SEARCH

SEARCH BY CITATION

Keywords:

  • hurricanes;
  • disturbance;
  • inventory plot;
  • remote sensing;
  • modeling

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Tropical cyclones disturb forest ecosystems and have the potential to alter forest structure and species composition as well as ecosystem functions including rates of nutrient cycling and biomass accumulation. Quantifying these forest disturbances is necessary to evaluate the extent and severity of damage for estimating biomass loss, developing regional carbon budgets, and making management decisions following hurricanes. In this study, we quantified forest disturbance (downed and dead and snapped trees) produced by hurricanes using a relationship between field-measured tree mortality and Landsat data that can be broadly applied to Gulf Coast forest ecosystems impacted by hurricanes. Field-measured tree mortality data was collected in Gulf Coast forests at 60 inventory plots established to monitor forest disturbance produced by hurricanes Katrina and Rita, which hit the region in 2005, and Hurricane Gustav, which hit the region in 2008. Large-scale disturbance estimates were obtained by regressing Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) data that in turn were associated with Forest Inventory and Analysis (FIA) data from the U.S. Forest Service. The use of the general relationship produced a biomass loss from dead trees of 43.9 ± 8.4 Tg C for Hurricane Katrina and 37.9 ± 6.4 Tg C for Hurricane Rita, which are near the upper limit of the expected values reported in our previous studies across a number of different forest types. Our results provide an important contribution for reliable assessments of large-scale disturbance produced by hurricanes in forest ecosystems. Improving our ability to accurately assess the impacts of hurricanes on forests and on terrestrial carbon cycles is particularly important given that climate projections suggest that hurricane intensity is likely to increase.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Hurricanes can cause high rates and complex patterns of tree mortality by wind-thrown and snapped trees [Lugo and Scatena, 1995]. This can alter patterns of forest regeneration [Lugo and Scatena, 1996; Lugo, 2000] by providing an optimal environment for colonizing vegetation [Allen et al., 2005; Denslow, 1985; Gagnon et al., 2007; Gagnon and Platt, 2008; Snitzer et al., 2005] leading to change in species composition [Xi et al., 2008]. Tree mortality associated with wind-driven disturbance may also weaken the terrestrial ecosystem sink of CO2 due to increased decomposition [Chambers et al., 2007a; Negrón-Juárez et al., 2008; Zeng et al., 2009]. Forest areas impacted by hurricanes are also characterized by more radiant energy reaching the soil surface, producing increased evaporation rates and reduced soil moisture [Carlton and Bazzaz, 1998], and changes in surface albedo which can modify regional climate [Negrón-Juárez et al., 2008].

[3] Hurricane-induced tree mortality is caused by wind, flooding, and storm surges [Negrón-Juárez et al., 2008] but the severity and extent depends on forest structure, species composition, topography and storm intensity [Brokaw and Walker, 1991; Boose et al., 1994; Zhao et al., 2006]. The immediate and direct damage to forests ranges from defoliation, which tends to be the most common effect, damage to small branches, loss of large branches and finally trunk twisting, snapping and uprooting [Brokaw and Walker, 1991]. Collateral damage is also produced when trees fall on adjacent trees. Taller trees with a larger diameter are more likely to be uprooted rather than snapped, which often results in mortality [Brokaw and Walker, 1991; Gresham et al., 1991; Snitzer et al., 2005]. Smaller trees tend to bend or twist and can also snap but are rarely uprooted [Gresham et al., 1991; Stanturf et al., 2007].

[4] While hurricane intensity has been discussed [Emanuel, 2005; Webster et al., 2005; Hoyos et al., 2006; Wu and Wang, 2008; Intergovernmental Panel on Climate Change, 2007] and the associated damage to forests is qualitatively known, quantifying large-scale forest disturbances (downed and dead, and snapped trees) has been limited. Observational studies to quantify forest disturbance produced by tropical cyclones are often constrained to small spatial scales because of the limited number of forest inventory plots that can be logistically handled postdisturbance. The small spatial scale of these studies does not allow for large-scale extrapolation. Remote sensing offers the opportunity to observe and quantify large-scale landscape disturbance. In combination with field work, remote sensing has been used to develop models to detect and quantify large-scale forest disturbance [Mildrexler et al., 2009; Chambers et al., 2007a], to predict the amount of coarse woody debris present following wildfire events [Huang et al., 2009], and to predict fire fuel conditions [Elmore et al., 2005]. Quantifying large-scale forest disturbance is a critical step in assessing the impact of tropical cyclones on landscape carbon balance and local climate [Chambers et al., 2007a; Running, 2008; Negrón-Juárez et al., 2008; Zeng et al., 2009] and in providing postdisturbance management strategies [Stanturf et al., 2007]. These methods can later be applied to areas with similar forest characteristics where field measurements are difficult to obtain [Elmore et al., 2005].

[5] In this study we estimate large-scale forest disturbances (downed and dead and snapped trees) resulting from hurricanes in the U.S. Gulf Coast region by combining field-measured tree mortality rates, remote sensing data and modeling. Our estimate is based on a relationship between observed disturbance data and a Landsat remote sensing metric across a range of U.S. Gulf Coast forest types affected by hurricanes Katrina (2005), Rita (2005), and Gustav (2008). Quantifying forest disturbance produced by hurricanes is necessary to evaluate the extent and severity of damage which in turn allows us to identify patterns of vulnerability, make management decisions following hurricanes, and design strategies for its mitigation. This quantification is also necessary to estimate biomass loss from hurricane-induced tree damage in order to obtain a proper assessment of carbon loss which can be used in climate models to add confidence to climate change projections. In addition, assessing the extent and severity of these disturbances will produce a baseline for studying the successional processes that follow.

2. Data and Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Site Descriptions

[6] Hurricane Katrina made its final landfall on 29 August 2005 at the mouth of the Pearl River at the Louisiana/Mississippi border, as a Category 3 hurricane (maximum sustained wind speed of 50–58 m s−1) on the Saffir-Simpson Hurricane scale (http://www.nhc.noaa.gov). The Pearl River Wildlife Management area (14000 ha), centered at (30.34°N, 89.67°W) is the largest area of intact bottomland hardwood forest in the southern United States (http://www.wlf.louisiana.gov) and is dominated by oaks (Quercus spp.), sweet gum (Liquidambar styraciflua), hickory (Carya spp.) and elm (Ulmus spp.). The region ranges from open to fully closed canopy and typically has a forest height of about 20 m [Chapman et al., 2008]. In this area we set up twenty-five 20 m × 20 m forest inventory plots and two 180 m × 30 m transects (divided in six 30 m × 30 m subplots) over a gradient of damaged forest.

[7] Hurricane Rita made landfall on 24 September 2005 on the Gulf Coastal Plain between Sabine Pass, Texas and Johnson's Bayou, Louisiana as a Category 3 hurricane (http://www.nhc.noaa.gov). Our study encompassed the Lance Rosier Unit (∼10100 ha), centered at (30.25°N, 94.42°W) in the Big Thicket National Preserve, Texas, an area characterized by large stands of loblolly pine (Pinus taeda), swamp chestnut oak (Quercus michauxii) and laurel oak trees (Quercus laurifolia) [Cozine, 2004]. In this area eight 20 m × 20 m inventory plots were installed to measure the forest disturbance associated with Hurricane Rita. In these plots we found sweet gum (Liquidambar styraciflua), loblolly pine (Pinus taeda), American holly (Ilex opaca), and oak (Quercus spp.). Loblolly pine dominates the upper elevations and a mixed pine hardwood mosaic is found throughout the remaining areas of study.

[8] Hurricane Gustav made its landfall on 1 September 2008 near Cocodrie, Louisiana as a Category 2 hurricane (maximum sustained wind speed of 43–49 m s−1), but weakened to a tropical storm as its forward motion slowed across southern and western Louisiana (http://www.nhc.noaa.gov). To measure the resulting forest disturbance, fifteen 20 m × 20 m inventory plots were installed in bottomland hardwood forests at the Indian Bayou, an area of ∼11500 ha centered at (30.39°N, 91.71°W), property of the U.S. Army Corps of Engineers (http://www.wlf.louisiana.gov/). The dominant trees in these plots were cottonwood (Populus deltoides), boxelder (Acer negundo), oaks, green ash (Fraxinus pennsylvanica), and sycamore (Platanus occidentalis).

[9] Figure 1 shows the maximum sustained wind at the surface (>18 m s−1) associated with Hurricanes Katrina, Rita and Gustav, and their respective tracks after landfall, as well as the areas encompassing the monitored plots. Figure 1 shows that the plots used to monitor forest disturbance produced by Hurricane Rita (2005) were located far from the area affected by Hurricane Katrina (2005). Also, the study area used to monitor the disturbance produced by Hurricane Gustav (2008) was located ∼200 km west (left) of Hurricane Katrina's main track and no damage due to Hurricane Katrina was reported at this study area. Wind surface data was downloaded from the Hurricane Research Division of the National Oceanic and Atmospheric Administration (http://www.aoml.noaa.gov/hrd/). This data is based on the H*wind model [Powell et al., 1998] and uses wind measurements from a variety of observation platforms to develop an objective analysis of the extent and strength of wind field speeds from a hurricane. The tracks for hurricanes correspond to data available at http://www.nhc.noaa.gov.

image

Figure 1. Wind field (wind data >18 m s−1), track and study area for (a) Hurricane Katrina, (b) Hurricane Rita, and (c) Hurricane Gustav. Wind field data was downloaded from http://www.aoml.noaa.gov/hrd/. Hurricane track data is available at http://www.nhc.noaa.gov/. TS, tropical storm (18–32 m s−1); H1, hurricane Category 1 (33–42 m s−1); H2, hurricane Category 2 (43–49 m s−1).

Download figure to PowerPoint

2.2. Remote Sensing Data Processing and Metric

[10] Landsat images (∼3.4 × 104 km2, each) from the U.S. Geological Survey (USGS) covering our study areas (scene p022r39, p025r039, and p023r039 for hurricanes Katrina, Rita, and Gustav, respectively) were collected before and after hurricane impact (prehurricane: 29 May 2003, 23 May 2005, 18 June 2008; posthurricane: 6 June 2006, 27 June 2006, and 20 May 2009, for hurricanes Katrina, Rita, and Gustav, respectively) at the time of maximum greenness (verified using Moderate Resolution Imaging Spectroradiometer, MODIS, data as in the work by Chambers et al. [2007a]). Landsat images are supplied as geometrically corrected radiance values and must be converted to reflectance values. First, the Carlotto technique [Carlotto, 1999] which accounts for correction due to haze and smoke contamination was applied over images as needed. Then, prehurricane Landsat images were atmospherically corrected and converted to reflectance values using the Atmospheric Correction Now (ACORN) software (ImSpec LLC, Boulder, CO). The resulting images comprised our images of reference. The posthurricane images were intercalibrated to the reference images by regressing the encoded radiance against the respective prehurricane images using invariant targets [Furby and Campbell, 2001].

[11] In wind-thrown areas, clearings are created, and material that is exposed to the satellite sensor (wood, dead vegetation, and surface litter) presents high values in the middle infrared reflectance band (band 5, 1.55–1.75 μm) from Landsat. This signal lasts for about a year; the time it takes for regrowing vegetation to cover the exposed wood and surface litter. In this work the posthurricane Landsat images were collected approximately 9 months after the hurricane landing. Spectral mixture analysis (SMA) [Adams et al., 1995] was applied to Landsat scenes (bands 1, 2, 3, 4, 5, and 7) to quantify the severity of the forest disturbance produced by each hurricane. SMA quantifies a per pixel fraction of composite end-members which sums to match the full pixel spectrum of the image [Adams et al., 1995]. Image end-members are spectra that when mixed produce spectra that fit the pixels in the image [Adams et al., 1995]. Scene-derived end-members of green vegetation (GV: photosynthetically active vegetation), nonphotosynthetic vegetation (NPV: wood, dead vegetation and surface litter), soil, and shade were obtained using a pixel purity index (PPI) algorithm [Boardman et al., 1995]. PPI and SMA tools from the Environment for Visualizing Images (ENVI, ITT industries, Inc., Boulder CO, USA) software were utilized. Change in NPV (ΔNPV) provide a quantitative measure of shifts in dead vegetation and woody biomass associated with disturbance [Chambers et al., 2007a], and were calculated by subtracting prehurricane NPV from posthurricane NPV images.

[12] Forest inventory plots were randomly established across the entire Landsat ΔNPV disturbance gradient over the study areas and statistical relationships were developed between these values and field-measured tree mortality. The gradient was determined by dividing the full range of ΔNPV into five classes (less than 0.15, 0.15–0.3, 0.3–0.45, 0.45–0.6, and higher than 0.6). To estimate disturbance across the entire impact region (forest areas impacted by wind speed higher than 18 m s−1) Landsat and MODIS (MOD09A1) ΔNPV were also correlated [Chambers et al., 2007a; Negrón Juárez et al., 2008].

2.3. Biometric Measurements

[13] Landsat ΔNPV images were used to determine the location of forest inventory plots randomly established across full gradients of ΔNPV (the disturbance metric) at each site. In each plot, for trees with diameter at breast height (DBH) ≥10 cm, we performed taxonomic identification and measurement of tree height, tree diameter, snapped tree height and number of dead, snapped and standing trees. Dead trees are defined here as uprooted trees and the term mortality quantifies the number dead trees per plot. Trees that had snapped and resprouted were not considered mortality, but the loss in biomass from damage was estimate from the height of the snapped bole. Table 1 shows the tree disturbance data collected in the field from all 60 plots used in this study as well as the dominant tree species per plot. These species are also the most dominant native species in the Gulf Coast region [Schneider and Sharitz, 1986; Megonigal et al., 1997; Schoenholtz et al., 2001; Battaglia et al., 2002; Denslow and Battaglia, 2002]. The plot data on tree mortality and forest disturbance was collected from July 2006 to August 2009 for Hurricane Katrina, in April 2008 for Hurricane Rita and from July to August 2009 for Hurricane Gustav.

Table 1. Location of Inventory Plots and Transects Used to Study the Effects of Hurricanes Katrina (K), Rita (R), and Gustav (G) on Forest Disturbancea
HurricaneLatitude (deg)Longitude (deg)ΔTTDTSTMMSH*Habitat DescriptionDominant Species
  • a

    The variables measured at each plot for trees with DBH ≥10 cm were total trees (TT), dead trees (DT), snapped trees (ST), mortality (M, %), mortality plus snapped trees (MS, %), maximum sustained wind from the H* model (H*, m/s). The Landsat ΔNPV (Δ) value, habitat description and the most common species per plot are also shown. Inv, invasive species. Transect NS (30°20′43.23″, −89°38′16.71″) (30°20′38.37″, −89°38′16.87″) and EW (30°20′25.34″, −89°38′2.72″) (30°20′25.49″, −89°38′8.33″) were established over areas of extreme disturbance.

K30.386233−89.6821250.0127060.000.2242.43Cypress tupeloTaxodium distichum, Nyssa aquatica
K30.385686−89.6985000.0317070.000.4142.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.369722−89.6923560.0419050.000.2642.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.386617−89.6864810.0321030.000.1442.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.374389−89.6935190.0023020.000.0942.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.390539−89.7025750.1719030.000.1642.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.391439−89.6953670.1832530.160.2542.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.387847−89.6817580.1812110.090.1742.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.373539−89.6934750.1618120.060.1742.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.374708−89.6862580.1617120.060.1842.43Cypress tupeloTaxodium distichum, Nyssa aquatica
K30.388944−89.6762890.2235200.060.0642.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.389953−89.7007220.2820500.250.2542.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.378769−89.6970420.289320.330.5642.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.373953−89.6884690.3028320.110.1842.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.386144−89.6783610.2014030.000.2142.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.376336−89.6971220.3619320.160.2642.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.386169−89.7019610.3618520.280.3942.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.377775−89.6867560.3613020.000.1542.43Cypress tupeloTaxodium distichum, Nyssa aquatica
K30.388092−89.7020330.4012330.250.5042.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.373247−89.6713330.3515340.210.4742.66Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.378019−89.6995640.5917600.350.3542.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.377194−89.6989530.5117530.290.4742.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.380422−89.6982330.5414330.210.4342.43Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.358653−89.6824330.5314610.430.5042.61Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
K30.361275−89.6792280.6218630.330.5042.61Bottomland hardwoodQuercus spp., Liquidambar styraciflua, Ulmus americana, and Carpinus caroliniana
G30.385550−91.6929670.1331200.060.0627.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.385033−91.6935830.2420340.150.3527.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.385550−91.6942170.2822210.090.1427.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.384500−91.6945330.4824460.170.4227.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.385033−91.6951330.2027130.040.1527.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.382633−91.6983000.1522110.050.0927.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.381833−91.6986330.3428320.110.1827.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.379667−91.6987500.0219030.000.1627.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.379383−91.6992830.3022230.090.2327.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.388117−91.7046330.0720000.000.0027.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.387283−91.7038670.1117020.000.1227.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.387800−91.7026170.3717300.180.1827.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.388250−91.7047330.2539240.050.1527.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.386750−91.7032500.4424320.130.2127.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
G30.386200−91.7029500.3843600.140.1427.39Bottomland hardwoodPopulus deltoides, Acer negundo, Quercus spp., and Triadica sebifera (Inv)
R30.290667−94.4032500.2214220.140.2921.44Pine/hardwoodLiquidambar styraciflua, Pinus taeda, Ilex opaca, and Triadica sebifera (Inv)
R30.271917−94.5400000.0020030.000.1521.42Pine/hardwoodLiquidambar styraciflua, Pinus taeda, Ilex opaca, and Triadica sebifera (Inv)
R30.272750−94.5410000.0916020.000.1321.42Pine/hardwoodLiquidambar styraciflua, Pinus taeda, Ilex opaca, and Quercus spp.
R30.267833−94.5402830.0726230.080.1921.42Bottomland hardwoodLiquidambar styraciflua, Ilex opaca, Quercus spp., and Carya spp.
R30.265667−94.5403000.1519210.110.1621.42Pine/hardwoodLiquidambar styraciflua, Pinus taeda, Ilex opaca, and Quercus spp.
R30.290750−94.4090330.1615220.130.2721.44Bottomland hardwoodLiquidambar styraciflua, Ilex opaca, Quercus spp., and Carya spp.
R30.290300−94.4027170.3418530.280.4421.44Pine/hardwoodLiquidambar styraciflua, Pinus taeda, Ilex opaca, and Quercus spp.
R30.289633−94.4033670.1118210.110.1721.44Bottomland hardwoodQuercus spp., Liquidambar styraciflua, and Acer rubrum
K-NS  0.4420520.250.3543.09Bottomland hardwoodNyssa sylvatica, Quercus spp., and Carpinus caroliniana
K-NS  0.44231820.780.8743.09Bottomland hardwoodNyssa sylvatica, Quercus spp., and Carpinus caroliniana
K-NS  0.5029600.210.2143.09Bottomland hardwoodNyssa sylvatica, Quercus spp., and Carpinus caroliniana
K-NS  0.42221120.500.5943.09Bottomland hardwoodNyssa sylvatica, Quercus spp., and Carpinus caroliniana
K-NS  0.49281430.500.6143.09Bottomland hardwoodNyssa sylvatica, Quercus spp., and Pinus glabra
K-NS  0.59271230.440.5643.09Bottomland hardwoodNyssa sylvatica, Quercus nigra, and Pinus glabra
K-EW  0.40433110.720.7443.09Bottomland hardwoodPinus glabra, Quercus spp., and Liquidambar styraciflua
K-EW  0.37241810.750.7943.09Bottomland hardwoodLiquidambar styraciflua, Nyssa sylvatica, and Quercus spp.
K-EW  0.51171410.820.8843.09Bottomland hardwoodLiquidambar styraciflua, Nyssa sylvatica, and Quercus spp.
K-EW  0.42181400.780.7843.09Bottomland hardwoodLiquidambar styraciflua, Nyssa sylvatica, Nyssa aquatica, and Quercus spp.
K-EW  0.14601820.300.3343.09Bottomland hardwoodNyssa aquatica, Fraxinus pennsylvanica, and Acer rubra
K-EW  0.1363410.060.0843.09Bottomland hardwoodNyssa aquatica, Nyssa sylvatica, and Taxodium distichum

2.4. Biomass Loss Estimation

[14] To estimate biomass loss from hurricane-related forest disturbance across the impacted forest areas, we used the approach introduced by Chambers et al. [2007a], and briefly described here. A scatterplot of minimum green vegetation (GVmin, in February) and maximum green vegetation (GVmax, in May) was used to show the ternary data with corners representing pure deciduous, pure evergreen (composed of conifers), and pure urban/nonvegetated areas using MODIS MOD09A1 data. The absolute perpendicular distance (d) of any pixel to the line connecting pure deciduous and pure evergreen behavior was calculated as a measure of proximity to the forest. If the pixel was classified as forest (d < 0.20) then a probability function determined if this pixel was deciduous or coniferous, and stem density was calculated based on a stem density probability distribution for deciduous and coniferous trees, respectively, using inventory data from the USDA Forest Service (Forest Inventory and Analysis National Program, FIA). The FIA describes forest location and area in the United States as well as tree species, size, health, total growth, biomass, removals by harvest, rates of wood production and fate of harvested wood (USDA 2007). The stem density (stem/ha) values of all the plots located within storm areas were used to generate the probability distribution of stem density. The trees with DBH larger than 12 cm were used to simulate the distribution of stem density. FIA data is available at http://199.128.173.17/fiadb4-downloads/datamart.html.

[15] Across each hurricane impact zone the number of dead and snapped trees was estimated by multiplying stem density (randomly sampled from its distribution function) by the disturbance value. The disturbance value is the percentage of dead and/or snapped trees estimated from ΔNPV. For each dead tree, an amount of biomass was generated based on biomass distributions (using FIA data) which were empirically determined to be a logarithmic normal distribution. For simplification, biomass loss from snapped trees was calculated by multiplying the biomass of the entire tree by the snapped proportion sampled from a snapped distribution function obtained from the field data. A Monte Carlo model was developed to stochastically select the density and biomass values from their respective probabilistic distribution functions and calculated the number of dead and snapped trees and their respective biomass loss at each pixel from MODIS ΔNPV. The Monte Carlo model also calculated the standard error of these calculations and the spatial autocorrelation of forest mortality.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[16] Figure 2 shows the relationship between disturbance data collected at all inventory plots (60 plots) versus Landsat ΔNPV. The largest numbers of plots are located at the Hurricane Katrina site (Table 1) due to the different research activities we are developing in the area. The lowest and highest average ΔNPV values were associated with hurricanes Rita and Katrina, respectively (Table 1). A strong relationship was observed between mortality and mortality plus snapped trees and ΔNPV (r2 = 0.4 and p < 0.0001, respectively) (Figures 2a and 2b). The statistical summary of the regressions are shown in Table 2.

image

Figure 2. (a) Tree mortality and (b) tree disturbance (mortality plus snapped trees) rates from 60 plots located at different study areas (Figure 1) and Landsat-derived ΔNPV.

Download figure to PowerPoint

Table 2. Statistical Results of Relationships of Mortality and Mortality Plus Snapped (y) Versus ΔNPV (x)a
 CoefficientStandard Errortp
  • a

    The respective relationship is y = a*x + c. The results were obtained using the statistical libraries from SigmaPlot 11 (Systat Software, Inc.).

Mortality
a0.8460.1266.718<0.0001
c−0.0370.041−0.8950.3747
 
Mortality Plus Snapped
a0.7790.1286.077<0.0001
c0.0920.0422.2210.0303

[17] Two transects were installed to monitor the mortality associated with Hurricane Katrina. These transects, one NS (30°20′43.23″, −89°38′16.71″) (30°20′38.37″, −89°38′16.87″) and the other EW (30°20′25.34″, −89°38′2.72″) (30°20′25.49″, −89°38′8.33″), were deliberately established over gradients that encompassed the largest range in ΔNPV values. These transects have different vegetation characteristics and species that can affect directly the ΔNPV values even though they were affected by a similar wind speed of 155 km/h The NS transect was dominated by oaks, spruce pine (Pinus glabra), water tupelo (Nyssa aquatica), black gum (Nyssa sylvatica), and sweet gum, while the EW transect was dominated by Nyssa spp., bald cypress (Taxodium distichum), sweet gum, and oaks. The NS transect presented 25 ± 4 (±SD) trees per 30 m × 30 m subplots and had a mortality of 45 ± 21%. The EW transect had 38 ± 21 trees per subplot and a mortality of 57 ± 31%.

[18] As the main component of carbon loss from hurricanes is related to dead trees, we used dead trees estimates to evaluate the general relationship as shown in Figure 3. The use of the general relationship produced 161.7 ± 30.9 million dead trees (equivalent to a biomass loss of 43.9 ± 8.4 Tg C) for Hurricane Katrina and 115.4 ± 24.7 million dead trees (equivalent to 37.9 ± 6.4 Tg C) for Hurricane Rita. The number of dead trees for hurricanes Katrina and Rita was 45% and 6% higher, respectively, than the mortality values reported in our previous works [Chambers et al., 2007a; Negrón-Juárez et al., 2008]. The overestimate may be related to the inclusion of data from NS and EW transects as discussed below.

image

Figure 3. Number of dead trees obtained from the relationship reported in this work (Figure 2a) and similar estimations from previous studies for Hurricane Katrina [Chambers et al., 2007a] and Hurricane Rita [Negrón-Juárez et al., 2008].

Download figure to PowerPoint

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[19] Figure 2 shows that the general relationships of ΔNPV and mortality and mortality plus snapped trees explains 40% of the variance. However, for the mortality relationship we observed that the exclusion of the EW and NW Katrina transects produced an explained variance of 60%. This suggests that the general relationship for mortality and ΔNPV could produce higher values than those previously reported [Chambers et al., 2007a; Negrón-Juárez et al., 2008]. On the other hand, the exclusion of data from the EW and NW Katrina transects did not produce changes in the explained variance for the mortality plus snapped trees and ΔNPV relationship. This may be related to the 9 month gap in time between the hurricane landing and the satellite disturbance monitoring, which allowed enough time for the loss of the entire crown of downed trees and the resprouting of snapped trees. Since the spectral reflectance is affected by the chlorophyll content and water absorption in the leaf, as well as the scattering processes between leaves and background [Roberts et al., 1998a], then the resprouting processes would make it difficult to predict the number of snapped trees.

[20] SMA is based on the spectral characteristics of a set of end-members which are assumed to be representative of the landscape. Thus, over areas with different vegetation types and tree species the use of a single set of end-members may not account for the variability of the spectral contrast and the characteristics of the landscape [Roberts et al., 1998b]. For instance, the EW and NS Katrina transect, mostly dominated by medium to low and low wind resistance tree species (using the classification presented by Duryea et al. [2007]) have subplots predominantly with high ΔNPV and high mortality values. However, few subplots presented high ΔNPV values but low mortality. Besides the differences in trees species previously mentioned, physical characteristics of the two transects could also contribute to the variability of observed mortality: (1) the transects were located 10 m from the banks of the Pearl River therefore they experienced the impact of tidal surge before the hurricane winds and (2) elevated soil in two of the six plots in the EW transect related to the deposition of dredge material from the East Pearl 40 years ago causing species changes along a gradient from loblolly pine at higher elevations to cypress tupelo forest at lower elevations over 180 m (where low mortality and low ΔNPV values were observed). Fortunately, advances in remote sensing may overcome this limitation and eventually enable the monitoring of changes in forest structure, stand characteristics, and even the possibility of monitoring individual trees over large heterogeneous areas [Chambers et al., 2007b; Frolking et al., 2009]. We also observed that the selection of deciduous and evergreen forests affects the probability of image pixels to be classified as forest pixels and or our Monte Carlo model simulation was sensitive to this selection. Nevertheless, the approach used in this study can provide important baseline estimates of disturbance from hurricanes.

[21] Our results show the lowest ΔNPV values over the plots established to monitor the impacts of Hurricane Rita. The lower ΔNPV values may be explained by the following.

[22] 1. The plots at the Rita site were located on the left side (weaker side) of the hurricane track (Figure 1). The area on the right side (relative to the direction of travel) of a hurricane is affected by the strongest winds due to the additive effect of the hurricane's wind speed and the speed of the larger atmospheric flow (http://hurricanes.noaa.gov/pdf/hurricanebook.pdf). On average the study areas for hurricanes Katrina, Gustav, and Rita were affected by winds of 42.6, 27.3, and 21.4 m s−1, respectively.

[23] 2. A higher topographic relief in the area that was affected by Hurricane Rita with respect to the areas affected by hurricanes Katrina and Gustav (as verified by using the GTOPO30 data and available at http://edc.usgs.gov/products/elevation/gtopo30/w100n40.html). In general, damage decreases at higher elevation, although it is also associated with the vegetation type [Boose et al., 1994].

[24] 3. By using the classification presented by Duryea et al. [2007], we determined that the Hurricane Rita plots were mostly dominated with a higher number of wind resistance species (64.38%) than the Katrina (39.74%) and Gustav (20.01%) study areas.

[25] Flooded forests were not severely damaged by Hurricane Katrina [Chapman et al., 2008], Rita, or Gustav but nonflooded and seasonally flooded forests were damaged by all three hurricanes. This is because flooded areas were of the cypress tupelo association, dominated by black gum and bald cypress which have a strong stem taper that may provide structural stability against bole snap [McNulty, 2002]. Nonflooded and seasonally flooded areas were of the bottomland hardwood forests association. Figure 4 shows the tree species with the highest mortality following each hurricane, as obtained from plot data. For Hurricane Katrina, five species (four species of oaks with average DBH = 49.93 ± 15.5 cm, average ± sd, and one specie of sweet gum, DBH 27.36 ± 9.9 cm) out of 11 species comprised 63% of the observed mortality. Over the area affected by Hurricane Rita, four species (three of oaks, average DBH = 21.8 ± 7.8 cm and one of sweet gum, average DBH = 23.05 ± 6.8 cm) out of a total seven species, comprised 53% of the observed mortality. Two species (Cottonwoods average DBH = 27.81 ± 7.9 cm, and boxelder, average DBH = 15.65 ± 5.1 cm), out of eight species with mortality, comprised 77% of the observed mortality associated with Hurricane Gustav. A comprehensive analysis of all factors contributing to the observed mortality is out of the scope of this paper.

image

Figure 4. Two species with the highest mortality over the study areas following Hurricane Katrina, Hurricane Rita, and Hurricane Gustav. The standard error bar is given when multiple species are present.

Download figure to PowerPoint

[26] The general relationships reported in this work were based on an extensive data set from inventory plots established to study forest disturbance from hurricanes. This included new forest plots for hurricanes Gustav, Rita, and Katrina, and existing forest plots used in other studies [Chambers et al., 2007a; Negrón-Juárez et al., 2008]. Although the number of tree species used to construct our general model is different from site to site, the significance of the models made the general relationships presented in this work suitable for estimation of damaged trees and biomass loss. Future work should encompass a greater diversity of tree species, a higher tree count within species, and a resurvey of existing plots since trees that were resprouting or damaged can die slowly over time. For example, it has been shown that pines can die slowly over a period of 6 months to 2 years after hurricanes [Duryea and Kampf, 2007]. In comparison with our previous studies, the use of the general relationship produced an increase of biomass loss, but the increment was proportional to upper limit of the expected values. In spite of differences in hurricane intensity, topography, and vegetation characteristics, we found that that the relationship between tree disturbance and ΔNPV was significant (p < 0.001), and given the moderate r2 (r2 = 0.4), could be considered as a useful general relationship in other large-scale studies. Overall, our general model produced values of forest disturbance across a number of different forests types consistent with those previously reported [Chambers et al., 2007a; Negrón-Juárez et al., 2008]. This is an important contribution and advancement in the assessments of large-scale disturbance produced by hurricanes in forest ecosystems.

[27] Although developed for forested areas in Louisiana and Texas, our model may be applicable to much of the coastal forests in the Gulf Coast, up the Atlantic coast as far north as Southern Virginia and up the Mississippi River into Arkansas, Mississippi and Tennessee where similar forest habitats occur [Batista and Platt, 2003; Conner et al., 2002; Stanturf et al., 2007; Twedt, 2006; Warren et al., 2004; Zhao et al., 2006].

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[28] We presented a model based on observed tree mortality data and Landsat-derived ΔNPV. The observational data was collected over forested areas disturbed by Hurricane Katrina (The Pearl River Wildlife Management), Hurricane Rita (the Lance Rosier Unit of the Big Thicket National Preserve), and Hurricane Gustav (Indian Bayou). Despite differences in hurricane intensity, topography, and vegetation characteristics, the general relationship relating observed tree mortality and the remote sensing metric was highly significant, enabling its use to estimate hurricane-related tree disturbance in the U.S Gulf Coast. The use of this general relationship resulted in higher tree mortality values than reported in our previous work [Chambers et al., 2007a; Negrón-Juárez et al., 2008] but that may be considered as the upper range boundary of the expected values. Overall, our general model produced values of forest disturbance across a number of different forests types consistent with disturbance values previously reported. The tree mortality from these disturbances results in large transfers of biomass from live to dead pools. If disturbances such as hurricanes increase in intensity or frequency in a warming climate, this shift in biomass can reduce the terrestrial ecosystem sink of CO2, resulting in a potentially important positive feedback to a warming climate. Thus, methodologies that can quantify forest disturbance are a critical step to understanding the biospheric impacts of large-scale disturbances.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[29] We would like to thank comments from the Editor and two anonymous reviewers. This study was supported by a grant from the National Institute for Climatic Change Research Coastal Center, which is sponsored by DOE's Office of Biological and Environmental Research.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrg651-sup-0001-t01.txtplain text document9KTab-delimited Table 1.
jgrg651-sup-0002-t02.txtplain text document0KTab-delimited Table 2.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.