You are here:
SAMPLE SELECTION OF MRLC'S NLCD LAND COVER DATA FOR THEMATIC ACCURACY ASSESSMENT
Citation:
Wickham, J D. AND T G. Wade. SAMPLE SELECTION OF MRLC'S NLCD LAND COVER DATA FOR THEMATIC ACCURACY ASSESSMENT. Presented at 2001 ESRI Conference Software Application Fair, San Diego, CA, July 9-13, 2001.
Impact/Purpose:
Our research objectives are to: (a) develop new methods using satellite remote sensor data for the rapid characterization of LC condition and change at regional to national scales; (b) evaluate the utility of the new NASA-EOS MODIS (Moderate Resolution Imaging Spectrometer) leaf area index (LAI) measurements for regional scale application with landscape process models (e.g., biogenic emissions and atmospheric deposition); (c) provide remote sensor derived measurement data to advance the development of the next generation of distributed landscape process-based models to provide a predictive modeling capability for important ecosystem processes (e.g., nutrients, sedimentation, pathogens, etc.); and (d) integrate in situ monitoring measurement networks with UAV and satellite based remote sensor data to provide a continuous environmental monitoring capability.
Description:
The Multi-Resolution Land Characteristics (MRLC) consortium was formed in the early 1990s to cost- effectively acquire Landsat TM satellite data for the conterminous United States. One of the MRLC's objectives was to develop national land-cover data (NLCD) for the conterminous United States. The NLCD data set was completed in early 2000. The land-cover categories were based on the Anderson et al. (1976) thematic classification system. An important and final step in land-cover mapping is development of accuracy estimates (Scepan 1999). A national-scale accuracy assessment poses several challenges for meeting the criteria of a well-constructed probability-based sampling design: 1) non-zero and known inclusion probabilities for each pixel, 2) sufficient samples for each land-cover class to pemit estimates of variance and hence precision of each class' accuracy, and 3) cost-effective collection of reference data (Stehnian 2001). Perhaps the most challenging of these is cost- effectiveness.
A two-stage cluster sampling design was adopted in order to cost-effectively collect reference data and also satisfy the other criteria of a well-constructed sampling design. A two-stage cluster design splits a region into nested geographic partitions for collection of reference data. For NLCD, the partitions are: 1) a geographic stratum, 2) primary sampling units (PSUs), and 3) 30-meter pixels. Accuracy estimates are based on agreement between map and reference classifications for the 30-meter pixels. The nesting of 30-meter pixels within PSUs and PSUs within a geographic stratum provide the framework for random selection of samples (30-meter pixels). The sampling design partitions each mapping region into a geographic stratum with straum cells of AxB dimensions. Each stratum cell is then further subdivided into PSUs of equal size. One PSU is selected at random from each stratum cell, and then I 00 sample elements (pixels) from each land-cover class are selected at random from within the randomly drawn PSUs. For NLCD accuracy assessments, the geographic stratum cell size is 6Ox3O km, and PSUs are 6x6 km (i.e., each PSU is a 2% sample of a stratum cell). The two-stage cluster design has significant cost advantages. Sample elements are drawn only within selected PSUs and cannot occur elsewhere within the entire region.