B33D-01
Modeling Pine Plantation NEP Using Landsat
The CASA (Carnegie Ames Stanford Approach) ecosystem process model predicts terrestrial ecosystem fluxes using satellite-based inputs at a maximum geographic resolution of 30 meters to infer variability in forest carbon fluxes. We are using CASA to model pine plantation net ecosystem production (NEP) under a range of standard silvicultural prescriptions, primarily thinning by fertilization interactions. Landsat scenes from WRS path/row 14/35, 21/37, and 16/34 are being used. Within each frame, all available cloud-free scenes within a two- to three-year period have been obtained from the USGS EROS Data Center processed to L1T, and subsequently converted to top-of-atmosphere reflectance using standard methods and the latest calibration parameter files. Atmospheric amelioration started with dark object subtraction (band minimum) and only proceeded to more complex techniques as necessary. Subsequent to preprocessing, the reduced simple ratio (RSR; using global min/max) was calculated for all images for each WRS path/row. Pure pine pixels in each frame were identified using unsupervised classification of the most recent leaf-off scene. We developed four age classes using two decades of Landsat data over each WRS path/row. CASA runs, which require soil parameters, and gridded climate/solar radiation in addition to satellite-derived vegetation indices, are now complete. Soil respiration and productivity estimates are being evaluated using a regionwide network of validation sites spanning the range of loblolly pine (Texas to Virginia). Preliminary results indicate that Landsat-based process modeling (1) is necessary for the scale at which land is actually managed and (2) produces estimates with an accuracy and precision affording improved understanding and management of forest ecosystems.
B33D-02
Generation of Temporally Complete Daily Nadir MODIS Reflectance Time Series
The MODIS nadir BRDF-adjusted reflectance (NBAR) algorithm provides consistent reflectance time series suitable for global terrestrial monitoring but the time series are often temporally incomplete due to persistent clouds and missing observations. This paper presents research that builds on the MODIS BRDF retrieval strategy to generate temporally complete, daily NBAR time series needed for numerous remote sensing applications, including phenological and change detection studies. Temporally complete daily 500m NBAR data are generated in a semi-physical model based manner in three steps: (1) reflectance outlier detection based on BRDF model inversion, (2) full BRDF model inversion and NBAR retrieval, (3) NBAR gap filling in periods with insufficient MODIS observations (≤ 6) for full model inversion. Two NBAR gap filling approaches are presented (i) the standard MODIS BRDF magnitude inversion approach, which scales the NBAR derived from static land cover specific archetypal BRDF parameters against the 3-6 available observations, ii) an adaptive form of the magnitude inversion approach which scales the NBAR from BRDF parameters defined by the temporally closest preceding and subsequent full inversions against the 3-6 available observations. Two scaling approaches are considered: standard ordinary least squares and a median-based robust least square regression. A daily rolling compositing approach is used whereby 16 day inversion periods are moved on a daily temporally overlapping basis providing daily NBAR, more reliable outlier detection due to the consideration of BRDF model fitting from neighboring overlapping periods, and the dynamic characterization of the surface BRDF in adaptive NBAR gap filling. Results are illustrated for two years of daily MODIS Terra and Aqua land surface reflectance data at four locations in South Africa that represent different phenological variations, degrees of missing data, and rapid reflectance change caused by fire. Derived daily NBAR and NDVI time series are shown to capture vegetation phenological variations in a coherent temporally consistent manner and the adaptive gap filling approach is shown to outperform the archetype approach when there are at least 4 MODIS observations every inversion period.
B33D-03
Retrieving LAI from Remotely Sensed Images: Spectral Indices vs. Spatial Texture
Leaves are the interface where energy and gas exchanges between the atmosphere and forest ecosystems occur. Accurate knowledge of the amount leaves is essential to successfully modeling the fluxes of water and carbon through the earth's forests. Leaf area index (LAI) is a parameter used to quantify the abundance of leaves in a given stand. Remote sensing offers the only feasible way to quantify LAI over large areas. Tremendous efforts have been devoted to this task by remote sensing scientists, but there is still a lack of concensus on how LAI can be best retrieved. Though global LAI products are available, their accuracy has remained unsatisfactory for regional applications. Previous work has primarily focused on using the spectral information in remotely sensed imagery. In this study, we compared the potential of LAI retrieval from various spectral indices derived from Landsat TM images with retrieval using the spatial information, image texture, derived from the Ikonos images. LAI on the ground was derived from allometry, LAI-2000 and the TRAC device in the Duke Forest area in central North Carolina. Our results show that the commonly used spectral indices, normalized difference vegetation index (NDVI) and simple ratio vegetation index (SRVI) were not the best choice for LAI retrieval. We found that Landsat TM derived Structural Index (SI=TM4/TM5) and normalized difference water index (NDWI), as well as Ikonos image texture are much better alternatives.
B33D-04
Vegetation Canopy Structure from NASA EOS Multiangle Imaging
We used red band bidirectional reflectance data from the NASA Multiangle Imaging SpectroRadiometer
(MISR) and the MODerate resolution Imaging Spectroradiometer (MODIS) mapped onto a 250 m grid in a
multiangle approach to obtain estimates of woody plant fractional cover and crown height through adjustment
of the mean radius and mean crown aspect ratio parameters of an hybrid geometric-optical (GO) model. We
used a technique to rapidly obtain MISR surface reflectance estimates at 275 m resolution through
regression on 1 km MISR land surface estimates previously corrected for atmospheric attenuation using
MISR aerosol estimates. MISR data were used to make end of dry season maps from 2000-2007 for parts of
southern New Mexico, while MODIS data were used to replicate previous results obtained using MISR for
June 2002 over large parts of New Mexico and Arizona. We also examined the applicability of this method in
Alaskan tundra and forest by adjusting the GO model against MISR data for winter (March 2000) and summer
(August 2008) scenes.
We found that the GO model crown aspect ratio from MISR followed dominant shrub species distributions in
the USDA, ARS Jornada Experimental Range, enabling differentiation of the more spherical crowns of
creosotebush (Larrea tridentata) from the more prolate crowns of honey mesquite (Prosopis glandulosa).
The measurement limits determined from 2000-2007 maps for a large part of southern New Mexico are ~0.1
in fractional shrub crown cover and ~3 m in mean canopy height (results obtained using data acquired
shortly after precipitation events that radically darkened and altered the structure and angular response of
the background). Typical standard deviations over the period for 12 sites covering a range of cover types
are on the order of 0.05 in crown cover and 2 m in mean canopy height.
We found that the GO model can be inverted to retrieve reasonable distributions of canopy parameters in
southwestern environments using MODIS V005 red band surface reflectance estimates at ~250 m spatial
resolution accumulated over 16 day periods. The MODIS (N=895) and MISR (N=576) estimates of forest
height and cover both showed agreement with USDA, Forest Service estimates, with MODIS mean absolute
errors (MAE) of 0.09 and 8.4 m respectively; and MISR MAE of 0.10 and 2.2 m, respectively, noting that a
sub-optimal background was used for the MODIS inversions. The MODIS and MISR MAE for estimates of
aboveground woody biomass via regression against Forest Service estimates were both 10.1 Mg.ha-1.
We found that red band MISR data for central Alaska can be used to obtain first-order estimates of forest
cover and height using a snow-free summer scene and shrub cover using a winter scene with full snow cover.
The GO model inversion results are often physically unrealistic but spatial distributions correspond to high
resolution images and reflect the potential for the multiangle/GO method to retrieve meaningful information
that is qualitatively different to that obtained using vegetation indices.
http://csam.montclair.edu/~chopping/wood/aguF08
B33D-05
Mapping Spatial Patterns of Woody Plant Cover Expansion in Chihuahuan Desert Grasslands, USA
Chihuahuan Desert grasslands are highly managed systems which support rich biodiversity and many endemic species as well as provide a valuable economic resource for cattle ranching livelihoods, with 90% of the grasslands open to grazing. Chihuahuan Desert grasslands share many characteristics with other managed grazing systems, which occupy 25% of the global land surface and are the most extensive form of land use. Grasslands around the globe, including those located within the Chihuahuan Desert ecoregion, are experiencing land cover modification from woody encroachment and increasing woody plant cover. This research used the Landsat Thematic Mapper (TM/ETM+) record from 1984-2008 to map changes in woody plant cover and identify spatial patterns and temporal trends of woody plant cover expansion in Chihuahuan Desert grasslands of the southwestern United States. Images were acquired during the dry season (May- June timeframe) in order to hold phenology constant, spectrally separate woody plant cover from grass cover, and avoid precipitation variability present during the wet monsoon. Woody plant cover was mapped annually using spectral mixture analysis (SMA) to isolate the sub-pixel occurrence of woody plant cover, quantify the percent of woody plant cover in each pixel, and track change over time using a trend analysis. Results were validated using two approaches: 1) the spatially explicit root mean-square error (RMSE) image produced by the SMA model; and 2) comparison to ground-collected measurements, estimates from aerial photography, and the Multi-angle Imaging Spectro-Radiometer (MISR) woody plant cover product. Validation results indicated good fit of endmembers within the SMA model and strong agreement between derived woody plant cover fractions and reference data. The trend analysis revealed an overall increase in woody plant cover over the 25-year period with rates and amounts of woody plant cover expansion varied spatially across the region. The next steps in this research are to correlate spatial patterns of woody plant cover expansion with land use and land management in the region and explore linkages between increases in woody plant cover and climate variability.
B33D-06
Improved Retrieval of Chlorophyll and Carotenoids Contents at the Canopy Scale Using Hyperspectral CAO Data and PROSAIL Model
The sustainability of biodiversity requires frequent and spatially detailed assessment of species number and distribution, among other information. Remote sensing is one of the most promising way to achieve this environmental management due to reasonable cost and accuracy. However, the use of airborne and spaceborne data remains challenging: sensors must combine the appropriate spatial and spectral resolutions to retrieve pertinent environmental parameters that will permit identification of specific properties of organisms present in an ecosystem. Leaf area index (LAI), canopy structure and pigment composition of vegetation are valuable information to study ecosystem dynamics and distinguish between many species. Chlorophyll and carotenoid pigments are of particular interest because they are involved in photosynthesis, but until now, remote sensing was unable to assess these pigments separately and accurately. Some advances have been made recently with the separation of these pigments in PROSPECT-5, a radiative transfer model that simulates leaf spectral reflectance and transmittance at 1 nm resolution. PROSAIL, the joint vegetation canopy reflectance model associating PROSPECT-5 with 4SAIL, the latest version of the SAIL model, was first run in direct mode to design vegetation indices sensitive to leaf pigments. Subsequently, we retrieved chlorophyll and carotenoid contents using PROSAIL with CAO (Carnegie Airborne Observatory) data acquired in Hawaii. The CAO, an imaging spectrometer coupled with a 3-D laser scanner, has already demonstrated its ability to manage biodiversity in various ecosystems like tropical rainforests or savannah. Its performance make it particularly adapted to assess vegetation structure, biochemistry, and then fluxes. The first results obtained when processing CAO images with PROSAIL are promising in terms of chlorophyll and carotenoid retrieval at the canopy scale. They show that our approach can provide original information on vegetation and hopefully new indicators to help manage and conserve biodiversity. They open new fields of investigation in the domain of mapping and monitoring Earth ecosystems from present operational (Aviris, Hyperion) or future (EnMap, HyspIRI) airborne and spaceborne sensors offering high spectral resolution data.
B33D-07
Integration of multi-temporal spectral and structural information to map wetland vegetation in a brackish Connecticut marsh
This study utilizes multitemporal QuickBird and single date LiDar canopy height data to classify the common plant communities of a tidal marsh at the mouth of the Connecticut River. A specific goal was to map the expanding distribution of non-native Phragmites australis (Cav.) Trin ex Steud (common reed), which has been outcompeting native species, particularly in disturbed marshes. P. australis spreads vigorously, forming dense monocultures that result in reduced biodiversity of plant, avian and macroinvertebrate species. We collected visible to near-infrared (VNIR) reflectance spectra of the dominant plant species S. patens (salt meadow grass), Typha spp. (cattail), and P. australis over two growing seasons to develop metrics that maximize phenological spectral and canopy height variability to distinguish these plants within a complex marsh community containing >100 plant species. Relative to other species, P. australis is best distinguished by its high NIR response and height late in the growing season. Typha spp. was well distinguished from other species by its high red/green ratio and S. patens by a unique green/blue ratio and low heights throughout the growing season. The field spectra and LiDar-derived heights were used to guide an object-oriented classification methodology using multitemporal QuickBird data collected over the same time interval as the field spectra. The classification was validated using a field inventory of marsh vegetation. Overall maximum fuzzy accuracy for the classification was 97% for Phragmites, 63% for Typha spp. and 80% for S. patens meadows; this improved to 97%, 76%, and 92%, respectively, using a fuzzy acceptable match measure. Image acquisition timing was critical for the identification of targeted plant species in this heterogeneous marsh. These datasets and protocols may provide coastal resource managers, municipal officials and researchers a set of recommended guidelines for remote sensing data collection for marsh inventory and monitoring.
B33D-08 INVITED
Dissecting the Species Energy Curve Geographically for Conservation Implications: A Case Study for North American Birds
Ecosystem energy is now recognized as a primary correlate and potential driver of global patterns of species
richness. The increasingly well tested species energy relationship (SER) is now ripe for application to
conservation management, and recent advances in satellite technology make this more feasible. The first
step in using species energy theory (SET) as a management tool requires that we recognize the best energy
correlates of species richness and the nature of the relationship across a wide range of energy levels. While
this question has been addressed many times previously, this research utilizes recent advances in satellite
data that show promise in improving our understanding of potential underlying mechanisms.
We found that MODIS Annual average Gross Primary Production was the strongest correlated with avian
richness, with a quadratic model as the strongest model. This negative slope of the quadratic model was
tested and confirmed to have a significant negative slope at the highest energy levels. This finding
demonstrates that there are three different slopes to the SER across the energy gradient of North America:
positive, flat and negative. If energy has a causal relationship with richness, then in low energy areas energy
causes richness to increase, energy has little effect in intermediate energy areas, and energy depresses
richness in the highest energy areas. This information provides a basis for applications to conservation
management, such as prioritizing land allocation to favor places of high conservation value. Knowledge of
the mechanisms underlying the SER may provide a basis for manipulation of nutrients, vegetation structure,
and/or disturbance regimes to favor higher levels of diversity in a given place. These strategies will likely be
most effective if tailored to the local energy gradient.
http://www.homepage.montana.edu/~hansen/documents/currentr2004/lindaeos.htm