Phase I : "North American Forest Disturbance and Regrowth Since 1972"

In 2005, phase I of the NAFD project began. Building on the preliminary work of the protoype study, the science components of the project were expanded to the conterminous US. Also, a NASA Applied Sciences Program element was added. The primary goal of the Phase I science activity was to statisticaly describe US forest disturbance and regrowth dynamics for the past 30+ years using a sample of 23 Landsat scenes. This involved :

In order to better assess the observed Landsat spectral trajectories, multiple threads of validation are being carried out including, comparisons with FIA field inventory data , spectral canopy modeling, and stratified statistical map validation. To be redirected to the original web page of the Phase I NAFD study, please click HERE.


Sampling Design

NAFD samples

FIGURE: This map shows the location of the Eastern and Western stratum LTSS used in NAFD:Phase I over top the FIA forest type map (Ruefenacht et al. 2007. Conterminous US and Alaska Forest Type Mapping Using Forest Inventory and Analysis Data, Photogrammetric Engineering and Remote Sensing, In Press.)

The spatial grain size and temporal density of Landsat data necessary to accurately map forest disturbance and regrowth make wall to wall mapping impractical. For the NAFD project a new sampling design was created (see reference) with sampling units defined by Thiessen polygons. These are non-overlapping areas of each TM path/row address in the coterminous US, or Thiessen scene areas (TSAs).

Two stratum were used to divide the Eastern and Western halves of the country and areas with minimal forest cover were excluded. A number of objectives were defined as important for the selection of the TSAs and were calculated into inclusion probabilities for each. Two of these objectives include capture of diverse forest types and includsion of TSAs with high forest area. The final samples were randomly choosen and the sampling scheme allows for national estimations based on unequal-probability sampling statistics, along with estimates of varaince for each stratum.

Landsat Time Series Stacks (LTSS)

The Landsat Record provides a unique opportunity for understanding forest disturbance history over the last two decaces. The NAFD project uses roughly one Landsat image eery year or every two years from 1984 to 2006 for roughly 30 locations acorss the coterminous US. Inorder to efficiently uutilize these dense data, a streamlined process for assembling the Landsat Time Series Stacks (LTSSs) was developed. This processing stream includes an image selection protocol, high level preprocessing algoritms and verification procedures.

The high level preprocessing routines have been implemented in the Landsat Ecosystem Disturbance Adaptive System (LEDAPS). This system was designed for processing large quantities of Landsat imagery, and it performsre-calibration and atmospheric correction routines for calculating surface reflectance imagery, and precision registration and orthorectification routines for improving geolocation accuracy.

Verification procedures were used as a first order indicator of geometric and radiometric integrity. One method of verification was to generate movie loops of specific locations within each LTSS. For an example of these scroll over the TSA polygons in the map below. These are the LTSS that were used for the statistical validation. NAFD validation sites

National Statistics

Spectral Canopy Modeling

Accurate information on forest regrowth after disturbance is essential for understanding North American carbon dynamics. A vegetation canopy model inversion approach was proposed as a robust method for estimating forest biophysical properties. However, a fundamental requirement for successful inversion is that the model accurately represents the remotely sensed spectral response of the canopy. Three different models SAIL (Verhoef 1984), GeoSAIL (Hummerich 2001) and Flight (North 1996) were used to simulate the reflectance spectra of forest stands. The objectives were to:

Model inputs were field measurements of forest plots from the Forest Inventory and Analysis (FIA) database, ancillary data estimated using region specific allometric models (e.g. crown radius, live crown ratio and crown shape), and spectral measurements of leaf and bark reflectances compiled from the OTTER (Miller et al 1990), SNF (Hall et al 1996), LOPEX (Hosgood et al 1993), and the ASTER spectral library (Baldridge, in press).

Map Validation

For Phase 1 two complementary methods of map validation were developed. One method incorporates FIA nation‐wide field observations by plotting stand age derived from disturbance maps against FIA determined stand age. Another approach uses visual analysis of the Landsat time series along with at least one date of high‐resolution aerial photography, for a selected sample of sites. This methodology was applied to the six validation sample stacks from NAFD Phase I.

We designed an approach to identify 1) how common non‐stand clearing disturbance is for a particular year in the time series, and 2) what the underlying VCT “threshold” is for non‐stand clearing disturbance. The percentage of stand‐clearing disturbances for five of the sample sites range from 65% to 83%. Stand clearing events include fire, harvest, and urbanization, while the majority of non‐stand clearing disturbances seen in this analysis are due to partial thinning or partial fire damage.