Development and optimization of a wildfire plume rise model based on remote sensing data inputs – Part 2

Abstract. Biomass burning is one of a relatively few natural processes that can inject globally significant quantities of gases and aerosols into the atmosphere at altitudes well above the planetary boundary layer, in some cases at heights in excess of 10 km. The "injection height" of biomass burning emissions is therefore an important parameter to understand when considering the characteristics of the smoke plumes emanating from landscape scale fires, and in particular when attempting to model their atmospheric transport. Here we further extend the formulations used within a popular 1D plume rise model, widely used for the estimation of landscape scale fire smoke plume injection height, and develop and optimise the model both so that it can run with an increased set of remotely sensed observations. The model is well suited for application in atmospheric Chemistry Transport Models (CTMs) aimed at understanding smoke plume downstream impacts, and whilst a number of wildfire emission inventories are available for use in such CTMs, few include information on plume injection height. Since CTM resolutions are typically too spatially coarse to capture the vertical transport induced by the heat released from landscape scale fires, approaches to estimate the emissions injection height are typically based on parametrizations. Our extensions of the existing 1D plume rise model takes into account the impact of atmospheric stability and latent heat on the plume up-draft, driving it with new information on active fire area and fire radiative power (FRP) retrieved from MODIS satellite Earth Observation (EO) data, alongside ECMWF atmospheric profile information. We extend the model by adding an equation for mass conservation and a new entrainment scheme, and optimise the values of the newly added parameters based on comparison to injection heights derived from smoke plume height retrievals made using the MISR EO sensor. Our parameter optimisation procedure is based on a twofold approach using sequentially a Simulating Annealing algorithm and a Markov chain Monte Carlo uncertainty test, and to try to ensure the appropriate convergence on suitable parameter values we use a training dataset consisting of only fires where a number of specific quality criteria are met, including local ambient wind shear limits derived from the ECMWF and MISR data, and "steady state" plumes and fires showing only relatively small changes between consecutive MODIS observations. Using our optimised plume rise model (PRMv2) with information from all MODIS-detected active fires detected in 2003 over North America, with outputs gridded to a 0.1° horizontal and 500 m vertical resolution mesh, we are able to derive wildfire injection height distributions whose maxima extend to the type of higher altitudes seen in actual observation-based wildfire plume datasets than are those derived either via the original plume model or any other parametrization tested herein. We also find our model to be the only one tested that more correctly simulates the very high plume (6 to 8 km a.s.l.), created by a large fire in Alberta (Canada) on the 17 August 2003, though even our approach does not reach the stratosphere as the real plume is expected to have done. Our results lead us to believe that our PRMv2 approach to modelling the injection height of wildfire plumes is a strong candidate for inclusion into CTMs aiming to represent this process, but we note that significant advances in the spatio-temporal resolutions of the data required to feed the model will also very likely bring key improvements in our ability to more accurately represent such phenomena, and that there remain challenges to the detailed validation of such simulations due to the relative sparseness of plume height observations and their currently rather limited temporal coverage which are not necessarily well matched to when fires are most active (MISR being confined to morning observations for example).