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
Landscape scale vegetation fires, on average, burn an area equivalent to that
of India plus Pakistan every year
Unlike most emissions to the atmosphere (apart from aircraft emissions or
volcanic plumes), biomass burning is potentially able to loft and release its
burden of gases and particles at various altitudes and not just at the
surface. Because atmospheric transport is dependent on altitude, releasing
smoke emissions at different heights into the atmosphere has a considerable
influence on their region of impact, and may also alter their chemical
evolution as, for example, the advection resulting from the interaction of
the plume and the atmosphere can modify the ambient conditions within the
developing plume (upon which its evolution in part depends). Examples of
biomass burning plumes reaching the high troposphere, or even the lower
stratosphere, were first shown comprehensively by
As mentioned in the companion review to this work
Since plume dynamics are highly coupled to atmospheric processes, the
development and testing of a InjH parametrization that represents the
vertical transport of the emitted material is a challenging task
The prerequisites required to initialize these three plume rise models are
generally information on fire size (usually the “active fire area” that
denotes the area of active fuel consumption and fire energy emission), the
Convective Heat Flux (CHF), along with the ambient atmospheric conditions
(i.e. stratification, and also relative humidity when microphysical processes
are considered). While the host CTM, or global reanalysis atmospheric model,
can be used to extract the ambient atmospheric profiles at the fire location,
satellite remote sensing data is generally required to characterize the
active fire area (AF-area), which equates to the surface area of a black body
having the same spectral emission properties as does the observed fire at the
measurement wavebands
The Moderate-resolution Imaging Spectroradiometer (MODIS) sensor, operated on
the Aqua and Terra satellites, can provide multispectral observations from
which the FRP and AF-area of the detected fires (and individual fire pixels)
can be derived. Since MODIS and MISR are both available on the Terra
satellite, it is relatively easy to obtain collocated active fire and smoke
plume height information. Indeed, a large number of fire events are available
with such information, for example via the MISR plume height
project https://www-misr.jpl.nasa.gov/getData/accessData/MisrMinxPlumes/
In this study, we enhance the
A detailed description, together with an overview of
recent developments of the
In an attempt to improve the performance of PRMv0, we previously derived
a scheme to estimate both AF-area and CHF from satellite EO data
The processes involved in plume dynamics are dependent
upon ambient atmospheric conditions (e.g. entrainment, wind shear), and
therefore on altitude. In the boundary layer, entrainment and wind shear act
against the buoyancy, while for plumes that make it to the free troposphere,
if entrainment, wind shear and stratification are still a break to the
residual buoyancy, the ambient cooling can generate latent heat via
condensation of the entrained water vapour and therefore potentially
re-accelerate plume rise. Combining all these effects, the detrainment of the
plume in the atmosphere is most certainly happening at all altitude levels
PRMv2 formulation is based on
The en/de-trainment coefficients are inspired from shallow convection parametrizations
The microphysical scheme of the model remains unchanged from PRMv0 of
The PRMv2 model is run on a
According to energy budget measurements conducted on small-scale vegetation fires, convection represents around half of the total energy released during the combustion process
With the above definitions, the PRMv2 model is fully defined with the inputs of (i) AF-area (
The end point of the time integration (
As mentioned in Sect.
To adjust the measured radiances for atmospheric effects, we computed the atmospheric transmittance for each fire and waveband of observation, which is particularly important when combining multispectral information to retrieve fire parameters, such as when using the
We used an iterative solver to provide solutions to the
Equation (
It is important to note that the fire cluster information as derived from the
To improve the convergence of
This section introduces two independent methods to derive FRP, based on the outputs of the
This section describes how the dataset containing fire characteristics (AF-area and FRP), ambient atmospheric condition, and observed plume top height is computed. It is based on the same fire and plume observations made by MODIS and MISR contained in the official MISR plume injection height project
As in
We used the MINX-derived smoke plume contour reference point for the plume location, and applied our methodology to every MODIS active fire cluster found within a radius of
As mentioned previously, a good proxy for InjH is the final altitude of the smoke plume. Previous studies using products from the MINX-MISR plume height project
The top of the stack (
Figure
Together with
We use ECMWF analysis runs to define, for each fire cluster, ambient atmospheric profiles of pressure, temperature, humidity and wind, re-sampled to a
Based on the different definitions of FRP, AF-area and plume height mentioned previously, the “raw fire cluster” MISR dataset, which is based on the original MINX-derived plume height the “filtered fire cluster” MISR dataset, which is based on the previous data set with full application of the methodologies for FRP, AF-area and stack height the “Good” cluster dataset, which include FRP, AF area and InjH layer as defined in previous section. It is specifically developed for the optimization of the PRMv2 model and detailed below. The rejection of fires on agricultural land, since these are often quite small, controlled fires with very changeable characteristics caused by human intervention, and which may therefore be less likely to reach a steady state. According to the MISR data set land cover map shown in A correct match between the wind information extracted from the ECMWF analysis and from the MISR products. The stereo-matching algorithm used in MINX corrects for plume displacement, and each pixel in the MISR plume height product is characterized by a plume top height and local wind speed across and along the plume direction A limited number of fire clusters near the plumes origin. Only plumes with less than A plume is clearly visibly observable. The objective here is to remove fires that fail to show a well developed plume, possibly because it is too early in the fires lifetime. Each fires plumes development was assessed using simulated true colour composite MODIS imagery covering
To further maximise the appropriateness and quality of the match-up “fire-atmosphere-plume height” dataset used to optimize and evaluate the PRMv2 model, in addition to the criteria of the filtered fire cluster MISR dataset, the following requirements were also set in the good fire data set:
Using the selection criteria outlined above, the MINX-MISR dataset was reduced to
As mentioned previously, and in
The NIR colour composite provides qualitative information about both the fire and the plume, providing an insight to the plumes relative location with respect to the fire, to the plumes constituent make-up (e.g. ice vs. water droplets). In the case of the fire O18779-B36-P1, since detected active fire pixels only appear on the edge of a large iced fluffy cloud, either the cloud mask in the MODIS fire product is masking the plume as cloud and preventing detection of fire pixels underneath, and/or the plume is absorbing or scattering enough of the fires MIR radiation to limit the number of fire pixels detected.
Looking at the
To understand some of the interaction between plume and fire behaviours, Fig.
As in
Considering only plumes above the boundary layer (
The filtered clusters data set (Fig.
The good clusters dataset contains only
The variables contained within the PRMv2 model parameter vector (
We used an objective function (
Although related by their shared use of the Metropolis criterion
We used the SA algorithm to first assess the physical representativeness of the observed fire plumes modelled by PRMv2. Several
instances of SA were run with different training data sets
After running the SA algorithm, we used an adaptive McMC algorithm
to analyse the error of the optimal vector,
For most non-trivial cases, it is not possible to analytically solve
A particular obstacle to using the McMC method is finding an
optimal proposed distribution for selecting new parameter values. If
this distribution is poorly tuned, then the ratio of accepted
to rejected parameter values becomes too low, and the chain (samples)
does not properly converge to the stationary distribution. The
For PRMv2, we used McMC to compute the joint posterior distribution of the
parameter space.
A total of
Considering the entire InjH layer (not only its top height), Fig.
To further validate PRMv2, as well as to study its sensitivity to the input parameters, we used the Markov chain Monte Carlo (McMC) uncertainty test, introduced in Sect.
The marginal parameter distributions
It is interesting to note that the parameter vector found via SA
(
Due to the limited number of observations that came out of our match-up dataset selection criteria (see Sect.
Using the Collection
Following the results of Fig.
PRMv2 is the only approach that produces behaviour similar to those of the
observations, i.e. with a power law relationship (
Here we present a first gridded product of plume InjH distribution, based on the PRMv2 model optimised in Sect.
The seasonal maps of wildfire plume InjH resulting from use of PRMv2 with the MODIS active fire data are shown in Figs.
The fire events used here are the same as those in Fig.
In the vertical distribution of fire-consumed mass for Summer
To improve the modelling of biomass burning (BB) emissions transport, several parametrizations of smoke plume injection height (InjH) for implementation in atmospheric chemical transport models have been proposed
In this work, we therefore use a subset of the current North American MISR data of
The limited number of fire events available did not allow us to properly validate PRMv2 using modelled vs observed height comparisons, but when applied to a year of MODIS active fire observations for North America PRMv2's response to the effect of atmospheric stability is consistent with previous findings showing a direct relationship between plume height and FRP for fire events in unstable atmospheres
After validation over more geographical locations (e.g. deforestation fire in South America), the application of PRMv2 to global fire inventory would be considered to set up “injection height climatology”, which could be either used as a model validation dataset if transport models have a sub-grid fire plume model, or as climatological database to represent the effect of fire plumes in transport models (for models not having sub-grid parametrization).
From this work we conclude that plume rise models for application to landscape scale fires are still very much worth developing, since they may help us understand plume dynamics and in particular interpret the relatively sparse plume observations available from instruments such as MISR. However, as recommended by
Using mass flux conservation, the Boussinesq approximation and introducing the variable
The vertical momentum equation is derived as in
The energy conservation is ensured by the temperature equation, where source terms are: the dry adiabatic lapse rate, latent heat (see
Finally, as in PRMv0, the amplitude of the horizontal flow
In PRMv2, the entrainment (
The effect of vertical plume transport on horizontal flow is modelled following the work of
The main difference between our approach to en/de-trainment, and that of
To account also for wind shear effects in PRMv2, the approach of
To validate our implementation of the
To make PRMv2 most easily applicable to the largest number of fire events seen
with MODIS, we also applied the
The Dozier algorithm is then applied to
As a first test of our implementation, we compare the FRP retrieved via the outputs of the
When using BIRD or MODIS data with the
To pursue the validation of our Dozier algorithm implementation, Figs.
Assuming that (Bi
Figure
When removing cool pixels, clusters should end up with higher active fire temperatures and smaller areas. Figure
Figure
In the scope of this study, we suggest that our implementation of the Dozier algorithm based on the MOD14 data is able to characterize fires clusters by their effective active fire temperature and area, and that when applied to flaming dominated clusters where the TIR band signal is well characterized from the background (
The colors in the NIR colour composite image shown in Fig. Red – this represents a pixel containing an actively burning fire. The peak wavelength of spectral emittance of a landscape scale fire is within or close to the shortwave infrared (SWIR) spectral region. Therefore, a consequently increased White – this represents a pixel containing large particles: e.g. water droplets or ice particles. These scatter electromagnetic radiation at the wavelengths used to create the NIR colour composite approximately equally, due to their large size and resultant non-selective scattering. For this reason, the same pixels also appear white in the simulated true color composite. Blue/cyan for pixels in the plume, coupled with bright pixels in the simulated true color composite image: pixel containing more ice than water droplets, because reflectance of ice is lower than water at Blue – this represents pixels containing smaller scattering particles, since the Black – this represents pixels where the land surface is either a lake, river or ocean, or is a recently burned area (water and the black ash typically laid down by fires has a low reflectance at IR wavelengths). Green – this represents a pixel containing substantial live vegetation, which has a high reflectance in the NIR spectral region.
Various components of this study were supported by the NERC grant NE/E016863/1, by the NERC National Centre for Earth Observation (NCEO), and also by the EU in the FP7 and H2020 projects MACC-II and MACC-III (contracts 283576 and 633080). The authors thank ECMWF for providing meteorological data, and the DLR BIRD Team and Zebris GMbH for providing the BIRD HSRS datasets used herein. NASA is acknowledged for the MODIS and MISR data used within this study, and the MISR Plume Height Project using the MINX tool is acknowledged for its provision of the plume height and other information used.
Input parameters of the PRMv2 model. See Sect.
Number of high fire plumes predicted by the
Sketch of fire plume dynamics as parametrized in the PRMv2 plume rise model developed herein.
Example of the input and output of the PRMv2 model for the fire O13289-B39-P3 of the MISR plume height dataset
Methodology used here to derive injection height estimate from MISR plume top height observations, as applied to the fire O13 289-B39-P3 of the MISR-derived plume height data set of
Comparison between the smoke plume injection height measured at the location of the plume stack (
Example of horizontal wind amplitude profile at a fire location, as extracted from the
ECMWF analysis (black dashed line) and as derived from the MISR plume height product (triangular points). The red line shows the best fit that minimizes the residual error defined by Eq. (
Temporal evolution of fire O18779-B36-P1 of the MISR plume height project of
Relationship between fire and plume observables: FRP, AF-area, FRP density and plume height defined in Sect.
One-to-one relationship between fire plume InjH layer (“detrainment zone”) estimated from MISR observations (Sect.
Result from the PRMv2 parameter uncertainty test based on the Markov chain Monte Carlo (McMC) algorithm presented in Sect.
.
Relationship between modelled injection height top layer height and FRP, for fire plumes reaching the free troposphere. Fires are those detected by the Collection
Seasonal maps of the distribution of landscape fire smoke plume injection height (InjH) over North America, calculated for the year
Same as for Fig.
Geo-referenced, co-located imagery collected by
Comparison between FRP
Comparison of fire radiative power as derived from active fire effective temperature and area measures output from implementation of the
Comparison of effective active fire area (AF-area; ha) as estimated from implementation of the
Comparison of effective active fire area temperature (
Action of the TIR radiance signature test (