John S. Iiames

Mapping Cropland and Major Crop Types across the Great Lakes Basin using MODIS-NDVI Data

This research evaluated the potential for using the MODIS Normalized Difference Vegetation Index (NDVI) 16-day composite (MOD13Q) 250 m time-series data to develop an annual crop type mapping capability throughout the 480,000 km 2 Great Lakes Basin (GLB). An ecoregion-stratified approach was developed using a two-step processing approach that included an initial differentiation of cropland versus non-cropland and subsequent identification of individual crop types. Major crop types were mapped for the calendar years of 2002 and 2007. National Agricultural Statistics Service (NASS) census data were used to assess county level accuracies on a unit area basis (2002), and the NASS Crop Data Layer (C DL ) was used to generate 231,616 reference data points to support a pixel-wise assessment of the MODIS crop type classification (2007) accuracy across the US portion of the CLB. County level comparisons for 2002 indicated 2.2, ―6.8, ―6.0, and ―5.8 percent of area bias errors for corn, soybeans, wheat, and hay, respectively. Detailed pixel-wise accuracy assessments resulted in an overall crop type classification accuracy of 84 percent (Kappa = 0.73) for 2007. Kappa coefficients ranged from 0.74 to 0.69 for individual ecoregions. The users accuracies for corn, soybean, wheat, and hay were 87, 82, 81, and 70 percent, respectively. There were spatial variations of classification performances across ecoregions, especially for soybean and hay. Field sizes had a direct impact on the variable classification performances across the GLB.

A Quantitative Assessment of a Combined Spectral and GIS Rule-Based Land-Cover Classification in the Neuse River Basin of North Carolina

mentation processes. Of particular importance, is the applicaThe 14,582 km 2 Neuse River Basin in North Carolina was tion of LC data for the generation of landscape-based assesscharacterized based on a user-defined land-cover (LC) classi- ment metrics to evaluate relative ecosystem condition over a fication system developed specifically to support spatially wide range of analysis scales (i.e., watershed to national) to asexplicit, non-point source nitrogen allocation modeling studies. sess impacts attributable to human land-use activities (WickData processing incorporated both spectral and GIS rule-based ham and Norton, 1994; Jones et al., 1997; Riitters et al., 1997). analytical techniques using multiple date SPOT 4 (XS), Landsat Currently, high priority non-point-source (NPS) issues are 7( ETM + ), and ancillary data sources. Unique LC classification focused on nutrient and sediment transport from the landscape elements included the identification of urban classes based to receiving streams. These NPS loadings are used to support the on impervious surfaces and specific row crop type identifi- development of total maximum daily loads (TMDL) determinacations. Individual pixels were aggregated to produce variable tions of streams and rivers (USEPA, 1999). These dynamic, ecominimum mapping units or landscape “patches” correspond- system NPS processes function at multiple analytical scales ing to both riparian buffer zones (0.1 ha), and general watershed and require relatively high-resolution geospatial data to supareas (0.4 ha). An accuracy assessment was performed using port watershed-scale modeling efforts. Landscape parameters reference data derived from in situ field measurements and required to support these spatially explicit modeling approaches, imagery (camera) data. Multiple data interpretations were used include the identification and delineation of individual LC eleto develop a reference database with known data variability ments or “patches.” Landscape “patches” typically represent to support a quantitative accuracy assessment of LC classi- the primary modeling unit of a spatially explicit landscape fication results. Confusion matrices were constructed to incor- model. They are defined in this study as contiguous and relaporate the variability of the reference data directly in the tively homogeneous LC types that can be repetitively mapped accuracy assessment process. Accuracies were reported for using remote sensor data. hierarchal classification levels with overall Level 1 classiThe characterization of riparian buffer zones is required to fication accuracy of 82 percent (n 825) for general watershed evaluate their functional capacity and ecosystem value. Typiareas, and 73 percent (n 391) for riparian buffer zone cally, riparian buffer zones are defined as areas directly adjacent locations. A Kappa Test Z statistic of 3.3 indicated a significant to the top-of-the-stream bank and extending outward in a perdifference between the two results. Classes that performed pendicular direction for a distance of approximately 20 to 30 m. poorly were largely associated with the confusion of herba- Riparian buffer zones play important functional roles in nutriceous classes with both urban and agricultural areas.

Leaf Area Index (LAI) Change Detection Analysis on Loblolly Pine (Pinus taeda) Following Complete Understory Removal

The confounding effect of understory vegetation contributions to satellite-derived estimates of leaf area index (LAI) was investigated on two loblolly pine (Pinus taeda) forest stands located in Virginia and North Carolina. In order to separate NDVI contributions of the dominant-codominate crown class from that of the understory, two P. taeda 1 ha plots centered in planted stands of ages 19 and 23 years with similar crown closures (71 percent) were analyzed for in situ LAI and NDVI differences following a complete understory removal at the peak period of LAI. Understory vegetation was removed from both stands using mechanical harvest and herbicide application in late July and early August 2002. Ikonos data was acquired both prior and subsequent to understory removal and were evaluated for NDVI response. Total vegetative biomass removed under the canopies was estimated using the Tracing Radiation and Architecture of Canopies (TRAC) instrument combined with digital hemispherical photography (DHP). Within-image NDVI change detection analysis (CDA) on the Virginia site showed that the percentage of removed understory (LAI) detected by the Ikonos sensor was 5.0 percent when compared to an actual in situ LAI reduction of 10.0 percent. The North Carolina site results showed a smaller percentage of reduced understory LAI detected by the Ikonos sensor (1.8 percent) when compared to the actual LAI reduction as measured in situ (17.4 percent). Image-to-image NDVI CDA proved problematic due to the time period between the Ikonos image collections (2.5 to 3 months). Sensor and solar position differences between the two collections, along with pine LAI increases through multiple needle flush, exaggerated NDVI reductions when compared to in situ data.

Analyst variation associated with land cover image classification of Landsat ETM + data for the assessment of coarse spatial resolution regional/global land cover products

This study examined analyst variation associated with land cover (LC) image classification using 30 × 30 m Landsat ETM+ data for the assessment of coarse spatial resolution regional/global LC products. The study was designed to test the effect of varying training site selections (location and number) among six analysts performing a supervised classification on a Landsat ETM + image. Design constraints maintained other aspects of the classification process constant (i.e., type of classifier, choice of band combinations, etc.). Results indicated that training site selection alone did not provide a predictive measure of classification accuracy. Only when training data selection was combined with variations in spatial resolution did significant differences occur. Differences in classification accuracies between analysts increased threefold in the aggregation process from 90 × 90 m to 1200 × 1200 m. Error sources (i.e. analyst differences) and the dynamics of the spatial aggregation process can potentially account for differences in environmental modeling outcomes.

Basal Area and Biomass Estimates of Loblolly Pine Stands Using L-band UAVSAR

Fully polarimetric L-band Synthetic Aperture Radar ( SAR ) backscatter was collected using NASA ’s Unmanned Aerial Vehicle ( UAV ) SAR and regressed with in situ measurements of basal area ( BA ) and above ground biomass ( AGB ) of mature loblolly pine stands in North Carolina. Results found HH polarization consistently displayed the lowest correlations where HV and VV exhibited the highest correlations in all groups for both BA and AGB . When plantation stands were analyzed separately (plantation versus natural), correlation improved signifi cantly for both BA (R 2 = 0.65, HV ) and AGB (R 2 = 0.66, VV ). Similarly, results improved when natural stands were analyzed separately resulting in the highest correlation for AGB (R 2 = 0.63, HV and VV ). Data decomposition using the Freeman 3-component model indicated that the relative low correlations were due to the saturation of the L-band backscatter across the majority of the study area.

Uncertainty Analysis in the Creation of a Fine-Resolution Leaf Area Index (LAI) Reference Map for Validation of Moderate Resolution LAI Products

The validation process for a moderate resolution leaf area index (LAI) product (i.e., MODIS) involves the creation of a high spatial resolution LAI reference map (Lai-RM), which when scaled to the moderate LAI resolution (i.e., > 1 km) allows for comparison and analysis with this LAI product. This research addresses two major sources of uncertainty in the creation of the LAI-RM: (1) the uncertainty associated with the indirect in situ optical measurements of southeastern United States needle-leaf LAI and (2) the uncertainty in the process of classifying land cover (LC). Uncertainty within the loblolly pine (Pinus taeda) in situ data collection was highest for the assessment of the plant area index (PAI), Le (27.2%), and the woody-to-total ratio, α, (30.6%). The needle-to-shoot ratio, λE, and the element clumping index, ΩE, contributed 14.9% and 9.3%, respectively, to the uncertainty in the calculation of LAI. Combining LC differences (3.4%) with the uncertainty within the loblolly pine component resulted in doubling the LAI-RM variability (σ = 0.50 to σ = 0.97) at the 1 km2 validation site located in Appomattox, Virginia, USA.