ABSTRACT

In 1990, two different projects simultaneously mapped old growth in National Forests in the northern spotted owl region in the Pacific Northwest. The two projects used significantly different technology and techniques, and found amounts of old growth acreage that differed by a factor of two.

This paper investigates the procedures used in each project to determine the sources of the differing results, using four of the forests mapped as a case study. The project carried out by Pacific Meridian Resources (PMR) for the U.S. Forest Service (USFS) used Landsat TM imagery and sophisticated image processing techniques, whereas The Wilderness Society project relied primarily on photo-interpretation but also used Landsat MSS imagery. The USFS project quantitatively mapped different size and structure classes, but The Wilderness Society's project attempted to qualitatively categorize several variations of old growth. The projects operated under tremendous deadline pressure, had vastly different budgets, used very different hardware and software systems, and had very different techniques for ground truth and accuracy assessment. Thus, despite the fact that their goals were identical, these factors caused significantly different results.

INTRODUCTION

In 1990, two different projects were conducted to map old-growth forests in the Pacific Northwest. Both used the same, ecological, definition of old growth. Both mapped the same region, using data of similar vintage. Both used remote sensing and GIS. They were conducted simultaneously and took about the same amount of time to complete. However, the two studies came out with different results that differed by more than a factor of two. How could two such similar projects disagree by such a large amount? This paper compares the methods used, in an attempt to understand this difference. A companion paper [1] takes a step back and examines the institutional context in which the projects were conducted, including the contentious atmosphere that surrounded old growth, and how that context might have affected the methodologies described here (e.g., budget, timeframe).

DEFINITION OF OLD GROWTH

Determining what constitutes old growth was an essential starting point for each project. Each mapping project adopted the same definition for old growth forest, one based on ecological characteristics. This definition, known as "PNW-447" specified a number of criteria that a stand must meet to qualify as old growth, including size, density, age, number of snags and downed logs, structure of the canopy, and species mix. There are different threshold values for these criteria for different parts of the region, to take into account species variation in the different parts of the range, due to climate and topography. The definition considered but did not include a minimum stand area criterion because different management objectives require different minimum sizes of stands. It did emphasize that any inventory using the definition would have to determine a minimum acreage for a stand based on management objectives [2].

However, even though both projects used PNW-447 as their old-growth definition, they actualized the definition differently. A major difficulty for both projects was that not all of the components of the definition - in particular, snags and downed logs - can be ascertained by the remote sensing devices employed by the projects. Thus, PMR's definition simply specifies that a stand must be multistoried and have a minimum crown closure among trees of a certain size (dependent on elevation and latitude), and be a minimum of 10 acres to be considered as old growth.

The Wilderness Society mapped what it called "Ancient Forest". For four national forests, the Society defined and mapped separately three categories of ancient forest: "old growth", "high elevation ancient forest", and "other ancient forest". For the other eight national forests, Morrison mapped "ancient forest" (with no distinctions among the subcategories) and "high elevation ancient forest." For old growth, they mapped stands that had a high probability of meeting the PNW-447 criteria. The "other ancient forest" category consists of stands that meet only some of the characteristics of old growth. Morrison used 5 acres in the photo-interpreted forests and one 80m square MSS pixel (about 11/2 acres) in the forests mapped using satellite imagery.

PROJECT METHODOLOGIES

This section outlines the methodologies employed by each of the projects.

The USFS/PMR project

PMR used an innovative and rigorous technique of interpreting Landsat TM imagery to conduct its mapping project. Specifics about the techniques are available in a variety of sources, e.g., [3].

The technique involved several stages. During the image classification stage, PMR used both supervised and unsupervised classifications and merged the results using cluster analysis. PMR then manually compared the resulting classified image to aerial photographs, plot data, and other similar data. The manual editing corrected about 15% of the pixels. The procedure was repeated four times to produce four pixel maps: species, canopy structure, size, and crown closure.

The next stage was polygon creation. Because USFS personnel were familiar with maps that were products of photo interpretation, i.e., maps that had vegetation polygons, the deliverables for the project were not raster maps but rather polygon maps. PMR used decision rules similar to those used by photo interpreters to create polygons.

The crown closure pixel map was processed first. The map was scanned using two different scanning algorithms to smooth the data into polygons. For each crown-closure polygon, structure pixels were scanned, to produce sub-polygons of homogeneous crown c losure and structure. The procedure was then repeated for size-class pixels, creating homogeneous polygons of crown closure, size, and structure. The species pixels were grouped together into classes of like species using USFS classifications, then the grouped species were scanned, converted to polygons, and overlaid on the first set of polygons. Density scans were also used throughout this process to determine a threshold for homogeneous versus heterogeneous areas.

Lastly, PMR used several methods to check the resulting polygon maps for accuracy. Draft maps were distributed to USFS personnel for review. Also, both guided and random approaches were used to compare subsets of the draft maps to photo-interpretation.

All the raster processing was performed in ERDAS Imagine on Sun workstations. The polygons were created by conversion to GRID and then Arc. SAS was used for the cluster analysis, and Oracle was used for some database work.

It is notable that PMR used the title "Potential Old Growth Map" in the first set of maps they drafted. The USFS insisted that the term "potential" be dropped from the title of the final version of the maps. This indicates that PMR felt that the stands identified by their methodology were areas where old growth was likely, but not certain, to be.

The Wilderness Society project

Peter Morrison directed the old growth mapping projects for The Wilderness Society, and used a much more qualitative technique. He first worked on the Olympic National Forest (NF) [4]. He felt it important to map past as well as current forest conditions, to determine how the forest had changed over time. Thus, four sets of photographs were interpreted, from 1940, 1962, 1980, and 1982. Morrison selected "reference stands" that represented each of the classes that were being mapped. Two photo-interpreters each interpreted every quadrangle on the forest, comparing each stand to the nearest corresponding reference stand, and Morrison checked each quadrangle as well.

After some field verification, Morrison used a 1988 Landsat MSS image to determine which areas had been logged since 1982.

A variety of software was used in the process. Volunteers digitized the interpreted stands in Atlas-Draw. From Atlas-Draw, the polygons were transferred to workstation Arc/INFO and rasterized in GRID at the Washington State Department of Wildlife. These coverages were transferred to PC-Arc/INFO and EPPL7 on a PC at The Wilderness Society Seattle office. Morrison used ERDAS to perform the classification of the Landsat image.

The Society was quite pleased with the results of Morrison's work on the Olympic, but they were not pleased with the time that it took (over a year) nor the amount that it cost. The northern spotted owl was proposed as a candidate for listing as an endangered species the month after the Olympic study was released, and a final decision was due within a year. They asked him to step up the pace.

Morrison thus used less intensive procedures for the next three forests that he mapped. Only one year's worth of photos was examined. Only the Mt. Hood NF was dual-photo-interpreted; Gifford Pinchot and Mt. Baker-Snoqualmie NFs were examined by only one interpreter, plus checking by Morrison. The amount of field verification was very limited and focused on borderline areas.

With continued pressure to get work completed even more quickly, Morrison used Landsat MSS data for all of the work inventorying old growth on the remaining eight forests. He used ERDAS to do a stratified, unsupervised classification for these forests, using orthophotos and aerial photos to check the classification. The images were topographically normalized. However, he ran out of time on the last National Forest that he worked on (the Siuslaw) and no topographic correction was performed there [5].

COMPARISON OF RESULTS

Project Results: Following the methods described above, the results of the two projects were issued in early 1991 with total old growth acreages that differed by more than a factor of 2. Total old growth acreage for the study area was reported by the USFS as 4.3 million acres, and was estimated by The Wilderness Society as 2.0 million acres. The Wilderness Society reported 3.7 million acres of the broader ancient forest category.

Discrepancies: However, these numbers do not represent a direct comparison. Despite the similarities of the projects (e.g., goals, timeline), several factors complicate a direct comparison of the two studies. These factors include the nature of their old growth definitions and the differing geographic areas over which the studies were conducted.

Because Morrison mapped "old growth" as a subset of "ancient forest" for only the four forests that were photo-interpreted, a direct comparison of old growth acreage reported by the two projects is possible only for those four forests. For the other forests, only Morrison's larger category of ancient forest can be compared to PMR's old growth figures. Thus, the comparison in this paper will focus on those four: Olympic, Mount Baker-Snoqualmie (MBS), Gifford Pinchot, and Mt. Hood National Forests.

Within those forests, there are still some discrepancies about what parts of those forests were mapped. These inconsistencies include border adjustments, inholdings, and different treatment of areas administered by other national forests

Direct comparison: Table 1 summarizes the results of an "apples-to-apples" comparison, including only areas mapped by both studies. The overlap between the two studies of the old growth classification is quite small. Clearly there is significant disagreement between the projects not only over how much old growth there is, but where it is. A graphical comparison (Fig. 1) illustrates this - the overlay shows no consistent pattern or bias. Areas mapped as old growth or ancient forest by one study and not the other are randomly distributed across the forests.

Table 1

SUMMARY

Many factors contributed to the differences between the results of these two projects. Simplistically, each project used the same definition of old growth, and remote sensing and GIS technologies. However, the methodologies were completely different -- Landsat TM and statistical models versus photo-interpretation and Landsat MSS, and neither were completely appropriate for assessing all of the characteristics of old growth in the definition. Unfortunately, the lack of any discernable pattern in the comparison of the projects makes it impossible to assess specifically how the differing methodologies affected the results.

Many other factors surrounding these projects also had an effect on their results. The short timeline that each of these large projects had, and the different budgets (as reflected in the hardware and software that each project used) with which they operated had profound effects on the methodologies used. Furthermore, the institutional agenda of the projects' sponsors cannot be discounted. See [1] for an assessment of the influence of these factors on the projects.

FURTHER INFORMATION

More information, including further results of the direct overlay between the two projects, including confusion matrices and maps, a full bibliography, the text of [1], details of the geographic discrepancies alluded to, and other results from this work, are available on the World-Wide Web at http://purl.oclc.org/net/norheim/oldgrowth.

REFERENCES CITED

[1] R. A. Norheim, "Dueling databases: examining the differences between two old growth mapping projects," presented at 13th International Symposium on Computer-assisted Cartography (Auto-Carto 13), Seattle, 1997. Available on the web at http://purl.oclc.org/net/norheim/oldgrowth/dueling-db.html
[2] Old-Growth Definition Task Group, Interim definitions for old-growth Douglas-fir and mixed conifer forests in the Pacific Northwest and California. Research Note PNW-447. Portland: USDA Forest Service, 1986.
[3] R. G. Congalton, K. Green, and J. Teply, "Mapping old growth forests on national forest and park lands in the Pacific Northwest from remotely sensed data," Photogrammetr. Eng. and Remote Sens., vol. 59, pp. 529-535, 1993.
[4] P. H. Morrison, Ancient forests on the Olympic National Forest: Analysis from a historical and landscape perspective. Washington, D.C.: Wilderness Society, 1990.
[5] P. H. Morrison, D. Kloepfer, D. A. Leversee, C. M. Socha, and D. L. Ferber, Ancient forests in the Pacific Northwest: Analysis and maps of twelve national forests. Washington, D.C.: The Wilderness Society, 1991.

Last updated: July 20, 1998

Copyright 1996-98 Robert A. Norheim and IEEE