In the southern Amazon relationships have been established among drought,
human activities that cause forest loss, fire, and smoke emissions. We
explore the impacts of recent drought on fire, forest loss, and atmospheric
visibility in lowland Bolivia. To assess human influence on fire, we consider
climate, fire, and vegetation dynamics in an area largely excluded from human
activities since 1979, Noel Kempff Mercado National Park (NK) in northeastern
Bolivia. We use data from five sources: the Moderate Resolution Imaging
Spectroradiometer Collection 6 active fire product (2001–2015) (MODIS C6),
Global Fire WEather Database (GFWED) data (1982–2015), MODIS land cover data
(2001–2010), MODIS forest loss data (2000–2012), and the regional
extinction coefficient for the southwestern Amazon (i.e.,
Observations from the southern Amazon reveal fire emissions increased from 1987 through the early 2000s (van Marle et al., 2017). During this time, humans used fire in the southern Amazon while logging timber, and to clear land for building infrastructure and agriculture (Moran, 1993; Nepstad et al., 1999, 2009; Fearnside, 2005; Morton et al., 2008; Cochrane and Barber, 2009; van der Werf et al., 2010), suggesting human activities had a significant impact on increased smoke emissions from fire (van Marle et al., 2017). To minimize the impacts of deforestation and fire on deforestation in the Amazon, and in turn on carbon emissions and global climate, restrictions on land expansion and policies regulating beef and soy production in the Brazilian Amazon were implemented during the early 2000s (Nepstad et al., 2014). While these restrictions and policy changes helped reduce deforestation from 2004 to 2013 (Nepstad et al., 2014), others suggest a decrease in demand for Amazon resources was the primary driver of reduced deforestation and fire from 2004 to 2012 (Fearnside, 2017). Outside of the Brazilian Amazon, forest loss from deforestation and fire has continued in parts of the southern Amazon, particularly in lowland Bolivia (Chen et al., 2013b; van Marle et al., 2016, 2017). The Cerrado biome in particular has experienced increased deforestation since 2010, which could be due to a shift in agriculture to the southern Amazon and Cerrado (Soares-Filho et al., 2014). Likely amplifying the effects of deforestation and fire on forest loss in the southern Amazon were drought conditions during the early 2000s (Brown et al., 2006; Aragão et al., 2007; Marengo et al., 2008), raising the question of what the relationships among recent drought, fire, and forest loss in lowland Bolivia are.
Both paleofire investigations (Bush et al., 2008; Marlon et al., 2008; Power et al., 2013) and modern fire records (Brown et al., 2006; Aragão et al., 2007; Marengo et al., 2008) link drought to fire in the southern Amazon. Our study considers relationships between recent drought and fire in lowland Bolivia using the Global Fire WEather Database (GFWED) (Field et al., 2015) and the Moderate Resolution Imaging Spectroradiometer Collection 6 (MODIS C6) active fire product (Giglio et al., 2016). The GFWED Drought Code (DC) in particular captures net drying of deep fuels, with lower DC values observed during the wet season and higher DC values during the dry season (Field et al., 2015). We interpret high (low) DC values during the fire season from August to October in lowland Bolivia as an indicator of antecedent dry (wet) conditions during the preceding wet and dry seasons.
Fires and smoke observed by the Moderate Resolution Imaging
Spectroradiometer (MODIS) over Bolivia and Noel Kempff Mercado National Park
(NK) on 17 August 2010
In addition to understanding links between drought and fire in lowland Bolivia, we also consider where past fires occurred spatially in relation to land use and biome type using data from the MODIS-based global land cover product (Broxton et al., 2014) and the Landsat-based forest loss product (Hansen et al., 2013). Considering humans have had a significant impact on forest loss and fire activity in unprotected biomes (Morton et al., 2013), we compare fire distribution in unprotected biomes in lowland Bolivia to biomes in Noel Kempff Mercado National Park (NK), an area in lowland Bolivia protected from deforestation since 1979. Specifically, we explore climate and fire relationships in NK to determine when fire activity is high in relation to interannual climate variability, and where fire occurs spatially in relation to different biomes and land uses.
Finally, to extend our fire record for lowland Bolivia prior to the onset of MODIS C6 in 2001, we analyze horizontal visibility data from surface weather stations at the World Meteorological Organization (WMO) level. Visibility data have been used as a fire emissions proxy to understand fire activity over the southern Amazon from 1973 to 2015 (van Marle et al., 2017). We test relationships between MODIS C6 active fire data for lowland Bolivia and regional WMO visibility data (locations of weather stations in Fig. 1a), to determine how well visibility data correspond to the MODIS C6 fire record from 2001 to 2015 and to extend the fire record for lowland Bolivia prior to 2001.
Our results are useful and relevant when considering uncertainties regarding the fate of the southern Amazon in response to climate change (Zhang et al., 2015). Here, we show how recent interannual climate variability has impacted fire activity across different biomes in lowland Bolivia. A further understanding of relationships among interannual climate variability and biome type and fire in lowland Bolivia is valuable when considering fire weather (Bedia et al., 2015) and fire season severity are expected to increase in the southern Amazon during the 21st century (Flannigan et al., 2013). Further, fire in the southern Amazon can cause increased smoke emissions that negatively impact human health (e.g., Brown et al., 2006), and can impact carbon emissions and global climate (Fearnside, 2005; Aragão et al., 2014). Seasonal covariation between fire and horizontal visibility data are explored here to provide further information on how fire is related to visibility and smoke emissions in the southern Amazon.
MODIS C6 offers a tool to adequately answer questions related to recent fire activity from November 2000 to present (Morton et al., 2011). MODIS C6 has been used globally to explore a variety of fire-related questions ranging from biomass burning (Wooster et al., 2003) to fire detection in the Amazon (Chen et al., 2013a). High-spatial-resolution (1 km) near-real-time MODIS C6 data used in our analyses are provided by the Land, Atmosphere Near real-time Capability for Earth Observing System Fire Information for Resource Management System, and are operated by the NASA Goddard Space Flight Center Earth Science Data and Information System (Giglio et al., 2016). Two key limitations of the previous MODIS C5 product were small forest clearings causing false active fires, and thick smoke obscuring large fires (Giglio et al., 2016). For tropical ecosystems, these two key limitations were addressed and errors were reduced from MODIS C5 to MODIS C6 (Giglio et al., 2016).
For our study, MODIS C6 data for Bolivia (Fig. 1b) were downloaded using the
NASA Earth Observation Data archive download tool. Data were subset by
location (e.g., NK), year, and by fire detection confidence
Landsat forest loss data 2000–2012 (Hansen et al., 2013) and MODIS-based Collection 5.1 MCD12Q global land cover data (Broxton et al., 2014) were obtained to determine the spatial coherency of fire, land use, and forest loss. High-spatial-resolution figures (i.e., Figs. 1, 4, 5) were created for detailed spatial analyses. Considering the high spatial resolution of certain figures, to view forest loss displayed as white pixels (i.e., Fig. 5), or detailed biome and fire spatial variability (e.g., Fig. 1), readers will need to increase the zoom. To simplify in-text discussions on the spatial distribution of fire in lowland Bolivia in relation to various biomes, MODIS land cover types seen in figure legends (i.e., Figs. 1b, 4, 5) will be hereafter referred to in the paper as the Cerrado, METF, SDTF, and seasonally inundated wetlands. The Cerrado biome includes open shrubland, woody savanna, savanna, and grassland MODIS land cover types. The SDTF biome includes deciduous broadleaf forest, mixed forest, and closed shrubland MODIS land cover types. The METF biome includes the evergreen broadleaf forest MODIS land cover type. Seasonally flooded wetlands in lowland Bolivia are hydromorphic climatic savannas that are periodically flooded during the wet season and desiccate during the dry season (Junk et al., 2011). For readers interested in a more detailed land cover classification (Broxton et al., 2014), the original land classification was maintained in some figures (i.e., Figs. 1b, 4, 5) for detailed spatial analyses of fires and biomes over lowland Bolivia.
In the absence of long-term fire data, horizontal visibility has been used as
a proxy for fire emissions in Indonesia (Field et al., 2009, 2016) and
Amazonia (van Marle et al., 2017). Here, horizontal visibility observations
(1973–2015) were taken from the NOAA National Climatic Data Center
Integrated Surface Database
(
Mean monthly Pearson's correlations (January 2001–December 2015)
between Moderate Resolution Imaging Spectroradiometer C6 (MODIS C6) active
fires with
To correct for limitations of the human eye and imperfections of the
landmarks used to estimate the maximum distance seen, the observations in
meters are usually expressed as the extinction coefficient (
To examine climatic controls on fires in NK, we used meteorological parameters and components of the Fire Weather Index (FWI) system, which integrates different surface weather parameters influencing the likelihood of fires starting and spreading. The FWI system consists of moisture codes for three generalized fuel classes and three fire behavior components, computed each day from surface temperature, relative humidity (RH), wind speed, and precipitation. Because of its flexibility, it is the most widely used of such systems in the world and has been adapted for use in different fire environments operationally and for research purposes (de Groot and Flannigan, 2014).
Mean monthly Moderate Resolution Imaging Spectroradiometer C6 (MODIS
C6) active fires with
To explore interannual climate variability related to fire activity in
lowland Bolivia, FWI data for the period 1982–2015, were obtained from the
GFWED (Field et al., 2015) and processed for
a 50 km
Several sets of linear correlations were performed in R to better understand
seasonal and interannual relationships among MODIS C6, GFWED, and
WMO visibility data. For each set of correlations, correlation coefficients
were estimated using a Pearson's coefficient, with a standard transformation
to a
The spatial distribution of MODIS C6 active fires in lowland Bolivia and NK
is seen from 2001 to 2015 (Fig. 1b). For lowland Bolivia from 2001 to 2015, a
significant relationship was found between mean monthly MODIS C6 fire data
and
Mean monthly (January–December) time series of MODIS C6 active
fires
Mean monthly Pearson's correlations (January 2001–December 2015)
between
NA – not available
For lowland Bolivia from 2001 to 2015, higher-than-normal (i.e.,
Inverse relationships between MODIS C6 and precipitation across MERRA-2, CPC,
GPCP, and TRMM were all comparably low (Table 1). Of the precipitation sources
analyzed, the strongest observed relationship was between Bolivia MODIS C6
and MERRA 2 precipitation data, with a 95 % correlation confidence
interval of
Mean fire season (i.e., August–October) Pearson's correlations
(January 1982–December 2015) between
For NK from 2001 to 2015, higher-than-normal (i.e.,
Precipitation and fire inverse relationships were even lower over NK than
over Bolivia across all different precipitation estimates (Table 1). The
strongest observed relationship for precipitation was between NK MODIS C6 and
MERRA 2 precipitation, with a 95 % correlation confidence interval of
Moderate Resolution Imaging Spectroradiometer C6 (MODIS C6) active
fires with
During 2010, fire was widespread (Fig. 4a) and September GPCP DC values were higher and spatially coherent in northeastern Bolivia (Fig. 4b). During 2014, fire was less active in lowland Bolivia (Fig. 4c) and GPCP DC during September was lower than in 2010 (Fig. 4d). Outside of NK, fire occurred in the Cerrado, SDTF, METF, and seasonally inundated wetland biomes (Fig. 4a, c). Within NK, 223 MODIS C6 active fires were observed in 2010 (Fig. 4a), and 17 MODIS C6 active fires were observed in 2014 (Fig. 4c). During both years, fires in NK occurred primarily in the Cerrado biome on the Huanchaca plateau.
Forest loss from 2000 to 2012 (Hansen et al., 2013) displayed in
white
Forest loss from 2000 to 2012 largely corresponded to areas where MODIS C6 fires occurred from 2001 to 2015 (Fig. 5). Within NK, the majority of forest loss can be found on the Huanchaca plateau where the Cerrado biome is found, along the Cerrado–METF boundary, and in seasonally inundated wetlands. Overall, forest loss is minimal within NK compared to the unprotected areas adjacent to NK (Fig. 5). For unprotected areas outside of NK, forest loss occurred in the Cerrado, SDTF, METF, and seasonally inundated wetland biomes, and in urban and agriculture land use zones (Fig. 5). Forest loss outside of NK largely corresponded to areas where fire also occurred (Fig. 5).
Overall, mean monthly correlations among monthly MODIS C6 fire and
mean monthly GFWED variables (Table 1) were stronger than mean monthly
correlations among
Over the longer record from 1982 to 2015, significant correlations were
observed between mean fire season
Mean fire season (August–October) time series (1982–2015) of daily
Global Fire WEather Database (GFWED) variables for Bolivia including MERRA-2
precipitation
From 2001 to 2015, our analyses reveal strong fire seasonality in lowland Bolivia (Fig. 2a) and NK (Fig. 2b). Within NK, drought conditions were the main driver of fire (e.g., Table 1; Fig. 3j), while fires in unprotected areas of lowland Bolivia were controlled by a combination of drought (e.g., Figs. 3e, 4b, 6d), biome type (e.g., Fig. 5), and forest loss likely influenced by human activities (Fig. 5). Driving drought conditions in lowland Bolivia are oceanic oscillations including El Niño–Southern Oscillation (Aragão et al., 2007; Bush et al., 2008; Asner and Alencar, 2010; Marengo et al., 2011), the Madden–Julian Oscillation (Marengo et al., 2011), and Atlantic sea-surface temperature (SST) variability (Vera et al., 2006; Aragão et al., 2007; Yoon and Zeng, 2010). Oceanic oscillations alter atmospheric circulation associated with the South American monsoon (Vera et al., 2006; Aragão et al., 2007; Yoon and Zeng, 2010; Marengo et al., 2011), causing precipitation deficits and region-wide drought in the southwestern Amazon during the wet season (October–November to April–May) (Aragão et al., 2007; Yoon and Zeng, 2010).
High-fire years in 2005, 2007, and 2010 (Fig. 3a, f) correspond to years of
drought (Lewis et al., 2011; Chen et al., 2013b) and high-fire years in the
southern Amazon identified by others (Chen et al., 2013b; van Marle et al.,
2017). Prolonged drought conditions in the southern Amazon are caused by
reduced rainfall, higher-than-normal temperatures, and reduced atmospheric
moisture during the wet and dry seasons (Marengo et al., 2008). Our results
suggest the CPC DC (Fig. 3e) and regional
While our results and paleosedimentary records (Burbridge et al., 2004; Maezumi et al., 2015) show fires are frequent on the Cerrado landscape in NK (Killeen et al., 2002), determining if drought, human activities, or a combination of both were the dominant drivers of recent fire activity in the SDTF and METF biomes in lowland Bolivia is less understood. Considering unfragmented tropical Amazon forests are more resilient to fire and drought conditions (Cochrane, 2003; Davidson et al., 2012) than fragmented tropical Amazon forests (Nepstad et al., 1999; Laurance and Williamson, 2001; Fearnside, 2005; Chen et al., 2013b), we suggest the spatial correspondence of forest loss and fire in METF biomes outside of NK (Fig. 5) was associated with forest clearing activities for economic practices common in the area (e.g., Killeen et al., 2008; Fearnside, 2017).
Seen as forest loss (Fig. 5), the logging of Amazon forests has increased dry surface fuels, suppressed soil moisture, and created an environment susceptible to fire during drought (Nepstad et al., 1999; Fearnside, 2005; Chen et al., 2013a, b). Large geometric rectangular areas of forest loss observed in METF biomes and other biomes surrounding NK indicate forest loss from fire was not from lighting ignitions alone. While we did not directly monitor human activities, our results suggest human activities further amplified forest loss and fire during high-DC years. Spatial coherency between fire and forest loss seen in our results, and the results of others, suggests fire in the SDTF and METF biomes is a function of human and lighting ignitions amplified by drought during the historical record (Killeen et al., 2002; Brown et al., 2006) and the paleorecord (Bush et al., 2008). However, an area of fire and forest loss seen in the METF west of NK appears spatially random, and not necessarily because of human activities alone (Fig. 4a).
Spatial correspondence among fire, high DC, and forest loss in the METF biome west of NK is observed in 2010 (Fig. 4a, c), indicating high DC (i.e., drought) was the dominant control on fire and forest loss. Drought was not limited to lowland Bolivia in 2010. Rather, severe drought was observed across much of the Amazon basin (Lewis et al., 2011). The drought in 2010 was linked to high Atlantic SSTs and intensified El Niño conditions (Lewis et al., 2011). While our research focuses on small-spatial-scale interactions between fire and local to regional climate variability, oceanic oscillations impact drought in the Amazon (e.g., Aragão et al., 2007), and likely influenced high-fire years in lowland Bolivia during 2004, 2005, 2007, 2008, 2010, and 2011 (Fig. 3a).
Fire within NK is a function of biome type (Figs. 1b, 4, 5). In NK, fires ignited by lighting are frequent in the Cerrado biome from August to October (Killeen et al., 2002). Consistent with Cochrane (2003), a lack of fire activity was observed in the METF biome in NK from 2001 to 2015. Fire during the MODIS C6 record in lowland Bolivia and in NK was largely restricted to the Cerrado and seasonally inundated wetland biomes.
Biome boundary dynamics among the Cerrado and other biomes influence fire in the southern Amazon (Power et al., 2016). Extreme seasonal droughts can amplify the role of fire on biome boundaries among the Cerrado, SDTF, and METF. The drying out of plant biomass and soil moisture increases the potential for Cerrado grassland fire propagation into neighboring biomes (Power et al., 2016). Amazon forest boundaries are vulnerable to positive feedbacks linked to forest loss and climate-induced drought that increase forest fragmentation and fire propagation (Laurance and Williamson, 2001). Unprotected areas outside of NK show evidence of biome boundary dynamics related to fire and forest loss (Fig. 5). The majority of fires in the METF are observed at biome boundary interfaces. Both biome boundary dynamics and human-caused forest loss seem to have affected fire in the SDTF and METF biomes during our study.
We suggest
Statistical relationships further suggest
We speculate antecedent dry conditions linked to precipitation and
temperature anomalies prior to the fire season impacted high DC values and
fire in Bolivia and NK. When southern Amazon wet season drought is severe,
terrestrial water storage deficits can amplify drought and fire severity
during the subsequent dry season (Chen et al., 2013a). Wet season drought
leading to drought and fire seems plausible considering the DC is used here
as an indicator of heavy surface fuel drying over several months and deep
organic soil moisture content (Field et al., 2015). Further, this would
explain weaker linear correlations observed among fire,
Knowing the importance of DC as an indicator of antecedent dry conditions
that influence fire and visibility (e.g.,
In addition to the GFWED data used in our study, errors associated with the WMO visibility data are important to consider. Visibility data are vulnerable to errors related to human-observed measurements and are derived from spatially inconsistent weather stations distributed in the Amazon region (van Marle et al., 2017). Because of the role variable smoke transport has in relationships between fire activity and visibility, we have limited our visibility to only the broadest relationships across the large lowland area of Bolivia. We note, however, that the regional visibility signal was relatively insensitive to whether it was calculated from 4, 6, 8, or 11 stations (Fig. S1 in the Supplement).
Given these limitations, our results demonstrate (i.e., 2001–2015)
connections among fire in lowland Bolivia,
While increased FWI and fire severity are a concern for lowland Bolivia and for carbon emissions and global climate, fire leading to forest loss in the METF biome within NK was not observed from 2001 to 2015 (Figs. 2b, 4, 5). Our results suggest if human activities that amplify fire in the southern Amazon were restricted, recent fire activity could have been reduced in the METF biome. Considering the spatial distribution of fires in NK, and the spatial coherence of forest loss and fire in the unprotected METF biome outside of NK (Fig. 5), a major limitation of our study is that we did not quantify the amount of forest loss in lowland Bolivia from human activities. As mentioned by others (e.g., Bedia et al., 2015), to better understand potential impacts of fire on southern Amazon tropical forests, human activities causing forest loss and fire need to be considered. To minimize deforestation and fire in the southern Amazon, our results (e.g., Fig. 5) and those of others (Flannigan et al., 2013) suggest human ignitions need to be reduced. Considering deforestation in the Brazilian Amazon has increased since 2012 (Fearnside, 2017), and in the southern Amazon Cerrado biome since 2010 (Soares-Filho et al., 2014), land use incentives and agricultural policies implemented to reduce deforestation in parts of the Brazilian Amazon (Nepstad et al., 2014) need to be enforced throughout the Amazon. In lowland Bolivia, if land use incentives and agricultural policies to reduce deforestation are not implemented and enforced, and the demand for Amazon resources continues to increase (Fearnside, 2017), future anthropogenic deforestation and fire could worsen, particularly when drought occurs (e.g., Fig. 4a, c).
We have demonstrated how multiple data can be used to explore seasonal and interannual relationships among climate, fire, land use, forest loss, and smoke emissions. A key finding, high DC and low humidity were dominant causes of recent fire activity in unprotected and protected areas of lowland Bolivia. In addition, fire was likely enhanced by fragmented biomes because of human activities, seen as forest loss in our results. Of interest to biogeographers, fires in NK from 2001 to 2015 occurred primarily in the Cerrado biome and in seasonally inundated wetlands, and were absent from the NK METF biome with the exception of Cerrado–METF biome interfaces. Considering fire was minimal in the NK METF biome from 2001 to 2015, we recommend tropical forests in the southern Amazon and lowland Bolivia need further protection from human ignitions and deforestation. Further, considering Cerrado and seasonally inundated wetland susceptibility to fire when drought occurs, attention should be given to Cerrado expansion into seasonally inundated wetlands and METF biomes.
In addition to exploring climate, fire, land use, and biome relationships, our
results demonstrate how differences among precipitation estimates used to
calculate DC can bias DC values (e.g., MERRA-2, CPC, GPCP, and TRMM).
Differing DC values because of precipitation estimate uncertainties
demonstrate the importance of using multiple data sources when considering
relationships among climate, fire, land use, forest loss, and smoke
emissions. By using multiple data sources, we were able to extend the
historical fire record for lowland Bolivia using
MODIS C6 data can be obtained at
All authors contributed to writing and revising the manuscript. JPH was responsible for collecting MODIS C6 and Landsat data, performing statistical analyses, and analyzing all data used. MJP provided expertise on fire and climate data. RDF collected and provided expertise on GFWED and horizontal visibility data. MJEvM collected and provided expertise on horizontal visibility data.
The authors declare that they have no conflict of interest.
This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under grant 1256065. Any opinion, findings, and conclusions or recommendations expressed in this material are those of the authors(s) and do not necessarily reflect the views of the National Science Foundation. Robert D. Field was supported by the NASA Precipitation Measurement Missions Science Team and the NASA Modeling, Analysis and Prediction Program. Edited by: Jochen Schöngart Reviewed by: three anonymous referees