seven, they calibrated fire counts from two earlier sensors — VIRS on the TRMM platform and ATSR — against the MODIS maps, deriving burned-area-per-count ratios from the overlap period and applying them backward in time. Those burned-area maps were fed into the CASA biogeochemical model. CASA doesn't treat fuel loads as fixed — it computes them dynamically. The core idea is that net primary production, the carbon plants actually fix from the atmosphere, equals incoming photosynthetically active radiation multiplied by fAPAR — the fraction of that radiation that plants actually absorb and use — multiplied by a light-use efficiency that gets scaled down when temperature or moisture conditions are poor. fAPAR was derived from satellite vegetation indices, and precipitation, temperature, and solar radiation were updated monthly, so fuel buildup and loss tracked actual climate conditions year by year. The novel addition in this study was peat and organic soil burning. Van der Werf and colleagues adjusted CASA's decomposition so modeled soil carbon matched observed values to thirty centimeters, and they allowed deeper carbon pools to become fuel in wetland grid cells. In tropical peat areas, fires were assumed to consume at least fifty percent of the soil carbon pool, with moisture controlling how much of the rest could burn. Combustion completeness for peat was set between zero point nine and one point zero, reflecting the efficient smoldering combustion of deep organic material. This wasn't a cosmetic addition. It turned out to be central to the entire story. Zoom out to the eight-year average and the global budget looks like this. Plants fixed roughly fifty-eight petagrams of carbon per year as net primary production. About ninety-five percent of that returned to the atmosphere through heterotrophic respiration — microbes and decomposers breaking down organic matter. Fire accounted for roughly four percent or about two point five petagrams of carbon per year. That's fire's share of the global carbon cycle on average. But the average conceals everything interesting. Burned area is overwhelmingly a savanna story. Africa, Australia, and South American grasslands together accounted for roughly eighty percent of global burned area. But when you look at carbon emissions, the picture flips. Africa contributed forty-nine percent of emissions, but that's partly because African savannas are so vast. The real surprise is what drives the spikes. Emissions per unit area in forest grid cells averaged two point twenty-two kilograms of carbon per square meter compared to just zero point fifty-two in herbaceous savanna grid cells. Forests burn less often, but when they burn, they release more than four times as much carbon per area. Peatlands are in a category of their own. That asymmetry is the conceptual core of this paper, and it explains everything that follows. Burned area and carbon emissions are largely decoupled. A year with massive savanna fires can look enormous by area and modest by carbon. A year with intense tropical forest and peat burning can look unremarkable by area and catastrophic by carbon. Which brings us to nineteen ninety-eight. Global fire emissions that year hit three point two petagrams of carbon compared to two point zero petagrams in two thousand — a swing of more than one petagram in two years. The nineteen ninety-seven to nineteen ninety-eight El Niño produced severe drought across tropical Asia, particularly in Indonesia, where vast peat deposits that had been drained for agriculture became vulnerable to ignition. Equatorial Asia has the highest modeled fuel loads in the dataset, on the order of ten kilograms of carbon per square meter, driven by aboveground biomass plus peat. When those peatlands burned, they didn't just release surface carbon — peat can smolder meters deep, consuming ancient organic material that took centuries to accumulate. Van der Werf and colleagues found that global fire emissions were negatively correlated with the Southern Oscillation Index at a correlation coefficient of negative zero point five three, a clear statistical fingerprint of ENSO's influence. Nineteen ninety-eight wasn't a single-region anomaly. Almost every major biomass burning region showed elevated emissions that year — boreal North America, boreal Asia, Central and South America, and equatorial Asia all contributed. That breadth is what pushed nineteen ninety-eight so far above the mean. Equatorial Asia showed the most dramatic interannual variability of any region — a standard deviation one point three times its own annual mean. Africa, by contrast, had a standard deviation just zero point one times its mean. Savannas burn reliably, year after year, driven by seasonal rainfall patterns that don't shift much with ENSO. Forests and peatlands are the wild card. The validation picture is meaningful but imperfect. The seasonality in the Global Fire Emissions Database that van der Werf and colleagues produced — now called GFED version two — matched atmospheric carbon monoxide and aerosol observations more closely than their earlier inventory had. That seasonal agreement is a useful check: the timing of when emissions peak and trough tracks what the atmosphere actually recorded. At the same time, the updated emissions from this version, which now included peat and organic soil burning, were higher than GFED version one, particularly in boreal regions and equatorial Asia. Adding peat moved the numbers substantially. That alone shows how sensitive global totals are to whether peat combustion is in the model or not. Remaining uncertainties are concentrated in the hardest-to-observe places: deforestation fronts where cloud cover limits burned-area detection, and peatlands where burn depth is poorly constrained. Van der Werf and colleagues recommend continued comparison with top-down atmospheric inversions — including multi-species inversions to reduce errors from uncertain emission factors — and finer-resolution bottom-up modeling in heterogeneous regions. They also point to improved satellite burned-area products and better fuel-load calibration as the clearest paths to tighter estimates. The practical implication is this: fire emissions can swing by roughly one billion tonnes of carbon per year depending on climate conditions and where fires occur. That's not background noise in the global carbon budget — it's a climate-sensitive signal large enough to shift the atmospheric carbon dioxide growth rate in years when El Niño aligns with vulnerable peatlands and dense tropical forests. Any honest accounting of Earth's carbon cycle has to treat fire as a dynamic variable, not a fixed term. Van der Werf and colleagues gave us the first rigorous framework for doing exactly that — and in doing so, showed just how much we had been missing. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.