Accurate near real time fire emissions estimates are required for air quality forecasts. To date, most approaches are based on satellite-derived estimates of fire radiative power (FRP), which can be converted to fire radiative energy (FRE) which is directly related to fire emissions. Uncertainties in these FRE estimates are often substantial. This is for a large part because the most often used low-Earth orbit satellite-based instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS) have a relatively poor sampling of the usually pronounced fire diurnal cycle. In this paper we explore the spatial variation of this fire diurnal cycle and its drivers using data from the geostationary Meteosat Spinning Enhanced Visible and Infrared Imager (SEVIRI). In addition, we sampled data from the SEVIRI instrument at MODIS detection opportunities to develop two approaches to estimate hourly FRE based on MODIS active fire detections. The first approach ignored the fire diurnal cycle, assuming persistent fire activity between two MODIS observations, while the second approach combined knowledge on the climatology of the fire diurnal cycle with active fire detections to estimate hourly FRE. The full SEVIRI time series, providing full coverage of the fire diurnal cycle, were used to evaluate the results. Our study period comprised of 3 years (2010–2012), and we focused on Africa and the Mediterranean basin to avoid the use of potentially lower quality SEVIRI data obtained at very far off-nadir view angles. We found that the fire diurnal cycle varies substantially over the study region, and depends on both fuel and weather conditions. For example, more “intense” fires characterized by a fire diurnal cycle with high peak fire activity, long duration over the day, and with nighttime fire activity are most common in areas of large fire size (i.e., large burned area per fire event). These areas are most prevalent in relatively arid regions. Ignoring the fire diurnal cycle generally resulted in an overestimation of FRE, while including information on the climatology of the fire diurnal cycle improved FRE estimates. The approach based on knowledge of the climatology of the fire diurnal cycle also improved distribution of FRE over the day, although only when aggregating model results to coarser spatial and/or temporal scale good correlation was found with the full SEVIRI hourly reference data set. We recommend the use of regionally varying fire diurnal cycle information within the Global Fire Assimilation System (GFAS) used in the Copernicus Atmosphere Monitoring Services, which will improve FRE estimates and may allow for further reconciliation of biomass burning emission estimates from different inventories.
Landscape fires are a global phenomena, and the annually burned area is roughly equivalent to the area of India (Giglio et al., 2013). Most burned area occurs in the savannas of Africa, Australia, and South America, where they shape ecosystem dynamics and due to their scale are an important source of global emissions of (greenhouse) gases and aerosols (Seiler and Crutzen, 1980; Bowman et al., 2009). Fires affect air quality both locally and regionally (Langmann et al., 2009), with recent studies putting mortality rates over 300 000 annually due to exposure to smoke (Johnston et al., 2012).
Traditionally, the amount of dry matter burned and quantity of trace gases and aerosols emitted have been calculated using biome-specific fire return intervals and estimates of the total amount of biomass as well as the fraction of biomass burned, the combustion completeness (Seiler and Crutzen, 1980). Thanks to new satellite input streams that better capture the spatial and temporal diffuse nature of fires, the estimated fire return intervals have been replaced by direct estimates of monthly, weekly or even daily area burned (Roy et al., 2005; Giglio et al., 2009). In addition, satellite information and biogeochemical modeling have been used to improve estimates of biomass and combustion completeness. However, uncertainties in these bottom-up fire emission estimates are still substantial (Reid et al., 2009; Zhang et al., 2012; Larkin et al., 2014), and they are generally inappropriate for use in near real-time systems partly because the burned area signature is only detectable days to weeks after the actual fire occurrence.
Hot-spot observations from satellites have been used as a proxy for burned area and emissions fluxes in near real time (Freitas et al., 2005; Reid et al., 2009; Wiedinmyer et al., 2011). Another promising and relatively new bottom-up approach uses estimates of fire radiative power (FRP) observed from satellites to calculate daily fire radiative energy (FRE). Wooster et al. (2005) found that these FRE estimates scale directly to dry matter burned, potentially circumventing the uncertainties associated with estimating area burned, fuel loads, and the combustion completeness. In addition, FRP observations can be observed and processed in near real time (Xu et al., 2010; Kaiser et al., 2012; Zhang et al., 2012) and can be measured for small fires that remain undetected in burned area products (Roberts et al., 2011; Randerson et al., 2012).
Hot-spot and FRP observations are currently the only available options when
modeling exercises require near real time observations, for example in
chemical weather forecasts used to predict air quality. The Global Fire
Assimilation System (GFAS; Kaiser et al., 2012), for example, is used to
estimate global near real time daily fire emissions within the EU-funded
project Monitoring Atmospheric Composition and Climate III (MACC-III). GFAS
is currently using fire observations from the polar orbiting
Moderate Resolution Imaging Spectroradiometer (MODIS) instruments aboard the
Terra and Aqua satellites (Giglio et al., 2006). Due to their relative
proximity to the Earth, the Terra and Aqua MODIS instruments have a high
sensitivity to even quite low FRP (smaller and/or lower intensity) fires.
However, they only provide four daily observations under ideal conditions but
less when optically thick clouds are present, which may not be enough to
fully characterize how fire activity varies over the course of the day.
Observations with a much higher temporal resolution are available from
geostationary satellites. However, as a consequence of their geostationary
position, these satellites individually do not provide global data and are
located at greater distance from the Earth resulting in typically coarser
pixel sizes than polar orbiting instruments. Since the threshold of
detectability of a fire is not only dependent on the instrument but also a
function of the pixel area, geostationary sensors have a higher minimum FRP
detection limit (typically
Previous studies found that fire activity shows a strong diurnal cycle, and one that is both temporally and spatially variable (Prins and Menzel, 1992; Giglio, 2007; Roberts et al., 2009). The ideal setup to detect fires would be a high temporal resolution imaging system, sensitive to even the lowest FRP fires, and providing global coverage, but due to the limitations of the orbital characteristics outlined above there is no single platform available to provide this. Therefore the estimation of FRE at a global scale is difficult, with polar orbiting satellites lacking observations to accurately represent the fire diurnal cycle and geostationary satellites being limited to certain regions of the globe and omitting the (rather common) low FRP fires. However, previous studies have developed approaches to estimate FRE based on the combination of data from different satellite systems (Boschetti and Roy, 2009; Ellicott et al., 2009; Freeborn et al., 2009, 2011; Vermote et al., 2009).
Some of these mixed approaches used both burned area and active fire data
(Boschetti and Roy, 2009; Roberts et al., 2011), which may provide benefits
in terms of more accurate FRE determination, but cannot be used easily in near
real time systems because of the latency in burned area observations.
Alternatively, FRP observations of polar orbiting and geostationary
satellites can be blended to combine the sensitivity of the MODIS instruments
to lower FRP fires and the diurnal sampling characteristics of SEVIRI.
Freeborn et al. (2009) developed a database for matching SEVIRI and MODIS FRP
observations based on frequency-magnitude statistics, but the samples had to
be accumulated over significant spatial areas (5
Global fire emissions estimates at high spatial and temporal resolutions, ideally produced in near real time, are required to feed into atmospheric models which are under continuous development and run at improved resolutions thanks to increased computational power (Zhang et al., 2012). Higher temporal resolution may also help to reconcile bottom-up and top-down emission estimates (Mu et al., 2011). None of the approaches mentioned above are, however, suitable for providing this. Due to these limitations current state of the art global near real time emission inventories still ignore possible effects of fire diurnal cycle on their emission estimates (e.g., Wiedinmyer et al., 2011; Kaiser et al., 2012) and may therefore be structurally biased due to the fire diurnal cycle and the MODIS sampling design (e.g., Ichoku et al., 2008; Ellicott et al., 2009; Freeborn et al., 2011).
The purpose of the work presented here is to better understand the fire
diurnal cycle and its spatiotemporal dynamics, in order to develop a new way
to include this into a near real time fire emissions estimation framework.
First, the spatial distribution and dependencies of the fire diurnal cycle
and their effect on active fire detections at MODIS overpasses were explored.
Then, data assimilation was used to compare two different methods to derive
hourly FRE estimates at 0.1
To explore the spatiotemporal dynamics of the fire diurnal cycle, we used hourly temporal resolution FRP data derived from 15 min observations made by the SEVIRI instrument hosted onboard the geostationary Meteosat satellite (Sect. 2.1). However, to drive the models developed here we only used SEVIRI FRP observations made at the overpass times of the MODIS polar orbiting sensors (Sect. 2.2), whilst the hourly temporal resolution SEVIRI time series were used to evaluate the results. Land cover characteristics (Sect. 2.3), along with data on fire size (Sect. 2.4), were used to better understand the spatial distribution of fire diurnal cycle. These data sets are described in more detail below, followed by the methods used in Sect. 3.
The SEVIRI instrument aboard the geostationary Meteosat Second Generation
(MSG) series of satellites provides coverage of the full Earth disk every 15
min in 12 spectral bands (Schmetz et al., 2002). The Meteosat SEVIRI
FRP–PIXEL product contains per-pixel fire radiative power data along with
FRP uncertainty metrics produced from the full spatial and temporal
resolution SEVIRI observations (Wooster et al., 2015). The FRP–PIXEL product
is produced using an operational version of the geostationary fire thermal
anomaly (FTA) algorithm described in Roberts and Wooster (2008), and the
product and its performance characteristics are described in Wooster et
al. (2015). The FRP–PIXEL products are freely available from the Land
Surface Analysis Satellite Applications Facility (LSA SAF;
The two MODIS sensors on board of the Terra and Aqua satellites provide four
daily overpasses in most Earth locations, albeit sometimes at view angles in
excess of 45
The dominant land cover type was derived from the MODIS MCD12C1 land cover
product, which provides 0.05
Here we define the fire size for a certain grid cell as the mean burned area
per fire event, weighted by their total area burnt (when calculating the
mean, a fire event burning 100 km
Our overall goal within GFAS is to provide hourly estimates of FRE at
0.1
Using data assimilation we combined the discrete actual SEVIRI observations, made at the time of the MODIS detection opportunities, with hourly predictions of fire activity – using their combination to create continuous hourly best estimate FRE time series (Sect. 3.3). We developed two prediction methods. The first method assumed persistent fire activity until the next satellite detection opportunity, and provides further insights into the combined effect of the fire diurnal cycle and the MODIS sampling design on hourly FRE estimates when the fire diurnal cycle is ignored (Sect. 3.4). The second method followed previous studies and used a Gaussian function to predict fire development over the day (Vermote et al., 2009). By combining prior knowledge about the climatology of the fire diurnal cycle with active fire observations at MODIS overpasses to estimate the parameters of the Gaussian function, this approach provides a possible pathway to implement the fire diurnal cycle into the near real time fire emission modeling framework (Sect. 3.5). Comparing the results of the two approaches to those from the full hourly SEVIRI time series allowed us to determine and discuss their strengths and limitations (Sect. 3.6).
We started exploring the fire diurnal cycle and its drivers. A Gaussian
function was optimally fitted (least squares) to the hourly SEVIRI
observations
Although the fire diurnal cycle as observed by SEVIRI is comparable to that which would be observed by MODIS if it had the same temporal sampling ability, it is a little different due to SEVIRI's inability to discriminate the lowest FRP fire pixels which typically dominate more towards the start and end of the daily fire cycle, but which are also present along with often higher FRP pixels towards the diurnal cycle maxima (Freeborn et al., 2009). To gauge the magnitude of the effect Freeborn et al. (2009) derived the “virtual MODIS” fire product that has the temporal sampling of SEVIRI and the sensitivity to fire of MODIS. They found that the full-width at half maximum height (i.e., the width of the diurnal cycle at half of the daily FRP maximum value) of the diurnal cycles derived from the SEVIRI and the “virtual MODIS” data sets are very similar, it is the amplitude and the full-width at base height of the two cycles, which are more different. In terms of total FRE emitted, the latter is of less importance, here we followed Freeborn et al. (2011) in assuming that the diurnal cycles from SEVIRI and MODIS are sufficiently similar.
In order to visualize the spatial distribution of the fire diurnal cycle, the climatological diurnal cycle was calculated for each grid cell depending on the mean parameter values of the Gaussian function weighted for daily FRE, including all days of fire activity during the study period without cloud obscurance. To get a better understanding of the drivers of the fire diurnal cycle these results were compared to land cover and aspects of the fire regime (fire size, total annual FRE, and the annual number of days with fire activity), see Sect. 2.
During the data assimilation, SEVIRI observations at MODIS detection
opportunities were used to drive the models. Here, SEVIRI observations for a
given hour
SEVIRI data sampled at MODIS detection opportunities were compared to the full SEVIRI hourly time series to explore the effect of the fire diurnal cycle on the daily sampling at MODIS overpasses. More specifically we calculated the percentage of FRE emitted on days without any active fire detection at MODIS detection opportunities, and the total daily number of MODIS overpasses during the fire season. The latter was calculated by weighing the mean number of monthly detection opportunities at MODIS overpasses by monthly total detected FRP, thus giving the largest weight to the month with most fire activity (ignoring cloud cover).
We used a modified version the fire data assimilation methodology of GFAS to
allow representation of the fire diurnal cycle. GFAS assumes that the
availability of observations dominates the error budget of the global FRP
fields. It approximates these errors by further assuming the FRP variance to
be inversely proportional to the fraction of observed area
Our modifications affected the step size and the FRP prediction model. The
former was set to 1h to be able to represent a diurnal cycle. For calculating
the FRP prediction
Applying the daily persistence approach of Kaiser et al. (2012) to hourly
time resolution, we first explored the most parsimonious approach that
predicts FRP
In the second approach we followed previous studies of Vermote et al. (2009)
and Ellicot et al. (2009) and the recommendation in Kaiser et al. (2009) to
use a Gaussian function to describe a “standard fire diurnal cycle”.
Wooster et al. (2005) and Roberts et al. (2009) already demonstrated that
SEVIRI observations sample the diurnal cycle of large fires well, and for
some individual large fires show FRP time series that depict diurnal
characteristics appearing close to Gaussian in nature even at 15 min
temporal resolution. The prediction was calculated by fitting a Gaussian
function through the last 24 h of analysis:
The estimated hourly FRE fields (or analysis;
Finally, we compared daily regional aggregated FRE time series for several study regions of the two modeling approaches and SEVIRI. In order to compare daily regional time series to the model, a cloud cover correction needed to be carried out. Since persistent cloud cover is relatively rare during the burning season in most parts of Africa, we chose a simple gap-filling approach where the value of the last cloud-free observation is assumed to be valid until the next cloud-free observation, which is consistent with the observation gap filling in the daily GFAS.
First, we present the results related to the spatial distribution of the fire
diurnal cycle, and assess the impact of the fire diurnal cycle on active fire
observations made at times of MODIS overpasses. The spatial
distribution of the fire diurnal cycle was explored by optimally fitting a
Gaussian function to the hourly, 0.1
Hourly-mean FRP time series derived from SEVIRI data, the same data
but only sampled at MODIS detection opportunities, and an optimally fitted
Gaussian function fitted to the full SEVIRI FRP time series. These two
examples are for a 0.1
Figure 1 shows an example of two 0.1
The results shown in Fig. 1 indicate that high FRP, relatively long-lived
fire activity is rather well described by a Gaussian function, even at this
0.1
Weighted mean values of parameters of the optimally fitted Gaussian
function for each 0.1
The shape of the Gaussian function, and consequently the parameters: SD
(
Characteristics of the fire regime and fuel types based on
2010–2012 data.
In addition to an observed variability in the fire diurnal cycle seen on
different days, we found distinct spatial patterns in the optimal fitted
Gaussian parameters (Fig. 2). Some of these patterns were similar for the
different parameters. In particular, there were zones of generally more
intense fires (e.g., South Sudan, northern Central African Republic,
Botswana, Namibia and parts of Angola and the Democratic Republic of Congo
(DRC)), showing relatively high values of
Mean values of the parameters of the Gaussian function per land
cover type (excluding days of cloud cover and weighted by FRE), SD are shown
in parenthesis. Values of
Table 1 shows the land cover-averaged values and SD of the results presented
in Fig. 2. In addition we calculated the ratio of the mean SEVIRI FRP at
MODIS daytime detection opportunities to the maximum daytime FRP
In order to better understand the spatial distribution of the fire diurnal
cycle features, we studied characteristics of the fire regime that were
expected to be related to fuel properties and the diurnal cycle (Fig. 3a, c
and d). To guide the interpretation we have included a land cover map, partly
governing fuel loads, in Fig. 3b. Annual emitted FRE varied widely over the
study region, and highest values were found in the savannas and woody
savannas (compare Fig. 3a with b) and coincided with regions of large fire
size and/or a high number of annual fire days (compare Fig. 3a with c and d).
Similarities with characteristics of the fire diurnal cycle were also found,
the earlier mentioned zones of generally more intense fires (high values of
The relative fraction of FRE emitted on days that SEVIRI data sampled at
MODIS observation times did not observe active fires is an important factor
affecting model performance, and showed similar spatial patterns as
Detection of fire activity at MODIS detection opportunities.
Total fire radiative energy (FRE) estimated via the two modeling
approaches using SEVIRI observations taken at only the MODIS detection
opportunities, expressed as fraction of the total FRE calculated using the
entire set of hourly mean, 0.1
To evaluate the two modeling approaches that estimated FRE from SEVIRI data
only at the MODIS sampling times we started with comparing the spatial
distribution of mean estimated FRE for each method with the cloud corrected
SEVIRI FRE calculated using the entire hourly, 0.1
Daily FRE for four study regions (areas of 85 000 to
567 000 km
Equally important as the absolute FRE estimates shown in Fig. 5 and Table 2 are their temporal dynamics. Figure 6 shows regional daily budgets for several study regions with different geographical positions and land cover (see Fig. 7). Similar to the results in Fig. 5, we found a general overestimation by the persistent approach, and better overall estimation by the climatological approach. Overestimation of the persistent approach was occurring mostly in the tropics (e.g., Nigeria and DRC), where also stronger day-to-day variability was observed as compared to that derived with the complete SEVIRI data or the climatological approach (Fig. 5). The climatological approach showed a small delay in FRE estimations compared to the complete SEVIRI data set.
To further test the ability of the two modeling approaches to allocate FRE to
the individual grid cells at the right moment in time, correlation
coefficients were calculated. Table 3 shows Pearson's
Estimated annual FRE during 2010–2012 by the two model approaches as percentage of SEVIRI FRE (cloud cover corrected).
Pearson's
Unlike biomass burning emission inventories based on burned area, inventories using active fire observations from Earth observation satellites can be produced in near real time (Freitas et al., 2005; Reid et al., 2009; Sofiev et al., 2009; Wiedinmyer et al., 2011; Kaiser et al., 2012; Darmenov and da Silva, 2013). The near real time emissions inventories are, at present, generally based on active fire data from the MODIS instruments operating onboard the Terra and Aqua polar orbiting satellites. The FRP observations of MODIS are almost without saturation, operating day and night, with a reasonable spatial resolution and with new observations available for any location at least a few times every day – cloud cover permitting. However, it is well known that fire activity in most regions follows a clear daily cycle (e.g., Roberts et al., 2009; Vermote et al., 2009). Consequently, the FRP measures derived from intermittent polar orbiting MODIS observations are often not fully and directly representative of the actual daily fire activity (Fig. 1; Giglio, 2007; Vermote et al., 2009; Freeborn et al., 2011). Although several approaches have been developed to obtain more accurate estimations of FRE from the limited temporal sampling of FRP provided by MODIS (e.g., Ellicott et al., 2009; Freeborn et al., 2009, 2011; Vermote et al., 2009), they are all best suited to be used with previously collected and/or aggregated FRP data, and none can be readily implemented at high spatiotemporal resolution in near real time. For this reason, most current global emission inventories produced in near real time actually ignore fire diurnal dynamics completely (e.g., Kaiser et al., 2012), and this results in large biases in the FRE budgets (Ellicott et al., 2009; Zhang et al., 2012).
Study regions used in Fig. 6. Abbreviations refer to: Botswana (BWA), the Democratic Republic of Congo (DRC), Nigeria (NGA) and Portugal (PRT).
Here we start discussing the spatial distribution of the fire diurnal cycle, and its drivers (Sect. 5.1). Building on previous work, we explored two new methods to estimate hourly FRE in near real time from observations made by SEVIRI at MODIS detection opportunities. The results illustrate how MODIS observations might be used to calculate hourly FRE, and where errors can be expected due to the diurnal cycle and the limited temporal sampling provided by MODIS (Sect. 5.2).
The fire diurnal cycle characteristics were explored by fitting of a Gaussian
function to the hourly SEVIRI time series. Vermote et al. (2009) and Ellicott
et al. (2009) found that at a 0.5
Although the shape of the “average” fire diurnal cycle is scale dependent, regional patterns in the diurnal cycle characteristics (Fig. 2) remain similar over different scales, and therefore we found similar land cover dependent characteristics as previous studies. For example, shrublands and grasslands generally faced drier conditions when burning than did woody savannas or tropical forest, and therefore fire activity typically continued longer over the day and the hour of peak fire activity was generally located later in the afternoon (Fig. 2; Table 1; Giglio, 2007; Roberts et al., 2009). For the same reason, temperate and boreal forests have been reported to show a more pronounced diurnal cycle than grasslands (Fig. 2; Sofiev et al., 2013; Konovalov et al., 2014). Building on the land cover based analysis of Roberts et al. (2009), we provide a first analysis of the spatial distribution of the fire diurnal cycle.
The three parameters determining the shape of the Gaussian can be used to
visualize the spatial distribution of the fire diurnal cycle. The daily
FRP-maximum is given by
The peak hour of fire activity found here corresponds to the moment of day at
which 50 % of the total FRE has been emitted (assuming
Data assimilation and two modeling approaches, were used to estimate hourly FRE from SEVIRI FRP data sampled at the times of MODIS detection opportunities. Here we start discussing the performance of the different methods with respect to their total FRE estimates and daily regional FRE estimations. Then we discuss the more uncertain model performance at higher spatiotemporal resolutions.
The persistent approach is comparable to a direct hourly extension of the current GFAS methods (Kaiser et al., 2012), where the fire diurnal cycle is ignored and the predicted FRP for each hour is equal to that of the last FRP observation. This led to a general overestimation of daily FRE because the 13:30 LT temporal sampling time of MODIS is relatively close to the peak hour of daily fire activity, and therefore not very representative of the full period until the next observation at 22:30 LT (Figs. 2d and 5; Table 2). Moving away from the equator, the number of daily MODIS observations increases due to orbital convergence at higher latitudes, and consequently the model performance improved (Figs. 4b, 5 and 6; Giglio et al., 2006; Reid et al., 2009). Additional inclusion of daytime observations due to orbital convergence will typically be somewhat earlier or later in the afternoon and may therefore lower the FRE estimation. In the persistent approach, missing nighttime observations may cause an overestimation and missing daytime observation an underestimation of daily FRE, resulting in erroneous regional day-to-day variations in FRE estimates in the tropics (Fig. 6). Following previous research, we found that due to the spatiotemporal variation of the fire diurnal cycle FRE was overestimated more for some land cover types than for others (Table 2; Freeborn et al., 2011). Land cover classes that typically showed longer fire durations (Fig. 2c) with peak fire activity later in the afternoon (Fig. 2d) were not as much overestimated as land cover classes with more pronounced fire diurnal cycles (Figs. 5 and 6; Table 2). However, part of this effect likely stems from these land covers mostly being located in the more frequently observed higher latitudes of our study region. Although the persistent method is not directly comparable to the methods of widely used emission inventories like GFAS or QFED (Quick Fire Emissions Dataset) (Kaiser et al., 2012; Darmenov and da Silva, 2013), they likely introduce similar errors by ignoring the fire diurnal cycle.
The climatological approach showed better performance in terms of absolute FRE estimations, while also better able to reproduce its spatial variability (Fig. 5). In contrast to the persistent approach, the hourly predictions were based on the last 24h of fire activity, enabling more realistic gap filling during periods without observations. This resulted in an advantage during periods of cloud cover or missing observations due to the satellite orbits, but because of the low number of actual daily observations the climatological approach had the tendency to continue predicting fire activity after fires had ceased, seen as a small delay in the signals in Fig. 6.
An additional criterion to evaluate the model performance was the correlation
between the modeling approaches and the SEVIRI data at different
spatiotemporal scales. Correlation between the modeled and SEVIRI
time series improved considerably when moving from hourly to daily
resolution, showing that the models were better able to estimate daily
budgets than the distribution of fire activity over the day. These
differences may be explained by the inability of the models to correctly
estimate the hour of peak fire activity, a fire diurnal cycle that is not
well represented by a Gaussian function, or in the case of small fires the
fire diurnal cycle may not be fully detected by the SEVIRI instrument.
Because of the large day-to-day variation in the fire diurnal cycle and the
FRP measurements limited to the time of the MODIS overpasses, the individual
FRP observations have a low precision (i.e., large random error) and omission
(i.e., non detection) of fires is frequent (Figs. 1 and 4), resulting in low
correlation at high spatiotemporal scales (Table 3). Because fires rarely
occur on their own and generally form part of a regional pattern (Bella et
al., 2006), the correlation increased considerably when accumulating results
to a 1
Overall, using the climatological approach resulted in the best model
performance, although in specific cases using the persistent approach showed
better results. For example, at 0.1
Within GFAS, to handle the uncertainties introduced into the MODIS-derived FRE estimates by neglecting the diurnal cycle influence, the estimated FRE is converted into estimates of dry matter burned (DM) using land cover-specific conversion factors. These were derived via comparison of long-term monthly FRE estimates to the DM estimates calculated over the same period by the Global Fire Emissions Database (GFED 3.1; van der Werf et al., 2010; Kaiser et al., 2012). It is currently assumed that by allowing the conversion factors to vary with land cover type the impact of any land cover-varying diurnal cycle is also incorporated, reducing the influence of the diurnal cycle. The issues discussed above, along with the accuracy of the GFED DM calculations, which are for example affected by the quality of the burned area product and the biochemical models used, all influence values of the land cover-specific FRE-to-DM conversions factors presented by Kaiser et al. (2012).
Wooster et al. (2005) and Freeborn et al. (2008) previously explored the conversion factors between FRE and DM using small scale experiments, and found that they appeared relatively independent of vegetation type. However, when moving to the satellite-scale there are additional factors influencing this FRE-to-DM relationship, for example the fire regime of an area and the degree to which MODIS misses the lowest FRP fires, and the canopy density of trees that might obscure some of the thermal radiation being emitted by fires burning in the ground fuels (Freeborn et al., 2014). The thermal radiation recorded in satellite products is additionally reduced by cloud cover and erroneous flagging of smoke as clouds during data processing. Konovalov et al. (2014) nevertheless found FRE-to-DM relationships relatively similar to those of the earlier small-scale experiments when using atmospheric observations and biomass burning trace gas and aerosol emissions factors to estimate fuel consumption. Exploring methods to incorporate the fire diurnal cycle in the GFAS global FRP-based near real time emission inventory is a first step in taking into account some of these issues in order to improve global FRE estimates made at relatively high spatiotemporal resolutions, and hopefully also in reconciling some of the differences in current emission inventories.
Emission inventories based on FRP observations have great potential to
improve biomass burning emission estimates, by eliminating the need for
modeling of fuel loads and fuel consumption, and can be produced in near
real time. However, to date uncertainties in FRE estimates remain high when
using polar orbiting FRP data sets, largely due to difficulties in combining
the limited temporal resolution observations and knowledge about the fire
diurnal cycle. Geostationary data can alleviate this issue, but bring their own problems related to the non-detection of the lower FRP fires due to the
coarse spatial resolution of the geostationary observations. Geostationary
data sets are also not global in extent. Here we explored the spatial
dependencies of the fire diurnal cycle and its impact on active fire
detections made at the time of MODIS overpasses. Two modeling approaches
were developed to derive hourly FRE estimates based on data-assimilation and
SEVIRI FRP observations subsampled at MODIS detection opportunities. The
first approach ignored the fire diurnal cycle assuming persistent fire
activity between two MODIS detection opportunities, while the second
approach combined prior knowledge of the fire diurnal cycle with active fire
observations at MODIS detection opportunities to simulate the fire diurnal
cycle. Both approaches were evaluated against the actual hourly FRP
observations made by SEVIRI. Our main conclusions are the following.
We considered various drivers of the spatial distribution of
fire diurnal cycle (dominant land cover, fire size, annual number of
fire days, and diurnal climate conditions) and found that all played a role.
The strong relation between fire size and fire diurnal cycle for grass-fueled fires, and the climatic gradient in diurnal cycle, indicate that
using fuel characteristics rather than land cover alone to characterize
the fire diurnal cycle provides a potential pathway to improve these estimates.
Here we showed that this information can partly be obtained by studying the
fire characteristics, such as fire size, which are contained within the
remote sensing data themselves. Ignoring the fire diurnal cycle may cause structural errors in FRE
estimates, and likely results in a general overestimation of FRE due to the
timing of the MODIS overpasses. The errors vary regionally, mostly due to
variations in the fire diurnal cycle, while results get more accurate at
higher latitudes due to the increasing number of daily MODIS detection
opportunities caused by orbital convergence. Due to the large day-to-day variations in the fire diurnal cycle at
the grid cell level, and the scarce number of MODIS observations of any one
location per day, daily FRP fields calculated from observations made at MODIS
detection opportunities are characterized by low precision (i.e.,
observations are not representative for daily fire activity) and high
omission (i.e., non observation of fires). Therefore a sufficiently large
sample size of MODIS observations is required to accurately estimate FRE, as
shown earlier by Freeborn et al. (2011). In zones of frequent fires, where
fires are generally part of a regional biomass burning pattern, model
performance greatly improved when moving to a coarser scale, increasing the
sample size. Model performance was also considerably better for zones of
relatively large fires that were characterized by low omission. Production of
emission inventories at very high spatiotemporal resolution using data from a
limited number of low-Earth orbit satellite observations may therefore
provide somewhat restricted added value compared to those derived at coarser
spatiotemporal scales. Relative overrepresentation of day- or nighttime FRP observations may
cause large day-to-day variations in estimated FRE when the diurnal cycle is ignored. The way we observe the fire diurnal cycle is scale dependent, mostly
because of the large variation in fire diurnal cycle, even within the same
grid cell between different days.
We recommend implementing the climatological model within GFAS in Copernicus
Atmosphere Services in order to improve global and regional FRE estimates
and further reconcile emission estimates from the various different
inventories currently available.
We like to thank Samuel Remy at ECMWF for processing MODIS and SEVIRI data, and the data providing agencies: NASA and the EUMETSAT LSA SAF for making their data publicly available. This study was funded by the EU in the FP7 and H2020 projects MACC-II and MACC-III (contracts no. 283576 and 633080). Edited by: R. Engelen