Introduction
The austral winter in most of South America is typically dry with extensive
vegetation fires, mostly human induced, in areas of deforestation and
agricultural or pasture land management. The fire activity, especially in
Amazonia and cerrado areas (a savanna-like biome),
usually lasts for at least 3 months from August to October every year and
has typically been called the “biomass burning season”. Fire emissions can be
a significant source of carbon dioxide (CO2) to the atmosphere (e.g.,
Gatti et al., 2014) and a major source of several other trace gases to the
atmosphere (Andreae et al., 2002). In addition to trace gases, vegetation
fires also produce a large amount of aerosol particles, in particular in the
fine mode, which on average contribute to at least 90 % of the total AOD in
the visible spectrum in the case of South American regional smoke (Reid
et al., 2005; Rosário, 2011). The resulting smoke haze typically covers
areas of several million square kilometers over South America (Prins et al.,
1998). Outside the biomass burning season, mean AOD in the visible spectrum
varies from 0.10 to ∼ 0.15 across the Amazon rainforest and from 0.12 to
∼ 0.20 in the cerrado areas (Schafer et al., 2008; Rosario, 2011). At
the peak of the burning season, during September, monthly mean AOD can reach
values up to 10 and 5 times higher than the clean wet season values in the
southern and the northern areas, respectively (Schafer et al., 2008;
Rosário, 2011). The Ångström exponent (AE), an optical property
used to characterize aerosol particle size based on the spectral dependence
of AOD, increases from ∼ 0.60 during the clean period to 2.0 during the
peak of the burning season (Schafer et al., 2008). Such a high value of AE is
representative of air masses dominated by fine-mode particles, a major
feature of the South American regional plume. Low values of AE indicate
a dominance of coarse-mode particles, a characteristic of the pristine region
of the Amazon dominated by biogenic particles. The absorption characteristics
of the particles, expressed by the optical property single scattering albedo
(SSA), ranges in the southern portion of the Amazon forest during the burning season
from 0.92 to 0.93 at 550 nm and reveal a dominance of moderate absorbing
particles (Schafer et al., 2008). This is unlike those from cerrado, which
are highly absorbing and present low values of SSA (∼ 0.89 ± 0.04). Cerrado has a substantially lower SSA than the mean values in the
southern Amazon forest due to the higher fraction of flaming-phase
combustion, which is typical of savanna-like vegetation (Schafer et al., 2008).
In Amazonia under heavy smoke conditions, surface cooling
can reach 3 ∘C, restraining the turbulent flows and consequently the
evapotranspiration of water and sensible heat flux. Thus, the result is a
drier and shallower boundary layer inhibiting the formation and development of
convective clouds and hence decreasing precipitation (Yu et al., 2002). In
particular, biomass burning aerosols reduce the net direct solar radiation
reaching the surface, while they increase the diffuse fraction of solar
radiation. The diffuse fraction of photosynthetically active radiation
(PAR) can increase from about 19 %, which is the typical value for a clean
atmosphere scenario, up to 80 % under biomass burning conditions (Yamasoe
et al., 2006). Biomass burning aerosols also act as cloud condensation nuclei
affecting cloud microphysical properties and therefore change the radiation
budget and hydrological cycle over disturbed areas (Kaufman, 1995; Rosenfeld,
1999; Andreae et al., 2004; Koren et al., 2004).
Changes in the total downward solar irradiance at the surface usually does
not impact the photosynthetic activity of the leaves on the top of the forest
canopy because those are usually light saturated around midday, closing the
plant stomata. On the other hand, sub-canopy leaves typically remain
under light-deficit conditions and do not fully achieve their photosynthetic
potential. Thus, increasing the diffuse light that penetrates deeper into
the canopy increases PAR availability to the sub-canopy leaves and the rate
of photosynthesis, and consequently atmospheric carbon assimilation
(Baldocchi, 1997; Misson et al., 2005; Oliveira et al., 2007; Knohl and
Baldocchi, 2008; Mercado et al., 2009; Doughty et al., 2010; Kanniah et al.,
2012; Rap et al., 2015). However, the increase in the diffuse radiation is
also accompanied by a decrease in the total radiation, therefore defining an
optimal diffuse to total radiation fraction that allows a maximum of carbon
assimilation. Under heavy pollution or cloudy skies, plant productivity
increases with the diffuse radiation, though it is still insufficient to
compensate for the reduction in the total irradiance reaching the surface.
Additionally, other authors suggest that the contribution of biomass burning
aerosols to CO2 assimilation can also be due to the cooling of the
air (Min, 2005; Doughty et al., 2010; Steiner and Chameides, 2011), causing
an increase in relative humidity (Collatz et al., 1991) near the earth's
surface, which reduces plant respiration and thermal stress of the leaves.
Indeed, field observations indicate that tropical forest productivity is
highly sensitive to temperature variations (Feeley et al., 2007), with
CO2 assimilation decreasing sharply during warmer periods. Doughty
et al. (2010) estimated that under dense biomass burning conditions in
Amazonia, 80 % of the increase in CO2 absorption is due
to the increase in sub-canopy light (diffuse radiation), and only 20 % is due
to the reduction in the canopy temperature. The gross primary production
(GPP) is the total carbon uptake resulting from photosynthesis by plants,
especially leaves, in an ecosystem over a land area. GPP responds to the
amount of photosynthetically active radiation of solar energy
reaching the plants, given limitations of soil moisture and nutrients. A
modeling study at the global scale exploring the aerosol impact on GPP
concluded that the positive effect of the diffuse radiation increase was
indeed larger than the negative effect of the irradiance reduction (Mercado
et al., 2009). JULES simulations forced with aerosol fields from the Hadley
Centre Global Environment Model version 2 (HadGEM2) by these authors
pointed to an increase in the diffuse fraction of irradiance and a
consequent increase in the global carbon uptake of 23.7 % from 1960 to 1999.
Rap et al. (2015) also used JULES but with a different offline aerosol
model: the 3-D GLObal Model of Aerosol Processes (GLOMAP; Mann et al., 2010).
They estimated that the biomass burning aerosols affected the diffuse radiation by
3.4 to 6.8 % and increased the net primary production (NPP) of 1.4 to 2.8 %
in Amazonia during the period between 1998 and 2007. The biomass burning aerosol indirect effect will also impact CO2
fluxes by changing the amount of rain (and soil moisture), solar
radiation availability, and diffuse fraction. Lastly, vegetation fires also
emit ozone precursors and promote tropospheric ozone production and surface
deposition. Ozone is highly phytotoxic, damaging plant stomata and reducing
CO2 uptake (Sitch et al., 2007).
Motivated by all these previously cited recent observational and theoretical
studies that have demonstrated the impacts of aerosols on CO2
fluxes, we applied an integrated in-line numerical atmospheric modeling
system to explore the following scientific questions: what is the relative
role of the main processes between soil, vegetation, and the atmosphere
controlling the carbon cycle in Amazonia? What are the effects
of biomass burning aerosols for each of these processes? What is the net
aerosol effect on CO2 fluxes? What is the relative effect of the
direct interaction of aerosol radiation against the aerosol impact on the
diffuse fraction? What is the regional dimension of the aerosol impact on
CO2 fluxes? Finally, how well can a state-of-the-art chemical
transport model (CTM) reproduce CO2 fluxes and mixing ratio
observations over the Amazon Basin?
The structure of this paper is as follows. In Sect. 2.1, we present a
description of the most relevant aspects of the integrated atmospheric
modeling system for this application. The adopted model configuration and the
input data sets, including emissions and boundary conditions, are described
in Sect. 2.2. The observational data sets, both from direct and remote sensing
observations, used for model evaluation and analyses are described in Sect. 2.3. Model results are presented and discussed in Sect. 3. Section 3.1 provides
an overview of the meteorology and fire activity in Amazonia
during the study period, including both model results and observations. We
then follow up in Sect. 3.2 with the model results for the regional biomass
burning plume. Finally, in Sect. 3.3 we examine the model results for energy
and CO2 fluxes related to several surface–atmosphere interaction
processes and the aerosol biomass burning impacts on them. The main
results are summarized in Sect. 4.
Methods and data sets
Description of the relevant parts of the modeling system
In this work, we employed the integrated atmospheric-chemistry model BRAMS
version 5.0 (Brazilian developments on the Regional Atmospheric Modeling
System; Freitas et al., 2005, 2009, 2017), which has been coupled in a
two-way mode with the Joint UK Land Environment Simulator v3.0 (JULES), the
land surface scheme of the UK Hadley Centre Earth System Model, as described
in Moreira et al. (2013). The coupling is two-way in the sense that, for each
model time step, the atmospheric component provides to JULES the current
near-surface wind speed, air temperature, pressure, condensed water and
downward radiation fluxes, and water vapor and carbon dioxide mixing
ratios. After its processing, JULES advances its state variables over the
time step and feeds back the atmospheric component with sensible and latent
heat and momentum surface fluxes, upward shortwave and longwave radiation
fluxes, and a set of trace gas fluxes.
Biosphere model: the Joint UK Land Environment Simulator (JULES)
JULES simulates the exchange of carbon, momentum, and energy between the land
surface and the atmosphere. Additionally, it represents subsurface
hydrological processes, plant photosynthesis and respiration, and vegetation
and soil dynamics (Best et al., 2011; Clark et al., 2011).
Atmospheric aerosols influence ecosystem functioning via effects on GPP
from changes in the quality and quantity of radiation but also indirectly via
temperature effects on GPP and plant and heterotrophic respiration.
The photosynthesis radiation scheme in JULES accounts for the effects of
diffuse radiation on canopy photosynthesis by splitting direct and diffuse
radiation and sunlit and shaded leaves at each canopy layer. Specifically,
the multilayer radiation scheme includes an explicit calculation of the
absorption and scattering of the direct beam and the diffuse radiation fluxes
in both visible and near-infrared wavebands at each canopy layer using the
two-stream approach from Sellers (1985). Additionally, the attenuation of
non-scattered incident direct beam radiation (sun flecks) is calculated using
the approach by Dai et al. (2004). At each canopy layer, JULES estimates the
fraction of absorbed direct and diffuse photosynthetic active radiation
(PAR),
thus providing a vertical profile of intercepted radiation fields, which
allows for the calculation of photosynthesis at each canopy level. At each canopy
layer, the fraction of sunlit and shaded leaves is estimated as a function of
the canopy beam radiation extinction coefficient (as explained in Clark et
al.,
2011), and it is assumed that shaded leaves absorb only diffuse radiation and
sunlit leaves absorb all types of radiation. Photosynthesis at each canopy
layer is then estimated as the sum of sunlit and shaded leaf photosynthesis
weighed by their respective fraction. Total canopy photosynthesis is
estimated as the sum of the leaf-level fluxes in each layer scaled by the leaf
area of each canopy layer.
Fraction of diffuse broadband solar irradiance reaching the surface
as a function of AOD at 670 nm and optical air mass intervals (m).
Temperature effects on photosynthesis are simulated in JULES via
biochemistry, leaf respiration, and effects of vapor pressure deficit (VPD) on
stomatal conductance in response to the temperature (see details in Clark et
al., 2011). The temperature response of leaf respiration is linked to the
temperature response of maximum carboxylation activity of Rubisco (Vcmax) in
JULES, which is described by a peaked response function. The temperature
response of remaining maintenance respiration components is also simulated
using the leaf respiration temperature function. Growth respiration is
estimated as a proportion of net primary productivity (NPP). Heterotrophic
respiration is simulated either using a Q10 temperature function or a RothC
temperature function (Jenkinson et al., 1990 as described in Clark et al., 2011).
Evaluation of the skill of JULES in simulating GPP under highly direct and
highly diffuse radiation conditions has been tested against flux sites in the
Amazon and in temperate forest sites where direct and diffuse radiation
measurements are available. This is shown in Fig. 2 of Rap et al. (2015) at
Tapajós and French Guiana in the Amazon and at two temperate forest sites in
Mercado et al. (2009; Fig. 1). Investigation of the response of
photosynthesis to changes in direct and diffuse radiation across relevant
plant functional types for the Amazon region is carried out within this
study.
Atmospheric model: Brazilian developments on the Regional Atmospheric Modeling System (BRAMS)
BRAMS is in-line coupled with a Eulerian transport model (CCATT) suitable to
simulate emission, transport, dispersion, chemical transformation, and removal
processes associated with trace gases and aerosols (Longo et al., 2013). In
CCATT, aerosol and trace gas transport runs consistently in-line with the
atmospheric state evolution using the BRAMS dynamic and physical
parameterizations. The tracer mass mixing ratio, which is a prognostic
variable, includes the effects of sub-grid-scale turbulence in the planetary
boundary layer (PBL) and convective transport by shallow and deep moist
convection in addition to grid-scale advection transport. For gaseous
species, CCATT-BRAMS can in principle employ several gaseous chemical
mechanisms. However, for this study, only carbon monoxide (CO),
CO2, and aerosol particles (biomass burning type) were emitted and
transported. The physical removal processes (dry and wet deposition) were
applied to all three tracers, and effective lifetimes were applied to
CO and aerosol particles. As the modeling of aerosol biomass burning
particles is the focus of the present study, only biomass burning emission
sources were considered.
The BRAMS model parameterizations chosen for the simulations performed in
this work are described as follows. The parameterization for the unresolved
turbulence in the PBL was based on the Mellor and Yamada (1982) formulation,
which predicts turbulent kinetic energy (TKE). For the microphysics, we used
the single-moment bulk microphysics parameterization, which includes cloud
water, rain, pristine ice, snow, aggregates, graupel, and hail (Walko et al.,
1995). It includes prognostic equations for the mixing ratios of rain and
each ice category of total water and the concentration of pristine ice. Water
vapor and cloud liquid mixing ratios are diagnosed from the prognostic
variables using the saturation mixing ratio of liquid water. The deep and
shallow cumulus convection schemes are based on the mass-flux approach and
described in Grell and Freitas (2014).
Parameters for a third-degree polynomial fit to the diffuse fraction of broadband solar irradiance
reaching the surface as a function of AOD at 670 nm for distinct air mass intervals.
Optical air mass
a
b
c
d
R2
m ≤ 1.10
0.0115
-0.1115
0.4693
0.1258
0.994
1.10 <m≤ 1.25
0.0129
-0.1235
0.4997
0.1304
0.994
1.25 <m≤ 1.40
0.0075
-0.1087
0.5035
0.1477
0.989
1.40 <m≤ 1.70
0.0052
-0.1031
0.5077
0.1795
0.990
1.70 <m≤ 2.00
0.0144
-0.1634
0.6207
0.1696
0.991
2.00 <m≤ 2.80
0.0166
-0.2237
0.7458
0.1851
0.981
m > 2.80
0.0736
-0.4631
1.0152
0.2005
0.985
Model domain with the map of land cover used in BRAMS, with the color
scale depicting the dominant biomes. The red contour line on the map delimits
the LBAR and the locations of CO2 and CO airborne
measurements: Santarém, PA (2.85∘ S, 54.95∘ W);
Rio Branco, AC (9.99∘ S, 67.80∘ W); Alta
Floresta, MT (9.87∘ S, 56.09∘ W); and Tabatinga, AM
(4.25∘ S, 69.94∘ W).
The radiation scheme is a modified version of the Community Aerosol and
Radiation Model for Atmospheres (CARMA; Toon et al., 1988), which includes the
aerosol–radiation interaction with feedback to the model heating rates (Longo
et al., 2013; Rosário et al., 2013). In addition, we included in CARMA a
parameterization to calculate the diffuse fraction of solar irradiance
specific to biomass burning aerosols in Amazonia. This
parameterization was based on measurements of broadband and narrowband solar
global and diffuse irradiance components performed with a multi-filter
rotating shadowband radiometer (MFRSR; Harrison et al., 1994). With the
narrowband measurements centered at 415, 670, 870, and 1036 nm, AOD was
estimated following the methodology of Harrison and Michalsky (1994) and
Rosário et al. (2008). The measurements were performed at Reserva
Biológica do Jaru, RO (10.145∘ S, 61.908∘ W) during the
dry biomass burning season of 2007. The diffuse fraction of broadband
irradiance reaching the surface as a function of AOD at a 670 nm wavelength for
distinct optical air mass intervals (m), expressed as a ratio of the optical
path length relative to the path length at zenith (Fig. 1), was analyzed;
a three-degree polynomial fit was proposed as follows:
D=aAOD6703+bAOD6702+cAOD670+d.
D represents the diffuse fraction; the values of the fitting parameters
a, b, c, and d of the function in
Eq. (1) are described in Table 1. These fittings were achieved after
filtering the data for clouds, which can be present even during the dry
season, especially during days with low AOD values. When clouds are present,
the diffuse fraction of radiation increases significantly with values as high
as a very polluted atmosphere. However, as discussed below, the analysis
presented here focuses only on areas and during hours without cloud cover;
i.e., the results were obtained by filtering out the points with cloudiness,
considering only the model grid boxes where the total column integrated
condensed water was equal to zero. The aim of this work is only to compute
the aerosol effect; therefore this filter was essentially used to exclude the
effects of clouds in the CO2 fluxes.
Model configuration and input data sets
BRAMS model simulations were conducted for a domain covering the northern
part of South America (southwest corner at 18∘ S, 90∘ W and
northeast corner at 15∘ N, 30∘ W) using a regular grid with 20 km of resolution, as illustrated in Fig. 2. JULES was configured with
10 canopy layers. The chosen model domain encompasses the Legal Brazilian
Amazon region (LBAR, depicted by the red line in Fig. 2), which is a region
of 5 016 136.3 km2 (59 % of the Brazilian territory). It hosts
approximately 24 million inhabitants, which is only around 12 % of the
Brazilian population. Brazilian Federal Law No. 5.173 (Art. 2)
established the LBAR in 1966 as an administrative unit to promote sustainable
development in one of the most, if not the most, resource-rich regions in
Brazil. The main tropical biomes in South America, the Amazon rainforest and
Cerrado and Pantanal wetlands, are all found in the LBAR. Despite
government protection, deforestation activities followed by vegetation
fires have led to extensive land use change to pasture and agricultural
fields and changes in the atmospheric aerosol load and characteristics. This
study aims to analyze the effect of these changes on the atmospheric
environment, radiation budget, and forest productivity in this important
region.
The NCEP Global Forecast System analysis
(http://rda.ucar.edu/datasets/ds083.2/), with 6-hourly time resolution
and 1–1∘ of spatial resolution, provided initial and
boundary conditions for the time integration of the meteorological fields.
Sea surface temperature (SST) was taken from the NOAA Optimum Interpolation (OI)
SST product version 2 with 1–1∘ of spatial resolution
(available at
http://www.esrl.noaa.gov/psd/data/gridded/data.noaa.oisst.v2.html;
Reynolds et al., 2002). Data from the RADAMBRASIL project (Rossato et al.,
1998) were used for the soil type in Brazil, and data from FAO (Zobler, 1999)
were used outside Brazil. The model was run with seven soil levels: 0.10,
0.35, 1.0, 2.25, 4.25, 7.25, and 12.25 m below the surface. Soil moisture was
initialized with the soil moisture estimation operational product developed
by Gevaerd and Freitas (2006) and available at CPTEC/INPE, and the soil
temperature was initialized assuming a vertically homogeneous field defined
by the air temperature closest to the surface from the initial atmospheric
data. The carbon data assimilation system, Carbon Tracker 2015 (Krol et al.,
2005; Peters et al., 2007) with 3–2∘ of spatial resolution
and 34 vertical levels (available at
http://www.esrl.noaa.gov/gmd/ccgg/carbontracker/), provided CO2
initial and boundary conditions. Initial and boundary conditions for carbon
monoxide (CO) were based on optimized fluxes with 1–1∘ of spatial resolution as calculated by the 4D-var system using the
Infrared Atmospheric Sounding Interferometer (IASI) data taken onboard the
Eumetsat Polar System (EPS) Metop-A satellite (Krol et al., 2013).
Biomass burning emissions of trace gases and aerosols were provided by the
Brazilian Biomass Burning Emission Model (3BEM; Longo et al., 2010). The 3BEM
emissions are based on a database of fire pixel counts and burned area
derived from the combination of the following: remote sensing fire products from
the Geostationary Operational Environmental Satellite-Wildfire Automated Biomass
Burning Algorithm (GOES WF-ABBA product; Prins et al., 1998); the Brazilian
National Institute for Space Research (INPE) burning points observed, which
are based on the Advanced Very High Resolution Radiometer (AVHRR) aboard the
NOAA polar-orbiting satellite series (Setzer and Pereira, 1991); and the
Moderate Resolution Imaging Spectroradiometer (MODIS) fire product (Giglio
et al., 2003). Fire emissions were split into smoldering and flaming emission
contributions, releasing trace gases and aerosol particles in the lowest
model layer and in the injection layer, respectively, as determined by the
in-line plume rise model (Freitas et al., 2007, 2010).
The land use data set from the United States Geological Survey (USGS) at 1 km
of resolution was merged with a land cover map for the Legal Brazilian Amazon
region (PROVEG; Sestini et al., 2003). PROVEG is based on the Landsat
Thematic Mapper (TM) images with a spatial resolution of 90–90 m from the year 2000 and deforestation data from the Amazon
Deforestation Monitoring Program (PRODES) for the year 1997. At the 20 km
resolution, each grid box has 400 specifications of vegetation, which were
reduced to nine patches of different land cover categories (broadleaf trees,
needle leaf trees, C3 and C4 grasses, shrubs, urban, inland water, soil, and
ice), each with its respective occupation fraction. JULES treats each
category separately and returns to BRAMS average fluxes weighted by the
occupation fraction. The model results are then discussed for the land cover
categories present in the considered model domain: broadleaf trees (tropical
forest), shrubs (cerrado), and C3 and C4 grasses (pasture). The land
use map in the model domain is illustrated in Fig. 2.
The model simulations were initialized on 15 August 2010 at 00:00 UTC and
conducted for 45 days. We discarded the first 15 days as spin-up and
restricted our analysis to the month of September to avoid model artifacts
related to the initial conditions.
A set of three experiments was performed. In the first one (hereafter named
NO-AER), the aerosol–radiation interaction was neglected. The direct aerosol
effect was taken into account in both the second (hereafter named DIR-AER)
and third (hereafter named DIR+DIF) experiments, but only in the latter was the
diffuse fraction of radiation passed to JULES. Otherwise it was set to zero.
Additionally, a long-term model run (2 years, from January 2010 to December
2011) was carried out only for the DIR+DIF model configuration.
Method of analysis of the model results
Comparing the carbon fluxes from the three model simulations should allow us
to assess the aerosol effect on CO2 uptake in Amazonia
and the relative roles of surface temperature, the direct aerosol effect, and
the increase in the diffuse fraction of PAR due to aerosol scattering. The
CO2 fluxes from the model were analyzed as the variation related to
the total aerosol effect (both on diffuse radiation and direct radiation):
ΔFluxtot=FluxDIR+DIF-FluxNO-AER.
Only with the direct radiation aerosol effect:
ΔFluxdir=FluxDIR-AER-FluxNO-AER.
And, only with the diffuse radiation aerosol effect:
ΔFluxdiff=FluxDIR+DIF-FluxDIR-AER.
We examined the spatial distribution and diurnal cycles of CO2
fluxes related to several surface–atmosphere interaction processes. In
addition to GPP, we also looked at the CO2 fluxes associated with
plant respiration (RP), soil heterotrophic respiration (RH), and the
net ecosystem exchange (NEE =RP+RH-GPP), which is a measurement of the
quantity of carbon entering and leaving the ecosystem (negative when the
ecosystem is a CO2 sink and positive when it is a CO2 source).
The spatial distributions of CO2 fluxes are presented as a monthly mean
in µmolCm-2s-1 for September 2010 at 16:00 UTC, which is
around noon local time for most of Amazonia. The diurnal cycles
of CO2 fluxes are presented as hourly monthly means in units of
µmolCm-2s-1 for September 2010, considering the model
grid cells circumscribed in the LBAR delimited in Fig. 2. The
monthly net values of CO2 fluxes in the LBAR are calculated as the
integral of the mean diurnal cycles and presented in TgCmonth-1,
considering the LBAR area of ∼5.2×1012 m2, which is the total
area of the model cells circumscribed in the LBAR delimited in
Fig. 2 multiplied by 30 to get the monthly total.
Because each biome has its own characteristics and responds differently to
changes occurring at or under the surface and in the atmosphere, we first
examined how each biome responds to the presence of biomass burning aerosol
both in terms of total irradiance attenuation near the surface and an increase in
the diffuse fraction of PAR. We also evaluated the relative contribution of
the diffuse to the total (diffuse + direct) aerosol effect on the CO2
fluxes for each biome type. After assessing the specific behavior of each
biome, model results for CO2 and energy fluxes from each biome were
then averaged and weighted by the biome fraction of each grid to address the
heterogeneity of the Amazon region in terms of land cover and local climates.
Data sets for model evaluation
The model results for precipitation and near-surface temperature were
contrasted with direct observations and products derived from satellite
observations. The aerosol biomass burning spatial distribution from the model
was validated with remote sensing products. Additionally, model performance
in simulating CO and CO2 mixing ratios was assessed using
measurements of carbon monoxide (CO) and CO2 concentration in
air samples collected over the Amazon during 2010 and 2011. The CO
concentration varies as a function of fire source, horizontal and vertical
transport, and deposition. It was not coupled with the biosphere model.
Therefore, the DIR-AER and DIR+DIF scenarios have similar concentration
distributions.
Biomass burning mainly releases water vapor and CO2 to the atmosphere
but is also a major source of other tracers, such as CO, volatile organic
compounds, nitrogen oxides, and organic halogen compounds (Andreae and
Merlet, 2001). An enhancement of CO has been historically observed in
Amazonia during the dry season, which is mostly attributed to
fire emissions because volatile organic compound (VOC) oxidation is
expected to have little seasonality in Amazonia (Holloway
et al., 2000; Duncan and Logan, 2008, Andreae et al., 2012). Therefore, the
evaluation of CO mixing ratios from the model against observations
provides an assessment of model skill to simulate fire emissions and their
transport and removal. We also looked into fire activity and used the soil
moisture and meteorological variables from the model as indicators of the spatial
scale of locations where fires were more likely to occur. The data sets used
for model evaluation are described below.
(a) Spatial distribution of mean 2 m temperature and 10 m
wind field from the model during September 2010 at 16:00 UTC. The LBAR and the
locations of 72 near-surface measurement ground stations used to evaluate
the model 2 m temperature are depicted in the map with white asterisks.
(b) Mean diurnal cycle of 2 m temperature (∘C) observed in the 72
near-surface ground stations (black line) and from the model grid point
nearest the respective station (red line). The standard deviation (shaded
gray) and the mean observed temperature values were calculated using
measurements at the 72 observational stations, while the model standard
deviation (red bars) and mean temperature were calculated using the model
temperatures at the grid boxes corresponding to the locations of the 72
stations.
Precipitation
Monthly mean precipitation
over the Amazon region was obtained from the algorithm 3B42 of the Tropical
Rainfall Measuring Mission (TRMM) merged high quality (HQ)/infrared (IR)
precipitation product at a spatial resolution of 0.25–0.25∘ (http://trmm.gsfc.nasa.gov/3b42.html; Kummerow et al.,
1998; Kawanishi et al., 2000). TRMM 3B42 is derived from retrievals of
3-hourly precipitation amounts from the precipitation radar (PR), TRMM
microwave imager (TMI), and visible and infrared scanner (VIRS) aboard the
TRMM satellite merged with rain gauge data from the Climate Anomaly
Monitoring System (CAMS) and the Global Precipitation Climatology Project
(GPCP). Satellite estimates of precipitation were used for model evaluation
due to their more complete spatial and temporal coverage compared to rain
gauge data. The latter was also complementarily used to evaluate modeled
precipitation and temperature, though not ignoring the low density and
heterogeneous distribution of the observational network in the geographical
model domain: 72 PCDs (automatic stations installed and maintained by the
Brazilian National Institute of Meteorology, INMET) in 5 016 136.3 km2.
Temperature
We evaluated the mean
diurnal cycle of 2 m temperature from the model with data from 72
near-surface measurement ground stations in the LBAR; the locations are depicted in
Fig. 3a.
Fire activity
We checked the coherence
of soil moisture results from the model with the burning points observed from
the Advanced Very High Resolution Radiometer (AVHRR) onboard the NOAA polar-orbiting satellite series. The fire detection used is based on the AVHRR
retrieval algorithm from the Brazilian National Institute for Space Research
(www.cptec.inpe.br/queimadas).
Biomass burning CO and CO2
Model performance in simulating CO and CO2
mixing ratios was assessed using measurements of carbon monoxide (CO)
and CO2 mole fraction (“concentration”) in air samples collected
over the Amazon during 2010 and 2011. The air samples were collected with
portable sampling systems consisting of separate compressor and flask units
(Tans et al., 1996) onboard a Cessna 206 aircraft in descending spirals from
4300 to 300 m over the four Amazon locations indicated in Fig. 2:
Santarém, PA (2.43∘ S, 54.72∘ W), Rio
Branco, AC (9.97∘ S, 67.81∘ W), Alta Floresta, MT
(12.54∘ S, 55.71∘ W), and Tabatinga, AM
(4.25∘ S, 69.94∘ W). The air samples collected in
Santarém, Rio Branco, and Alta Floresta are
characteristic of the moist tropical forest, both primary and secondary,
surrounding them. During the dry season, both local (forest) and remote
cerrado and pasture fire emissions affect these three sites. The
samples collected in Tabatinga, which is further west in a more
pristine area of the Amazon forest, respond to the influence of the intact
forest landscape upwind. During the dry season, fire emissions influence the
atmospheric chemistry in all sites, although the biomass burning impact in
Tabatinga is more related to episodic long-range transport events
and is not as persistent as the others. The air sampling was always between
12:00 and 14:00 local time and analyzed at the laboratory of the
Instituto de Pesquisas Energéticas e Nucleares (IPEN) in
São Paulo, Brazil. Measurement precision of CO2 and
CO is estimated to be around 0.04 ppmv and 0.5 ppbv, respectively. For
further details regarding air sampling and analytical methods, see Gatti
et al. (2010).
We also used CO2 measurements in the Tapajós National
Forest near kilometer 67 (2.85∘ S, 55.04∘ W) of the
Santarém–Cuiabá highway, just south of
Santarém, for model evaluation. The Tapajós
measurements are based on eddy covariance methods using the profile mixing
ratio data to estimate the change in vertical average mixing ratio between
the ground and flux measurement height to calculate the column average
storage of CO2 (Saleska et al., 2003). The CO2 mole fractions
were measured at eight levels along the tower (62.2, 50, 39.4, 28.7, 19.6, 10.4,
and 0.91 m). Sample air was drawn and analyzed with an infrared gas analyzer
(IRGA; LI-6262; Licor, Lincoln, NE). The pressure and temperature of the Licor
cells are controlled. The IRGA was automatically zeroed every 2 h and the
Licor profile every 20 min. All Licors are automatically calibrated with
span gases every 6 h.
Biomass burning aerosol
The AOD (aerosol
optical depth) product derived from MODIS sensors onboard the Aqua
satellite is used to evaluate the simulated biomass burning aerosol plume. In
this work, we used the MODIS Level 2.0 Collection 5.1 (051) data and Level 3
atmospheric product denominated MYD08_D3 (mean aerosol optical thickness at
550 nm).
Accumulated precipitation (mm) during September 2010 from the
(a) ground station observation network interpolated for the model grid point,
(b) TRMM rainfall product, and (c) model results.
Model results and discussion
Meteorology, fire activity, and regional biomass burning plume
Temperature
The simulated 2 m
temperature in the central portion of the Amazon Basin during September 2010
peaked around 32 ∘C at 16:00 UTC, while the northwest region was
slightly cooler, typically ranging from 30 to 28 ∘C (Fig. 3a).
Going southeast towards the cerrado region, the mean 2 m
temperature reached 35 ∘C at 16:00 UTC. The temperature gradient is
mostly associated with the gradient of soil moisture and land cover in the
region. The evaluation of the mean diurnal cycle of 2 m temperature from
the model against observations using 72 near-surface measurement ground
stations in the LBAR during the same period shows that the model results are
consistent with observations (Fig. 3.b), though the model mean temperature
was typically cooler (∼ 2.5 ∘C) during the night period and late
afternoon hours and was not far from the standard deviation of the observed
mean temperature. In addition, the model temperature has a diurnal cycle with
a gap of 1 h earlier than the observation.
Precipitation
The monthly mean rainfall
data from the ground station monitoring network, interpolated to the model
grid points (Fig. 4a) and the TRMM rainfall product (Fig. 4b), both
reveal a well-defined spatial distribution of the precipitation in the
northern region of South America during September 2010. First, in the
southeastern Amazon, accumulated rainfall was low, with values typically lower
than 50 mm for this month; the northwestern region was wetter, with
accumulated values around 200 mm. In addition, there were few areas of high
rainfall in the northern part of South America, mainly in the Guiana
Highlands, associated with the topographic forcing. The general spatial
distribution of model accumulated precipitation (Fig. 4c) compares well
with ground observations (Fig. 4a), with an indication that the model
overestimated the total precipitation. However, one must take into account
that the measurement stations are very scarce in this region, making detailed
comparisons difficult. Interpolation in the presence of limited information
usually reduces the intensity of precipitation, spreading the value observed
around the neighboring grid points without data available. Indeed, the
precipitation estimated by TRMM (Fig. 4b) agrees much better with the
model results (Fig. 4c).
Monthly mean soil moisture (m3m-3) during September 2010
at three soil levels: (a) 0.35 m, (b) 1.00 m, and (c) 4.25 m.
The rectangles depict areas with a predominance of forest and moist
soil (red), forest and dry soil (blue), and cerrado with dry soil
(gray).
Soil moisture and fire activity
As a result of the precipitation distribution, the soils are predominantly
wetter in the northwestern part of South America, with simulated volumetric
moist content ranging from 0.35 to 0.45 m3m-3 for all soil layers
(Fig. 5). The high soil moisture of the northwestern Amazonia contrasts
with the rest of the region's dryness; the moisture in the top soil
layer of the model is below 0.2 m3m-3 (Fig. 5a), and only
deeper soil layers remain fairly moist (∼ 0.3 m3m-3, below 4 m; Fig. 5c). Comparing soil moisture in two forest areas with different
rainfall regimes (red rectangle, 148 mm on average and blue
rectangle,
34 mm on average), the area receiving a higher volume of rainfall is
about 55 % wetter than the other in the shallow layers but only 12 %
in the deeper soil layer. By contrast, comparing areas of forest (blue
rectangle) and cerrado (gray rectangle) with similar rainfall regimes (34
mm), the forest region remains considerably wetter (∼ 15 % in both shallow
and deeper layers). According to Köchy and Wilson (2000), high rates of
water uptake per unit mass may reflect the high root density of the
vegetation. In fact, James et al. (2003) found at a site 20 km east
of Regina, Canada (58.47∘ N, 104.37∘ W) where the ability
of grass to reduce soil moisture is nearly 5 times higher than that of
woody vegetation expressed on a per-gram basis. For the forest region, the
soil moisture values from the model are consistent with the mean value of
0.39 m3m-3 measured at 0.2 m at the Tapajós site (near
Santarém; location indicated in Fig. 2) during the dry season
(Doughty et al., 2010). Previous measurements at the same site reported soil
moisture only slightly higher (∼ 0.44 m3m-3 from
0.15 to 0.30 m) and also confirmed the model results for higher moisture
in the deeper soil layers (∼ 0.42 m3m-3 at 4 m; Bruno et al.,
2006).
The period chosen for this study coincided with the peak of the biomass
burning season when a total of 439 297 fires were detected in the area of the
model domain by AVHRR-NOAA, which is 42 % of the total number of fires
detected by the same sensor during the entire burning season. The spatial
distribution of the fires resembles the pattern of precipitation and soil
moisture simulated by the model (Fig. 5). Nonetheless, most of the fires
were ignited by human activities (Nepstad et al., 1999; Cochrane and Laurance,
2008). Fires occurred mostly in the cerrado, C3 type, and C4 type
grass-covered areas, but forest fires with much higher biomass density are
typically responsible for the highest amount of biomass burning aerosols and
trace gases released into the atmosphere. Figure 6 depicts the spatial
distribution of vegetation fires detected by remote sensing over the Amazon
and central Brazil during September 2010.
Burning points observed by the AVHRR sensor during September 2010
(source: www.cptec.inpe.br/queimadas).
Time series (a, c) and scatterplots (b, d) of
CO (ppbv; a, b) and CO2 (ppbv; c, d) airborne
measured (black dots) and simulated (DIR+DIF experiment,
blue dots and line) at Alta Floresta, Santarém,
Rio Branco, and Tabatinga (indicated on each plot) at
about 2 km above the ground level from April 2010 to October 2011. On
the time series, the red dashed lines indicate the fire
season of 2010. On the scatterplots, the colors scale depicts the vertical
layers: < 900 m (purple), 900–1800 m (blue), 1800–2700 (green),
2700–3600 (orange), and > 3600 m (pink), and the error bars refer to
the standard deviation of the mean values. The locations of the measurement
sites are indicated in Fig. 2.
Smoke regional plume, CO, and
CO2 mixing ratios
Model results for the CO mixing ratio as
simulated by the DIR+DIF experiment, which are expected to be the most
realistic, were compared with CO vertical profile measurements from
four different sites in Amazonia. The time series of the CO
mixing ratio derived from observations and model results around 2 km
above the ground level in four different locations in Amazonia
(Alta Floresta, Santarém, Rio Branco, and
Tabatinga) are shown in Fig. 7 (top left). There was an
enhancement of CO from July to October for both years in all four
locations. In 2010, the CO mixing ratio values in the PBL during the
dry season increased from 100 to 150 ppb, which are typical wet season values (Andreae
et al., 2012), to up to 500 ppb in both the simulations and observations.
The dry season of 2010 had CO mixing ratios about twice as high as
the same period in 2011; this is consistent with the total number of fires,
which
approximately doubled, detected by remote sensing during the dry season of these two years (http://www.inpe.br/queimadas/estatisticas.php).
Scatterplots of the CO mixing ratio values from observation and model
results for the same four locations separated into several vertical layers
are depicted in Fig. 7 (top right). The CO background mixing
ratio values for the Tabatinga site are close to the 1:1 line, while
the biomass-burning-affected values are more scattered. Model results show
good agreement with observations but tend to underestimate CO and
CO2 observations, especially at low levels, in locations mainly
affected by fire emissions both locally (Alta Floresta, Rio
Branco, and Santarém) and by long-range transport
(Tabatinga). The black line on each scatterplot in Fig. 7 shows
the linear fit and the corresponding R-squared values. The largest
underestimation of CO occurred in Alta Floresta, with a slope
of 0.58, but the biggest dispersion occurred in Santarém, with
R2=0.58. This pattern is probably related to the 20 km model
resolution not picking up individual biomass burning plumes and fire
emission underestimation (Pereira et al., 2016). Previous studies indicated
that biomass burning emissions contribute more than 95 % to the variability
of CO over the Amazon and that the emissions used in this study (3BEM;
Longo et al., 2010) are about 20 % underestimated (Andreae et al., 2012).
On the other hand, the airborne vertical profiles analyzed in this study and
our modeling results indicated a lesser enhancement of CO2 related to
fire activity compared to CO. These results are in agreement with previous
measurements of fire biomass burning plumes that showed relatively small
enhancements of CO2 relative to the background in
Amazonia (Andreae et al., 2012). Figure 7 (bottom left)
shows the observed (airborne air sampling at around 2 km above the
ground level) and simulated time series of CO2 mixing ratios. The
major uncertainty of the CO2 mixing ratios is probably most strongly
related to vertical transport, fresh smoke plumes, and uncertainty in
the forest NEE. For example, the misplacement of convective systems of few
grid cells, which is very acceptable for a low-resolution atmospheric model, can
produce huge variations in the CO2 values near the surface. In
addition, the timing of the convection in tropical regions is a well-known
limitation of atmospheric models in general. Nonetheless, the CO2
scatterplots (Fig. 7, bottom right) evidenced much higher variability
in both observed and modeled values compared to CO and a
poorer model representation value close to the ground compared to the upper
levels. The low-level behavior is likely to be associated with local
convective processes but could also have a minor contribution from fresh
biomass burning plumes both venting CO2 and locally changing the
diffuse fraction of solar radiation. By contrast, the model tends to better
represent the upper levels in terms of observed CO2, which is due to
the fact that air circulation is more intense and mainly controlled by the
Carbon Tracker boundary conditions, and the fire emission contribution becomes
even less significant. However, the model is not sensitive to the CO2
concentration within the given range, and therefore the model problems in
reproducing point observations should not have a major effect on the final
results.
Monthly mean AOD at a 550 nm wavelength for September 2010
from the
(a) MODIS Aqua retrieval and (b) from the model as simulated in the DIR+DIF
experiment.
Biomass burning regional plume AOD
In Fig. 8, we show the mean regional biomass burning plume for September
2010 through the monthly mean AOD at a 550 nm wavelength from both the
MODIS Aqua retrieval (Fig. 8a) and the DIR+DIF model simulation
(Fig. 8b). A substantial portion of central Brazil, neighboring
countries, and the southern Amazon Basin were covered by smoke with a
resulting monthly mean AOD higher than 0.5, which is 3 to 5 times larger than
the typical values of clean conditions. Moreover, there were also large
sub-areas with a monthly mean AOD higher than 1, indicating persistent and
high loading of smoke particles. The model results fairly reproduced the
spatial distribution of the regional smoke plume from MODIS retrievals. A
scatterplot of AOD values from the model and MODIS retrieval (Fig. S5 in
the Supplement) presents a slope of 0.71 (with R2=0.73).
Conversely, MODIS retrievals tend to overestimate AOD in relation to the
AERONET retrievals in Amazonia, especially for high aerosol
loadings (Hoelzemann et al., 2009). Analysis of model results versus AERONET
retrieval in some sites in southern Amazonia (not shown)
confirmed that the order of magnitude of the model underestimation is about
the same 20 % as previously estimated.
(a) Mean diurnal cycle of the CO2 (ppmv) mixing ratio
(black line) during September 2010 in Tapajós forest tower at 39.6 m (2.85∘ S, 55.04∘ W; slightly southward
of Santarém; location indicated in Fig. 2). The gray shaded area
indicates the standard deviation of mean values observed. (b) Mean diurnal
cycle of the CO2 (ppmv) mixing ratio from the model during the
same period in the LBAR (red line in Fig. 2). In both plots, model results
are at the 39.3 m model level for the three simulations NO-AER (light
pink), DIR-AER (pink), and DIR+DIF (red). The red error bars on the model
curves refer to the standard deviation of the mean model values.
Site-level plant–atmosphere CO2
exchange
Figure 9a shows the mean diurnal cycle of the CO2 mixing
ratio in the first model layer of the three experiments together with the
mean diurnal cycle of the CO2 mixing ratio just above the canopy of the
Tapajós forest (near Santarém; location indicated
in Fig. 2) from measurements during September 2010. In
Tapajós, both observation and model results present a nighttime
increase in CO2 due to plant respiration, peaking shortly after
sunrise, and a daytime decrease due to photosynthetic processes, with the
lowest values before sunset. Despite the model difficulties in simulating the
CO2 mixing ratio near surface, on average, the discrepancies with
observations were only about 0.9 and 1.4 % for the maximum and minimum
values, respectively, with the model values lower than the observation values during
the peak hour and vice versa when the photosynthetic process dominates. The
model amplitude of the CO2 cycle is lower than that observed, though
the model cycle is still within the standard deviation of the mean observed
diurnal cycle. Model results without including aerosol effects (NO-AER) and
with the inclusion of the direct aerosol effect only (DIR-AER) produce a very
similar CO2 diurnal cycle. However, the inclusion of the diffuse
radiation effects due to biomass burning aerosols reduces the values of
the CO2 mixing ratio and brings model results much closer to observation,
especially during the day, even though the mean AOD modeled in
Tapajós was very low compared to the area mostly affected by
biomass burning and even underestimated by the model. A curious fact is that
at night the difference between NO-AER and DIR+DIF is greater than in the
daytime period. One possible explanation for this is the influence of the
neighborhood. Note in Fig. S6 that the
average wind at 00:00 UTC is from the east, a forest region, and has differences
between aerosol and nonaerosol simulations. However, the wind at 10:00 UTC is
from the northeast and crosses the river, where the influence of the aerosol
in the carbon fluxes is low. Considering the whole LBAR, which includes
the region with the highest aerosol load, the inclusion of the aerosol effect
on CO2, especially the effect of the diffuse radiation, reduced the
CO2 mixing ratio by about 10 ppmv in the CO2 mixing ratio all
day long (Fig. 9b). The shift of the diurnal cycle of the CO2 mixing
ratio from the model relative to the observation in Tapajós is
likely to be related to the AOD underestimation.
In the next section, we will present the model results for energy and carbon
fluxes and explore the role played by biomass burning aerosol in the
carbon cycle in Amazonia.
JULES sensitivity test
We performed sensitivity tests to assess the JULES response to several
atmospheric variables. We ran JULES offline (version 3.0) for September 2010
using as input BRAMS results for the NO-AER experiment considering the
nearest grid box to the tower at kilometer 67. Figure S1 shows monthly variation in downwelling shortwave (RSHORT) and
longwave irradiance (Rlong), air temperature near the surface, and specific
humidity near the surface, all used as input for the sensitivity test. The soil
carbon in this grid cell is 10 kgCm-2 during the entire month. In
addition to
using the BRAMS model results for each parameter, we varied each
parameter by reducing and increasing its original value to cover the standard
deviation of the monthly mean. We also varied the diffuse fraction of
the shortwave radiation, which was originally zero (NO-AER scenario), from 0
to 0.8 of the total radiation. Therefore, we ran 567 simulations for the
month of September 2010. For each simulation, we calculated the monthly mean
fluxes. Figures S2, S3, and S4 show the results for these sensitivity
tests. JULES results for soil respiration, and consequently NEE, are quite
sensitive to the prescribed soil carbon content (Fig. S2). In addition,
the GPP increases with the increase in soil moisture for all biomes (Fig. S3). However, RH and RP
also increase with soil moisture (Fig. S3a and S3m). Therefore, for the forest and cerrado biomes, the NEE
decreases until a certain value and then increases again with
increasing soil moisture (Fig. S3s). In summary, the sensitivity
analyses show that (i) for a 7 % decrease in shortwave radiation there are
minimal changes in GPP (Fig. S4a); (ii) a change in temperature of 1 ∘C (from current midday conditions) did not imply major
changes in the simulated GPP (Fig. S4b); and (iii) a 40 % increase in
the diffuse fraction of shortwave radiation increased the GPP by 39,
71, 4, and 72 % in forest, C3 grasses, C4 grasses, and cerrado
(shrubs) vegetation, respectively (Fig. S4c).
(a) Mean downwelling shortwave irradiance at the surface
(Wm-2) at 16:00 UTC (which is around midday in most of
Amazonia) for September 2010 from the DIR+DIF experiment; (b) the difference in
the mean downwelling shortwave irradiance at the surface (Wm-2)
during the same time period as simulated in DIR+DIF and NO-AER; and (c) the
difference in the 2 m temperature (∘C) during the same time
period as simulated in DIR+DIF and NO-AER. The darker black contour line on
the maps delimits the LBAR.
Impacts of biomass burning aerosol on energy and carbon fluxes
Incoming radiation
The modeled mean downwelling shortwave irradiance at the surface (RSHORT) at
16:00 UTC during September 2010 from the DIR-AER experiment ranged from
900 Wm-2 in the southwestern Amazon to 1000 Wm-2 in the
northeastern portion (Fig. 10a), with the biomass burning aerosol
direct impact (ΔRSHORT=RSHORTDIR-AER-RSHORTNO-AER)
reaching -100 Wm-2 in biomass burning areas (AOD > 0.5; Fig. 10b). As a consequence of the biomass burning aerosol direct effect, the
2 m temperature decreased by 1.2 ∘C on average in the biomass
burning areas around midday (ΔTemp=TempDIR-AER-TempNO-AER,
Fig. 10c). The noise in the northwestern region for both RSHORT and
temperature differences within the two simulations is related to expected
nonlinear aerosol perturbations on cloud distribution. These results are
consistent with previous modeling studies (Rosário et al., 2013) and with
estimations based on AERONET measurements (Procópio et al., 2004).
Additionally, observations in Tapajós during the dry season
indicate an average reduction of 80 and 123 Wm-2 for AOD >0.5 and
AOD >0.7, which corresponded to a decrease in the mean temperature of 0.26
and 0.41 ∘C, respectively.
(a) Mean PAR (µmolm-2s-1) at 16:00 UTC for
September 2010 from the DIR+DIF experiment and (b) the mean diffuse fraction of
solar radiation at 16:00 UTC for September 2010 as simulated by
DIR+DIF. The darker black contour line on the maps delimits the LBAR.
The presence of biomass burning aerosol in the atmosphere also impacts the
flux of PAR in Amazonia during the dry season. Monthly mean PAR
(µmolm-2s-1) at 16:00 UTC, which is around midday in most
of Amazonia, as simulated by the DIR+DIF model experiment for
September 2010 is depicted in Fig. 11a. The modeled PAR monthly mean
values at 16:00 UTC ranged between 900 and 1000 µmolm-2s-1 from southwest to northeast in the LBAR. The presence of biomass
burning aerosol increases the diffuse fraction of radiation by up to 40 % in
the biomass burning areas (Fig. 11b). Figure 12a shows the diffuse PAR as
a function of AOD.
(a) The diffuse PAR radiance (µmolm-2s-1) versus
AOD at 550 nm as simulated in the DIR+DIF experiment at 16:00 UTC during
September 2010 (temporal) in the LBAR (spatial). (b) The decrease in 2 m
temperature versus the decrease in downwelling shortwave irradiance
(Wm-2) as simulated in the AER-DIR and NO-AER experiments at 16:00 UTC
during September 2010 in the LBAR. The color scale refers to soil
moisture (m3m-3) at 0.35 m.
Decreases in surface temperature due to the direct effect of aerosol are also
influenced by the balance between latent and sensible heat fluxes or,
ultimately, on soil moisture. The difference in 2 m temperature (ΔTemp) and shortwave irradiance (ΔRSHORT) is, as expected, highly
correlated, though with a large band of ΔTemp for the same value of
ΔRSHORT (Fig. 12b). The ΔTemp bandwidth increases almost
linearly with the ΔRSHORT, with the values corresponding to higher
soil moisture populating the lower part of the curve (Fig. 12b). This means
that for regions with the same AOD, the ones with drier soil will suffer
higher surface cooling.
Carbon fluxes for the vegetation types in
Amazonia
Spatial fields for simulated GPP across the
northern part of South America are presented in Fig. 13a for September
2010 at 16:00 UTC for forest, C3G, C4G, and shrubs in rows 1–4,
respectively. Over forest, simulated GPP ranges from 20 to 25 µmolCm-2s-1, while over the regions occupied by cerrado,
pastures, and tinges of forest there was much higher variability, with
GPP widely varying from below 5 to above 20 µmolCm-2s-1. In column (b) of Fig. 13, we show the difference between monthly
mean GPP as simulated for the DIR+DIF and NO-AER experiments, i.e., the
relative impact of the total effect of aerosols on simulated GPP for the four
studied biome types: forest (b1), C3G (b2), C4G (b3), and
cerrado (b4). In column (c), we show the difference between the monthly
mean GPP of the simulation without the aerosol effect on the diffuse
radiation (DIR-AER) and the simulation without any aerosol effects (NO-AER);
i.e., we evaluate the relative impact on the direct solar radiation effect.
Mean GPP (µmolCm-2s-1) for September 2010 at 16:00 UTC as simulated in DIR+DIF (column a), the difference in the monthly
mean GPP (µmolCm-2s-1) as simulated in DIR+DIF and NO-AER
(column b), and DIR-AER and NO-AER (column c). The first to the
fourth rows present the results for the biomes of forest, C3G, C4G, and shrubs,
respectively. The darker black contour line on the maps delimits the LBAR.
Monthly mean values of net GPP, ΔGPPtot, and ΔGPPdir in the
LBAR during September 2010 for the three simulations and different biomes.
Biome
GPPDIR+DIF
ΔGPPtot
ΔGPPdir
[TgCmonth-1]
[TgCmonth-1]
[%]
[TgCmonth-1]
[%]
Forest
1206
293
32
8
1
C3G
850
195
30
24
3
C4G
2431
200
9
-69
-3
Cerrado
359
59
20
12
3
NEE measurements during the dry season in several locations in the Amazon region, Brazil. The mean values and standard
deviations of NEE from the modeling results from the DIR+DIF experiment (September 2010) are also presented in the last column.
NEE from flux measurements
NEE from the model (Sept 2010)
Data
Daytime
Daytime
Biome
collection
Daily total
Nighttime
peak
Daily total
Nighttime
peak
Location
Type
Period
molCm-2 day-1
µmolCm-2 s-1
µmolCm-2s-1
molCm-2 day-1
µmolCm-2 s-1
µmolCm-2 s-1
Jarua
Forest
Sept 1992
-0.090
–
–
-0.065 ± 0.002
+5.2 ± 0.8
-11.9 ± 2.4
Jarub
Dry season 1999–2002
-0.069
+7.1
-17.5
FNSc
Pasture
Dry season 1999–2002
-0.12
+3.0
-13.2
-0.14g± 0.1
+3.3 ± 0.8
-13.1 ± 5.4
Tapajósd
Forest
Sept 2002
+0.017
–
–
-0.032 ± 0.039
+5.6 ± 0.9
-11.9 ± 0.6
Sept 2003
+0.026
–
–
Sept 2004
-0.017
–
–
Sept 2005
-0.069
–
–
Sinope
Forest
Dry season 2005–2006
+0.008 ± 0.029
+5.2 ± 0.4
–
-0.23 ± 0.006
+3.0 ± 0.3
-14.4 ± 1.0
Dry season 2006–2007
-0.013 ± 0.024
+5.5 ± 0.4
–
Dry season 2007–2008
-0.041 ± 0.022
+5.6 ± 0.3
–
Cuieirasf
Forest
Dry season 1999–2009
–
∼+4.0
-20.0
+0.037 ± 0.015
+6.8 ± 1.0
-11.7 ± 1.4
a Jaru reserve (10∘05′ S, 61∘57′ W; Grace et al., 1995).
b Jaru reserve (10∘05′ S, 61∘57′ W; Von Randow et al., 2004).
c Fazenda Nossa Senhora (10∘45'44′ S,62∘21'27′ W; Von Randow et al., 2004).
d Tapajós National Forest (2∘51′ S, 54∘58′ W; Hutyra et al., 2007).
e 50 km NE of Sinop, MT (11∘25′ S, 55∘20′ W; Vourlitis et al., 2011).
f Cuieiras Biological Reserve (2∘36'33′ S, 60∘12'33′ W; Cirino et al., 2014).
g The pasture site was identified as C4 grass type.
In Fig. 14a, one can note that the mean solar irradiance that reaches the
Amazon region during September promotes a high GPP for the C4 plants even
into the local afternoon when C3 plants close their stomata to reduce water
loss. Thus, typically, the GPPs of the C4 plants are highly correlated
with the amount of irradiance received. Therefore their GPP resembles the
same diurnal cycle shape of the irradiance. By contrast, the other vegetation
types suffer, to a lesser or greater extent, a decrease in carbon assimilation
during the period of maximum irradiance, thereby reshaping their GPP
diurnal cycle. Still in Fig. 14a, the net increase in GPP due to biomass
burning aerosol over forest (ΔGPPtot; difference between the
curves in the red and green filled squares) is 3.8 µmolCm-2s-1, with the majority
(%Fluxdiff≅94%) of the impact related
to the increase in the diffuse fraction of solar radiation. The reduction in
direct solar radiation by biomass burning aerosols increases forest GPP
only up to 0.2 µmolCm-2s-1 (ΔGPPdir), which is
associated with the cooling of the leaves. Over cerrado areas, the
increase in GPP was up to 0.9 and 0.1 µmolCm-2s-1 due
to the aerosol effect on the diffuse fraction of radiation (ΔGPPdiff)
and the direct radiation (ΔGPPdir), respectively,
with the aerosol direct radiative effect much lower than the diffuse
radiation effect because the GPP of the cerrado is also severely
limited by the excess of irradiance. In the case of the C4 grass type, which
was not limited by irradiance, the direct radiative aerosol effect induced a
reduction in the GPP (-0.7 µmolCm-2s-1), but the
increase in the diffuse fraction of radiation more than compensates for the
reduction in GPP due the irradiance attenuation of the direct effect.
When including both direct and diffuse radiation effects, the GPP increases
from 43 to 47 µmolCm-2s-1. Table 2 summarizes the integrated values of GPP for each biome in
the LBAR during September 2010 and the variation related to the total
aerosol effect (both on diffuse radiation and direct radiation; ΔGPPtot)
and only with the direct aerosol effect (ΔGPPdir).
According to the model results, the net GPP of the forest biome in the LBAR
was 1206 Tg C during September 2010. The presence of biomass burning
aerosol was responsible for an increase of about 32 % of the GPP over the
forest, mainly associated with the impact of the aerosol on the diffuse
radiation. For the cerrado and C3 grass, the net GPP was 359 and
850 Tg C in the same region during the same period, with the biomass burning
aerosol acting to increase the GPP by about 20 and 30 %, respectively. We
estimated an average increase of 27 % in GPP for September 2010 in the LBAR
associated with the aerosol effect in Amazonia (Table 4).
However, Rap et al. (2015), using the JULES model forced with aerosol field from
another model, estimated an average increase in GPP of only 2.8 % for
August considering the period 1998–2007. Also, our estimate of net
primary production (NPP =GPP-RP) for the DIR+DIF simulation was
553 TgC month-1 (1113–560) and 363 TgC month-1 ((1113–240)–(560–50)) for NO-AER.
Therefore, we estimate an increase of 52 % in NPP for September 2010 due to the
aerosol in the LBAR, while Rap et al. (2015) estimated a increase in NPP
of only 5.4 % for August. Our results for the aerosol impact over
Amazonia is higher than the Rap et al. (2015) estimation.
However, one must keep in mind that the Rap et al. estimation was based on 9 years (1998–2007) and for a month (August) that typically has much lower
aerosol loading than September. Our work was based on September, the
peak for the biomass burning season, and 2010, a drier and smokier year.
The diurnal cycle of plant respiration for each biome is shown in Fig. 14b
for the three model runs. As expected, higher GPP leads to higher plant
respiration. Plant respiration peaks are 7.5, 2, 4.6, and 14 µmolCm-2s-1 for forest, cerrado, C3 grass types, and C4 grass types,
respectively, with the aerosol impact more pronounced for forest and C4
biomes. The mean soil respiration found in the LBAR is 2.78 µmolCm-2s-1, with a relatively mild diurnal cycle that basically depends
on soil temperature and hence has lower values in the morning and a tendency to
increase slightly in the afternoon. However, soil respiration is highly
variable in the LBAR, depending on the amount of carbon in the soil, which is as
low as 0.13 µmolCm-2s-1 in the cerrado and
grass plot areas and ranges between 2 and 8 µmolCm-2s-1
in forest areas. Regarding NEE, model results show that the forest biome
released around +5 µmolCm-2s-1 during the night and early
morning, and then when GPP compensates for the respiration, the net uptake
goes as low as -11 µmolCm-2s-1. The net effect is an
uptake of approximately 0.015, 0.565, and 0.060 molCm-2day-1
for the C3 grass, C4 grass, and forest biomes, respectively, and a release of 0.126 molCm-2day-1 for the cerrado. Grace et al. (1995),
with measurements of CO2 fluxes at the Jaru Reserve in Rondônia, Brazil
(10.08∘ S, 61.94∘ W) in September 1992, estimated
an accumulation of 0.09 molCm-2day-1 for forest during the
dry season. So, the forest uptake estimation based on our modeling results is
approximately 30 % lower than the estimations based on the Grace et al. (1995)
measurements. However, several measurements in the Amazon indicated both high
yearly and regional variabilities around the Amazon Basin due to several
factors, which include hydric stress, aerosol loads, topography, and differences
in soil carbon and forest physiology. Also, previous studies indicated that
there are high uncertainties in the magnitude of nocturnal NEE measurements
because of the lack of turbulence in and above the forest (Araujo et al.,
2002, 2010). In Table 3 we collected from the literature some
estimates of NEE (daily total, nighttime and daytime peak) based on
CO2 flux measurements in different sites in Amazonia
during the dry season in different years. For the same sites, we also
present in Table 3 the NEE from the DIR+DIF experiment for September 2010.
For example, measurements taken at the Jaru Reserve (the same site used by
Grace et al., 1995) and at a grass plot (Fazenda Nossa Senhora (FNS) at
10.762∘ S, 62.358∘ W) during the dry season
from 1999 to 2002 revealed an uptake of around 0.12 and 0.069 molCm-2day-1 for the pasture and forest site, respectively, with the
diurnal values of NEE reaching -13.2 and -17.5 µmolCm-2s-1 for the pasture and forest, respectively (von Randow, 2004).
More recently, Cirino et al. (2014) reported 10 years of CO2 flux
measurements carried out in the central Amazon Cuieiras Biological Reserve flux tower:
K34 LBA (Large Scale Biosphere-Atmosphere Experiment in
Amazonia; 2.609∘ S,
6.209∘ W) from 1999 to 2009 and also in the Jaru Reserve from
1999 to 2002. The measured diurnal cycle of NEE at both sites was, as
expected, positive during the nighttime and negative during the daytime.
During the dry season, the mean values of NEE measured during the night
were approximately +5.2 (±0.8) and +6.8 (±1.0) µmolCm-2s-1 at the Jaru (Von Randow et al., 2004) and Cuieras (Cirino et al.,
2014) reserves, respectively, due to differences in the physiology of the
forest and possibly the topography in the two sites. There were even more
significant differences between the maximum values of carbon uptake between
the two sites, which were around -17.5 and -20 µmolCm-2s-1 under biomass burning and cloudy sky conditions at the Jaru and
Cuieras reserves, respectively, and -18 µmolCm-2s-1 under
clean skies in both sites. The differences between the peak values of carbon
absorption are likely related to the presence of biomass burning, the variability
of cloudiness and rainfall, and soil characteristics, like water content and
nutrients (N, P). Measurements at the Tapajós National
Forest (2.85∘ S, 55.04∘ W) from 2002 to 2005 (during
September) led to estimations of NEE varying from +0.017 to -0.069 molCm-2day-1 (Hutyra et al., 2007). The mean value of NEE
from the model at Tapajós was -0.032 molCm-2day-1 for September 2010. The Jaru Reserve is systematically and intensely
affected by both local and long-range transported biomass burning aerosols,
with monthly means for AOD (550 nm) during the dry season typically above
0.5 but often above 1.0; in contrast, the biomass burning affects in the
northern part of the basin, where Cuieras and Tapajós are
located, are more episodic (Longo et al., 2009). In addition, the northern
part has more variability in terms of rainfall during the dry season compared
to the southern part, due mainly to the position of the Inter-Tropical
Convergence Zone (ITCZ). So, strong variability in terms of carbon
fluxes in the northern part of the Amazon is indeed expected and is really
challenging for a low-resolution model to match point measurements.
Nevertheless, our modeled monthly mean diurnal cycles of NEE for forest and
pasture biomes (Fig. 14c) are remarkably close to the diurnal cycle
reported for the Jaru Reserve and FNS, respectively, by von Randow et al. (2004), with the daily total, nighttime,
and daytime peak values (Table 3) of a similar order of magnitude within the variability observed.
Mean diurnal cycle of (a) GPP (µmolCm-2s-1),
(b) RP (µmolCm-2s-1), and (c) NEE (µmolCm-2s-1) during September 2010
for the different biomes in the LBAR.
Different symbols indicate the biomes of forest (filled squares), C3G
(hollow circles), C4G (hollow squares), and shrubs (filled circles).
Different colors indicate the modeling experiments: green is for NO-AER, blue
is for DIR-AER, and red is for DIR+DIF.
Scatterplots of GPP (µmolCm-2s-1) and PAR
(µmolm-2s-1) from the DIR+DIFF experiment for forest (filled
squares) and shrub biomes (filled circles) and grass types C4 (hollow squares) and
C3 (hollow circles) during September 2010 (temporal) in the LBAR (spatial).
The color scale depicted indicates the fraction of diffuse radiation. The
data were filtered for a soil water factor above 0.9.
Figure 15 depicts the model results of the DIR+DIF simulation for GPP
(µmolCm-2s-1) in response to the available PAR (µmolm-2s-1) for the four main different biome types in the domain
of study and in a range of 0.2 to 0.6 for the fraction of the diffuse
irradiance. The analysis is also limited to conditions in which the soil
wetness is above 0.9. The maximum values of GPP for forest and C3
grass are reached with PAR around 1600 and 1300 molm-2s-1, respectively, indicating that these are the
saturation point for these biomes relative to the amount of energy reaching
the surface. For cerrado (shrubs), the saturation point is much
lower, around 600 molm-2s-1, but the plants maintain their
carbon assimilation rate up to PAR around 1600 molm-2s-1
and
only then decrease with a further increase in PAR. For the C4 grass,
GPP increases almost linearly with PAR, not showing any evidence of
saturation with the amount of energy received. In general, the results in
Fig. 15 indicate that for a given biome and PAR value, a higher amount
of diffuse radiation implies higher GPP. However, the cerrado
(shrubs) biome is an exception since it saturates with relatively lower values
of PAR.
Mean net CO2 fluxes (µmolCm-2s-1) weighted
per biome type for September 2010 at 16:00 UTC (column a). The
differences in the mean CO2 fluxes as simulated in DIR+DIF and NO-AER
(column b) and in DIR-AER and NO-AER (column c) during the same time period.
The first to the fourth rows indicate the results for GPP, RP, RH, and
NEE processes, respectively. The darker black contour line on the maps
delimits the LBAR.
The mean spatial distribution of the relative impact of aerosol effects on
modeled fluxes at 16:00 UTC is shown in Fig. 16, considering the
changes both on direct and diffuse radiation and only on direct solar
radiation. In the LBAR where the forest biome dominates, there is an
increase in GPP (Fig. 16b1) ranging from 0.1 to 5.0 µmolCm-2s-1 related to the aerosol effect, the lower values being
associated with lower AOD values (Fig. 8) and drier soil (Fig. 5). In the
remaining regions, the aerosol impact on GPP is still positive but much
lower (0.1–0.5 µmolCm-2s-1). In all domains, the
majority of the aerosol impact is related to the increase in the diffuse
fraction of solar radiation due to the presence of aerosols
(%GPPdiff>95%). As a general rule, our model results indicate that the
increase in GPP leads to an increase ranging between 0.2 and 1.6 µmolCm-2s-1 in plant respiration (RP) for the forest and does
not affect the other biomes significantly. On the other hand, soil
respiration (RH; row three in Fig. 16) varies from 1 to 8 µmolCm-2s-1, with the higher values in the forest area with higher soil
moisture (Fig. 5). The aerosol impact on RH is somewhat noisy, varying
from -1 to +1 µmolCm-2s-1 over the forest and negative
otherwise (∼ -0.1 µmolCm-2s-1). The noise of RH is
associated with the nonlinear effects of aerosol on cloudiness (and thus
temperature) and precipitation. The total impact of aerosol in NEE (row four
in Fig. 16) ranged from -0.1 to -5 µmolCm-2s-1, with
the lower values found in the forest region with intense and/or persistent biomass
burning mainly related to the diffuse radiation effect, meaning that our
modeling suggests that the aerosol biomass burning effect creates a
CO2 sink in Amazonia.
(a) Mean diurnal cycle of CO2 fluxes (µmolCm-2s-1) during September 2010 in the LBAR, with different symbols
indicating the processes of GPP (hollow triangles), RP (plus symbols), and RH
(filled triangles). (b) Mean diurnal cycle of CO2 fluxes (µmolCm-2s-1) associated with NEE during the same period and in the
same area. In both plots, different colors indicate the modeling
experiments: green is for NO-AER, blue is for DIR-AER, and red is for DIR+DIF.
Total CO2 assimilation ratio for the DIR+DIF simulation, ΔFluxtot, and
ΔFluxdir in the LBAR during September 2010 for the different atmosphere–biosphere exchange processes.
Process
CO2 Flux
ΔFluxtot
ΔFluxdir
[TgCmonth-1]
[TgCmonth-1]
[%]
[TgCmonth-1]
[%]
GPP
1113
240
27
1
0
Rp
560
50
10
0
0
RH
449
-14
-3
-7
-2
NEE
-104
-205
-203
-8
8
Total carbon fluxes in
Amazonia weighted for the vegetation types
Figure 17 depicts
the monthly mean diurnal cycle of the CO2 fluxes in the LBAR for
September 2010, again related to GPP, RP, RH, and NEE averaged
over the four types of vegetation present in each atmospheric model grid box.
The presence of the biomass burning aerosol affects the CO2 flux
associated with GPP, and consequently Rp. Responding to the increasing
diffuse radiation, both GPP and RP rise, with a GPP enhancement
about 4 times higher than the RP. In contrast, RS has an opposite response
as the presence of aerosol implies a cooler soil (Fig. 10) and
consequently lower microbial activity. The net effect is a higher
CO2 daytime uptake with negligible nighttime variation. Moreover,
the biomass burning aerosol strongly impacts the NEE (Fig. 17b). Around
noon, the NEE decreases from -7 to -10 µmolCm-2s-1 in
the presence of biomass burning, mainly due to the diffuse radiation effect.
Nevertheless, it is interesting to note that the impact of the aerosol
influence on the relative contribution of the diffuse to the total (diffuse + direct) on the NEE (Eq. 4) has different behavior depending on plant
functional type and decays exponentially as the AOD increases for all biomes,
except for the C4 grass type. The contribution of the diffuse radiation
effect to NEE (ΔNEEdiff/ΔNEEtot) versus AOD for each
biome is depicted in Fig. 18 along with its fitting functions. Over
forest, the percentage of the diffuse radiation effect on CO2 uptake
decreases exponentially ([ΔNEEdiff/ΔNEEtot]forest≈e-0.9AOD; R2=0.7)
from 100 to 50 % with the increase in
aerosol loading, reaching a balance of 50–50 % between the diffuse and
direct effect for AOD above 0.5. For C3 grass and cerrado, as
expected, the contribution of the diffuse radiation effects tends to zero
with the increase in AOD
([ΔNEEdiff/ΔNEEtot]cerrado,C3≈0.7e-4AOD; R2=0.7). For the C4 grass type, the contribution of the
diffuse radiation to NEE exponentially increases with AOD
([ΔNEEdiff/ΔNEEtot]C4≈eAOD; R2=0.9),
and the C4 photosynthetic pathway does not rapidly saturate with the amount of
light received. Considering the AOD underestimation of about 20 % and the
exponential behavior of the relative contribution of the diffuse fraction to
NEE, it is reasonable to say that the contribution of the diffuse radiation
effect on CO2 uptake can reach 40 % over the forest and 10 % over
cerrado and the C3 grass type for high aerosol loads.
The contribution of the diffuse radiation effect to NEE
(ΔNEEdiff/ΔNEEtot) as a function of AOD in the
LBAR, but separated with different colors for different types of vegetation.
The model data were filtered for cloudiness and precipitation. Additionally,
only model points with the same soil water factor for all three
experiments and a soil moisture difference below 0.001 m3m-3 were
included. The fitting functions of the
ΔNEEdiff/ΔNEEtot versus AOD for each biome are
also shown in the figure.
The model results for the CO2 fluxes integrated for the month of
September (2010), which is the peak of the burning season in the LBAR, are
summarized in Table 4. Total modeled GPP in the LBAR is 1113 Tg C with
the aerosol being responsible for an increase of 240 Tg C, with less than 1 %
due to the aerosol radiation direct effect. Plant respiration is affected by
approximately 50 Tg C, related only to the increase in the diffuse fraction
of radiation. The impact of the aerosol on the soil respiration is only 3 %
but in the opposite direction, i.e., a reduction. Integrating throughout the
full month for September 2010, the NEE changed from +101 to -104 Tg C
when the aerosol effect is considered. The total aerosol effect on radiation
was responsible for about 96 % of the NEE change, while the temperature
reduction due to the direct aerosol effect on radiation accounts for only
5 %. That is, the aerosol effect, especially the change in the diffuse
fraction of radiation, is strong enough to invert the signal of NEE,
changing the ecosystem from being a source to a sink of CO2. Table 3
shows that the NEE observed during the dry season at the Amazon forest and
pasture biomes exhibits substantial site-to-site and interannual variability.
Nevertheless, for each site, the 2010 model results are within the observed
variability.
Conclusions and final remarks
We conducted a modeling study during the peak of the burning season in
Amazonia to assess the ability of a current state-of-the-art
integrated in-line numerical atmospheric modeling system to simulate the
CO2 fluxes in Amazonia. A set of three different
modeling experiments was conducted: first totally disregarding aerosol biomass burning
effect, then considering only the direct aerosol effect, and finally also
adding the aerosol effect on the diffuse fraction of radiation. The model
results allowed us to assess and quantify the impacts of biomass burning
aerosols on CO2 fluxes in the Amazon Basin during the dry season.
Moreover, the relative role of the main soil, vegetation, and atmosphere
interaction processes controlling the carbon cycle in Amazonia
was weighed, and the aerosol effect on each of them was measured separately.
Consistent with previous studies (Freitas et al., 2005, 2009, 2017; Longo
et al., 2010, 2013; Rosário et al., 2013; Moreira et al., 2013), BRAMS
performed well while modeling the meteorology and aerosol biomass burning
emission, transport, and removal processes in Amazonia, which has
resulted in accurate simulation of the major features of AOD variability
associated with the regional biomass burning plume over South America. The
model results for surface temperature, rainfall, and AOD were once again in
agreement with observations for the 2010 dry season case study, representing
the main characteristics of the spatial distribution and the diurnal cycle of
temperature and precipitation. BRAMS was also evaluated on its performance to
simulate CO and CO2 mixing ratios using measurements acquired
from air samples collected using light aircraft over the Amazon during the 2010
and 2011 burning seasons. Typically, the model tends to slightly
underestimate the CO mixing ratio, particularly in the lower levels,
in regions affected by fresh biomass burning and haze biomass burning layers.
Previous studies had already indicated an underestimation of the biomass
burning emissions database used in this work (3BEM; Longo et al., 2010) of
about 20 % (Andreae et al., 2012), mainly related to fire omission and
misrepresentation of the vegetation and carbon maps used (Pereira et al.,
2016). For CO2 mixing ratios, the comparison between model and
observation is highly scattered, again especially in the lower levels, though
in this case more likely related to convective activity pumping CO2
to the upper layers of the atmosphere and inaccurate modeling of surface
carbon net fluxes (NEE). In both cases, model inaccuracies are at
least partially related to the lower model resolution (20 km), suggesting
that further sensitivity studies on model resolution would be helpful.
Nevertheless, although the 20 km model resolution was not capable of
capturing CO2 point measurements in Amazonia, the order
of magnitude of the CO2 mixing ratio has been in general well
represented. Moreover, the diurnal cycle of CO2 measured above the
canopy of the Tapajós forest was represented in the model with
differences of only about -0.9 and +1.4 % between model results and
observations during the time of minimum and maximum values, respectively.
Our modeling results indicate that during the dry season in
Amazonia, regions with lower precipitation do not always have
high values of NEE because the lower soil respiration of a dryer soil can
compensate for the deficit of water available for plants (e.g., Saleska et al.,
2003). Being an equatorial region, Amazonia receives abundant
PAR. Therefore, areas with plenty of water availability in the soil have
higher GPP compared to dry soil areas. However, after noon local time, when
the radiation excess typically occurs, there is a drop in carbon assimilation
for all biomes, except for the C4 grass type that has maximum assimilation
coinciding with the peak of PAR.
The presence of an intense biomass burning aerosol layer during the dry
season over Amazonia reduces the solar energy reaching the
surface, consequently reducing near-surface temperature. The model results
show this cooling effect contributing to increasing the GPP in regions
covered by forest, grass C3, and cerrado. However, in addition to
reducing the surface energy, the aerosol layer also increases the diffuse
fraction of radiation. This is the major effect that contributes to
increasing GPP and, in this case, including the C4 grass type biome. These
two effects together increase GPP by about 32, 30, 9, and 20 % for
forest, C3 grasses, C4 grasses, and cerrado, respectively.
In the LBAR, the GPP increased about 27 %, reaching 1113 TgC during
September 2010, when the aerosol effects were included. Plant respiration
also increased from 510 to 560 TgC, with the aerosol biomass burning effect
as a response to the increase in GPP. The more CO2 the plant
assimilates to produce sugar, the more it needs to increase its respiration
for energy supply. On the other side, soil respiration dropped from 463 to
449 Tg C. Consequently, the NEE in the LBAR during September 2010 dropped
from +101 to -104 TgC when the aerosol effects were considered, mainly due to
the diffuse radiation effect. That is, the LBAR during the dry season, in the
presence of high biomass burning aerosol loads, changes from a source to
a sink of CO2 to the atmosphere. These results are also consistent
with the observations of Yamasoe et al. (2006), who found no correlation
between NEE and aerosol load for low AOD values (< 0.7); however, for
AOD > 0.7 NEE values became negative, and for AOD> 1.5–2 NEE started to
increase again. Our model results also indicate that the impact of the
aerosol on the NEE change is mainly related to the aerosol increasing the
diffuse fraction of radiation. For AOD higher than 0.5, the forest reaches a
balance of 50–50 % between the diffuse and direct aerosol effects. For
the C3 grass type and cerrado, as expected, the contribution of the
diffuse radiation effect is much lower than for the forest biome and tends to
near zero with the increase in AOD. Direct measurements at the
Tapajós site (Doughty et al., 2010) led to an estimation of the
relative aerosol contribution in CO2 uptake, for high values of AOD,
of 80 % as a result of increased shaded light in the sub-canopy related to
the effect of aerosol increasing the diffuse fraction of radiation.
Only 20 % of the aerosol impact on CO2 uptake was attributed to the
decrease in canopy temperature. These same authors, however, do recognize
that is “difficult to know whether this proportion is applicable to forest
biomes worldwide or limited to tropical forest”. So, based on our model
results, we go even further and say that it is difficult to even to affirm
that there is a unique rule applicable to the entire Amazon forest due to its
high diversity of plant and soil characteristics and microclimates.
Considering that the fire activity in Amazonia typically lasts
for about 3 months, we can estimate as a first approximation that the net
impact of biomass burning aerosols on the carbon cycle in
Amazonia is about -615 TgC yr-1. However, we must say that
the fire activity in 2010 was very intense (see Fig. S7), and therefore this estimation is not likely to be
representative of an average year. According to Espírito-Santo et al. (2014), the impact of the natural disturbance in the carbon cycle in
Amazonia is generally around 1300 TgC yr-1. Thus, the
aerosol (negative) impact can be of a similar order of magnitude as the
(positive) impact of the natural disturbances in the carbon cycle in
Amazonia.
Our model results emphasize the importance of considering the effects of
aerosol in numerical models of climate forecasting, especially when
investigating the intensification of the greenhouse effect due to the
atmospheric CO2 concentration. In general, the numerical results
obtained were in good agreement with observational data, including
meteorological, aerosol, and trace gas variables, which gives us confidence
in the estimation of the carbon fluxes. However, we do recognize that
including the effect of cloudiness on the diffuse fraction of radiation is an
essential model capability that will allow us to explore the relative impact
of biomass burning aerosol and clouds as well as the seasonality and
annual variability of the carbon cycle in Amazon. This is a work in
progress and we will soon report the inclusion of the cloud effect on the
diffuse fraction of solar radiation in the model, which certainly has a major
effect on the CO2 budget in Amazonia during the wet
season.
In addition, further model development based on the current level of knowledge
could still improve the representation of biomass burning aerosol effects in
the carbon cycle. As such, model studies that include a reduction in
photosynthesis due to the oxidation of plant leaves by high levels of ozone
secondarily produced in biomass burning plumes and the indirect
aerosol effect on CO2 is also a work in progress.