B23F-01 INVITED
Trends in Mapping, Measuring, and Monitoring Land Cover Change
The necessity for improved maps and statistics documenting the rates and characteristics of land cover change from local to global scales has been driven by the acceptance that land change has pervasive and substantial environmental consequences. The expanding need for accurate land cover change characteristics data over long time periods and large geographic areas has stimulated both methodological and mission advances. Two important methodological advances include the use of the continuous fields approach for quantifying key landscape characteristics including vegetation cover and surface imperviousness, and the increased emphasis on using probability sampling to precisely estimate land cover change rates. We are using Landsat-based sampling within ecoregions to assess 1972-2000 land change in the United States. An additional methodological trend is the use of multi-source remotely sensed data for land cover change assessments. An example that blends these three elements is our use of MODIS and Landsat to map and measure 2000-2005 global deforestation. MODIS forest fraction maps provide an annual source of forest change locations that provides an efficient means to stratify probable change. Automated classification of randomly sampled Landsat scenes provide a means for detecting forest change within forest biomes and generating precise estimates of period deforestation. Studies such as this are enabled by the NASA Earth Observing System program and the NASA-USGS Landsat Data Continuity Mission. Together, these missions extend the global Earth observation record to unprecedented levels and enable new generations of detailed land change assessments.
B23F-02
Monitoring National Forests in Eastern U.S. Using Landsat Observations
National Forests (NFs) are protected forests and woodland areas, comprising about 8.5 percent of the total land area of the United States. This study examines forest land dynamics within and around NFs using historical Landsat data. Selected NFs are distributed across eastern U.S., including the De Soto NF in Mississippi, the Francis Marion NF in South Carolina, the Cherokee NF and Pisgah NF in North Carolina, the Allegheny NF in Pennsylvania, the Hiawatha NF in Michigan, and the Superior NF in Minnesota. For each NF, a Landsat time series stack (LTSS) consisting of 1 image every year or every two years from 1984 to 2005 has been assembled. The LTSS are being analyzed using a highly automated multi-temporal change analysis algorithm. The algorithm extracts the two decadal forest disturbance history and tracks the recovery process that may follow from disturbance. Comprehensive validation of the derived disturbance and recovery products is being conducted. The validated products will be used to compare spatial and temporal patterns of forest dynamics among the selected NFs and to evaluate whether variations in such patterns can be linked to management policies. We will also evaluate the observed forest changes within the NFs compared with those occurred in the surrounding areas to further assess the impact of management practices on NFs. Results from this study will not only provide insights into the dynamics of NFs in eastern U.S. over the last two decades, but will also offer a new approach for monitoring the NFs using the spatially and temporally comprehensive Landsat record or similar remote sensing data archives.
B23F-03
Pan-Tropical Forest Clearing, 2000-2005
Quantifying rates of tropical forest cover clearing allows for improved biogeochemical cycle and climate change modeling, management of forestry and agricultural resources, and biodiversity monitoring. As a result, there is a critical need to monitor forest clearing over large areas in a timely manner. While the use of satellite-based observations for monitoring tropical deforestation is well established, consistent and timely monitoring of the entire humid tropics has not been implemented and limits the effective management of this important resource. This paper presents a probability sampling approach employing multi-resolution satellite data to provide timely, synoptic estimates of humid tropical forest cover loss. Biome-wide change indicator maps were created using moderate spatial resolution imagery for 2000 to 2005 from the MODerate Resolution Imaging Spectroradiometer sensor (MODIS). A sample of 183, 18.5km by 18.5km blocks of high spatial resolution image pairs from the Landsat Enhanced Thematic Mapper Plus sensor was used to determine biome-wide area of forest clearing. The sampling strategy employed the MODIS data in the design to stratify the blocks and also in the analysis via a survey sampling regression estimator of forest clearing. This statistically rigorous sampling strategy provides a biome-level clearing estimate with known uncertainty. Forest was defined as greater than 25% canopy cover and change was measured without regard to forest land use. All tree cover assemblages that met the 25% threshold, including intact forests, plantations, and forest regrowth, were defined as forests. Forest area cleared for the biome is estimated to be 1.53% with a standard error of 0.106%. This translates to an estimated area cleared of 29.4 million hectares with a standard error of 2.1 million hectares representing a 2.54% reduction in year 2000 forest cover. Rates of clearing are on a par with those from the 1990's. Regional variation is pronounced, with 48% of forest clearing occurring in Brazil and nearly two-thirds overall in Latin America. Indonesia accounts for 12% of total biome forest cover loss and Asia as a whole one-third. Africa, while a center of widespread, low-intensity selective logging, contributes only 5% to the estimated loss of humid tropical forest cover. Nearly one-third of all clearing occurs in less than 4% of the biome area. Forest clearing as a percentage of year 2000 forest cover for Brazil (3.9%) and Indonesia (3.5%) easily outpaces the rest of Latin America (1.5%), the rest of Asia (2.7%) and Africa (0.7%).
B23F-04
Scaling effects on forest area estimation in the three Lake States of USA
While fine resolution land-cover datasets (e.g. 30-m Landsat data) are appropriate and verifiable for local land use planning, coarse land characterization datasets (1-km resolution) are more suitable for large scale ecological analysis. A better understanding of scaling-up effects on estimating some important landscape characteristics (e.g. forest cover percentage) is critical for improving ecological applications at large scales. This study illustrated scaling-up effects on regional forest cover estimates in Minnesota, Wisconsin, and Michigan of the USA using 30-m land-cover maps (1992 and 2001) produced by the National Land Cover Dataset. The 30-m land-cover maps were scaled up to 1-km resolution and the forest cover percentages before and after the scaling process were compared at the county level. The mean difference of forest area estimates at county level was 8% ranging from 0 to 17% within a 95% confidence interval. A simple empirical model allowed prediction of the scaling effect from data at either resolution. Mean difference between observed and predicted scaling effects at 1-km resolution for a spatial cross-validation test of the model was 2.5% (Std. = 1.9%, standard error = 3.1%), compared to 2.8%, 2.2%, and 3.6%, for a temporal test. Cross-validation of the empirical model indicates that uncertainties in forest area estimates caused by the scaling-up process could be quantified in a simple and fast way with a standard error of 7.6%. Furthermore, scaling-up effects on forest area estimates appear to be both spatially and temporally consistent as well as projection independent. The identified empirical relationship may have broad applicability for large-scale ecological applications using coarse resolution data.
B23F-05
A Prototype of Updating National Land Cover Dataset by Using Landsat Imagery
The USGS National Land Cover Database (NLCD) 2001 represents the nation's land cover status based on a nominal date of 2001, and is now being widely used by many scientific research groups and government organizations. However, fundamental land cover changes due to many natural and anthropogenic factors continue to occur. Maintaining current land cover information is vital and relevant. Here, we report on a method to update the NLCD 2001 to a nominal date of 2006. Landsat scenes in the same season in 2001 and 2006 are first acquired according to satellite paths and rows. Images are then normalized to allow calculation of change vectors between the two dates. Conservative thresholds based on Anderson Level I land cover classes are used to determine areas of change and no-change. By assuming that the NLCD 2001 is accurate, data from NLCD 2001 is used as the baseline for updating land cover data to 2006. Land cover classification is accomplished within change areas by obtaining training datasets from unchanged NLCD 2001 areas. If no change has occurred, NLCD 2001 base information remains the same. Landsat scene pairs of 2001/2006 were tested in Seattle, Washington; San Diego, California; Sioux Falls, South Dakota; Jackson, Mississippi; and Manchester, New Hampshire, for their uses in generating Anderson Level II land cover classes and percent impervious surface. Results from the five tested sites show that the majority of changes of land cover characteristics are captured and updated. Percent impervious surface associated with new urban developments are also updated for all study areas. Proportional changes between the two time periods in these areas vary from 4.2 percent (New Hampshire) to 9.9 percent (Mississippi) for land cover, and from 6.9 percent (San Diego area) to 9.2 percent (Seattle area) for imperviousness.
B23F-06
Retrieving Moderate Resolution Biophysical Parameters by Fusing Landsat-like Data and MODIS Land Products
Landsat data have been successfully used for retrieving biophysical parameters and serve as an important validation source for coarser-resolution satellite instruments such as MODIS. A continuous Landsat data record is critical for many applications. However, the scan-line corrector problem on Landsat 7 and the age of Landsat 5 are threatening the continuity of Landsat data record before a new Landsat mission starts operation. Landsat- like data sources such as the Advanced Wide Field Sensor (AWiFS) aboard the Indian Remote Sensing Satellite (IRS), the Charge Coupled Device (CCD) camera aboard the China-Brazil Earth Resources Satellite (CBERS), and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) aboard EOS/TERRA have spatial resolutions and bandwidths similar to Landsat TM/ETM+. They are good candidates for mitigating a possible gap in Landsat operations and thus reducing Landsat program risk. However, different acquisition times and the lack of blue or thermal band limit their abilities in retrieving surface reflectance and biophysical properties. Two approaches have been developed to retrieve biophysical parameters for these Landsat-like data sources. First, a relative atmosphere correction approach was developed to correct AWiFS data to surface reflectance using MODIS surface reflectance as reference. The corrected surface reflectance is consistent with MODIS surface reflectance through one step processing and thus biophysical parameters such as leaf area index (LAI) can be retrieved based on the corrected surface reflectance using MODIS algorithms. The same approach can be extended to other Landsat-like data sources (CBERS, ASTER) for retrieving biophysical products. Second, a general empirical relation model (GERM) was developed to relate satellite digital number (DN) with MODIS LAI directly through MODIS homogeneous pixels. This approach will be demonstrated using Landsat/Landsat-like data and the MODIS LAI product. These results show the power of combining multi-resolution data sources for land parameter retrievals, using MODIS as a reference data set.
B23F-07
The Suitability of Landsat SLC-off Data to Characterize Mid-decadal Tropical Forest Cover in the Congo River Basin, Africa.
Accurate depictions of tropical land cover and land cover change are required for important applications relating to human livelihoods and ecosystem services. Landsat data has been used to map forest cover change in many tropical areas at high spatial resolution. Data loss due to persistent cloud cover and atmospheric contamination are frequent problems in these efforts. Contemporary forest cover characterization and change estimates in the tropics are further hampered by the Landsat 7 scan-line corrector (SLC-off) malfunction which creates scan gaps and increases data loss in every acquisition. This paper examines the effects of Landsat scan line corrector problems and cloud cover on data requirements necessary to produce a contemporary, mid-decadal forest cover characterization for the Congo River Basin. 30 Landsat SLC-off acquisitions from 2004 through 2006 were obtained for six unique path/row combinations for a forested area in the Democratic Republic of Congo. Images were combined into five final, six path/row mosaics of one, two, three, four and five scenes per path/row using an automated, pixel-based mosaic method in which the highest quality pixel from all image inputs is used for analysis. Data loss due to missing scan lines and clouds totaled nearly 25 percent in the single acquisition landscape mosaic; 23 percent due to SLC-off issues and two percent due to clouds. Total data loss decreased to nine percent with one additional acquisition per path/row but 4.5 percent of this was due to cloud cover. After compiling five acquisitions per path/row, total data loss was approximately one percent. Cloud cover impacts are present in each mosaic as data used to fill scan gaps frequently contained cloud. High spatial resolution forest characterization and change detection in the tropics of central Africa is not possible using a single best image approach nor likely to be mapped on a frequent time step as clouds and scan gaps will require multiple acquisitions over the same area in each time period. A conservative estimate of the number of Landsat SLC-off acquisitions needed to map the entire Congo basin for the mid-decadal time period is 550 spanning three years time.
B23F-08
The Availability of Cloud-free Landsat ETM+ Data Over the Conterminous U.S. and Globally
This research evaluates the availability of cloud-free ETM+ data over the conterminous U.S. and globally, specific to broad terrestrial remote sensing application requirements, including obtaining at least one cloud-free ETM+ observation in a year, a season, and two different seasons, or at least a pair of cloud-free observations no more than 16, 32, 48, 64, and 80 days apart within a year or a season. Probabilities of obtaining cloud-free observations are derived by analysis of three years of USGS Landsat Project ETM+ cloud fraction metadata (a total of 236,000 records for 2000-2002). Spatially explicit probability maps and summary statistics are presented. The results are relatively stable across the three years. Over the conterminous U.S., where every Landsat scene is acquired by the Landsat project, applications requiring at least one cloud-free observation in a year, a season, two different seasons, or at least two cloud-free observations occurring even within a 16-day period in a year, are largely unaffected by cloud cover (mean probabilities of cloud-free data availability > 0.94), except for certain Winter applications and cloudy scenes near the U.S.-Canada border and the Great Lakes. Cloud becomes a constraint for U.S. applications when at least two cloud-free observations within a season are required. Globally, only land applications requiring at least one cloud-free observation per year are largely unaffected by cloud cover and the reduced global ETM+ acquisition frequency; all the other application requirements considered are severely affected. The implications of these results for the development of new approaches to mitigating cloud contamination in the USGS ETM+ data archive and the global acquisition rate for the future Landsat Data Continuity Mission are discussed.