of 16 /16
Atmos. Chem. Phys., 17, 6423–6438, 2017 www.atmos-chem-phys.net/17/6423/2017/ doi:10.5194/acp-17-6423-2017 © Author(s) 2017. CC Attribution 3.0 License. Particulate emissions from large North American wildfires estimated using a new top-down method Tadas Nikonovas, Peter R. J. North, and Stefan H. Doerr Geography Department, College of Science, Swansea University, Singleton Park, Swansea, SA2 8PP, UK Correspondence to: Tadas Nikonovas ([email protected]) Received: 31 March 2016 – Discussion started: 3 August 2016 Revised: 3 February 2017 – Accepted: 14 April 2017 – Published: 30 May 2017 Abstract. Particulate matter emissions from wildfires affect climate, weather and air quality. However, existing global and regional aerosol emission estimates differ by a factor of up to 4 between different methods. Using a novel approach, we estimate daily total particulate matter (TPM) emissions from large wildfires in North American boreal and temper- ate regions. Moderate Resolution Imaging Spectroradiometer (MODIS) fire location and aerosol optical thickness (AOT) data sets are coupled with HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory) atmospheric dispersion simulations, attributing identified smoke plumes to sources. Unlike previous approaches, the method (i) combines infor- mation from both satellite and AERONET (AErosol RObotic NETwork) observations to take into account aerosol water uptake and plume specific mass extinction efficiency when converting smoke AOT to TPM, and (ii) does not depend on instantaneous emission rates observed during individual satellite overpasses, which do not sample night-time emis- sions. The method also allows multiple independent esti- mates for the same emission period from imagery taken on consecutive days. Repeated fire-emitted AOT estimates for the same emis- sion period over 2 to 3 days of plume evolution show in- creases in plume optical thickness by approximately 10 % for boreal events and by 40 % for temperate emissions. Inferred median water volume fractions for aged boreal and temper- ate smoke observations are 0.15 and 0.47 respectively, indi- cating that the increased AOT is partly explained by aerosol water uptake. TPM emission estimates for boreal events, which predominantly burn during daytime, agree closely with bottom-up Global Fire Emission Database (GFEDv4) and Global Fire Assimilation System (GFASv1.0) invento- ries, but are lower by approximately 30 % compared to Quick Fire Emission Dataset (QFEDv2) PM 2.5 , and are higher by approximately a factor of 2 compared to Fire Energetics and Emissions Research (FEERv1) TPM estimates. The discrep- ancies are larger for temperate fires, which are characterized by lower median fire radiative power values and more sig- nificant night-time combustion. The TPM estimates for this study for the biome are lower than QFED PM 2.5 by 35 %, and are larger by factors of 2.4, 3.2 and 4 compared with FEER, GFED and GFAS inventories respectively. A large underesti- mation of TPM emission by bottom-up GFED and GFAS in- dicates low biases in emission factors or consumed biomass estimates for temperate fires. 1 Introduction Large and often severe fires in boreal and temperate for- est regions alter atmospheric composition, considerably af- fecting the Earth’s radiative budget (Langmann et al., 2009; Bond et al., 2013) and degrading air quality (Johnston et al., 2012). The burning regime in these regions is domi- nated by episodic extreme events (Stocks et al., 2002) emit- ting continental-scale plumes (Colarco et al., 2004) with interhemispheric transport potential (Damoah et al., 2004; Dahlkötter et al., 2013). Future climate predictions indicate both drier conditions and greater than average warming for northern latitudes, projecting a likely increase in area burned (Liu et al., 2010) and soil carbon consumption (Turetsky et al., 2015). For the quantification of smoke radiative forc- ing and impacts on human health, a realistic representation of biomass burning emissions in climate and air quality models is needed. Disagreement between bottom-up and top-down Published by Copernicus Publications on behalf of the European Geosciences Union.

Particulate emissions from large North American wildfires ... · Particulate emissions from large North American wildfires ... integrated FRP to emitted particulate matter without

  • Author
    hadan

  • View
    215

  • Download
    2

Embed Size (px)

Text of Particulate emissions from large North American wildfires ... · Particulate emissions from large...

  • Atmos. Chem. Phys., 17, 64236438, 2017www.atmos-chem-phys.net/17/6423/2017/doi:10.5194/acp-17-6423-2017 Author(s) 2017. CC Attribution 3.0 License.

    Particulate emissions from large North American wildfiresestimated using a new top-down methodTadas Nikonovas, Peter R. J. North, and Stefan H. DoerrGeography Department, College of Science, Swansea University, Singleton Park, Swansea, SA2 8PP, UK

    Correspondence to: Tadas Nikonovas ([email protected])

    Received: 31 March 2016 Discussion started: 3 August 2016Revised: 3 February 2017 Accepted: 14 April 2017 Published: 30 May 2017

    Abstract. Particulate matter emissions from wildfires affectclimate, weather and air quality. However, existing globaland regional aerosol emission estimates differ by a factor ofup to 4 between different methods. Using a novel approach,we estimate daily total particulate matter (TPM) emissionsfrom large wildfires in North American boreal and temper-ate regions. Moderate Resolution Imaging Spectroradiometer(MODIS) fire location and aerosol optical thickness (AOT)data sets are coupled with HYSPLIT (Hybrid Single-ParticleLagrangian Integrated Trajectory) atmospheric dispersionsimulations, attributing identified smoke plumes to sources.Unlike previous approaches, the method (i) combines infor-mation from both satellite and AERONET (AErosol ROboticNETwork) observations to take into account aerosol wateruptake and plume specific mass extinction efficiency whenconverting smoke AOT to TPM, and (ii) does not dependon instantaneous emission rates observed during individualsatellite overpasses, which do not sample night-time emis-sions. The method also allows multiple independent esti-mates for the same emission period from imagery taken onconsecutive days.

    Repeated fire-emitted AOT estimates for the same emis-sion period over 2 to 3 days of plume evolution show in-creases in plume optical thickness by approximately 10 % forboreal events and by 40 % for temperate emissions. Inferredmedian water volume fractions for aged boreal and temper-ate smoke observations are 0.15 and 0.47 respectively, indi-cating that the increased AOT is partly explained by aerosolwater uptake. TPM emission estimates for boreal events,which predominantly burn during daytime, agree closelywith bottom-up Global Fire Emission Database (GFEDv4)and Global Fire Assimilation System (GFASv1.0) invento-ries, but are lower by approximately 30 % compared to Quick

    Fire Emission Dataset (QFEDv2) PM2.5, and are higher byapproximately a factor of 2 compared to Fire Energetics andEmissions Research (FEERv1) TPM estimates. The discrep-ancies are larger for temperate fires, which are characterizedby lower median fire radiative power values and more sig-nificant night-time combustion. The TPM estimates for thisstudy for the biome are lower than QFED PM2.5 by 35 %, andare larger by factors of 2.4, 3.2 and 4 compared with FEER,GFED and GFAS inventories respectively. A large underesti-mation of TPM emission by bottom-up GFED and GFAS in-dicates low biases in emission factors or consumed biomassestimates for temperate fires.

    1 Introduction

    Large and often severe fires in boreal and temperate for-est regions alter atmospheric composition, considerably af-fecting the Earths radiative budget (Langmann et al., 2009;Bond et al., 2013) and degrading air quality (Johnstonet al., 2012). The burning regime in these regions is domi-nated by episodic extreme events (Stocks et al., 2002) emit-ting continental-scale plumes (Colarco et al., 2004) withinterhemispheric transport potential (Damoah et al., 2004;Dahlktter et al., 2013). Future climate predictions indicateboth drier conditions and greater than average warming fornorthern latitudes, projecting a likely increase in area burned(Liu et al., 2010) and soil carbon consumption (Turetskyet al., 2015). For the quantification of smoke radiative forc-ing and impacts on human health, a realistic representation ofbiomass burning emissions in climate and air quality modelsis needed. Disagreement between bottom-up and top-down

    Published by Copernicus Publications on behalf of the European Geosciences Union.

  • 6424 T. Nikonovas et al.: Particulate emissions from large wildfires in North America

    emission estimates of particulate matter, however, remainslarge (Kaiser et al., 2012; Ichoku and Ellison, 2014).

    Bottom-up emission inventories use emission factors (EF)(Andreae and Merlet, 2001; Janhll et al., 2010; Akagi et al.,2011; Urbanski, 2014), ratios of gases and particulate matteremitted per unit of dry fuel burned, compiled for differentbiomes from a range of burning experiment measurementsacross the globe. Emission factors are applied to biomassburned estimates, which are typically based on satellite ob-servations of ubiquitous but highly variable fire activity. TheGlobal Fire Emission Database (GFED) (van der Werf et al.,2010) makes use of satellite burned area products (Rander-son et al., 2012; Giglio et al., 2013) and active fire pixelcounts, while the Global Fire Assimilation System (GFAS)(Kaiser et al., 2012) employs fire radiative power (FRP) mea-surements (Giglio et al., 2006). Burned area estimates areconverted to biomass burned using modelled carbon poolsand soil-moisture-dependent combustion completeness char-acteristic to the fuel types. FRP-based methods rely on ob-served relationships between observed FRP and biomasscombustion rates (Kaufman and Tanre, 1998; Wooster et al.,2003, 2005).

    The more top-down methods utilize satellite aerosol op-tical thickness (AOT) observations. The Quick Fire Emis-sion Database (QFED) uses regional AOT measurementsto scale emissions based on EFs (Darmenov and da Silva,2015). Similarly, atmospheric model assimilation of GFASemissions (Kaiser et al., 2012) suggested a 3.4 global en-hancement factor was needed to reconcile total particulatematter (TPM) estimates with observed AOTs. Purely top-down methods estimate emissions through inverse modellingof satellite AOT retrievals (Ichoku and Kaufman, 2005;Dubovik et al., 2008). A top-down global gridded Fire En-ergetics and Emissions Research (FEERv1) (Ichoku and El-lison, 2014) product is based on collocated satellite FRPand AOT observations. Inferred total particulate matter emis-sions rates are linked to observed FRP. The estimated TPMemission coefficients allow direct conversion from time-integrated FRP to emitted particulate matter without invok-ing the emissions factors.

    Global and regional particulate matter estimates from thebottom-up burned area and fire pixel-count-based GFEDagree well with the FRP-based GFAS estimates. Model as-similations of these bottom-up emissions, however, suggestTPM underestimation by a factor of 2 to 4 compared tosatellite AOT observations (Kaiser et al., 2012). EnhancedGFAS TPM estimates and scaled QFED agree better withtop-down FEER emission coefficients on global scales. No-table discrepancies, however, are present for individual re-gions. North American emissions are larger for enhancedGFAS TPM and QFED when compared to top-down FEER,while FEER agrees closely with the bottom-up GFED inven-tory.

    A number of uncertainties in both bottom-up and top-down estimates can contribute towards the apparent TPM

    discrepancies. Average EFs for different biomes concealsmall sample numbers for some areas, and large variability inindividual measurement results from within-biome inconsis-tencies in vegetation density, climatic and burning conditions(Van Leeuwen and van der Werf, 2011; van Leeuwen et al.,2014). Consumed biomass estimates inherit errors of satel-lite burned area (Randerson et al., 2012), fire location (Hyeret al., 2009) or FRP retrieval (Giglio et al., 2006), and dependon a range of assumptions on availability and consumption ofcarbon in aboveground and soil pools (French et al., 2004).Top-down approaches are affected by AOT retrieval errorand large uncertainties in assumed smoke particle proper-ties, which are required to relate aerosol extinction to partic-ulate mass (Reid et al., 2005b). Moreover, estimates of emis-sion rates based on near-source retrievals are representativeof burning conditions at the time of satellite overpass. A re-cent study indicated that night-time TPM emissions might beunderestimated by a factor of 2030 for a large temperate for-est fire in the western USA (Saide et al., 2015), stressing theneed for better representation of night-time emissions in theinventories. Methods based on regional AOT observations,on the other hand, must take into account poorly constrainedageing effects (Reid and Hobbs, 1998; ONeill et al., 2002).

    This study presents estimates of particulate matter emis-sions from large wildfires with identifiable plumes in NorthAmerican boreal and temperate regions. A newly developedtop-down method is applied which attributes satellite aerosolobservations to a specific fire event and emission period.Quantified daily fire-emitted AOT takes into account aerosolsinjected throughout the diurnal cycle and does not rely on in-stantaneous emission rates observed during a satellite over-pass. In some cases, AOT attribution for the same emissionperiod is achieved from satellite images taken on succes-sive days, allowing for an assessment of uncertainty and aninvestigation of systematic changes in plume optical thick-ness over time. Total particulate matter is quantified by ap-plying mass extinction efficiency, which is simulated usingAERONET particle properties, and accounts for inferred wa-ter uptake by aerosols. The results are compared with exist-ing estimates in order to investigate systematic differencesbetween the approaches.

    2 Data and methods

    Daily total particulate matter emissions for large and persis-tent fire events were estimated by combining Moderate Res-olution Imaging Spectroradiometer (MODIS) active fire ob-servations and aerosol optical thickness retrievals with plumedispersion simulated using the Hybrid Single-Particle La-grangian Integrated Trajectory (HYSPLIT) model.

    Atmos. Chem. Phys., 17, 64236438, 2017 www.atmos-chem-phys.net/17/6423/2017/

  • T. Nikonovas et al.: Particulate emissions from large wildfires in North America 6425

    Figure 1. An illustration of the method showing an example of fire-emitted AOT attribution for two diurnal cycles of a temperate fire. Rowsin the figure represent 3 successive days of satellite imagery from which the attribution was achieved. Columns from left to right showMODIS AOT retrievals for the day from a single platform with the highest coverage (ac), snapshots of HYSPLIT particle positions and agetaken at local noon (df), and AOT interpolated to 25 km equal-area grid (gi). The two right columns show fire-emitted AOT attributed to28 (j) and (k) and 29 (l) and (m) July 2007 determined from images taken on different days. Total attributed AOT is shown within the plots.

    2.1 Active fires

    To represent fire activity we used the active fire location dataset MCD14ML produced by the University of Maryland andprovided by NASA Fire Information for Resource Manage-ment System (Giglio et al., 2006). The data product is basedon MODIS mid-range and thermal infrared observations.MODIS sensors are flown on board the sun-synchronouspolar-orbiting Terra and Aqua satellites which respectivelypass the equator at 10:30 and 13.30 local time during the day-time hours and at 22:30 and 01:30 at night. The instrumentshave a wide swath of approximately 2330 km, each providingnear-global coverage daily. For high latitudes the coverage isbetter due to increasing overlap between consecutive over-passes. Each detection in the data set represents an active firein a 1 km2 pixel at the time of satellite overpass, and containsinformation on the retrieved fire radiative power.

    2.2 Fire event selection

    Large and long-lived fire events, likely strong emissionsources, were identified and selected for the analysis. Burn-ing episodes larger than 100 km2 are not numerous, but ac-count for more than 80 % of total burned area in borealNorth America (Stocks et al., 2002; Kasischke et al., 2002),and are a dominant mode of burning in parts of temperateregions as well (Strauss et al., 1989). In order to identify

    such events, individual MODIS active fire detections wereagglomerated into large wildfire events by performing two-step spatialtemporal clustering. First, any MODIS fire de-tections located closer than 10 km in space and 24 h in timewere grouped together. Single detections not assigned to anyof the formed clusters were removed from further analysis.The clusters were then filtered by selecting events with (i) aspatial bounding box containing all fire detections belongingto the cluster larger than 100 km2 and (ii) a duration longerthan 7 days. The duration was determined by the time spanbetween the first and the last MODIS active fire detectionsbelonging to the cluster. The burning was considered unin-terrupted if the largest temporal interval between subsequentMODIS fire observations was less than 24 h. During the sec-ond step of clustering, any of the selected events active at thesame time and located closer than 150 km were grouped intolarge burning episodes and assigned a unique source label.These events were classified into boreal and temperate firesusing the dominant emission source given in the GFEDv4 in-ventory for areas and periods in which the events were active.

    2.3 Plume dispersion modelling

    Smoke transport for the selected fire events was simulatedwith the HYSPLIT model (Draxler and Rolph, 2003). Plumedispersion from a source location was represented by the mo-tion of a large number of discrete particles moved by the

    www.atmos-chem-phys.net/17/6423/2017/ Atmos. Chem. Phys., 17, 64236438, 2017

  • 6426 T. Nikonovas et al.: Particulate emissions from large wildfires in North America

    wind field with mean and random components. Global DataAssimilation System (GDAS) meteorological archive datawere employed to drive the model.

    For each day of burning, particles were continuously re-leased into the model domain from the locations of the in-dividual active fire detections within the fire event. In orderto represent fire diurnal cycle, different MODIS active fireobservations were used to release particles for two 12 h in-tervals representing day and night emissions from 09:00 to21:00 and from 21:00 to 09:00 local time respectively. Emis-sion source number and locations for daytime periods weredetermined from the highest number of fire detections ob-served during a single Terra or Aqua daytime overpass with10.30 and 13.30 equatorial crossing time. Similarly, emit-ted particle source numbers for the night periods were de-termined by the largest burning extent observed during oneof the night-time overpasses with 22.30 and 1.30 equatorialcrossing times. Notably, the Terra overpass at 22.30 at highlatitudes makes observations of regions where local time isearlier than 21:00. In this study, however, all fires detectedduring this overpass were classed as night-time observations.If no valid observations were available for some of the timeintervals, the count and fire pixel locations were set to a min-imum non-zero value estimated for the burning episode fromall daytime or night-time observations. This was done toavoid total temporary shut-down of the emissions, which isan unlikely scenario for a long burning episode. Every hour,20 particles were released for each fire pixel. As a result,emitted particle number for a burning episode was deter-mined by the number of active fire pixels observed duringa given time period.

    Particles were uniformly distributed between the surfaceand the top altitude of the planetary boundary layer as givenin GDAS archive. Satellite-based plume height estimates(Val Martin et al., 2010; Peterson et al., 2014) indicate thatin up to 80 % of the events analysed, injection heights werelimited to the planetary boundary layer. While confinementof the emissions to the mixing layer underestimates injectionheight for the most energetic burning episodes, such config-uration should nonetheless represent the majority of burningepisodes.

    Throughout the simulations, modelled particle positions,their age and source burning event identifier were recordedeach day at local solar noon. The generated point clouds werelater used to compare against Terra and Aqua aerosol opticalthickness (AOT) observations.

    2.4 Satellite aerosol data

    MODIS AOT collection 5.1 data products M*D04_L2 wereused in this study. The dark target algorithm (Kaufman andTanre, 1998; Levy et al., 2009) retrieves AOT at 550 nmand 10km 10 km spatial resolution at nadir. MODIS pixelsize increases with view angle, and pixels at the edge ofthe swath are approximately 9 times larger. For this study,

    Figure 2. Changes in attributed AOT over time. Image shows 39boreal and 37 temperate diurnal emission cycles for which estimateswere obtained on 3 consecutive days, for both daytime and night-time periods.

    all M*D04_L2 AOT retrievals with quality assurance con-fidence> 0 were selected. To maximize coverage, no cloudfraction filtering was applied. The AOT product global vali-dation against ground-based AERONET AOT observationssuggest a 1 error which increases linearly with aerosolloading (0.05+ 0.20 %) (Levy et al., 2010) for overlandcases. A regional MODIS M*D04_L2 AOT product vali-dation (Hyer et al., 2011) indicates that performance variesgreatly within North America. The study found that for 0.2