INDEX

• Introduction
• Products
• Project Personnel
• Study Site
• Photography, maps, and equipment
• Classification system
• Project methodology
• Accuracy assessment
• Results
• Discussion
• Conclusion

The riparian lands of the Pike & San Isabel National Forest are used for a variety of diverse applications such as outdoor recreations, livestock grazing, water supply, wildlife habitat and scenery. In order to properly manage these valuable multiuse lands, the Forest is developing an accurate and useable set of riparian overlays registered to 7.5’ quadrangle maps in digital format for the entire Forest.

The photo interpretation portion of this project was undertaken through a USFSContract. It began in June 1991 and was completed in December 1991. Another contract (#40-82-BH-2-0334) was issued for the accuracy assessment and documentation.

The objective of the project was to identify and map selected riparian vegetation communities (comprised of one or more plant associations) in the Forest using high altitude color infrared large format camera (LFC) photography as the primary data source. Overlays will be used to inventory riparian areas on a per class basis. They will also serve as a baseline for monitoring the effect of management prescriptions.

The project deliverables included the interpreted LFC photographs and a set of riparian overlays registered to 7.5’ mylar quads. The overlays contain labeled polygons that define the specific riparian vegetation communities being classified. Future products, being created under a different contract, will be a digital database build from the overlays supplied by this project.

The project was a coordinated effort involving the Pike & San Isabel Supervisors Office (S.O.), the Leadville Ranger District (R.D.), and James Ward (Contractor). Responsibilities for each group were as follows:

Forest SO – Dave Winters served as the COR on the project, provided administrative support and guidance, developed the classification system, and provided field support.

Leadville RD – Staff members supplied field data for the accuracy assessment.

James Ward – Interpretation of riparian vegetation and transfer to quad registered overlays, analysis of field data, and documentation.

The study site encompassed the twenty-four quads comprising the north end of the Arkansas River watershed. The quads included are:

Homestake Reservoir         Leadville North           Leadville South  
Mt. Champion                      Mt. Massive                Granite
Independence Pass            Mt. Elbert                    South Peak
Pieplant                                Winfield                       Mt. Harvard
Buena Vista West               Buena Vista East       Castle Rock
St. Elmo                               Mt. Antero                    Cumberland Pass
Nathrop                                Cameron Mt.               Gribbles Park

It is planned that eventually the entire Forest will be done.

Aerial Photographs

The LFC 9x18-inch, color infrared project photographs were acquired during August 1988 with a NASA ER-2 aircraft. The following is a list of selected photo parameters.

Flight Number: 88-117, acquired by NASA, Ames Research Center

Sensor Package: A-4 configuration (HR-732 LFC and Wild RC-10)

Sensor Type: Itek HR-732 Large Format Camera

Focal Length: 24"

Film Type: High Definition Aerochrome Infrared (SO-131)

% Overlap: 60%

Aircraft Altitude: 65,000 ft. (MSL), yielding a nominal scale of 1:30,000 based on an average ground elevation of 9,000 feet.

The quality of the photography was excellent, although there were a few scattered clouds and shadows to contend with. The color balance and resolution of the film were outstanding.

Maps

The Pike & San Isabel NF supplied 1:24,000 scale topographic maps on mylar, orthophoto quad transparencies, and a secondary series Forest map (1:125,000). The map products were used for the following purposes:

1.  to serve as a base of reference for a photo index;
2.  to aid the interpreter with reference to hydrography, roads, cultural features, elevations, and contours;
3.  orthos were used to provide 1:24,000 scale base maps for photo to map transfer of riparian polygons;
4.  to layout the sampling system transects.

Equipment

All of the photointerpretation (PI) equipment used during the project was furnished by the interpreter. An ordinary table top light table (3’x2’) modified to handle film rolls provided an adequate light source. A Leitz MS-27 mirror stereoscope with 3x optics was used for the PI work. The only problem or inconvenience was the absence of stereo scanning capability, which resulted in either the stereoscope and/or the photo stereo model having to be moved about when it became necessary to view different parts of the photographs. A Bausch and Lomb Stereo Zoom Transfer Scope was used to transfer polygons from the interpreted photographs to the 1:24,000 scale orthophotoquads. A 10x tube magnifier was used occasionally when a more detailed inspection was required. To delineate and label the very small polygons on the photographs, a Pilot SC-UF pen was used.

The following classification system was developed by Dave Winters, Pike & San Isabel NF Fisheries Biologist. It was used for the identification and classification of riparian plant communities found within the project area. Plant communities were classified to a minimum mapping size of at least a one acre with many going well below an acre. The classification system worked very well in relation to the scale, resolution, and emulsion of the LFC photography.

CWA – cottonwood, 10-40% crown closure
CWB – cottonwood, 40-70% crown closure
CWC – cottonwood, 70-100% crown closure
ASA – aspen, 10–40% crown closure
ASB – aspen, 40-70% crown closure
ASC – aspen, 70-100% crown closure
W – willow
AW – alpine willow
M – wet meadow
GR – mesic grass
E – evergreen tree
X – non riparian islands within a riparian zone
NV – non vegetated riparian (sand or gravel bars)
L – pond or lake, a pooled body of water regardless of size or permanence
R – riparian area to narrow to delineate and create a polygon

The above vegetation classes were often found in combination with one another. For example, the interpreter would find a willow stand with spruce/fir growing among the willow and label it W/E. When combinations were found, the dominant type was placed first followed by a / which separated it from the co-dominant type.

The project was done in two stages, photointerpretation and geographic information system (GIS) database development, and was spread out over a 2 year period. This report is documenting the photointerpretation stage methodology only and not the GIS.

Seven general project tasks were performed in order to derive the final products described earlier:

1. Photograph preparation
2. Photointerpreter training
3. Polygon delineation
4. Polygon labeling
5. Polygon transfer from photo to map
6. Accuracy assessment
7. Documentation

Photograph preparation

Upon receipt of the LFC photographs, a quality inspection was made and a photo index was drafted. The project duplicate positive film transparencies were then laminated and the effective area was delineated on every other photograph.

Photointerpreter Training

Training the photointerpreter is a very important step in any PI project and should never be under emphasized. Ideally, the photointerpreter should visit the project site and view the objects on the ground that he will be interpreting on the aerial photographs. This project required the photointerpreter to visit the field site during the initial phase of the project and was accompanied by Dave Winters, COR. Mr. Winters provided a great deal of very valuable information to the interpreter about the ecology of the area that made the interpretation more accurate. He also made certain that the interpreter saw a sample of the diverse riparian areas that exist on the study site. The field trip lasted three days, which is rather short, but was all the time and money budgeted for that part of the project. The more field time the interpreter has, the more accurate the maps will be, but the three days were sufficient as verified by the accuracy assessment. David continued to answer occasional questions about the riparian vegetation by telephone throughout the project.

Polygon Delineation

After the photographs were laminated, indexed, and delineated for effective area, and after the interpreter felt he had sufficient knowledge of the area, the riparian vegetation delineation began. A mirror stereoscope with 3x optics and a light table were used to complete this phase of the project.

The first interpretation step was to delineate riparian from non-riparian areas on the project color infrared photographs – these photographs emphasized wetlands superbly. Next, within the riparian areas, the vegetation differences were delineated. Separation of riparian vegetation classes were identified first by color, then texture, and finally by association with riparian topography.

A standard rule for all PI projects is that the boundary lines of all polygons along the border of the effective area must smoothly match with their respective polygons on the adjacent interpreted photograph. This was easily accomplished on this project by removing the common frame between interpreted photographs and then stereoscopically matching the edges of their effective areas.

Polygon Labeling

The labeling of polygons on each photograph was performed as soon as all of the polygons on a particular photograph had been delineated, i.e. on a photo per photo basis. Doing it this way seems to add to the consistency of a project and keeps the interpreter from writing a label across a riparian area before it has been delineated.

Polygon Transfer

The next project step required that the riparian polygons and their labels be transferred from the LFC photographs to 1:24,000 scale orthophoto quads. This task was accomplished using the Bausch and Lomb Stereo Zoom Transferscope (ZTS). With this instrument the interpreter can optically stretch or "rubber sheet" the photo image to match map control points and thereby match the interpreted polygons to their proper map positions. This step is made even easier when using an orthoquad because the interpreter can match detail seen on the LFC stereo model with the same detail as seen on the ortho. Image matching provides for an extremely high spatial accuracy. As the orthophoto quads were completed, they were edge matched and examined for open or unlabeled polygons.

The accuracy assessment field work was very intensive and required approximately 120 person hours to complete. The main objective of the accuracy assessment task was to obtain a ground verification and to estimate the overall accuracy of the photointerpretation classification work. It was also used to make adjustments to the classification system for future riparian mapping on other sections of the Forest.

Sample Size and Cluster (Transect) Layout

The accuracy assessment sampling design employed a cluster sampling technique with transects serving as clusters. Each change of vegetation type along a transect was a sample point. There were 248 sample points spread out along 144 transects. Dave Winters selected 6 out of the 25 quads for verification. The sample points were allocated throughout the quads proportional to the area occupied by each class, i.e., if 30% of the riparian area were willow then 30% of the sample points were placed in willow polygons.

The transects began and ended at the riparian zone boundaries perpendicular to the stream channels and were drafted onto the LFC photos and the riparian overlays. A new sample point was established at each change of vegetation along the transect. The 144 transects allocated were distributed throughout the elevational range of the study site. To accomplish this, transects were placed at approximate one mile intervals, moving up-stream.

Field Data Collection Procedures

The accuracy assessment field data collection work was performed by personnel from the Pike & San Isabel National Forest SO and the Leadville RD. The individuals were highly qualified and extremely familiar with the riparian ecology. The interpreter spent 6 hours training the field crews in how to collect the field data. The following data collection procedures were used:

1.  One end of the transect was carefully located at the edge of the riparian zone. The direction of the transect was noted and the team proceeded to the opposite end of the transect noting changes in the vegetation as they crossed the riparian zone.

2.  A field data sheet was used to record information relating to each vegetation class. These sheets were very informative to the interpreter because, in addition to recording the vegetation class, the field teams also made valuable remarks about the site and the interpretation.

Photointerpretation Accuracy Assessments

Based on the analysis of the collected ground data, a percent correct classification (PCC) of 83.5% was obtained for the photointerpretation effort. The error matrix shown in table 1 identifies some areas of possible confusion.

Photointerpretation error can be placed within 5 major categories. Three of these are a result of subjective judgment differences between the interpreter and the field crews. One is a misidentification because of similar appearance on the photos and the last is illogical and extreme error that really makes no sense.

Most of the error found in this project falls into the first type which is confusion between combined classes and is a result of subjective difference. For example E/W was interpreted as W/E four times, W four times, E twice and AW twice. W/E was interpreted as W and W/ASB once each, and E/W twice. M/W was called W and M once each. E was called E/W three times and W/ASB was labeled as W once (this is a class with 50% error because there were only 2 samples). This type of error accounts for 61.0% of the total error. The interpreter will continue to improve with experience, but this type of error will never be completely overcome and the user will have to be mindful of the error.

Another subjective error is in estimating crown closure among the aspen and cottonwood stands. Usually, these estimates are more accurate when examined from the air because a vertical perspective offers a more advantageous view of the site. This type was sampled once and makes up 2.4% of the total error.

The third type of subjective error is in transition zones such as when wet meadows (M) move into mesic meadows (GR) or when mesic meadows (GR) migrate into non-riparian (X). This problem is magnified if the photos were acquired during a wet season and the field verification done during a dry season or vice versa. It accounts for 4.9% of the total error.

Occasionally some very strange and illogical errors occurs that do not make sense. These include two errors when the interpreter called willow (W) a mesic meadow (GR), and evergreen (E) a willow stand (W). Errors such as these may be due to spatial misplacement on the ground because they are so different in appearance on the photo they should not occur. These errors made up 12.2% of the total error.

A final type of error, which accounted for 19.5% of the total error, is when different vegetation appears the same on the photos. This occurred twice on this project. Once when a small dense aspen stand looked like willow and another time when a non-riparian irrigated field was called a wet meadow (M). In both cases the appearance on the photo between the confused classes were so much alike as to be indistinguishable.

Table 2 shows the commission and omission errors associated with jut the riparian and non-riparian groups. Riparian zones were correctly classified 97.9% of the time. Omission and commission error are relatively even. Five times the interpreter said it was riparian and field investigation proved it was not. On the other hand, the interpreter failed to delineate riparian six times when it was found to be riparian after field examination. Some of the error was calling irrigated fields a wet meadow (M) or a mesic meadow (GR). Other times the interpreter called it a mesic meadow (GR) when it was actually not riparian at all. The interpreter missed a significantly large area of riparian for some reason and it will be resolved as soon as possible.

The overall percent correct classification of 83.6% for vegetation communities and 97.9% for riparian non-riparian is about what the interpreter expected and what a literature search of similar projects would predict. There were no significant trends in error that can be globally adjusted.

Photointerpreter

A single photointerpreter performed the delineation, identification, and labeling work on the project photographs. This approach was employed to ensure the consistency of the interpretation work, which is a highly desirable result in any project of this type.

GIS

The delineated riparian zones will be placed into a GIS database by the Colorado Division of Wildlife. As this photo-derived riparian information is used, changes or errors that are observed in the field should be noted and updated in the GIS database.

Classification System

As a result of this initial project, a significant change as been made in the classification system. Willow is now grouped with all other shrub vegetation and labeled (S). This is due to the inability to separate willow from alder and shrub size aspen.

The primary objective of this project – to develop an inventory of the riparian ecosystems within the study site, thereby creating a baseline map for monitoring change, has been successfully met. The LFC color infrared photographs used to delineate the riparian areas has proven to be a very good data source, one that is well suited for the resource identification and mapping requirements of this project.