Philip A. Townsend

Application of imaging spectroscopy to mapping canopy nitrogen in the forests of the central Appalachian Mountains using Hyperion and AVIRIS

Earth Observing 1 (EO-1) Hyperion and Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) imagery were used to predict canopy nitrogen (N) concentration for mixed oak forests of Green Ridge State Forest in Maryland. Nitrogen concentration was estimated for 27 ground plots using leaf samples of the dominant tree species from each plot that were dried, ground and analyzed in the laboratory for foliar N concentration. Foliar N data were composited based on relative species composition to determine overall canopy N concentration for the plot. Hyperion and AVIRIS images were converted to surface reflectance and related to canopy N using partial least squares (PLS) regression of first-derivative reflectance for wavelengths reported in the literature to be associated with N absorption features. The PLS model for Hyperion employed four factors and accounted for 97.8% of the variation in N concentrations and 40.4% of the variation in the spectral data whereas the AVIRIS model used three factors accounting for 84.9% of the variation in N and 72.4% of the variation in the spectral information. In the area of overlap between the AVIRIS and Hyperion images, >70% of the estimates from the two sensors were within 0.25%N of each other, indicating a very close fit between the models generated using data from Hyperion and AVIRIS. This research indicates the applicability of hyperspectral data in general and Hyperion data in particular for mapping canopy nitrogen concentration.

Land use - land cover conversion, regeneration and degradation in the high elevation Bolivian Andes

Regional land-cover change affects biodiversity, hydrology, and biogeochemical cycles at local, watershed, and landscape scales. Developing countries are experiencing rapid land cover change, but assessment is often restricted by limited financial resources, accessibility, and historical data. The assessment of regional land cover patterns is often the first step in developing conservation and management plans. This study used remotely sensed land cover and topographic data (Landsat and Shuttle Radar Topography Mission), supervised classification techniques, and spectral mixture analysis to characterize current landscape patterns and quantify land cover change from 1985 to 2003 in the Altiplano (2535–4671xa0m) and Intermediate Valley (Mountain) (1491–4623xa0m) physiographic zones in the Southeastern Bolivian Andes. Current land cover was mapped into six classes with an overall accuracy of 88% using traditional classification techniques and limited field data. The land cover change analysis showed that extensive deforestation, desertification, and agricultural expansion at a regional scale occurred in the last 20xa0years (17.3% of the Mountain Zone and 7.2% of the Altiplano). Spectral mixture analysis (SMA) indicated that communal rangeland degradation has also occurred, with increases in soil and non-photosynthetic vegetation fractions in most cover classes. SMA also identified local areas with intensive management activities that are changing differently from the overall region (e.g., localized areas of increased green vegetation). This indicates that actions of local communities, governments, and environmental managers can moderate the potentially severe future changes implied by the results of this study.

Relationships between vegetation patterns and hydroperiod on the Roanoke River floodplain, North Carolina

This study quantified relationships between forest composition and flooding gradients on the Roanoke River floodplain, North Carolina. Because flooding is highly variable in time and space, the research was designed to determine the specific hydrological parameters that control woody species abundance on the landscape scale. I specifically tested the importance of spring vs. yearly flood duration, as well as flood duration during hydrologically wet vs. dry years. Field vegetation samples of woody species composition were integrated with spatial data from a Landsat Thematic Mapper (TM) classification and a flood simulation model derived in part from synthetic aperture radar (SAR) imagery. Flood simulations were output and summarized for the periods 1912–1950 (before dams were constructed on the river) and 1965–1996 (after all of the dams were completed). Tenth percentile (dry), median, and 90th percentile (wet) hydroperiod (flood duration) regimes were generated for the spring and year, both pre- and post-dam. Detrended correspondence analysis (DCA) was used to ordinate the plot data, and correlation/regression between ordination axis scores and the flood variables were used to explore the relationships between flooding and species composition. Nineteenth percentile hydroperiod (i.e., wet conditions) correlated most strongly with DCA axis 1 (r>0.9), indicating that inundation during extremely wet years strongly controls species composition on the floodplain. The results were used to quantitatively determine the niche width for both species and mapped vegetation classes in terms of number of days flooded annually and during the spring growth period. The results suggest that spring hydroperiod is an important mechanism that may drive competitive sorting along the flooding gradient, especially during the early years of succession (i.e., pre-dam, which represents the period during which most of the forests sampled were established), and that annual hydroperiod affects the relative dominance of species as the forests mature.

Comparison of EO-1 Hyperion to AVIRIS for mapping forest composition in the Appalachian Mountains, USA

We used classification and regression trees (CART) to map forest composition with Hyperion and AVIRIS in the Central Appalachian Mountains. Imagery from both sensors exhibited strong topographic effects, with AVIRIS also having a view-angle dependent brightness gradient across the image swath. A DEM-based empirical adjustment to reflectance levels was implemented to reduce apparent topographic effects in the imagery. In general, classification accuracy improved using the topographically normalized imagery, although it is possible that the adjustments to the AVIRIS imagery diminished the superior signal:noise performance of the AVIRIS imagery. Subtle distinctions in forest composition were detectable from both AVIRIS and Hyperion imagery, and despite the superior S:N and spatial resolution of AVIRIS, classification of Hyperion images was as accurate or more accurate than AVIRIS for most species. We therefore demonstrate the utility of Hyperion imagery, but note that further comparisons are still required. In particular, the effects of sensor artifacts (such as striping and smile) must still be addressed when using Hyperion data.

Assessing flooding and vegetation structure in forested wetlands using Radarsat SAR imagery

We used 32 Radarsat SAR images collected between 1996 and 2001 over the Roanoke River floodplain, North Carolina (USA), to assess the utility of C-HH Radarsat SAR for operationally mapping flooding beneath forest canopies. Our objective was to test the sensitivity of standard-beam Radarsat to flooded forests under leaf-on and leaf-off conditions, and at multiple incidence angles (ranging from 10-46/spl deg/). We found that winter (leaf-off images) could be used to accurately map flooding (>95% accuracy) regardless of incidence angle. Mapping accuracy was acceptable for leaf-on images (>85%) but was nevertheless significantly lower than for leaf-off images. Images at shallow incidence angles (e.g. S6 mode, 41-46/spl deg/) were especially affected by speckle in the interpretations, whereas images acquired at very steep incidence angles (extended low beam, 10-23/spl deg/) exhibited a substantial amount of geometric feature displacement that limited their utility for practical applications. For mature forests, forest structure (independent of leaf-on or leaf-off status) did not appear to affect the ability to map flooding; however, separate interpretations were required to delineate flooding in very young, dense successional stands (<15 years). When controlling for flooding status, we found that Radarsat images were sensitive (in a statistical sense) to some components of forest structure (basal area, canopy height, crown depth), especially if both leaf-on and leaf-off images and images acquired at multiple angles were used for the analysis.

Predicting tropical forest carbon from EO-1 hyperspectral imagery in Noel Kempff Mercado National Park, Bolivia

An increasing interest in the ability of tropical forests to sequester large amounts of carbon has pushed scientists to look for new ways to acquire accurate estimates of biomass and other forest structural attributes over large, remote areas. We used hyperspectral images acquired in the fall of 2001 from the Hyperion imaging spectrometer to predict aboveground biomass in Noel Kempff Mercado National Park in the Bolivian Amazon. Forest structure data was collected from plots throughout the park in 1997 and 1999. We employed multiple linear regression to predict total carbon, aboveground carbon, carbon in wood, understory carbon and soil carbon as a function of reflectance at wavelengths in the visible, near infrared and short-wave infrared. The results are promising, although some uncertainty remains due to striping and geolocation error.