ODEM v1.0: an offline dust emission model for reanalysis-driven source estimation

Abstract. Global dust emission estimates remain uncertain by a factor of two to three across models, with uncertainty arising from both the emission physics and the choice of meteorological forcing. Disentangling these two sources of uncertainty requires running the same emission scheme with different meteorological inputs, which is not possible in online models where the emission physics and the forcing are coupled within a single atmospheric model.

This paper presents ODEM v1.0 (Offline Dust Emission Model), a standalone Python implementation of the brittle fragmentation dust emission parameterization of Kok et al. (2014), following the implementation of Leung et al. (2023). ODEM accepts either ERA5 or MERRA-2 as meteorological forcing and produces gridded dust emission flux fields at the native spatiotemporal resolution of each reanalysis (0.25°, 1-hourly for ERA5; 0.5° × 0.625°, 1-hourly for MERRA-2). The model applies process-based emission physics – including soil particle size, moisture inhibition, aerodynamic drag partition, and turbulent intermittency corrections – to every land grid cell independently in a single forward pass with no spin-up requirement.

For the year 2006, ODEM driven by ERA5 at 1-hourly resolution produces a global PM20 emission of 15 539 Tg yr⁻¹ and ODEM driven by MERRA-2 produces 12 747 Tg yr⁻¹. These values exceed the observationally constrained PM20 budget of 5000±1600Tg yr⁻¹ by factors of 3.1 and 2.5, respectively. This overshoot is a known property of the Kok et al. (2014) emission equation at its default calibration: Leung et al. (2023) report a comparable factor of 2.3 for their unnormalised scheme using the same equation, indicating that current emission physics cannot constrain the absolute magnitude from first principles. ERA5 produces 22 % more emission than MERRA-2, consistent with known differences in reanalysis friction velocity fields, particularly over North Africa. A sensitivity experiment using ERA5 at 3-hourly resolution yields 15 521 Tg yr⁻¹ (0.1 % lower), confirming that the turbulent intermittency correction effectively accounts for sub-timestep wind variability. Both experiments produce roughly an order of magnitude higher emission than the MERRA-2 online GOCART scheme (1564 Tg yr⁻¹), consistent with the known underestimation by empirical schemes relative to observational constraints.