A fire model with distinct crop, pasture, and non-agricultural burning: use of new data and a model-fitting algorithm for FINAL.1

Abstract. This study describes and evaluates the Fire Including Natural & Agricultural Lands model (FINAL) which, for the first time, explicitly simulates cropland and pasture management fires separately from non-agricultural fires. The non-agricultural fire module uses empirical relationships to simulate burned area in a quasi-mechanistic framework, similar to past fire modeling efforts, but with a novel optimization method that improves the fidelity of simulated fire patterns to new observational estimates of non-agricultural burning. The agricultural fire components are forced with estimates of cropland and pasture fire seasonality and frequency derived from observational land cover and satellite fire datasets. FINAL accurately simulates the amount, distribution, and seasonal timing of burned cropland and pasture over 2001–2009 (global totals: 0.434×10⁶ and 2.02×10⁶ km² yr⁻¹ modeled, 0.454×10⁶ and 2.04×10⁶ km² yr⁻¹ observed), but carbon emissions for cropland and pasture fire are overestimated (global totals: 0.295 and 0.706 PgC yr⁻¹ modeled, 0.194 and 0.538 PgC yr⁻¹ observed). The non-agricultural fire module underestimates global burned area (1.91×10⁶ km² yr⁻¹ modeled, 2.44×10⁶ km² yr⁻¹ observed) and carbon emissions (1.14 PgC yr⁻¹ modeled, 1.84 PgC yr⁻¹ observed). The spatial pattern of total burned area and carbon emissions is generally well reproduced across much of sub-Saharan Africa, Brazil, Central Asia, and Australia, whereas the boreal zone sees underestimates. FINAL represents an important step in the development of global fire models, and offers a strategy for fire models to consider human-driven fire regimes on cultivated lands. At the regional scale, simulations would benefit from refinements in the parameterizations and improved optimization datasets. We include an in-depth discussion of the lessons learned from using the Levenberg–Marquardt algorithm in an interactive optimization for a dynamic global vegetation model.