MODIS fire products

The MODIS fire observations serve to advance global monitoring of fire process and its effects on ecosystems, the atmosphere, and climate. The MODIS instrument acquires data 4 times a day from the Terra (10:30 am and 10:30 pm) and Aqua (1:30 pm and 1:30 am) satellites. We use MOD 14, the level 2 swath data provided daily at 1-km resolution and includes the MODIS Thermal Anomalies/Fire products, primarily derived from the MODIS channels 21 (4 µm) and 31 (11 µm). The fire detection strategy is based on the absolute detection of a fire (when the fire strength is sufficient to detect) and the detection relative to its background (to account for the variability of the surface temperature and the reflection by the sunlight). The science data sets in this product include the fire-mask, algorithm quality, radiative power, and numerous layers, describing fire pixel attributes.

The MODIS fire radiative power (FRP) is widely used to estimate fire emissions. Val Martin et al¹⁶ showed the correlation between plume heights and the MODIS FRP. They reported large FRP (~1600 MW) in the free troposphere and low FRP (<500 MW) within the boundary layer. Freeborn et al¹⁷ showed the MODIS FRP uncertainty in satellite-based active fire characterization and biomass burning estimation, which can limit the confidence in flux estimates at the pixel-by-pixel resolution.

An average of FRP within the domain (Figure 1) is 80 MW with a range of 10 to 2300 MW. Many points more than 500 MW have low confidence (<70%). Kauffman et al (1998)¹⁸ suggested to identify the fire category by the intensity as follows:

The fire category distribution in the study domain described below contains more than 85% that are in C1 and C2. We select FRP with the high confidence (>70) and less than 500 MW. When converting to fire size, the 0.5 km² corresponds to 500 MW. Figure 1 shows the distribution of FRP with the location of the aircraft flight.

Aircraft data

We have used the data acquired by the Atmospheric Radiation Investigations and Measurements group at the National Center for Atmospheric Research (NCAR) during the field project “Studies of Emissions and Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys” (SEAC4RS). The objectives of this mission included studying modeling results and measurements, to investigate how emissions change in deep convective flows under changing dynamical and chemical conditions, and examining the influences of aerosol particles on weather and climate. Two aircraft were involved in the project: the NASA DC-8, providing observations from near the surface to 12 km, and the NASA ER-2, providing high-altitude observations extending into the lower stratosphere. The mission was conducted from August to September 2013. The location of the aircraft track for the case study described here is shown in Figure 1 with the blue line along with the location and FRP of fires in the area.

The spectrally resolved actinic flux was measured by the charge-coupled device actinic flux spectroradiometer (CAFS) systems. The CAFS instruments are primarily used to calculate photolysis frequencies in airborne, chemistry-focused missions, including SEAC4RS. However, the spectrally resolved actinic flux can also be exploited for radiative aerosol properties, as in this study.

Table 1.

The spectral complex refractive index for each smoke component.

Spectral complex refractive indices of smoke aerosols

The light absorption of OC is strongly wavelength dependent across the UV, whereas BC has relatively linear absorption. Several experiments for examining the spectral optical properties of smoke have been performed. Depending on the sampling method, the real and imaginary parts of refractive indices are different not only in magnitude but also in spectral behaviors. Lee and Tien¹⁹ reported that the real part of the refractive index is decreasing to the wavelength of 0.8 µm and increasing to 3 µm, whereas Chang and Charalampopoulos²⁰ showed the continuously increasing real part of the refractive index with the increasing wavelength. Kirchstetter et al¹² showed that OC extracted from biomass smoke samples has the imaginary part of the refractive index decreasing with the wavelength. We define the spectral refractive index for OC, BC, and SO4 (350-700 nm) using the data from the following studies: Chang and Charalampopoulos,²⁰ Kirchstetter et al,¹² and Massie and Hervig²¹ (Table 1). Sulfate aerosols attenuate solar radiation by scattering, which is why no value in the imaginary part of the refractive index of sulfate is given, whereas carbonaceous aerosols absorb and scatter the UV and solar radiation.

The optical properties vary depending on smoke types and their components. Reid et al²² showed the existence of the different fractions of the mass of each component in different types of fires; the BC/OC ratio varies 0.06 to 0.23 for grass/savanna, 0.03 to 0.15 for boreal forest, and 0.08 to 0.2 for tropical forest biomass burning. Also, the fraction between the flaming phase and the smoldering phase forest fires is 40%/60%, savanna/grass is 93%/7%, woody savanna, and cerrado is 75%/25%. We assume that smoke aerosols from the biomass burning consist of BC, OC, and sulfate, and all smoke particles undergo rapid aging within a couple of hours.

The configuration of WRF-Chem-SMOKE

The WRF-Chem-SMOKE model is based on WRF (version 8.1.1), which is widely used for weather forecasting and regional meteorological studies, as well as to simulate gas-phase chemistry, and aerosol-cloud-radiation interactions.^23,24 Studies have shown that WRF-Chem is applicable to a wide range of atmospheric conditions, supported by the in situ and remote sensing data.^25,26 Wang et al²⁶ showed that when FLAMBE smoke emissions are injected within 800 m above the surface, and good agreement can be found between the WRF-Chem and satellite/ground-based observations of the surface particulate matter mass, the aerosol vertical profile, and the smoke transport path.

Table 2 lists the model configuration options considered in this study. The domain of our study is the southwestern United States (125W-90W, 27N-40N) with 350 × 170 grid points and the 9 km grid spacing (red rectangles in Figure 1). The 3-hourly, 32 × 32 km North American Regional Reanalysis (NARR) data are used for the initializing and specifying the boundary conditions. The microphysics and cumulus parameterizations used are the Morrison and the Grell-Freitas ensemble scheme, respectively. The Yonsei University (YSU) boundary layer scheme and unified Noah land surface model are used.

Table 2.

WRF-Chem model configuration with chemical schemes.

The simulation of aerosol processes is performed using the Second-Generation Regional Acid Deposition Model (RADM2) and the Modal Aerosol Dynamics Model for Europe with Secondary Organic Aerosol Model (MADE/SORGAM).^27,28 Because the main reason for using the WRF-Chem for this study is to track smoke from fire emissions, only the background aerosol database is used. No biogenic and anthropogenic emission databases are considered here.

Tropospheric ultraviolet and visible modeling

To estimate the impact of smoke on UV fluxes, we use the 1-dimensional (1D) tropospheric ultraviolet and visible (TUV) radiative transfer model developed at NCAR,²⁹ which is incorporated in the WRF-Chem-SMOKE model. The TUV version 4.2 includes 115 photo-dissociation reactions with the most recent data on the absorption cross sections and quantum yields, based on regular publications of the evaluation panels for the kinetic data from the NASA Jet Propulsion Laboratory and the International Union of Pure and Applied Chemistry (IUPAC). The radiative transfer model was run in a 2-stream approximation and calculations were performed from 350 to 700 nm. We selected the surface albedo of 0.1 for the UV and visible spectral region because the variation of the surface albedo is insignificant in the spectral region for considered surface types.^30,31

The selection of smoke aerosol loading conditions

Fire properties, determined by the emission intensity and the thermodynamic stability of the atmosphere, are used to distinguish between a flaming phase and a smoldering phase. The public version of the WRF-Chem model includes a 1D plume rise model developed by Freitas et al,³² which accounts for these properties. The smoke injection height is calculated by solving governing equations.³² The ability of the WRF-Chem 1D plume rise model to produce realistic injection heights has been demonstrated by Sessions et al.³³

We distinguish between flaming and smoldering phases of biomass burning events based on the injection height of the layer of smoke using a simple scheme: if a plume is located below planetary boundary layer (PBL), it represents the smoldering phase; if a plume is located above PBL, it represents the flaming smoke (illustrated on the bottom of Figure 2).

Figure 2.

The procedure to perform the calculations of radiative impacts of smoke.

An approach to computing spectral UV/VIS fluxes

To estimate spectral actinic fluxes, the WRF-Chem, Mie code (Lorenz-Mie scattering calculation), and the TUV model are used. The smoke aerosol size distribution and complex refractive index were used to develop aerosol optical models within the UV and visible spectral range. We assume that the smoke consists of OC, BC, and SO₄ with the spectral refractive indices shown in Table 1. The index for each component is provided to the Mie code together with size distribution to calculate the spectral optical properties, including single-scattering albedo (SSA) (ω₀), asymmetry factor (g), and extinction coefficient (k_e). Then, the optics of external mixtures is computed by summing up the weighed (based on the proportion assigned from the WRF-Chem-SMOKE simulation) optical properties of individual species. The optical properties are used as an input in the TUV code to simulate actinic fluxes and irradiances. Figure 2 illustrates the procedure to perform the calculations of radiative impacts of smoke.

Validation of simulated smoke plumes with satellite observation data

The simulated smoke from the biomass burning modeled with WRF-Chem-SMOKE is validated before the radiative flux computations are applied. Figure 3 shows MODIS aerosol optical depth (AOD) and simulated AOD for box (a) and (b) in August 16, 2013. There were several fires over southern California and the smoke plumes were transported to Arizona where MODIS measured 0.4 AOD in Arizona (Figure 3A). Figure 3C shows the simulated AOD due to smoke (~0.2) over southern California but the model does not appear to capture the transported plumes. There was high AOD in Louisiana (Figure 3B), matching the fire emission locations (in Figure 1) and the model also simulated high AOD (Figure 3D). Overall, the simulated aerosol spatial distribution matched with the observations, but the correlation by each pixel is low (~0.28) due to the uncertainty of emission data and meteorological fields (eg, wind speed/direction, temperature profiles), as expected. The presence of other aerosol types, ie, the pollution, may also contribute to the low correlation.

Figure 3.

An image of MODIS AOD (top) and simulated AOD (bottom) for box (a) (left) and box (b) (right) on August 16, 2013.

Figure 4 shows PM2.5 mass concentrations representing the simulated smoke plumes on August 16, 2013, and it matches to the fire locations used for the emission data shown in Figure 1. Extremely high FRP in northern California and Louisiana is eliminated because most smoke have small FRP (<40 MW). It is noticeable that high smoke plumes are detected in Louisiana, which is corresponding to high AOD in Figure 3. However, it is apparent that smoke concentration from fire emissions is still insufficient to have comparable AOD with the satellite data. The fire size for the emission inputs as well as meteorological fields must be adjusted for the future study.

Figure 4.

Simulated smoke plume using PM2.5 mass concentration (µm/m³).

Analysis of optical properties of smoke types

The analysis of the spectral optical properties of different smoke types assists to understand spectral aerosol characteristics due to light absorption and scattering. Figure 5 shows the averaged extinction coefficient (k_e) and SSA, asymmetry factor (g) for smoldering and flaming phases over land.

Figure 5.

Spectral optical properties for smoldering smoke (blue) and flaming smoke (red) over the land with the standard deviation on August 16, 2013.

The SSA albedo for the flaming smoke is higher than for the smoldering one across the UV and visible wavelengths due to the difference of the fraction of smoke components. As shown in Table 3, flaming smoke has a larger fraction of nonabsorbing component, sulfate, and a smaller fraction of absorbing components, OC and BC. This relates to the flaming smoke being loaded above PBL, where it can react and be oxidized by other gases and particles (so called “aged smoke”). The extinction coefficient is larger in the smoldering phase in the visible but larger in the flaming phase in the UV. The difference in the asymmetry factor between the 2 smoke types is insignificant because we assumed the same size distribution for the flaming and smoldering phases. Thus, the scattering angle (forward or backward) is similar.

Table 3.

The averaged fraction of the smoke component for smoke pixels.

Comparison of the calculated fluxes with the aircraft data

We select 2 subdomains, box (a) and box (b) in Figure 1, to compare the calculated actinic fluxes with the aircraft data. The aircraft data provide 2 types of actinic fluxes: “nadir” to measure upwelling radiation and “zenith” to measure the downwelling radiation. The net actinic flux defines the sum of direct and diffusive fluxes for all directions. The aircraft passed a location of 32.84N and 115.03W in box (a) at 7.81-km altitude at 17:25 LT and the comparable calculated flux is chosen at the nearest pixel (32.53N, 115.3W) at 7.87-km altitude. As shown in Figure 6A, the calculated net actinic flux (red star) trend increasing with wavelength is well matched with the measurement (black line), which shows the large increment after 400 nm and consistency after 500 nm. The net actinic flux from TUV is underestimated with the largest error in 450 nm (−1014 photons cm⁻² s⁻¹ nm⁻¹), which may be due to less diffusive flux in the atmosphere at the large solar zenith angle (SZA) in the calculation. Simulated AOD is very small (<<0.1), whereas MODIS AOD shows 0.2. The AOD is related to the mass concentration, and the simulated mass concentration is not sufficient to produce the observed actinic flux.

Figure 6.

Net actinic fluxes from the aircraft (black line) and from the calculation (red star) for a location in box (a) (left) and box (b) (right).

Another location for the validation of the calculated actinic flux is 29.84N and 96.02W at 7.1 km at 14 LT from the aircraft and the nearest pixel (31.04N, 93.23W) of the simulation at 7.77 km in box (b). The calculation has agreement with the observation in net actinic flux. The actinic flux with the increasing wavelength at small solar angle is simulated better than at the large solar angle, which implies the uncertainty of the diffuse flux at 14 LT (SZA: 29°) is less. Simulated AOD and MODIS AOD have both high AOD ~0.4.

The differences in actinic flux between the observation and the calculation (Figure 6) may be due to other aerosol components, which were not included in the calculation. Although the model considers only available aerosols due to biomass burning, the measured actinic flux will also be affected by aerosols from other natural and anthropogenic sources. The accuracy of the modeled injection height will also affect the comparison. The simulated flaming smoke is loaded at 2 to 4 km at the considered locations, and actual smoke height is about 5 km at 17:25 from the JPL Digital Image Animation Laboratory (DIAL) images ( https://science.larc.nasa.gov/lidar/seac4rs/data/20130816_bsr_1064_ff.png). Thus, further analysis related to the injection height and smoke components may be required to reduce the uncertainty of the net actinic flux from the simulation.

Analysis of the change of actinic fluxes caused by smoke

The actinic flux (photons cm⁻² s⁻¹ nm⁻¹) is the quantity of light available to molecules at a particular location which drives photochemical processes in the atmosphere. The actinic flux may be represented by 3 components: direct radiation, downward diffusive radiation, and upward diffusive radiation. The downward actinic flux is the sum of the direct solar radiation and the downward diffusive radiation, and the net actinic flux is the sum of all 3 components.

To investigate the smoke impact on the actinic fluxes, we selected a location of 33.19N/100.07W, which had both flaming and smoldering smoke cases in the column. Figure 7 shows the calculated downward and net spectral actinic fluxes for flaming and smoldering smoke cases for different times at the smoke plume altitude. Both flaming and smoldering smoke cases have a similar spectral impact on the flux. However, the difference of the actinic flux between 19 and 60 SZA is approximately 2 times larger for the flaming case than for the smoldering smoke case, which illustrates that light scattering/absorption of flaming smoke is sensitive to solar angle.

Figure 7.

Spectral actinic fluxes (downward: left, total: right) for the flaming smoke case (top) and the smoldering smoke case (bottom) at 33.19N, 100.07W at 0.5 km) for different times (UTC).

Figure 8 shows the changes of spectral actinic fluxes $(\mathrm{\Delta }F(\mathrm{lgr;}))$ for flaming and smoldering types of smoke at different SZAs. Here, we define the change of fluxes as follows:

Figure 8.

The change of spectral actinic fluxes (total) caused by flaming smoke (left) and smoldering smoke (right) (31.04N, 93.23W above smoke height [5 km for flaming and 0.2 km for smoldering smoke]) for different solar zenith angles (SZA = 17.783°, 38.2°, and 57.103°).

A positive ∆F indicates enhanced actinic flux, and negative ∆F indicates reduced actinic flux due to smoke. The type of smoke has different impacts on the radiative forcing in terms of a transition point where ∆F = 0. At the location of 31.04N and 93.23W at 4.5 km (located above the smoke plume height), the flaming smoke reduces actinic fluxes at UV wavelengths and increases the flux at visible wavelengths. The wavelength of the transition point of ∆F becomes shorter with increasing solar angle in the flaming case, which means the flaming smoke aerosol cause more diffusive radiation by the light scattering at the larger solar angle. The impact of smoldering smoke on the flux is relatively small at this location due to the low AOD (<0.1), resulting in small ∆F. At visible wavelengths, ∆F is positive at small solar angle (SZA = 17.783) and negative with increasing solar angle, indicating that the light absorption by aerosols dominates.

The impact of smoke at selected wavelengths is illustrated in Figure 9. It shows the vertical profile of ∆F at 13:20 LT for flaming and smoldering cases. Flaming smoke was loaded at 2 to 4 km at of 31.04N and 93.23W, and ∆F values increase above the smoke altitude. The flaming smoke case has an impact more than 10 times larger than for smoldering smoke. While actinic fluxes are reduced in the UV wavelength, actinic fluxes are increased in the visible wavelength due to absorption by OC. Also, the flaming smoke case significantly changes the actinic flux with altitudes because the flaming smoke is loaded above the boundary layer.

Figure 9.

Vertical profiles of ∆F for flaming smoke (left) and smoldering smoke (right) (31.04N 93.23W) for the different wavelengths at SZA = 17.78°.