Study area
On May 4, 2011 the Pains Bay Fire was ignited by lightning on the United States Fish and Wildlife Service’s (USFWS) Alligator River National Wildlife Refuge, adjacent to Pains Bay on the southern peninsula boundary the Dare County, North Carolina (35.588707°, -75.803814°). The wildfire ignition occurred in a Pinus serotina Michx (pond pine) woodland with no nearby accessible roads. The greatest rate of spread occurred during the first two days following ignition and was characterized by a rapidly spreading crown fire in the pond pine woodlands. By June 14, 2011 the wildfire had reached 17,925 ha in size and there were significant areas of organic soil ignition, smoldering, and associated smoke emissions. Organic soil consumption occurred in both the low and high pocosin vegetation throughout the duration of the flaming and smoldering stages of the wildfire. Fire crews conducted fire suppression tactics to extinguish the organic soil fires throughout the smoldering stage of the wildfire with helicopter water drops and hose lay sprinklers supplied by ground pumping stations from nearby water sources. The Pains Bay Fire was declared out after 112 days on August 24, 2011, at which time the wildfire perimeter encompassed 18,329 ha.
The mainland of Dare County, North Carolina, consists of numerous fire-adapted ecosystems. The majority of the land area in mainland Dare County is managed by the USFWS (61,512 ha) as a wildlife refuge and the U.S. Air Force (18,866 ha) as a training area. The Dare County mainland is a peninsula 22.5 km across, bordered on the north by the Albemarle Sound, on the east and south by the brackish Croatan and Pamlico Sounds, respectively, and on the west by the freshwater Alligator River. The long axis of the peninsula extends 46.7 km from north to south.
The climate in the study area is humid subtropical with an average annual temperature of 16.9°C and an average annual precipitation of 126.9 cm. The vegetation of mainland Dare County has been profoundly affected by both wildfire suppression, commercial logging, and sea level rise over the past several centuries [31, 32]. The study area has a pronounced east-west fire frequency gradient based on vegetation influenced by soil elevation above mean sea level, surface hydrology, and groundwater salinity (Fig. 1).
A generalized vegetation gradient from east to west across the peninsula consists of Spartina patens (Aiton) Muhl. (saltmeadow cordgrass), Distichlis spicata (L.) Greene (saltgrass), and Juncus roemerianus Scheele (black needlerush) that form a continuous saltmarsh shoreline band stretching from the Albemarle Sound to the south along the shoreline to the Pamlico Sound. The next vegetation band immediately to the west is comprised of low pocosin vegetation that includes pond pine woodland and an understory of Arundinaria gigantea (Walter) Muhl. (giant cane), Ilex glabra (L.) A. Gray (little gallberry), and Lyonia lucida (Lam.) K. Koch (shining fetterbush). The beginning of salt-intolerant canebrake marks the western limit of storm overwash. The highest elevation central region of the peninsula consists of a low pocosin dome dominated by shrub vegetation (little gallberry, shining fetterbush, and Zenobia pulverulenta (W. Bartram ex Willd.) Pollard (honeycup)) and surrounded by a high pocosin saturated conifer and hardwood forests dominated by Pinus taeda L. (loblolly pine), Nyssa sylvatica Marshall (blackgum), and Acer rubrum L. (red maple). In contrast, the western shore is dominated by non-pyrophytic swamp forests of Taxodium dichitum (L.) Rich. (bald cypress), Chamaecyparis thyoides (L.) B.S.P. (Atlantic white cedar), and blackgum which fringe the fresh waters of the Alligator River in a narrow band. The high fire frequency saltmarsh and canebreak communities of the eastern side and the low fire frequency river swamp forest on the west comprise the extremes of a cross-peninsula natural fire frequency gradient that ranges from 1–3 years to 100–300 years [33].
Pre-burn vegetation mapping
A Dare County, North Carolina USA aerial photography mission in spring 2004 collected color-infrared photographs with a spatial resolution of 7.5 inches per pixel [34]. The digitized photographs were orthorectified and used to develop an orthophoto mosaic for use as a base layer during vegetation community mapping. Using the orthophoto mosaic, stereo blockfile, a digital elevation model, surface hydrology data, and a digital soil survey, polygons representing distinct vegetation communities were delineated into fourteen association level communities of the USNVC within the fire perimeter (Fig. 1). The USNVC is an ecosystem-based classification scheme in which vegetation communities are grouped by their characteristic physiognomy and floristic composition [35]. To differentiate vegetation types on the orthophoto mosaic and stereo analyst block files, seven photogrammetric interpretation attributes were used: size, shape, shadow, color, texture, pattern, and association with other objects [36]. The heads-up stereo photography allowed differentiation of vegetation communities with differing dominant tree heights, canopy shapes, and canopy closure, the critical strata used to discriminate between USNVC Associations [37]. Soil series classification, above mean sea level elevation, and hydrologic soil groups were used to further inform the vegetation classification.
The fourteen associations were grouped into four forest, woodland, shrub, and herb vegetation classes: pine/hardwood swamp forest, pine woodland, shrubland, and saltmarsh. The four vegetation classes comprise the highest to lowest elevation gradient and the dry to wet surface hydrology regimes within the study area (Table 1).
The pine/hardwood swamp forests were associated with Hyde series loam and Roper series muck soils, consisting of poorly drained soils formed over loamy marine sediments. Woodlands
Table 1
Burned area (ha) of major vegetation classes and their corresponding dNBR burn severity classes
|
dNBR Fire Severity Classes (ha)
|
|
Vegetation Classes
|
Unburned/Very
Low Severity
|
Low
Severity
|
Moderate Severity
|
High
Severity
|
Total Burned Area (ha)
|
Shrubland
|
13.7
|
44.4
|
1,074.4
|
4,549.7
|
5,682.2
|
Pine woodland
|
219.0
|
1,144.1
|
3,012.2
|
4,697.1
|
9,072.4
|
Pine/hardwood swamp forest
|
465.7
|
198.1
|
339.3
|
182.7
|
1,185.8
|
Salt marsh
|
822.7
|
433.4
|
659.1
|
129.9
|
2,045.1
|
were found on Belhaven and Ponzer series muck soils, poorly drained soils that formed in organic material over loamy marine sediments. The shrubland vegetation was located on Pungo series muck soil, poorly drained soils that formed in organic material over loamy or clayey marine sediments. The saltmarsh vegetation was associated with the Currituck series mucky peat soil, a frequently tidally flooded poorly drained soil that formed in organic material over sandy marine sediments.
Differential Normalized Burn Ratio (dNBR) burn severity damage classes
Fire damage categories were defined based upon burn severity and USNVC Association vegetation classifications. Burn severity categories were selected ranging from 0 (no damage) to 3 (most severe damage). dNBR values [38] were classified into four BARC-A fire damage categories using the Jenks Natural Breaks Classification method [39, 40]. For our study site, the four burn severity damage class thresholds were: unburned/very low severity = 0-134, low severity = 135–168, moderate severity = 169–246, and high severity = 247–255. The mosaic image ArcGIS grid cells were intersected with land use/land cover polygons to illustrate burn severity within each of fourteen USNVC associations. ArcGIS® 10.1 [41] was used to carry out spatial processing for burn severity damage classes in a UTM 18 North projection with the WGS84 datum. Additional tabular processing was done using SAS® 9.2 [42].
The normalized burn ratio or Burned Area Reflectance Classification (BARC) grids based on Landsat images from 5 dates were supplied by USGS, along with grids containing the processed burn ratios. The pre-fire image was taken on May 8, 2010, and the normalized burn ratios were calculated relative to this pre-fire image. Images taken during the 2011 Pains Bay Fire on May 20th, July 4th, 20th, and 28th were used to calculate the normalized burn ratios. The initial burn ratio grids were scaled from 0 to 255. Since the wildfire was of long duration, some areas were exhibiting herbaceous vegetation regrowth while other areas within the fire perimeter were in flaming and smoldering stages. We used burn ratio grids from all four dates to calculate a new combined image product of maximum burn ratio within the fire perimeter from May 20th to July 28th (Fig. 2.). The maximum value was identified for each collocated grid cell in the four image dates. The grids, coded from 0 to 255, were smoothed with a 3x3 median filter. The grids were then reclassified to the four burn severity damage classes and again smoothed with a 3x3 median filter. Visual inspection indicated that other than smoothing there were no distortions of the original values. The final combined image product combined maximum burn severity for each cell within the entire Pains Bay Fire perimeter.
Organic soil ignition detection
The measurement of soil organic-matter emission losses during and following wildfires assists in quantifying changes in C cycling. Organic soil ignitions were determined using aerial thermal imagery from the US Fish and Wildlife Service (USFWS) and North Carolina Division of Forest
Resources (NCDFR), and Firehawk infrared data accessed from the US Department of Agriculture Forest Service’s National Infrared Operations (NIROPS). The NCDFR provided vector layers representing the occurrence of intense ground fires and scattered fires in the study area for 21 dates during the fire. Most dates had separate data representing these two classes. The intense class indicates a visible solid block of burning fuel. The scattered fire class represents an area where there were pockets of ground fire and possibly some active surface fire. The early imagery (5 dates) was collected by the NCDFR via helicopter. The later 15 dates were based on thermal imagery accessed from NIROPS. Data from one date was not used because the scattered and intense classes were not available separately. The fire layers were rasterized individually using a 5x5m cell size aligned with the normalized burn grids. The grids were assigned a 1 where there was intense or scattered fire on that date, and a 0 where no fire was identified. The grids were overlaid and summed for each fire-detect location. For the 20 dates with ground fire detects, there were up to 19 days where ground fire was identified in an adjoining cluster of grid cells. Each cell was classified into one of five classes based on the number of days with infrared detected ground fire ignition. All fire detect grid cells were merged into one ESRI ArcGIS layer and polygon overlaps were analyzed into five classes: class 0 = no detects, class 1 = 1-5 detects, class 2 = 6-10 detects, class 3 = 11-15 detects, and class 4 = 16-19 detects. A mosaic image of the fire detects (Fig. 3.) was developed to illustrate the duration of ground fire and to remove the effects of vegetation greening following the active flaming and smoldering stages of the wildfire over the duration of the 21 observation dates (Fig. 3).
Pre- and post-burn bare earth elevation measurements
Pre- and post-fire elevation difference measurements were used to determine organic soil loss. Pre-fire elevation data were derived using bare earth points from the North Carolina Flood Mapping Program [43]. LIDAR returns were acquired for January to March of 2001. Vertical accuracy was < 20 cm Root Mean Square Error (RMSE). A random point was selected on the ARNWR for each vegetation class and each burn severity damage class within the vegetation class that was greater than or equal to 100 acres. East to west transects were drawn in ArcGIS and a rectangular polygon was drawn along each transect to include a minimum of 50 LIDAR ground points. Post-fire soil elevation was measured at 50 randomly-selected bare earth points
along each of the east-west transects using survey-grade Global Positioning System (GPS). The system employed a Trimble R4 GPS Receiver - A Base Station and Rover Receiver for RTK GPS / GNSS Surveying, a Trimble TSC2 Controller and Trimble Survey Controller Software, and a Trimble RTX Verizon Cellular Data Correction Services – Cellular Network of GNSS Reference. Field measurements were corrected to the National Spatial Reference System using the NOAA Online Position User Service (OPUS). Similar approaches to assess LIDAR elevation accuracy have been conducted in other contexts [44, 45] and in the soil and vegetation assessments for the Evans Road Fire [27]. Photo images were collected for each USNVC association and each damage class to document the post-fire vegetation and soil damage.
Estimation of below-ground organics soil C emissions
Below-ground soil C emissions were calculated from GIS-derived area of land cover category combinations for the North Carolina Dare County and the SSURGO soil database elements (soil series, organic soil horizon depth, bulk density, and C content) [46]. Mean depth of organic soil horizon depth changes were calculated based on the difference of pre-burn LIDAR data points and post-burn field survey of co-located points.
Soil series within the fire perimeter consisted of seven organic soil series [46] with a thickness of the organic layers that ranged from 129.5 cm to 203.2 cm: Belhaven (Loamy, mixed, dysic, thermic Terric Haplosaprists); Currituck (Sandy or sandy-skeletal, mixed, euic, thermic Terric Medisaprists ), Hobonny (Euic, thermic Medisaprists), Hyde (Fine-silty, mixed, thermic Umbraquults), Ponzer (Loamy, mixed, dysic, thermic Terric Haplosaprists ), Pungo (Dysic, thermic Typic Haplosaprists); and Roper (Fine-silty, mixed, semiactive, acid, thermic Histic Humaquepts).)
The Microsoft Office Access database was used to import the map unit, soil physical and chemical components, and soil component horizon tables. These tables are related in the following manner: each map unit has one or more components, and each component has one or more horizons. We calculated averages for representative bulk density and representative percent organic matter within components from the data for the uppermost horizon within each component. The average values for each component were used to calculate weighted averages for each map unit using the component percentage value. The map unit averages were manipulated in ArcGIS 10.0 [41] and joined to a GIS shapefile showing the extents of each soil map unit, using the common map unit key column in each dataset. Soils GIS data were downloaded from the Soil Data Mart. The soil data were clipped to the fire boundary, and intersected with the damage and ownership classes. After calculating acreage for the clipped and intersected polygons, soil organic C was scaled up to the polygon area.
Soil organic C was calculated as adapted from Rasmussen [47] and Tan et al. [48] as follows:
SOCi = [Di x ρbi x (OMi x 0.5)/100] x 10 (1)
where SOCi is the soil organic C content (kg/m2) for the O or Oa horizon; Di is the depth of soil consumed (cm); ρbi is the soil bulk density (g/cm3); and OMi is the organic matter weight percentage. Soil C emissions (tons C) were estimated for each of the damage classes delineated for the Pains Bay Fire. Within each damage class, C emissions were summed for each of the NVCS Association, soil series, area, and mean depth of consumption.
Estimation of pre- and post-burn above-ground biomass
The estimation of C emissions from wildfires was first described by Seiler and Crutzen [49] and modified by French et al. [50]. We modified the multistep process to include: 1) determination of the area burned by relating the fire perimeter to the USNVC vegetation class and land cover class; 2) geospatial linkage of C stocks associated with land cover classifications to the C stocks in soil series and their horizons; 3) estimation of above- and below-ground C consumption and emissions the C fractions for 14 USNVC vegetation associations; and 4) estimation of C emissions from 7 soil series. Consumed C fractions were estimated for USNVC association and soil series. The estimation of the total C release (Ct) from burning of both above-ground biomass and ground layers was based on the equation modified by French et al. [51] from Seiler and Crutzen [49]:
Ct = A(Caβa + Cgβg) (2)
Where A is the total area burned (ha); Ca is the average C content of above-ground biomass (kg C m− 2) assuming the C fraction of the above-ground biomass is about 0.50; βa is the fraction of above-ground biomass consumed during a fire; Cg is the C content (kg C m− 2) of soil horizons exposed to a fire, and βg is the fraction of the soil horizon consumed by the fire.
Fire intensity on the Pains Bay Fire varied across space, resulting in heterogeneous fuel consumption across the USNVS association vegetation types. Estimates of the above-ground C consumption were conducted within each vegetation Association using differences for each of the four dNBR values. Higher dNBR values indicate higher above-ground combustion fraction. Litter, shrub, and tree foliage combustion fractions were determined for each dNBR value within each of the USNVC associations.
Above-ground C emissions were estimated using area burned, fuel loading (biomass per unit area), and consumption proportions following various studies addressing biomass combustion [49, 51, 52]. Biomass calculations were then multiplied by 0.5 to attain estimates of C. Land cover classification [34) estimates informed (1) the area burned, and (2) the basis of the fuel loading figures. Above-ground C emissions were calculated from estimates of tree density, and tree foliage, litter, and shrub biomass.
Foliage biomass estimates employed allometric biomass equations multiplied by a foliage ratio equation for eastern conifers and red maple: the dominant evergreen and deciduous species in the study area system. The eastern conifer biomass equation per tree was:
tree biomass = (0.5 + ((15000 * d2.7/(d2.7 + 364,946))) (3)
where d was the average stand diameter [53]. Equation results were multiplied by a foliage ratio using an equation for softwoods:
foliage ratio = exp(-2.9584 +(4.4766/d)) (4)
where d was the average stand diameter [54]. Red maple biomass was derived from the following equation for individual trees:
log10 tree biomass = ( -0.8602 + 1.7963 *(log10(d))) (5)
where d was the average stand diameter [55]. Equation results were multiplied by a foliage ratio using an equation for hardwoods:
foliage ratio = exp(-4.0813 +(5.8816/d)) (6)
where d was the average stand diameter [54]. Average stand diameters and tree densities were obtained through field measures of low and high pocosins and applied to equations in order to obtain biomass on a per area basis. Land area was applied to equations in order to obtain biomass on a per area basis [56].