Boise National Forest

US Department of Agriculture Forest Service

Photo Science Inc.

September 2012

Existing Vegetation Mapping Summary: Boise National Forest

Boise National Forest lntermountain Regional Office

Photo Science Inc.

The Forest Service, United States Department of Agriculture (USDA), has developed this information for the guidance of its employees, its contractors, and its cooperating Federal and State agencies and is not responsible for the interpretation or use of this information by anyone except its own employees. The use of trade, firm, or corporation names in this document is for the information and convenience of the reader. Such use does not constitute an official evaluation, conclusion, recommendation, endorsement, or approval by the Department of any product or service to the exclusion of others that may be suitable.

The USDA prohibits discrimination in all its programs and activities on the basis of race, color, national origin, sex, religion, age, disability, political beliefs, sexual orientation, or marital or family status (Not all prohibited bases apply to all programs). Persons with disabilities who require alternative means for communication of program information (Braille, large print, audiotape, etc.) should contact USDA's TARGET Center at 202-720-2600 (voice and TDD).

To file a complaint of discrimination, write USDA, Director, Office of Civil Rights, Room 326-W, Whitten Building, 1400 lndependence Avenue, SW, Washington, D.C. 20250-9410 or call202- 720-5964 (voice and TDD). USDÀ is an equal opportunity provider and employer.

Key Words: Boise National Forest, Forest Plan Revision, existing vegetation mapping, image segmentation, data mining, accuracy assessment

TABLE OF CONTENTS lntroduction 1 Overview 1 The Boise National Forest 2 Background 2 Partnership 4

Methods 6 Project Planning 6 Geospatial Data Acquisition 7 lmage and Geospatial Pre-Processing 8 Field Data Collection 10 Segmentation and Mapping 11 Draft Map Review and Revision 13 Accuracy Assessment Design 14

Map Products 16 Existing Vegetation Map 16 Map Groups and Map Units 17 Canopy Cover Class 19 Tree Size Class 20 Value-Added Products 20

Map Applications 21 Appropriate Uses 21 lnappropriate Uses 21

Accuracy Assessment Results 22

Conclusion 23

References 25

Appendix A: Project Planning A1-16

Appendix B: Geospatial Data Acquisition and Pre-Processing B1-5

Appendix C: Field Data Collection C1-24

Appendix D: Segmentation and Mapping D1-9

Appendix E: Draft Map Review and Revision E1-17

Appendix F: Accuracy Assessment Report F1-47

Appendix G: Guide to Data Drive G1-4

INTRODUCTION

Overview

Existing vegetation was mapped on over 2.6 million acres of the Boise National Forest (BNF) and surrounding area in southwestern Idaho (Figure 1) to standards established by the Intermountain Region Vegetation Classification, Mapping, and Quantitative Inventory (VCMQ) team. A key purpose of the project was to provide up-to-date and more complete information concerning the vegetative components on a landscape important to addressing the National Forest Management Act (NFMA) obligations such as providing for a diversity of vegetation and associated habitat for terrestrial wildlife species. The record resulting from this VCMQ project will be included in the Forest Plan record and each project record to support that the data relied upon to meet the NFMA obligations are “reasonably reliable and accurate.”

In 2007, the forest completed a needs assessment and determined that the existing vegetation structure maps from 1995 (Redmond, et al., 1997) were not sufficient for conducting analyses to support NFMA or the Forest Plan revision effort. The BNF formed a partnership with the Payette National Forest (PNF) and the Intermountain Regional Office (RO) VCMQ team and then contracted with Photo Science to produce a seamless, forest-wide vegetation layer for both the Boise and Payette National Forests. While both forests were mapped simultaneously, this report covers production and details of the BNF vegetation layer products; a separate report is available for the PNF. Multiple sources and scales of remotely sensed imagery and geospatial data layers were used along with image segmentation and data-mining technologies to develop the maps. Vegetation maps were characterized by vegetation map group (MG), dominant cover- type or map unit (MU), canopy cover class (CC), and tree size class (TS). The maps were designed to meet a minimum polygon size of 2 acres for riparian and aspen features and 5 acres for all other cover-types.

Figure 1. Boise National Forest Administrative Boundary with Ranger Districts.

The Boise National Forest

The Boise National Forest is located north and east of Boise in southwestern Idaho. The project area covered almost 3.0 million acres of Forest Service and private lands and encompassed 127 7.5-minute quadrangles. All five of the BNF Ranger Districts were mapped: Cascade, Emmett, Idaho City, Lowman, and Mountain Home. The mapped area followed the Forest Service administrative boundary including a 300 meter buffer outside the boundary.

The BNF was created in 1908 from portions of the Payette National and Sawtooth Forest Reserves, and currently covers 2,529,009 acres. Most of the land supports an evergreen forest that includes pure or mixed stands of ponderosa pine, grand fir, Douglas fir, Engelmann spruce, lodgepole pine and subalpine fir with small areas of western larch and whitebark pine. Shrubs and grasses grow in the non-forested areas. The Forest contains large numbers of big game species, such as mule deer and Rocky Mountain elk. Trout are native to most streams and lakes, while anadromous salmon and steelhead inhabit the Salmon River and its many tributaries. Much of the forest lies within the Idaho Batholith, a large and highly erodible geologic formation. Through uplift, faulting, and subsequent dissection by hydrologic action, a mountainous landscape has developed. Elevations on the forest range from 2,600 to 9,800 feet. The major river systems on the Boise National Forest are the Boise and Payette Rivers and the South and Middle Fork drainages of the Salmon River. The average annual precipitation ranges from 15 inches at lower elevations to 70 inches at higher elevations (Boise National Forest website).

Background

Vegetation is the primary natural resource managed by the Boise and Payette National Forests in the Intermountain Region (R4). Leadership teams on each National Forest are responsible for managing vegetation within their respective administrative boundaries to meet a variety of uses while maintaining the integrity of ecosystem components and processes at the mid-, fine- and project scales. The 2003 Land and Resource Management Plan (Forest Plan) for each National Forest provides the strategic management framework that guides project planning and implementation. In addition, management of resources within the administrative boundary must also be done in concert with Regional resource and operational goals, objectives and priorities outlined in the R4 Business Plan.

The management framework defined in the 2003 Forest Plan was based, in part, on a vegetative classification, mapping and inventory system updated through calendar year 2000. Since 2000, wildfires alone affected over 20% of the NFS lands administered by these National Forests. Due to this significant wildfire activity in both forests, the leadership teams believed it was important to update their current vegetation classification, mapping and inventory system so that they better reflect present day ground conditions. This update was necessary to insure projects designed to implement and achieve 2003 Forest Plan goals and objectives resulting in the desired mid- and fine-scale outcomes. In addition, this update was needed in order for the Forest Service (i.e. Agency) to demonstrate that it has used the best available data and science concerning the existing vegetative resource. Using the best available vegetative data and science is key to substantiating effects disclosures pertaining to how the Agency has met its

National Forest Management Act’s (NFMA) obligations concerning vegetative and species diversity.

Therefore, the purpose of this project was to update the existing mid-level vegetation classification, mapping and inventory system for the Boise and Payette National Forests. For purposes of this document, it should be noted that classification, mapping and inventory of existing vegetation are considered three separate, but related, processes. The general relationship of these three processes is as follows:

Classification is the process of grouping similar entities into named types or classes based on shared characteristics. Vegetation classification defines and describes the types and/or structural characteristics. In other words vegetation classification answers the question “what is it?”.

Mapping is the process of identifying the geographic distribution, extent and patterns of vegetation types and/or structural characteristics. Vegetation mapping entails the spatial delineation of vegetation patches and assigning labels to those patches. In other words, vegetation mapping answers the question “where is it?”.

Inventory is the process of applying an objective set of sampling methods to quantify the amount, composition, and condition of vegetation within specified limits of statistical precision. Vegetative inventory quantifies the amount, composition and condition of vegetation, and the reliability of the estimates. Vegetative inventory answers the question “how much is there?”.

The scope of work for the VCMQ project recognized that development of a system must be flexible and allow for additions, modifications and continuous refinement. This is necessary because a one-to-one relationship between vegetative types from a classification and vegetative map units is uncommon given the limitations of mapping technology and the level of floristic detail in most classifications. Classification and mapping, therefore, usually entails consideration of trade-offs among thematic resolution, spatial resolution, accuracy, and cost as a project progresses.

To begin looking at how to balance these trade-offs in light of latest technologies and imagery, the Boise and Payette National Forests initiated a pilot project in 2005 with the Remote Sensing Applications Center (Goetz et al., 2006). Outcomes from that pilot, including lessons learned, were used as a starting point for this project. The goal of this project was to develop a classification, mapping and inventory system that builds upon, or at least compliments, historical Forest Plan efforts.

Partnership

The mapping team was comprised of the BNF, PNF, RO VCMQ team, and Photo Science (PSI), a geospatial services contractor available to the Forest Service through the National Geospatial Data Services Contract. PSI provided project management, image processing and segmentation, aerial photo-interpretation, rule-based decision-tree modeling, GIS modeling, and accuracy assessment analysis. PSI utilized two sub-contractors to conduct specific aspects of the project. Chestnut Ridge Forestry (CRF) provided field crews to collect field site observations for classification in 2009 and 2010, and for accuracy assessment in 2011. RedCastle Resources (RCR) provided vegetation mapping support in the form of geospatial data acquisition, data preparation, and separability analysis on the BNF classification data set. They also provided input and updates during bi-weekly conference calls for mapping progress in the PNF that was occurring simultaneously with the BNF vegetation mapping project.

The RO was instrumental in the design of a fìeld-based classification system, field keys and procedures, developing map unit descriptions, applying classifications to existing datasets, and providing guidance to assure that the final map products meet VCMQ standards. The RO also reviewed and provided existing data sets for use in the accuracy assessment of the final vegetation maps.

The BNF coordinated Forest Service efforts for the project, provided key ancillary data, and access to existing field data. Staff of both the BNF and PNF provided expert ground knowledge in the form of field support for training and accuracy assessment data collection. In addition, the BNF supervisor's office staff and Ranger District staff provided feedback and timely review of draft and final map products.

Primary team members included:

Boise National Forest: Randy Hayman, Forest Planner Kathy Geier-Hayes, Forest Ecologist Elaine Alexander, Forester Lisa Nutt, Wildlife Biologist Carey Crist, GIS/Information Resource Manager Mike Willamson, GIS Specialist Terry Hardy, Soil Scientist and Contracting Officer Technical Representative

Payette National Forest: Patty Soucek, Forest Planner Susan Miller, Forest Ecologist Kim Johnson, Forest Silviculturist Ana Egnew, Wildlife Biologist Mickey Pillers, Data Manager Chans O'Brien, Geospatial Analyst

Intermountain Regional Office: Don Fallon, NRM-NRIS Resource Information Coordinator and VCMQ Team leader Sanford Moss, Regional Remote Sensing Specialist Dave Tart, Regional Ecologist Larry DeBlander, Forester Brian Wharton, Contracting Officer

Photo Science: Brad Weigle, Project Manager Nathaniel Morton, Remote Sensing/GIS Analyst Andrew Brenner, Accuracy Assessment Manager Emily Foster, Remote Sensing Analyst Richard Eastlake, Image Interpreter Chin Chan, Image Interpreter

RedCastle Resources: Paul Maus, Task Manager Wendy Goetz, Remote Sensing/GIS Analyst

Chestnut Ridge Forestry: Joel Fyock, Field Data Collection Manager

METHODS

The mapping phases for this project included: project planning, geospatial data acquisition, image pre-processing, field data collection, segmentation and classification, draft map review and revision, and accuracy assessment. Each phase of the project is discussed in more detail in the sections that follow.

Project Planning

In September 2008, staff of the BNF, PNF, RO, and PSI met to discuss map unit design and prepare a project plan. Since one of the goals for the project was to provide a template for other forests in the region, efforts were made to ensure that spatial and thematic characteristics of the maps as well as processes used would fulfill regional requirements. Vegetation classes were reviewed and a classification system was proposed that balanced budget and time constraints. The final map units, canopy cover, and tree size classes conformed to the mid-level mapping standards referenced in the Existing Vegetation Classification and Mapping Technical Guide (Brohman and Bryant, 2005). To minimize variation in ecological and vegetation characteristics, and to ease computer processing constraints, the study area was divided into two mapping regions: GA1 and GA3. (Figure 2). The most northerly portions of GA1 were originally included in 3 GAs in the PNF (2,4, 5) so the field samples in those areas were included with field samples for GA1.

For more information about project planning, map unit design, and the classification system, see Appendix A.

Figure 2. Boise NF Geographic Areas

Geospatial Data Acquisition

Geospatial data acquisition is a major activity in most vegetation mapping efforts using digital image processing methods. This project involved assembling multiple sources and scales of remotely sensed imagery and geospatial data layers. See Appendix B for the complete data list and more detailed information about acquisition activities.

A requirement of the mapping process was that any data layer used must be available across each forest to ensure consistency and efficiently. Newly acquired data included Landsat Thematic Mapper (TM) satellite imagery, digital quarter quad tiles (DOQQs) aerial imagery from the National Agriculture Imagery Program (NAIP), Digital Elevation Models (DEMs), Monitoring Trends in Burn Severity (MTBS), and Daymet surface weather information. Data loading and image Fall processing tasks were completed with the use of Forest Service Image Processing Systems—ERDAS Imagine and ArcGIS. In addition, ancillary data such as roads, trails, hydrology, harvest activities, landownership, landtype, geology, and administrative boundaries were provided by the forests. All data was reprojected to the UTM/Zone 11, GRS 1980, Summer NAD83 coordinate system. A discussion of the newly assembled geospatial data follows.

Landsat satellite imagery was acquired from the USGS Global Visualization Viewer (GloVis) for each path/row covering the Payette and Boise National Forests (Figure 3). Multi-temporal Landsat images Spring captured spring, summer, and fall vegetation conditions providing for “leaf on” and “leaf off” periods. Such multi-temporal image sets have proven useful for mapping vegetation (Mickelson et al., 1998; Wolter et al., 1995). A total of 20 orthorectified Landsat scenes were acquired for the project area. Figure 3 – Multiple dates of Landsat NAIP color infrared aerial imagery from 2009 was imagery capture phonological changes assembled for both the BNF and PNF. This imagery in vegetation. is acquired at a 1 meter ground sample distance with a horizontal accuracy that matches within 6 meters of photo-identifiable ground control points. The digital ortho quarter quad tiles were resampled to 10 meters and mosaiced into a seamless image for each geographic area (Figure 4). The imagery was used for field data acquisition maps in the BNF.

Figure 4 – NAIP CIR DOQ image (1m) 7

A DEM is a raster array of terrain elevations for ground positions at regularly spaced horizontal intervals. They are commonly used in mapping projects to help identify the elevation and aspect at which different vegetation communities occur. In this project, 10 meter DEMs were downloaded from DataDoors catalog retrieval system which is available to forest service users to access a variety of raster based geospatial data. (Figure 5).

MTBS is a multi-year project designed to consistently map the burn severity and perimeter of fires occurring since 1984. Burn severity data for the project area was available for the years 1984 through 2008. Figure 5 – Hydrology layers overlaid on a 10-m DEM. Daymet is a climatic model that generates daily surfaces of temperature, precipitation, growing days, and radiation over large regions. DEMs and daily observations of minimum and maximum temperatures and precipitation from meteorological stations over 18 years (1980 – 1997) are used to produce weather and climatological summary datasets at a 1 km resolution.

Image and Geospatial Data Pre-Processing

Landsat scenes were reviewed for haze, clouds, and cloud shadows in the project area. If any of these anomalies were found, they were clipped out from the image. New imagery was generated using a process called Model II Regression (Maiersperger, 2004; Beaty et al., 2011) (Figure 6). This process calibrates an overlapping image from a different date, which is cloud- free in the anomalous areas, to the original images. These new calibrated predicted scenes were the final images.

Figure 6 – On the left is the original Landsat TM image with clouds and shadows (summer). The predicted cloud free image is shown on the right.

Three seasonal (spring, summer, and fall) seamless mosaics of Landsat imagery were produced using this same process for the entire Boise and Payette project area. Once the mosaics were complete, spectral indices were produced: Normalized Difference Vegetation Index (NDVI), Tasseled Cap, and Principle Component Analysis (PCA) (Figure 7). Such indices can be useful in discriminating between vegetation cover-types.

Original Landsat TM

NDVI

Tasseled Cap

1st Principal Component Figure 7 - Summer Landsat imagery and a suite of standard indices.

DEMs were used to create topographic derivatives including slope, aspect, curvature, compound topographic wetness index, valley bottom, and a fully illuminated hillshade (Figure 8). Such topographic data depict environmental parameters that can help predict land cover-types in the mapping process. In addition, a texture layer was produced from the 1 foot resource digital photography. For more information about the geospatial data acquisition see Appendix B.

Elevation Slope

Aspect Heatload

Fully Illuminated Hillshade Figure 8 – Topographic derivatives

Field Data Collection

An effective field sample design was developed for the collection of image classification training data. Representative sample sites were located in relatively homogenous areas based on an initial landscape stratification of each of the five geographic mapping areas in the Boise and Payette National Forests. Training data were collected in the field and by visual aerial photo interpretation using accepted protocols, vegetation type keys, and forms. See Appendix C for protocols and field forms. In addition, existing field information was evaluated for potential use as training data.

Satellite imagery and elevation data derivatives were used to produce an initial spectral- topographic landscape stratification. For example, Landsat Tasseled Cap brightness, greenness, and wetness vegetation indices and topographic derivative data were assembled. An unsupervised classification was performed to cluster image pixels containing similar spectral and topographic properties. For each GA, 500 training samples were proportionally distributed within the resulting feature space classification. The number and location of sample sites was determined based on the quantity and distribution of pixel clusters output from the initial unsupervised image classification. The total number of samples was influenced by time and funding constraints. To maximize efficiencies, sample sites were placed in relatively accessible locations, typically within ¼ mile buffer of roads. The training sample design was reviewed by all project partners. Final field plots, GPS waypoints, aerial photos, and navigation and plot maps were assembled and delivered to field crews (Figure 9).

Figure 9. Field plot forms and site map

Field data for image classification of vegetation types were collected by PSI sub-contractor Chestnut Ridge Forestry field crews during the summers of 2009 and 2010. Field data collection protocols and corresponding data forms were developed for detailed plot samples as well as less rigorous observation polygons encountered while hiking between field plots. The

intent was to collect up to five observations polygons for each field plot. Field tests were conducted to finalize the protocols, dichotomous keys, and forms. An Access database was created for training data management and subsequent digital input to the map attribution procedure. During the week prior to conducting field work, training was provided to field crews by RO and forest staff to ensure consistent data collection procedures and to minimize sample variance. Crews conducted field work to record vegetation composition, canopy cover, tree size, and site identification data. Quality control field data was collected and assessment results reported by the individual forests. For each field site, crews were provided different scales of maps to aid in navigation and to add observation polygons. Following field surveys, percent canopy cover was determined for all field plots and observation polygons through visual aerial photo interpretation by PSI image interpreters. Data recorded on the forms was reviewed for quality and consistency and entered into the database.

To reduce sampling costs and facilitate data collection in inaccessible areas, aerial photo interpretation was performed on a portion of the training plots. Photo interpreters coordinated with BNF and PNF experts and site visits were conducted to familiarize themselves with local vegetation characteristics/conditions and to calibrate interpretations. In addition, field sample data entered into the database were forwarded to interpreters in a timely fashion. To maintain sampling consistency, the same photo interpreters measuring field plot and observation polygon percent canopy cover also interpreted the training plots. A quality assessment was conducted by comparing a subset of the photo interpreted data to data collected by a separate party.

In addition to the collection of new field and photo interpreted training samples, existing field data from sources such as land system inventories and other surveys were evaluated for appropriate use as training data. All field, photo interpreted, and existing training data were assembled to generate a single geospatial data layer.

For more information about field data protocols and collection forms, refer to Appendix C.

Segmentation and Mapping

A geographic object-based image analysis (GEOBIA) was performed to produce segments on approximately 5 million acres (2 million hectares) of the Boise and Payette National Forests of Idaho, USA, for the purpose of vegetation classification, mapping and quantitative inventory. Over 18,000 very-high resolution aerial resource images collected by the RO in 2008 from Zeiss/Intergraph’s Digital Mapping Camera (DMC), with 1-foot (0.35m) resolution are the base for this mid-level existing vegetation map product.

An image mosaic was created for each GA using a procedure that utilized only the most nadir portion of each image to eliminate distortion and object lean. From these "sweet spot" chips, a 5.25m resolution mosaic was created for each GA using ERDAS Imagine. Due to gain problems with the infra-red band in the imagery, no image balancing was performed while building the mosaic to retain the original spectral signatures for each chip in the mosaic.

The use of Trimble eCognition and its Cognition Network Language created a Size-Constrained Region-Splitting Multiresolution Segmentation Routine which facilitated the automatic delineation of homogenous objects of interest across the varying forested and non-forested

landscape scales. Initially, objects were created to approximate a known Forest Service product called TEUI (Terrestrial Ecological Unit Inventory; Winthers et al, 2005), which classifies ecological types and maps terrestrial ecological units. The large objects (polygons) were created with eCognition through the use of large scale parameter multi-resolution segmentation routines and were based on the fully illuminated hillshade to constrain objects within logical watershed boundaries. These TEUI-like objects maintain criteria for objects not crossing valley bottoms nor ridgelines, maintain homogeneity within each segment, maintain heterogeneity outside of segment, and are within a range for minimum mapping unit size. An eCognition routine was employed to find objects above a 5 acre threshold for additional region-splitting. This large object removal allowed for an iterative region-splitting of objects based on acreage and spectral characteristics. To achieve additional splitting, a multi-resolution segmentation region grow was performed at iteratively smaller scale parameters until a desired mean acreage size was obtained. The region-splitting occurred for all polygons which were labeled as upland. Additional polygons that intersected a valley bottom DEM derived product were selected and identified as candidate polygons for additional region splitting at a an even smaller scale parameter for smaller acreage riparian segments that met a smaller MMU. An example of both upland and riparian segments displayed over an infra-red image is shown in Figure 10.

Figure 10. Polygon segments developed from eCognition.

An image stack was developed from 58 different geospatial data layers representing topographic, spectral, textural, climatic, and other ancillary information. These base layers were summarized for each segment and the reference data sites to produce an image stack comprised of zonal means or zonal majorities of the 58 original base layers. For each polygon

segment and reference data plot, a mean/majority value of all pixel values within the segment boundary was calculated and recoded to the zonal mean value within eCognition.

Customized algorithms were developed using a software program called See5. This program uses data mining to generate rule-based decision trees. Survey-wide geospatial data layers and site specific field-based measurements are analyzed for predictable relationships. The relationships are converted to a classification and regression-tree, which is tested, ranked, and finally used to predict, or map, ground-based information across the entire study area. Multiple modeling runs were conducted to create masks for each Map Group. Each mask was reviewed and edited by image interpreters to eliminate obvious errors. Finally, the mask areas were individually modeled in See5 again using only Map Unit reference data plots that occurred in each map group.

A similar set of models were used to create canopy cover and tree size products. A GIS layer was produced containing all training data, how data were used, and the final map attributes associated with each site.

Additional detailed information about segmentation and mapping can be found in Appendix D.

Draft Map Review and Revision

Draft hardcopy maps of the vegetative map units were distributed to each of the BNF ranger districts during the week of July 25, 2011. Meetings were held in each of the five districts where staff from the Forest and PSI presented the mapping results and worked with local experts to assess the maps and provide input on what was needed to improve the drafts before the final maps were produced.

A total of 16 maps were produced at a scale of 1:42,000, with nine USGS quadrangles represented on each plot (Figure 11). Each Ranger District received a complete set of maps for their area and staff were given time during the meetings to review the draft maps. While some initial information was conveyed during the meetings, maps were left with the staff for the week to review in depth and provide detailed comments. Comments were compiled for each district and forwarded to the Supervisor's office for consolidation. The hardcopy maps and comments were returned to PSI with comments and notes. Recommended changes and manual edits were incorporated into the final map. For more information about the review and revision process involving local resource personnel, refer to Appendix E.

Figure 11. Draft vegetation map showing map units over 9 quadrangles that was delivered to Ranger District staff for review.

Three final map products were produced for delivery to the FS: 1) map groups and map units; 2) canopy cover for trees and shrubs; and 3) tree size. For the MG-MU map, segments were first dissolved to eliminate adjacent polygons having exactly the same labels. To achieve the MMU of 2 acres for riparian and aspen and 5 acres for all other MUs, segments below the MMU were merged based on a set of rules developed by the FS and project staff. These rules are included at the end of Appendix E. For the CC and TS final maps, segments were only dissoved and no merging occurred. Therefore, polygons are present in the CC and TS GIS layers that are smaller than the MMU.

Accuracy Assessment Design

Two approaches were used to collect accuracy assessment reference sites. The first approach reviewed existing inventory data for the purpose of using these data sets in the accuracy assessment. Following this review, the second approach identified MU classes that were under represented in the existing inventory data and field crews were dispatched to sample new accuracy assessment sites during the summer of 2011.

The FIA data reviewed were collected between 2004 and 2009 on forested lands. Non-forest plots had only a MG attribute. All plots that were determined to be non-representative or too mixed in composition were dropped. This corresponded to the dropping of 2 out of 204 plots in the Boise National Forest. In addition, secondary calls were made on plots where there were two possible labels for a single plot. These secondary calls were included in the site fuzzy accuracy assessment.

B-Grid data were initially collected over forested and non-forested areas and the plots were collected between 2006 and 2008. As with the FIA plots, the inventory data may sometimes be difficult to link to a specific segment since the plot may cross MU class boundaries. The number of B-Grid plots dropped on the Boise was 3 (1.2% of 238).

The CDC plots were collected for an ecological inventory of dominance types on the forests and are divided between two major datasets: Riparian and Upland. The objective of collecting these sites was very different from accuracy assessment and they were collected opportunistically along transects with little or no randomization. Plot areas were very small and clustered together and measure patches of vegetation that have a small extent, often significantly smaller than the 2 acre MMU for riparian and 5 acre MMU for upland. The number of CDC plots that were dropped was 37 for the Boise NF (10% of 373 plots).

Labels from the three existing databases (FIA, B-grid, and CDC) were created and attached to each of the reference sites. These sites were then assessed to ensure that they did not fall too close to a training site. A distance of 300 meters was chosen as the required separation between training site and accuracy assessment site with the same MU. Sites were then assessed with respect to their distribution of sites within each MU class. The MU classes that had less than 30 sites were identified as classes requiring additional collection of accuracy assessment sites. The classes were generally those that were rare in the landscape. Using the draft MU map that was delivered in July 2011, sites that fell within the selected classes were selected. The segments available for sampling were randomly selected from available segments that were:  Within ¼ mile of a motorized road or trail.

 On FS lands  > 300 meters from existing AA sites or training site locations  > 1 acre in size  Not inaccessible because of water or steep gradient.

Accuracy assessment field staff, provided by Chestnut Ridge Forestry, visited the accuracy assessment sites between August and early October of 2011. Accuracy Assessment field collection protocols, location maps, and data forms were similar to those used for field site sampling. Field crews were provided polygon boundaries for each location, but had no indication of the MU for each polygon. Representative sites within the polygons were assessed using field keys to determine the map unit, canopy cover, and size class information. A digital photo was taken at each site. Please refer to Appendix F for details on protocols and data forms.

Accuracy assessment analyses based on these data sets are briefly discussed in the Accuracy Assessment Results section below and fully discussed in Appendix E.

MAP PRODUCTS

A suite of map products were completed in December 2011. The package included the existing vegetation map and numerous value-added products. The team also prepared technology transfer materials including a technical report, PowerPoint presentations, digital maps, and field photos and databases.

Existing Vegetation Map

The final map product provides for continuous land cover information for the Boise National Forest. This map shows existing vegetation types and their structural characteristics and is formatted as a feature class within an ArcGIS 9.3 Geodatabase compatible with Forest Service corporate GIS software. The mapped area encompasses the entire BNF at 2.6 million acres. Categories mapped included Map Group (MG), Map Unit (MU), Canopy Cover (CC), and Tree Size Class (TS) (Table 1).

Table 1. Map Attributes and abbreviations

MU MAP UNIT MG MAP GROUP The vegetation map is AGR Agriculture B Burned consistent with mid-level AS Aspen C Coniferous Forest mapping standards set forth in BB Bitterbrush D Deciduous Forest the Existing Vegetation BFS Burned Forest Shrubland H Herbland Classification and Mapping BHE Burned Herbaceous N Non-Vegetated Technical Guide (Brohman and BSV Burned Sparsely Vegetated R Riparian Bryant, 2004). In conformance DEV Developed S Shrubland with these standards, small DF Douglas Fir polygons were aggregated up DFL Douglas Fir/Lodgepole CC CANOPY COVER to five acres with the exception DFP Douglas Fir/Ponderosa NC No Cover Class of riparian and deciduous, ES Engelmann Spruce SC1 Low Shrub Cover: 10-24% which were aggregated to 2 FO Forbland SC2 Medium Shrub Cover: 25-34% acres. Agriculture, water, and FS Forest Shrubland SC3 High Shrub Cover: >= 35% urban areas were not GF Grand Fir Mix TC1 Low Tree Cover: 10-19% aggregated. Map units are GFP Grand Fir/Ponderosa TC2 Low-Medium Tree Cover: 20-29% mutually exclusive and clearly GR Grassland TC3 Medium Tree Cover: 30-44% defined by local dominance LP Lodgepole Pine TC4 Medium-High Tree Cover: 45-59% types. MB Mountain Big Sagebrush TC5 High Tree Cover: >= 60% MS Mountain Shrubland PP Ponderosa Pine TS TREE SIZE RHE Riparian Herbaceous NS No Tree Size Class RSH Riparian Shrubland/Deciduous Tree TS1 Seedling: <4.5' Tall SV Sparsely Vegetated TS2 Sapling: 0.1-4.9'' DBH SA Subalpine Fir Mix TS3 Small: 5-9.9'' DBH WA Water TS4 Medium: 10-19.9'' DBH WH Weedy Herbaceous TS5 Large: 20-29.9'' DBH WL Western Larch TS6 Very Large: >= 30'' DBH WB Whitebark Pine Mix

Map Groups & Map Units

A total of 28 map units were mapped (Table 2). These dominance type classes ranged from specific vegetation species (i.e. mountain big sagebrush) to general land use type (i.e. agriculture). Map Groups were made up of logical aggregations of the map units and consisted of 7 classes: conifer, deciduous, shrubland, herbaceous, riparian, non-vegetated, and burned.

Table 2: Land cover type acreage summary

Boise National Forest Vegetation Map Unit Summary Map Boise NF Cascade RD Emmett RD Idaho City RD Lowman RD Mtn. Home RD Unit Acres % Map Acres % Map Acres % Map Acres % Map Acres % Map Acres % Map AS 30,117 1.2% 2,602 0.6% 5,641 1.6% 7,290 1.3% 3,581 0.8% 11,003 1.5% DF 404,343 16.0% 63,345 15.8% 45,703 12.9% 89,675 15.8% 116,124 24.8% 89,496 12.2% DFL 10,193 0.4% 4,002 1.0% 576 0.2% 1,687 0.3% 3,454 0.7% 474 0.1% DFP 100,154 4.0% 11,721 2.9% 33,263 9.4% 24,590 4.3% 11,779 2.5% 18,801 2.6% ES 14,979 0.6% 9,725 2.4% 3,276 0.9% 297 0.1% 1,631 0.3% 50 0.0% GF 13,988 0.6% 4,608 1.1% 9,366 2.6% 0 0.0% 15 0.0% 0 0.0% GFP 4,321 0.2% 3,003 0.7% 1,305 0.4% 0 0.0% 12 0.0% 0 0.0% LP 170,671 6.8% 68,820 17.2% 9,958 2.8% 8,202 1.4% 79,410 16.9% 4,281 0.6% PP 606,302 24.0% 24,530 6.1% 167,697 47.4% 233,471 41.1% 49,133 10.5% 131,471 17.9% SA 187,986 7.4% 57,847 14.4% 13,545 3.8% 24,442 4.3% 75,242 16.1% 16,909 2.3% WB 5,930 0.2% 2,171 0.5% 1,025 0.3% 734 0.1% 1,330 0.3% 669 0.1% WL 1,593 0.1% 1,456 0.4% 137 0.0% 0 0.0% 0 0.0% 0 0.0% RHE 13,453 0.5% 4,153 1.0% 676 0.2% 329 0.1% 3,018 0.6% 5,277 0.7% RSH 36,800 1.5% 3,982 1.0% 1,340 0.4% 6,702 1.2% 7,843 1.7% 16,933 2.3% BB 49,618 2.0% 0 0.0% 0 0.0% 4,919 0.9% 0 0.0% 44,700 6.1% FS 162,416 6.4% 5,236 1.3% 17,458 4.9% 44,029 7.8% 27,156 5.8% 68,537 9.3% MB 248,876 9.9% 861 0.2% 4,123 1.2% 17,662 3.1% 6,839 1.5% 219,391 29.9% MS 61,557 2.4% 389 0.1% 11,979 3.4% 15,039 2.6% 10,103 2.2% 24,047 3.3% GR 42,566 1.7% 943 0.2% 8,547 2.4% 10,520 1.9% 2,936 0.6% 19,621 2.7% FO 13,411 0.5% 1,809 0.5% 1,652 0.5% 3,793 0.7% 1,273 0.3% 4,885 0.7% WH 16,832 0.7% 0 0.0% 3,013 0.9% 1,121 0.2% 444 0.1% 12,255 1.7% AGR 8,604 0.3% 200 0.0% 86 0.0% 49 0.0% 18 0.0% 8,251 1.1% DEV 7,682 0.3% 704 0.2% 411 0.1% 1,937 0.3% 719 0.2% 3,911 0.5% WA 17,758 0.7% 2,204 0.5% 870 0.2% 908 0.2% 4,052 0.9% 9,723 1.3% SV 12,533 0.5% 979 0.2% 508 0.1% 6,053 1.1% 2,666 0.6% 2,326 0.3% BFS 73,582 2.9% 12,729 3.2% 3,116 0.9% 36,633 6.5% 6,980 1.5% 14,123 1.9% BHE 131,905 5.2% 62,637 15.6% 5,020 1.4% 23,781 4.2% 34,001 7.3% 6,467 0.9% BSV 76,798 3.0% 50,442 12.6% 3,342 0.9% 3,967 0.7% 18,977 4.0% 71 0.0%

Total 2,524,968 100% 401,097 100% 353,633 100.0% 567,828 100% 468,736 100% 733,673 100%

Conifers were the most abundant map group, covering 60% of the entire forest. Across the forest, ponderosa pine was the most abundant dominance type (24%) followed by Douglas fir (16% ), subalpine fir (7.4%), and lodgepole pine (6.8%). The shrub map group accounted for 21% of the forest, dominated by mountain big sagebrush (10% of entire forest) and the conglomerate MU forest shrub (6.4%). Eleven percent of the forest was mapped into the burned map group, with the herbaceous (3%), riparian (2%), non-vegetated (2%), and deciduous (1%) map groups comprising the remaining forest area. Summaries by Ranger District and by MG are shown below (Table 3).

Table 3. Map Group acreage summaries

Boise National Forest Vegetation Map Group Summary Map Boise NF Cascade RD Emmett RD Idaho City RD Lowman RD Mtn. Home RD Group Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area C 1,520,460 60.2% 251,228 62.6% 285,851 80.8% 383,098 81.7% 338,130 72.1% 262,152 35.7% D 30,117 1.2% 2,602 0.6% 5,641 1.6% 7,290 1.6% 3,581 0.8% 11,003 1.5% S 522,468 20.7% 6,486 1.6% 33,560 9.5% 81,648 17.4% 44,099 9.4% 356,675 48.6% H 72,810 2.9% 2,751 0.7% 13,211 3.7% 15,434 3.3% 4,653 1.0% 36,760 5.0% R 50,253 2.0% 8,135 2.0% 2,016 0.6% 7,031 1.5% 10,861 2.3% 22,210 3.0% B 282,285 11.2% 125,808 31.4% 11,478 3.2% 64,380 13.7% 59,957 12.8% 20,661 2.8% N 46,576 1.8% 4,087 1.0% 1,875 0.5% 8,947 1.9% 7,455 1.6% 24,211 3.3%

Total 2,524,968 100% 401,097 100% 353,633 100% 567,828 121% 468,736 100% 733,673 100%

Canopy Cover Class

A canopy cover map was generated by independently processing the following cover types: conifer and deciduous (TC1-TC5), and shrubland (SC1-SC3). All other areas were mapped as NC (no canopy cover class). The Canopy cover categories were assembled by Ranger District for the entire forest (Table 4). Deciduous and coniferous MGs were combined for the tree canopy analysis. Shrub and riparian shrub MGs were combined with the burned shrub MU for the shrub canopy analysis.

Table 4: Canopy cover acreage summary. Boise National Forest Vegetation Canopy Cover Summary Boise NF Cascade RD Emmett RD Idaho City RD Lowman RD Mtn. Home RD Class Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area NC 341,542 13.4% 124,071 30.9% 24,124 6.8% 52,457 9.2% 68,103 14.5% 72,787 9.9% SC1 294,938 11.6% 14,502 3.6% 12,345 3.5% 54,504 9.6% 23,829 5.1% 189,759 25.9% SC2 95,044 3.7% 1,624 0.4% 4,236 1.2% 14,965 2.6% 5,965 1.3% 68,254 9.3% SC3 242,867 9.6% 7,070 1.8% 21,436 6.1% 55,514 9.8% 29,128 6.2% 129,718 17.7% TC1 500,527 19.7% 62,520 15.6% 64,020 18.1% 158,428 27.9% 104,230 22.2% 111,329 15.2% TC2 355,524 14.0% 55,100 13.7% 74,311 21.0% 98,201 17.3% 85,165 18.2% 42,748 5.8% TC3 329,669 13.0% 50,661 12.6% 74,779 21.1% 75,953 13.4% 80,283 17.1% 47,993 6.5% TC4 291,577 11.5% 67,542 16.8% 59,131 16.7% 48,442 8.5% 63,051 13.5% 53,411 7.3% TC5 73,279 2.9% 18,007 4.5% 19,251 5.4% 9,364 1.6% 8,984 1.9% 17,674 2.4%

Total 2,524,968 99% 401,097 100% 353,633 100% 567,828 100% 468,736 100% 733,673 100%

Tree Size Class

A tree size map was generated by independently processing conifer and deciduous forest covertypes. Forested lands were classified into one of six tree size classes (TS1-TS6). All other areas were mapped as NS (no tree size class). Size class summary acreages by Ranger Distriect are shown in the following table. (Table 5).

Table 5: Tree Size acreage summary.

Boise National Forest Vegetation Tree Size Summary Size Boise NF Cascade RD Emmett RD Idaho City RD Lowman RD Mtn. Home RD Class Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area Acres % Area NS 974,392 38.3% 147,268 36.7% 62,141 17.6% 177,440 31.2% 127,025 27.1% 460,518 62.8% TS1 1,524 0.1% 102 0.0% 206 0.1% 433 0.1% 195 0.0% 589 0.1% TS2 98,856 3.9% 15,951 4.0% 15,584 4.4% 30,428 5.4% 23,303 5.0% 13,590 1.9% TS3 330,281 13.0% 75,964 18.9% 53,499 15.1% 66,947 11.8% 93,359 19.9% 40,513 5.5% TS4 1,004,103 39.5% 149,346 37.2% 196,325 55.5% 260,837 45.9% 207,698 44.3% 189,897 25.9% TS5 113,561 4.5% 12,277 3.1% 25,658 7.3% 30,754 5.4% 16,884 3.6% 27,988 3.8% TS6 2,251 0.1% 190 0.0% 221 0.1% 990 0.2% 272 0.1% 579 0.1%

Total 2,524,968 99% 401,097 100% 353,633 100% 567,828 100% 468,736 100% 733,673 100%

Value-Added Products

The value-added products developed as part of the vegetation mapping project include field- collected information and mosaics of standard geospatial data sources, as well as numerous image derivatives and indices. Listed below are some of the products.

 Field-collected information on the Boise National Forest:  1304 field visited samples  1304 digital ground photographs linked to field visited samples  4305 field observation plots

 Photo-interpreted information on the Boise National Forest:  1411 sites

 Enhanced image product mosaics:  Multiple dates of Landsat TM imagery and indices (NDVI, Tasseled Cap, and PCA)  Digital Elevation Models and derivatives (slope, wetness, etc...)

 Additional products:  2008 aerial resource photography  External hard drive containing all geospatial & field datasets, maps, etc...

MAP APPLICATIONS

A GIS land-cover map is a useful product for addressing specific resource and land management issues. The existing vegetation maps provide consistent baseline information about current vegetation composition, structure and landscape patterns that support many Forest Plan Revision needs. Vegetation maps are used to address a variety of important land management issues involving fuel loads, watersheds, rangelands, and wildlife habitat. They are also used for modeling threatened and endangered species habitat, conducting benchmark analysis, and monitoring the sustainability of resource management practices. However, both resource specialists and land managers should acknowledge that there are appropriate and inappropriate uses for a mid-level existing vegetation map product.

Appropriate Uses

The final maps should be used based on the scale of this map: 1:100,000. As previously stated, these map products can be used in the preparation of forest plans, NFMA consultations, and planning for monitoring activities. Other appropriate uses of these mid-level vegetation maps include:

 Forest/multi-forest level planning and monitoring  Landscape/watershed level ecosystem assessments  National Fire Plan implementation  Forest and rangeland risk & health assessments  Terrestrial and aquatic habitat assessment & monitoring  Update previous forest level multi-resource assessments  Target areas for further investigation for potential projects

Inappropriate Uses

Mid-scale existing vegetation maps are not designed for detailed project planning & implementation. They are not a substitute for field-based inventories and assessment. Primary inappropriate uses include:

 Project level planning and land treatments  Fuel treatments  Grazing, timber, habitat management  Assessing riparian area conditions  Project level monitoring & evaluation, e.g.  Range analysis  Stand exams  Determining historical conditions

ACCURACY ASSESSMENT RESULTS

The accuracy assessment followed standard procedures to assess the accuracy of the maps; the three steps involved (1) verifying whether photo-interpretation (PI) could reliably separate all the map units (MU), (2) developing the reference databases and (3) analyzing the data.

The conclusions from the first task were that all MUs could not be photo interpreted and so accuracy assessment sites would need to be visited in the field. The second task involved the integration of Forest Inventory and Analysis plots (FIA), B-Grid (an intensification of FIA), CDC (a vegetation inventory) and specific sites visited in 2011 for the purpose of collecting accuracy assessment plots. Each of the databases was quality controlled and non-representative plots were removed. The reference database was compared to their corresponding map segments.

Two main analyses were made. First, the area for each class on the map was compared with area estimates from FIA. At the map group (MG) level the numbers agreed reasonably well. However, the numbers varied at the MU, CC and TS levels. Low area estimates for the large tree classes was particularly noticeable.

The second analysis was the creation of confusion matrices, based on four sources of field data collection, FIA, B-Grid, field data collected for ecological surveys and field data collected directly as part of the project, that show which classes are confused. These analyses were conducted using a deterministic (where only primary calls were considered correct), site fuzzy (where acceptable calls were also considered correct), and class fuzzy (where similar classes were considered correct) assessments. A summary of the overall accuracies of the Boise NF vegetation maps are shown below.

Boise National Forest Map Accuracy Assessment

Assessment Type Producers Users Map Group Deterministic 80.5% 73.2% Site Fuzzy 82.4% 76.0% Class Fuzzy 83.3% 79.6% Map Unit Deterministic 44.4% 40.5%

Site Fuzzy 47.4% 44.8% Class Fuzzy 65.5% 60.5% Canopy Closure Shrubs Deterministic 28.5% 28.6% Class Fuzzy 61.6% 65.6% Trees

Deterministic 25.2% 22.5% Class Fuzzy 70.8% 67.8% Tree Size Deterministic 39.5% 34.8% Class Fuzzy 76.9% 70.9%

The overall accuracies represent the level expected from a project of this type, especially given the classification system, as discussed in Section 6.2 of Appendix F. At the MG level, an accuracy of 80% or higher would be expected, and at the MU level an accuracy of 60 – 65% would be expected using a fuzzy matrix as was conducted here. The canopy closure accuracy of 67 - 71% (see table above) is reasonable, although this is a little lower than expected. The reason for this may be that the estimates were in many cases made directly from the inventory data which generally estimates lower CC than when CC is photo interpreted. The tree size accuracy overall is higher than would normally be expected with this type of technology. TS5 and TS6, classes that were very small parts of the forest, were probably under-mapped. Very few training sites were available for these classes because they probably exist in the more remote areas. The classification of large trees was a concern in Boise National Forest and this may require further analysis of the data to differentiate these important classes.

A detailed analysis of class accuracies is contained in the full accuracy assessment report found in Appendix F.

CONCLUSION

A set of existing mid-level vegetation maps, as an ArcGIS file geodatabase, was produced for the Boise National Forest at a cost of $0.25 per acre. Each digital data layer was clipped to the forest ownership boundary. Individual feature classes were created that depict map units, canopy cover for trees and shrubs, and tree size. These data and maps created from the data can be used individually or in combinations to support Forest Plans and NFMA evaluations. The data meet standards established by the Regional VCMQ team and are consistent with standards established for mid-level mapping as defined in the Forest Service's Existing Vegetation Classification and Mapping Technical Guide (Brohman and Bryant, 2005).

In addition to the vegetation data, all of the GIS data layers used to create the individual polygons have been delivered. A series of maps was produced that depict the final vegetation coverages at both at the Forest and Ranger District scales. Some of the data used for accuracy assessment (FIA and B-grid) are classified as sensitive by the Forest Service and are not available to the public. The remaining field data sites for accuracy assessment were delivered along with a detailed Accuracy Assessment Report (Appendix F).

Users of the data are encouraged to read and re-read the Map Applications section of this report as a guide for appropriate and inappropriate uses of the data. Since GIS allows the display of data at any scale over georectified imagery of any resolution from any time, data users should be cognizant that these products relate to a 2008-2010 time frame and a scale of 1:100,000. The National Map Accuracy Standard (NMAS) define the requirements for meeting horizontal accuracy as 90% of all measurable points must be within 1/30th of an inch for maps larger than 1:20,000 in scale. For 1:100,000 maps, the NMAS is +/- 166.67 feet. The data delivered for this project are considered to be much closer to ground features than the NMAS for a 1:100,000 scale map.

The GIS data layers delivered from this project are considered state-of-the-art products created using current data in the latest versions of data processing software that include eCognition 8.0, ArcGIS 10, ERDAS 2011, and See5. These new vegetation data layers serve as an extended snapshot of vegetative conditions in the forest that occurred from 2008 to 2010. The vegetation polygons (aka segments) were created using aerial imagery collected in 2008. The polygons were classified using primary data sets collected from 2008 to 2010: data from the 2008 aerial imagery; Landsat imagery primarily from 2008 with one additional scene from 2007 and from 2009; field data from 2009 and 2010; and a variety of other data layers (Appendix B).

The final maps were assessed using new field data from 2011 along with some historic data that range back to around 2005. Overall the map is of good quality; it represents the landscape well and may have confusion with some boundaries and at low canopy closures but will provide a valuable resource for future land management planning. Therefore, the final maps can be considered indicative of the existing vegetation found in the forest at the end of the 2000 decade. Updates, based on changing conditions like fire, silviculture, insect infestations, and weather events, need to be scheduled on a regular basis to keep the data layers related to conditions on the ground and applicable to management decisions.

REFERENCES

Beaty, M.; Finco, M.; Brewer, K. 2011. Using Model II Regression to radiometrically normalized Landsat scenes for the purpose of mosaicking. RSAC-10012-RPT1. Salt Lake City, UT: U.S. Department of Agriculture, Forest Service, Remote Sensing Applications Center. 6 p.

Brohman, R.; Bryant, L. eds. 2005. Existing Vegetation Classification and Mapping Technical Guide. Gen. Tech. Rep. WO–67. Washington, DC: U.S. Department of Agriculture Forest Service, Ecosystem Management Coordination Staff. 305 p.

Goetz, W. 2001. Developing a predictive model for identifying riparian communities at an ecoregion scales in Idaho and Wyoming. Masters thesis. Utah State University, Logan, UT.

Goetz, W.; Werstak, C.E.; Maus, P.; Lachowski, H.; Miller, S. 2006. Developing a midlevel vegetation map, a test of NAIP in the Intermountain region. RSAC-0079-RPT1. Salt Lake City, UT: U.S. Department of Agriculture Forest Service, Remote Sensing Application Center. 11 p.

Li, M.; Huag C.; Zhu, Z.; Wen, W.; Xu, D.; Anxing, L. 2009. Use of Remote Sensing Coupled with a Vegetation Change Tracker Model to Assess Rates of Forest Change and Fragmentation in Mississippi, USA. International Journal of Remote Sensing. 30(24).

Maiersperger, T.; Finco, M.; Helmer, E. 2004. Eliminating Cloud Contamination from satellite imagery: A review in support of FIA remote sensing initiatives.RSAC-4016-RPT1. Salt Lake City, UT: U.S. Forest Service, Remote Sensing Applications Center. 11 p.

McCune, B. and Keon, D. 2002. Equations for potential annual direct incident radiation and heat load. Journal of Vegetation Science. 13:603-606.

Mickelson, U.G.; Civco, D.L.; and Silander, J.A. 1998. Delineating forest canopy species in the northeastern United States using multi-temporal TM imagery. Photogrammetric Engineering and Remote Sensing. 64:891–904.

Moore, I.D.; Gessler, P.E.; Nielsen, G.A.; Peterson, G.A. 1993. Soil attribute prediction using terrain analysis. Soil Science Society of America Journal. 57:443-452.

Redmond, R.L.; Tady, T.P.; Fisher, F.B.; Thornton, M.; and Winne, J.C. 1997. LANDSAT vegetation mapping of the southwest and central Idaho ecogroups. Final Report: Contract # 53-0261-6-25. Wildlife Spatial Analysis Lab, Montana Cooperative Wildlife Research Unit, University of Montana, Missoula, MT 59812. 139p.

Winthers, E.; Fallon, D.; Haglund, J.; DeMeo, T.; Nowacki, G.; Tart, D.; Ferwerda, M.; Robertson, G.; Gallegos, A.; Rorick, A.; Cleland, D. T.; Robbie, W. 2005. Terrestrial Ecological Unit Inventory technical guide. Washington, DC: U.S. Department of Agriculture, Forest Service, Washington Office, Ecosystem Management Coordination Staff. 245 p.

Wolter, P.T.; Mladenoff, D.J.; Host G.E.; Crow, T.R. 1995. Improved forest classification in the northern lake states using multi-temporal LANDSAT imagery. Photogrammetric Engineering and Remote Sensing. 61:1129-1143.

Appendix A: Project Planning

The following vegetation key is intended for the identification of existing vegetation types, not for identification of potential vegetation (PNV) types. This key is intended for use only within the boundaries of the Payette and Boise National Forests.

R4 Key to Vegetation Formations 3/28/2008

This key does not apply to lands used for agriculture or urban/residential development. It applies only to natural and semi-natural vegetation dominated by vascular plants. Semi-natural vegetation includes planted vegetation that is not actively managed or cultivated.

All cover values in this key to formations are absolute cover, not relative cover, for the life form. See Appendix A for a discussion of absolute versus relative cover. In this key tree cover includes both regeneration and overstory sized trees, so that young stands of trees are classified as forest.

1a 22a All vascular plants total < 1% canopy cover………………………………………... Non-Vegetated (p.13) 1b All vascular plants total ≥ 1% canopy cover………………………………………... 2 22a 2a All vascular plants total < 10% canopy cover………………………………………. Sparse Vegetation1 2b All vascular plants total ≥ 10% canopy cover………………………………………. 3

3a Trees total ≥ 10% canopy cover……………………………………………………... 4 3b Trees total < 10% canopy cover……………………………………………………... 5

4a Stand located above continuous forest line and trees stunted (< 5m tall) by harsh alpine growing conditions……………………………………………………... Shrubland Key (p.4) 4b Stand not above continuous forest line; trees not stunted………………………... Forest Key (p.2)

5a Shrubs total ≥ 10% canopy cover…………………………………………………… Shrubland Key2 (p.4) 5b Shrubs total < 10% canopy cover…………………………………………………… 6

6a Herbaceous vascular plants total ≥ 10% canopy cover…………………………… 7 6b Herbaceous vascular plants total < 10% canopy cover…………………………… 8

7a Total cover of graminoids ≥ total cover of forbs……………………………………. Grassland Key2 (p.7) 7b Total cover of graminoids < total cover of forbs……………………………………. Forbland Key2 (p.10)

8a Trees total ≥ 5% canopy cover……………………………………………………..... Sparse Tree1 8b Trees total < 5% canopy cover……………………………………………………..... 9

9a Shrubs total ≥ 5% canopy cover…………………………………………………….. Sparse Shrub1 9b Shrubs total < 5% canopy cover…………………………………………………….. 10

10a Herbaceous vascular plants total ≥ 5% canopy cover…………………………….. Sparse Herb1 10b Herbaceous vascular plants total < 5% canopy cover…………………………….. Sparse Vegetation1

1Record the map group as B (Burned) and the map unit as BSV (Burned Sparse Vegetation) if you are in an area that was forested and has been affected by wildfire as evidenced by burned, standing dead trees over at least 2/3 of the plot. If the area does not meet the definition of burned forest, record the map group as V and the map unit as SV (Sparse Vegetation). 2Record the map group as B (Burned) if you are in an area that was forested and has been affected by wildfire as evidenced by burned, standing dead trees over at least 2/3 of the plot. Follow the instructions in the appropriate key to determine how to record the map unit on plots that meet the definition of burned forest.

Appendix A: 1 Key to Forest and Woodland Dominance Types and DT Phases 05/19/2009 DLT

Instructions:

1. Preferably, plots or polygons should be keyed out based on overstory canopy cover (trees forming the upper or uppermost canopy layer) by tree species. 2. Plots or polygons lacking such data or lacking an overstory layer should be keyed out using total cover by species. 3. If a plot or polygon does not key out using overstory cover, then it may be keyed using total tree cover. 4. If two trees are equally abundant, the species encountered first in the key is recorded as the most abundant.

Map Map DT or DT Phase Code Unit Group

1a Black cottonwood is the most abundant tree species……………….. POBAT d.t. RDE R 1b Black cottonwood is not the most abundant tree species…………… 2

2a White alder is the most abundant tree species……………………….. ALRH d.t. RDE R 2b White alder is not the most abundant tree species…………………... 3

3a Sitka alder is the most abundant tree/shrub species………………… ALVIS d.t. RSH R 3b Sitka alder is not the most abundant tree/shrub species……………. 4

4a Thinleaf alder is the most abundant tree/shrub species…………….. ALINT d.t. RSH R 4b Thinleaf alder is not the most abundant tree/shrub species………… 5

5a Water birch is the most abundant tree/shrub species……………….. BEOC2 d.t. RSH R 5b Water birch is not the most abundant tree/shrub species…………… 6

6a Quaking aspen is the most abundant tree species………………….. 7 6b Quaking aspen is not the most abundant tree species……………… 8

7a Conifer species total at least 10% absolute canopy cover………….. POTR5-Conifer d.t.p ASC D 7b Conifer species total less than 10% absolute canopy cover………... POTR5-POTR5 d.t.p. AS D

8a Western larch is the most abundant species…………………………. LAOC d.t. WL C 8b Western larch is not the most abundant species…………………….. 9

9a Whitebark pine is the most abundant tree species…………………... PIAL d.t. WB C 9b Whitebark pine is not the most abundant tree species………………. 10

10a Ponderosa pine is the most abundant tree species…………………. PIPO d.t. PP C 10b Ponderosa pine is not the most abundant tree species…………….. 11

11a Lodgepole pine is the most abundant tree species………………….. PICO d.t. LP C 11b Lodgepole pine is not the most abundant tree species……………… 12

12a Douglas-fir is the most abundant species…………………………...... 13 12b Douglas-fir is not the most abundant species………………………… 15

13a Ponderosa pine with at least 10% absolute canopy cover………….. PSME-PIPO d.t.p. DFP C 13b Ponderosa pine with less than 10% absolute canopy cover………... 14

14a Lodgepole pine with at least 10% absolute canopy cover………….. PSME-PICO d.t.p. DFL C 14b Lodgepole pine with less than 10% absolute canopy cover………… PSME-PSME d.t.p. DF C

15a Engelmann spruce is the most abundant tree species………………. PIEN d.t. ES C

Appendix A: 2 Map Map DT or DT Phase Code Unit Group 15b Engelmann spruce is not the most abundant tree species………….. 16

16a Grand fir is the most abundant tree species………………………….. 17 16b Grand fir is not the most abundant tree species……………………… 19

17a Western larch with at least 10% absolute canopy cover…………….. ABGR-LAOC d.t.p. GFW C 17b Western larch with less than 10% absolute canopy cover………….. 18

18a Ponderosa pine with at least 10% absolute canopy cover………….. ABGR-PIPO d.t.p. GFP C 18b Ponderosa pine with less than 10% absolute canopy cover……….. ABGR-ABGR d.t.p. GF C

19a Subalpine fir is the most abundant tree species……………………… 20 19b Subalpine fir is not the most abundant tree species…………………. 23

20a Western larch with at least 10% absolute canopy cover…………… ABLA-LAOC d.t.p. SAW C 20b Western larch with less than 10% absolute canopy cover…………. 21

21a Whitebark pine with at least 10% absolute canopy cover…………… ABLA-PIAL d.t.p. SAB C 21b Whitebark pine with less than 10% absolute canopy cover…………. 22

22a Douglas-fir with at least 10% absolute canopy cover……………….. ABLA-PSME d.t.p. SAD C 22b Douglas-fir with less than 10% absolute canopy cover……………… ABLA-ABLA d.t.p. SA C

23a Curlleaf mountain mahogany is the most abundant tree/shrub species……………………………………………………….. CELE3 d.t. MM S 23b Curlleaf mountain mahogany is not the most abundant tree/shrub species……………………………………………………….. 24

24a Another or an unknown conifer is the most abundant tree species… UNKNOWN UNK C 24b The most abundant tree species is a broadleaf ……………………... 25

25a Stand is located in a riparian setting as indicated by proximity to a stream or lake, topographic position, plant species that require or tolerate free or unbound water, and/or soil properties associated with seasonally high water tables……………………………………………………………… UNKNOWN RDE R 25b Stand not located in a riparian setting as described above…. UNKNOWN UNK D

Appendix A: 3 DRAFT Key to Shrubland Dominance Types 05/18/2009 DLT Instructions:

Plots or polygons should be keyed out based on total cover by species. This key is divided into riparian, alpine, and upland sections. First identify the physical setting of the plot, stand, or polygon using the key below.

For the purposes of this key, a riparian setting is defined as an area (typically transitional between aquatic and terrestrial ecosystems) identified by soil characteristics associated with at least seasonally high water tables, distinctive vegetation that requires or tolerates free or unbound water (Manning and Padgett 1995), proximity to a stream or lake, and/or topographic position (e.g. valley bottom). The alpine setting includes the area above the upper limit of continuous forest. Above this limit trees occur only in scattered patches and become increasingly stunted at higher elevations (Arno and Hammerly 1984). In this key the alpine setting takes precedence over the riparian setting. The upland setting includes non-riparian areas below the continuous forest line.

It is likely that some dominance types occur in more than one of these settings. If your plot does not key out successfully in one setting, then try another setting. For example, basin big sagebrush is in the upland key but may occur in degraded riparian areas with downcut streams.

Key to Physical Habitat Setting

Key Leads: 22a 1a Stand is located in a riparian setting as indicated by proximity to a stream or lake, topographic position, plant species that require or tolerate free or unbound water, and/or soil properties associated with seasonally high water tables……………………………………………………………………………………. Go to Riparian Key (p.5)

1b Stand not located in a riparian setting as described above………………………. Go to Upland Key (p.6)

Appendix A: 4 Key to Riparian Shrubland Dominance Types

Instructions:

1. Plots or polygons should be keyed out based on total cover by species. 2. Codes for dominance type and map unit can be found using Table 2a. Find the name of the most abundant shrub in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 3. When two or more shrub species are equal in abundance, the species listed first in Table 2 is used to assign the dominance type and map unit. 4. If the most abundant shrub species is not listed in Table 2a, then record the dominance type as UNKNOWN.

Table 2a. Most Abundant Riparian Shrub and Indicated Dominance Type and Map Unit.

(1) (2) (3) (4) (5) Rank Most Abundant Shrub (Dominance Type) Dom. Type Map Unit Map Code Code Group 1 Alnus viridis ssp. sinuata Sitka alder ALVIS -R RSH R 2 Alnus incana ssp. tenuifolia thinleaf alder ALINT RSH R 3 Betula occidentalis water birch BEOC2 RSH R 4 Salix commutata undergreen willow SACO2 RSH R 5 Salix boothii Booth’s willow SABO2 RSH R 6 Salix drummondiana Drummond’s willow SADR RSH R 7 Salix exigua coyote willow SAEX RSH R 8 Salix wolfii Wolf’s willow SAWO RSH R 9 Salix planifolia planeleaf willow SAPL2 RSH R 10 Vacciunium uglinosum bog blueberry VAUL RSH R 11 Cornus sericea redosier dogwood COSE16 RSH R 12 Philadelphus lewisii Lewis’ mockorange PHLE4 RSH R 13 Dasiphora fruticosa shrubby cinquefoil DAFR6 RSH R 14 Artemisia cana ssp. viscidula mountain silver sagebrush ARCAV2 RSH R 15 Species not listed above UNKNOWN RSH R

Appendix A: 5 Key to Upland Shrubland Dominance Types Instructions:

1. Plots or polygons should be keyed out based on total cover by species. 2. Codes for dominance type and map unit can be found using Table 2b. Find the name of the most abundant shrub in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 3. When two or more shrub species are equal in abundance, the species listed first in Table 2 is used to assign the dominance type and map unit. 4. If the most abundant shrub species is not listed in Table 2b, then record the dominance type as UNKNOWN. 5. Where map group is “S or B” in Table 2b, record the map group as B (Burned) if you are in an area that was forested and has been affected by wildfire as evidenced by burned, standing dead trees over at least 2/3 of the plot. Otherwise record it as S. Record the dominance type according to the vegetation that is alive on the site. If the area does not meet the definition of burned forest, record the map unit according to the dominance type (FS or MS). If the area meets the definition of burned forest, record the map unit as BFS. Table 2b. Most Abundant Upland Shrub and Indicated Dominance Type and Map Unit. (1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Group Most Abundant Shrub (Dominance Type) Code Code 1 Vaccinium cespitosum dwarf bilberry VACE FS or BFS S or B 2 Menziesia ferruginea rusty menziesia MEFE FS or BFS S or B 3 Alnus viridis ssp. sinuata Sitka alder ALVIS-U FS or BFS S or B 4 Vaccinium scoparium grouse whortleberry VASC FS or BFS S or B 5 Vaccinium membranaceum thinleaf huckleberry VAME FS or BFS S or B 6 Physocarpus malvaceus mallow ninebark PHMA5 FS or BFS S or B 7 Acer glabrum Rocky Mountain maple ACGL FS or BFS S or B 8 Rubus parviflorus thimbleberry RUPA FS or BFS S or B 9 Sambucus racemosa red elderberry SARA2-F FS or BFS S or B 10 Salix scouleriana Scouler willow SASC-F FS or BFS S or B 11 Spiraea betulifolia White spiraea SPBE2 FS or BFS S or B 12 Symphoricarpos albus common snowberry SYAL FS or BFS S or B 13 Holodiscus discolor oceanspray HODI FS or BFS S or B 14 Lonicera utahensis Utah honeysuckle LOUT2 FS or BFS S or B 15 Physocarpus monogynus mountain ninebark PHMO4 FS or BFS S or B 16 Ribes lacustre prickly currant RILA FS or BFS S or B 17 Mahonia repens creeping barberry MARE11 FS or BFS S or B 18 Ribes viscosissimum sticky currant RIVI3 FS or BFS S or B 19 Ceanothus velutinus snowbrush ceanothus CEVE FS or BFS S or B 20 Arctostaphylos uva-ursi kinnikinnick ARUV FS or BFS S or B 21 Amelanchier alnifolia Saskatoon serviceberry AMAL2 MS or BFS S or B 22 Prunus virginiana common chokecherry PRVI MS or BFS S or B 23 Rosa woodsii Wood’s rose ROWO MS or BFS S or B 24 Prunus emarginata bitter cherry PREM MS or BFS S or B 25 Symphoricarpos oreophilus mountain snowberry SYOR2 MS or BFS S or B 26 Ribes cereum wax currant RICE MS S 27 Rhus glabra smooth sumac RHGL MS S 28 Cercocarpus ledifolius curl-leaf mtn. mahogany CELE3 MM S 29 Purshia tridentata bitterbrush PUTR2 BB S 30 Artemisia spiciformis snowfield sagebrush ARSP8 MB S 31 Artemisia tridentata ssp. vaseyana mountain big sagebrush ARTRV MB S 32 Artemisia tridentata ssp. tridentata basin big sagebrush ARTRT MB S 33 Artemisia trid. ssp. wyomingensis Wyoming big sagebrush ARTRW8 MB S 34 Chrysothamnus viscidiflorus yellow rabbitbrush CHVI8 MB S 35 Ericameria nauseosa rubber rabbitbrush ERNA10 MB S 36 Ericameria suffruticosa singlehead goldenbush ERSU13 MB S 37 Artemisia rigida stiff sagebrush ARRI2 LS S 38 Artemisia arbuscula ssp. thermopola cleftleaf sagebrush ARART LS S 39 Artemisia arbuscula ssp. arbuscula low sagebrush ARARA LS S 40 Artemisia nova black sagebrush ARNO4 LS S 41 Species not listed above UNKNOWN UNK S

Appendix A: 6 Key to Grassland Dominance Types 05/11/2010 DLT

Instructions:

Key to Physical Habitat Setting

Key Leads: 1a 22a Stand is located in an alpine setting above the upper elevation limit of continuous forest………………………………………………………………………. Go to Alpine Key (p.7) (Map unit = AL) 1b Stand is located below the upper elevation limit of continuous forest…………… 2 22a 2a Stand is located in a riparian setting as indicated by proximity to a stream or lake, topographic position, plant species that require or tolerate free or unbound water, and/or soil properties associated with seasonally high water tables……………………………………………………………………………………. Go to Riparian Key (p.8)

2b Stand not located in a riparian setting as described above………………………. Go to Upland Key (p.9)

Key to Alpine Grassland Dominance Types

Instructions:

1. Codes for dominance type and map unit can be found using Table 3. Find the name of the most abundant graminoid in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 2. When two or more graminoid species are equal in abundance, the species listed first in Table 3 is used to assign the dominance type and map unit. 3. If the most abundant graminoid species is not listed in Table 3, then record the dominance type as UNKNOWN.

Table 3. Most Abundant Alpine Graminoid and Indicated Dominance Type and Map Unit. (1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Most Abundant Graminoid (Dominance Type) Code Code Group 1 Juncus parryi Parry’s rush JUPA AL A 2 Deschampsia cespitosa tufted hairgrass DECE-A AL A 3 Species not listed above UNKNOWN AL A

Appendix A: 7 Key to Riparian Grassland Dominance Types

Instructions:

Table 4. Most Abundant Riparian Graminoid and Indicated Dominance Type and Map Unit. (1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Most Abundant Graminoid (Dominance Type) Code Code Group 1 Carex simulata analogue sedge CASI2 WET R 2 Eleocharis palustris Fewflower spikerush ELPA6 WET R 3 Calamagrostis canadensis bluejoint reedgrass CACA4 WET R 4 Deschampsia cespitosa tufted hairgrass DECE-R RHE R 5 Danthonia intermedia timber oatgrass DAIN RHE R 6 Carex microptera smallwing sedge CAMI7 RHE R 7 Carex spectabilis showy sedge CASP5 RHE R 8 Poa nevadensis Nevada bluegrass PONE3 RHE R 9 Juncus arcticus ssp. littoralis mountain rush JUARL RHE R 10 Species not listed above UNKNOWN

Appendix A: 8 Key to Upland Grassland Dominance Types

Instructions:

1. Codes for dominance type and map unit can be found using Table 4. Find the name of the most abundant graminoid in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 2. When two or more graminoid species are equal in abundance, the species listed first in Table 4 is used to assign the dominance type and map unit. 3. If the most abundant graminoid species is not listed in Table 5, then record the dominance type as UNKNOWN. 4. Where map group is “H or B” in Table 5, record the map group as B (Burned) if you are in an area that was forested and has been affected by wildfire as evidenced by burned, standing dead trees over at least 2/3 of the plot. Otherwise record it as H. Record the dominance type according to the vegetation that is alive on the site. If the area does not meet the definition of burned forest, record the map unit as GR. If the area meets the definition of burned forest, record the map unit as BGR.

Table 5. Most Abundant Upland Graminoid and Indicated Dominance Type and Map Unit. (1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Most Abundant Graminoid (Dominance Type) Code Code Group 1 Luzula glabrata (hitchcockii) smooth woodrush LUGL2 GR or BGR H or B 2 Calamagrostis rubescens pinegrass CARU GR or BGR H or B 3 Carex geyeri elk sedge CAGE2 GR or BGR H or B 4 Carex rossii Ross’ sedge CARO5 GR or BGR H or B 5 Trisetum canescens tall trisetum TRCA21 GR or BGR H or B 6 Bromus marginatus mountain brome BRMA4 GR or BGR H or B 7 Melica bulbosa oniongrass MEBU GR H 8 Carex hoodii Hood’s sedge CAHO5 GR H 9 Festuca idahoensis Idaho fescue FEID GR or BGR H or B 10 Achnatherum (Stipa) occidentale western needlegrass ACOC3 GR H 11 Carex pachystachya chamisso sedge CAPA14 GR H 12 Pseudoroegneria (Agropyron) spicata bluebunch wheatgass PSSP6 GR or BGR H or B 13 Achnatherum (Stipa) thurberianum Thurber’s needlegrass ACTH7 GR H 14 Elymus elymoides (Sitanion hystrix) bottlebrush squirreltail ELEL5 GR H 15 Poa secunda Sandberg’s bluegrass POSE GR H 16 Poa compressa Canada bluegrass POCO GR H 17 Thinopyrum (Agropyron) intermedium intermediate wheatgrass THIN6 GR H 18 Bromus tectorum cheatgrass BRTE WH H 19 Poa bulbosa bulbous bluegrass POBU WH H 20 Perennial species not listed above UNKNOWN GR H 21 Annual species not listed above UNKNOWN GR H

Appendix A: 9 Key to Forbland Dominance Types 05/11/2010 DLT

Instructions:

Key to Physical Habitat Setting

Key Leads: 1a 22a Stand is located in an alpine setting above the upper elevation limit of continuous forest………………………………………………………………………. Go to Alpine Key (p.11) (Map unit = AL) 1b Stand is located below the upper elevation limit of continuous forest…………… 2 22a 2a Stand is located in a riparian setting as indicated by proximity to a stream or lake, topographic position, plant species that require or tolerate free or unbound water, and/or soil properties associated with seasonally high water tables……………………………………………………………………………………. Go to Riparian Key (p.11)

2b Stand not located in a riparian setting as described above………………………. Go to Upland Key (p.12) (Map unit = FO)

Appendix A: 10 Key to Alpine Forbland Dominance Types

Instructions:

1. Codes for dominance type and map unit can be found using Table 6. Find the name of the most abundant forb in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 2. When two or more forb species are equal in abundance, the species listed first in Table 6 is used to assign the dominance type and map unit. 3. If the most abundant forb species is not listed in Table 6, then record the dominance type as UNKNOWN.

Table 6. Most Abundant Alpine Forb and Indicated Dominance Type and Map Unit.

(1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Most Abundant Forb (Dominance Type) Code Code Group 1 Arenaria aculeata prickly sandwort ARAC2 AL A 2 Polygonum phytolaccifolium poke knotweed POPH AL A 3 Species not listed above UNKNOWN AL A

Key to Riparian Forbland Dominance Types

Instructions:

Table 7. Most Abundant Riparian Forb and Indicated Dominance Type and Map Unit. (1) (2) (3) (4) (5) Rank Most Abundant Forb (Dominance Type) Dom. Type Map Unit Map Code Code Group 1 Senecio triangularis arrowleaf ragwort SETR RHE R 2 Maianthemum stellatum starry false lily of the valley MAST4 RHE R 3 Species not listed above UNKNOWN RHE R

NOTE: No riparian forb dominance types have been documented on the Boise and Payette NFs to date.

Appendix A: 11 Key to Upland Forbland Dominance Types

Instructions:

1. Codes for dominance type and map unit can be found using Table 8. Find the name of the most abundant forb in column 2 and move to column 3 for the dominance type code, column 4 for the map unit code, and column 5 for the map group code. 2. When two or more forb species are equal in abundance, the species listed first in Table 8 is used to assign the dominance type and map unit. 3. If the most abundant forb species is not listed in Table 8, then record the dominance type as UNKNOWN. 4. Where map group is “H or B” in Table 8, record the map group as B (Burned) if you are in an area that was forested and has been affected by wildfire as evidenced by burned, standing dead trees over at least 2/3 of the plot. Otherwise record it as H. Record the dominance type according to the vegetation that is alive on the site. If the area does not meet the definition of burned forest, record the map unit as FO. If the area meets the definition of burned forest, record the map unit as BFO.

Table 8. Most Abundant Upland Forb and Indicated Dominance Type and Map Unit. (1) (3) (4) (5) (2) Rank Dom. Type Map Unit Map Group Most Abundant Forb (Dominance Type) Code Code 1 Balsamorhiza sagittata arrowleaf balsamroot BASA3 FO or BFO H or B 2 Helianthella uniflora oneflower helianthella HEUN FO or BFO H or B 3 Geranium viscosissimum sticky geranium GEVI2 FO or BFO H or B 4 Pteryxia terebinthina turpentine wavewing PTTE FO H 5 Penstemon attenuatus sulphur penstemon PEAT3 FO H 6 Chamerion angustifolium fireweed CHAN9 FO or BFO H or B 7 Illiamna rivularis Streambank wild hollyhock ILRI FO or BFO H or B 8 Rudbeckia occidentalis western coneflower RUOC2 FO or BFO H or B 9 Wyethia amplexicaulis mule ears WYAM FO or BFO H or B 10 Eurybia (Aster) integrifolia thickstem aster EUIN9 FO H 11 Hackelia micrantha Jessica sticktight HAMI FO H 12 Pyrrocoma uniflora plantain goldenweed PYUN2 FO H 13 Xeraphyllum tenax beargrass XETE FO or BFO H or B 14 Valeriana sitchensis Sitka valerian VASI FO or BFO H or B 15 Polemonium pulcherrimum Jacob’s-ladder POPU3 FO or BFO H or B 16 Thalictrum occidentale Western meadow-rue THOC FO or BFO H or B 17 Pteridium aquilinum western brackenfern PTAQ FO or BFO H or B 18 Trautvetteria caroliniensis Carolina bugbane TRCA FO or BFO H or B 19 Polygonum phytolaccifolium poke knotweed POPH FO H 20 Potentilla glandulosa sticky cinquefoil POGL9-U FO or BFO H or B 21 Apocynum androsaemifolium spreading dogbane APAN2 FO or BFO H or B 22 Arnica cordifolia Heartleaf arnica ARCO9 FO or BFO H or B 23 Fragaria virginiana Virginia strawberry FRVI FO or BFO H or B 24 Lotus unifoliolatus American bird’s-foot trefoil LOUN FO H 25 Hieracium cynoglossoides houndstongue hawkweed HICY FO or BFO H or B 26 Lupinus argenteus silvery lupine LUAR3 FO H 27 Lupinus sericeus silky lupine LUSE4 FO H 28 Lupinus arbustus longspur lupine LUAR6 FO H 29 Lupinus wyethii Wyeth’s lupine LUWY FO H 30 Eriogonum heracleoides parsnipflower buckwheat ERHE2 FO H 31 Balsamorhiza incana hoary balsamroot BAIN FO H 32 Eriogonum flavum alpine golden buckwheat ERFL4 FO H 33 Allium acuminatum tapertip onion ALAC4 FO H 34 Eriogonum umbellatum sulphur-flower buckwheat ERUM FO H 35 Phlox hoodii spiny phlox PHHO FO H 36 Eriogonum douglasii Douglas’ buckwheat ERDO FO H 37 Madia gracilis grassy tarweed MAGR3 FO H 38 Chondrilla juncea rush skeletonweed CHJU WH H 39 Epilobium brachycarpum tall annual wilowweed EPBR3 FO H 40 Sisymbrium altissimum tall tumblemustard SIAL2 WH H 41 Gayphytum diffusum Spreading groundsmoke GADI2 FO H 42 Polygonum douglasii Douglas’ knotweed PODO4 FO H 43 Species not listed above UNKNOWN FO H Appendix A: 12

Key to Non-Vegetated Land Cover and Land Use Types 05/11/2010 DLT

Map Group 1a. Area is currently used for agricultural activity (e.g. a fallow field)...... Agriculture (AGR) N

1b. Area is not currently used for agricultural activity ...... 2

2a. Area is currently developed for urban, residential, administrative use . . . . . Developed (DEV) N

2b. Area is not currently developed for urban, residential, administrative use ...... 3

3a. Area is dominated by open water or a confined water coarse ...... Water (WA) N

3b. Area is not dominated by open water or a confined water coarse ...... 4

4a. Area is dominated by unburned barren land (e.g. bare ground, bedbrock, scree/tallus, mines/talings) ...... Barren/Rock (BR) N

4b. Area is not dominated by unburned barren land ...... Unclassified

5a. Area is recently burned with little or no live vegetation; standing dead trees present...... Standing Dead Tree (SBT) B

5b. Area not as above...... Unclassified

Appendix A: 13

Appendix A. Absolute and Relative Cover

Absolute cover of a plant species is the proportion of a plot’s area included in the perpendicular downward projection of the species. These are the values recorded when sampling a vegetation plot. Relative cover of a species is the proportion it comprises of the total plant cover on the plot (or the proportion of a layer’s cover). Relative cover values must be calculated from absolute cover values. For example, we estimate overstory canopy cover on a plot as follows: lodgepole pine 42%, Engelmann spruce 21%, and subalpine fir 7%. These values are the absolute cover of each species. The relative cover of each species is calculated by dividing each absolute cover value by their total (70%) as follows:

Absolute Cover Calculation Relative Cover Lodgepole pine 42% 100 x 42 / 70 = 60% Engelmann spruce 21% 100 x 21 /70 = 30% Subalpine fir 7% 100 x 7 /70 = 10% Total of values 70% 100%

We calculate relative cover of 60% for lodgepole pine. This means that lodgepole pine makes up 60% of the overstory tree canopy cover on the plot. Relative cover always adds up to 100%, but absolute cover does not. Because plant canopies can overlap each other, absolute cover values can add up to more than 100%. In our example, the total of the absolute cover values is 70, but this does not mean that overstory trees cover 70% of the plot. Overstory tree cover would be 70% if there were no overlap between the crowns of the three species, but only 42% with maximum overlap. The actual overstory cover must be determined when sampling the plot if the information is desired, but the sum of the species cover values is used to calculate relative cover.

If the absolute cover values in our example were all halved or all doubled, the relative cover of each species would not change even though overstory tree cover would be very different. Halving the absolute values would mean overstory cover would be between 21 and 35%, depending on the amount of overlap. Doubling the values would mean overstory cover could range from 84 to 100% (not 140%). Each of these scenarios would be very different from the original example in terms of wildlife habitat value, fuel conditions, fire behavior, and silvicultural options; but the relative cover of the tree species would be exactly the same. We should also note that they also could vary widely in spectral signature. The key point here is that relative cover values by themselves provide limited ecological information and may be of little value to resource managers. Relative cover can be derived from absolute cover, but absolute cover cannot be derived from relative cover values. This is why absolute cover is recorded in the field.

Appendix A: 14 Boise/Payette - Vegetation Type & Structure Classes

VEGETATION TYPE MAP UNIT GROUPS: Deciduous Forests Coniferous Forests Shrublands Herblands Riparian Sparse Vegetation Non-Vegetated

VEGETATION TYPE MAP UNITS: Deciduous Forests: Aspen Aspen/Conifer Mix

Coniferous Forests: Douglas Fir Douglas Fir with Lodgepole Pine Douglas Fir with Ponderosa Pine Grand Fir Grand Fir with Ponderosa Pine Engelmann Spruce Lodgepole Pine Ponderosa Pine Subalpine Fir Subalpine Fir with Douglas Fir Whitebark Pine (includes Subalpine Fir with Whitebark) Western Larch

Shrublands: Bitterbrush (not mapped on the Payette NF) Forest Shrub Low Sagebrush (not mapped on the Boise NF) Mountain Big Sagebrush Mountain Shrub

Herblands: Grasslands (Perennial & Annual) Forblands Weedy Herbaceous (not mapped on the Payette NF) Alpine (not mapped)

Riparian: Riparian Shrub (includes Riparian Deciduous) Riparian Herbaceous (includes Wetlands)

Sparse Vegetation: Sparse Vegetation (includes Barren/Rock)

Appendix A: 15 Non-vegetated: Agriculture Developed Water

Burned Area (with standing dead trees): Burned Herbaceous Burned Forest Shrubland Burned Sparse Vegetation

CANOPY COVER CLASSES: Forested Map Units (Deciduous & Coniferous) TC1 - 10%-19% TC2 - 20%-29% TC3 - 30%-44% TC4 - 45%-59% TC5 - ≥60%

Shrubland Map Units SC1 - 10%-24% SC2 - 25%-34% SC3 - ≥35%

TREE SIZE CLASSES Forested Map Units (Deciduous & Coniferous) TS1 - <4.5 ft. tall TS2 - <5” DBH TS3 - 5”-9.9” DBH TS4 - 10”-19.9” DBH TS5 - 20”-29.9” DBH TS6 - ≥30” DBH

Appendix A: 16 Appendix B: Geospatial Data Acquisition and Pre-Processing

Landsat Data Acquisition and Processing  Acquired Landsat 5 Thematic Mapper (TM) imagery (30m multi-spectral and 15m pan) from USGS Global Visualization geospatial data clearinghouse (http://glovis.usgs.gov/). The following chart outlines the scenes and dates used for each season:

Spring Landsat scenes: Date Path Row July 09, 2008 42 28-29 July 02, 2008 41 28-30 July 13, 2006 (cloud replacement) 41 29

Summer Landsat scenes: Date Path Row August 24, 2007 42 28-29 August 22, 2009 41 28-30 August 14, 2006 (cloud replacement) 41 30

Fall Landsat scenes: Date Path Row October 28, 2008 42 28-29 October 13, 2008 42 28-29 October 15, 2009 41 28-30 October 22, 2008 41 28-30

 Individual bands were georeferenced raster data in GeoTiff file format which were subsequently layer stacked in ERDAS Imagine to produce a single image file. The default spatial reference information was in WGS 84 UTM Zone 11N  Image anomalies (clouds, cloud shadows, haze, and smoke from fires) were identified and clipped out.

Appendix B: 1

 A single scene was selected from each season to be the primary reference scene to which all adjacent scenes were mosaicked. Sometimes this was a single scene or a strip of multiple scenes with the same path date.  An atmospheric correction was applied to the target scene.  Common-usable areas which overlapped between the primary reference scene and adjoining target scenes were outlined.  The mean and standard deviation of the date within the common-usable-area were calculated and the model II regression formula was applied to the target scene using the calculated results.  New calibrated scenes were mosaicked for each season.  The seasonal mosaics were clipped to the project area boundary, which included a buffer around the administered Boise and Payette National Forest boundaries.  Reprojected the final clipped mosaics to UTM, Zone 12, NAD27, GRS1980.  Clipped the mosaiced imagery to each GA boundary.

Landsat Derivatives (NDVI, PCA, and Tassel Cap)  ERDAS Imagine modules “Normalized Difference Vegetation Index (NDVI), Principal Components Analysis (PCA), and Tasseled Cap” were used to generate layers for each season’s (spring, summer, fall) Landsat image. (Model options included: 1) stretch to unsigned 8-bit; 2) ignore zero in stats; and 3) clip imagery to the project area boundary AOI.  Vegetation Change Tracker (VCT) and Monitoring Trends in Burn Severity (MTBS) are both Landsat derivatives designed to detect change due to disturbance. These layers were primarily used to identify and map areas that were impacted by recent forest fires.

DEM Acquisition & Processing  DEMs were retrieved from DataDoors, where derived value-added products are distributed to application specialists and decision-makers within the USFS. The data provided was a seamless, reprojected, 10-meter, grid, elevation data layer.

DEM Derivatives  Slope – Used ERDAS slope model to generate a continuous degree slope layer. This layer was then converted to a thematic layer with the following 12 classes:

Appendix B: 2

1. < 1 7. 26 to 30 2. 1 to 5 8. 31 to 35 3. 6 to 10 9. 36 to 40 4. 11 to 15 10. 41 to 45 5. 16 to 20 11. 46 to 60 6. 21 to 25 12. > 61  Aspect – Used ERDAS aspect model to generate a continuous degree aspect layer. This layer was then converted to a thematic layer with the following 9 classes: 1. ≥ 1 and < 23 north 2. ≥ 23 and < 68 north-east 3. ≥ 68 and < 113 east 4. ≥ 113 and < 158 south-east 5. ≥ 158 and < 203 south 6. ≥ 203 and < 248 south-west 7. ≥ 248 and < 293 west 8. ≥ 293 and < 338 north-west 9. ≥ 338 and ≤ 360 north 10. = 361 (flat areas) no aspect –flat areas

 Fully Illuminated Hillshade – Used a customized model to generate a shadow-free topographic relief composite. This layer is a composite of three equal-intervals of 120 degrees (120, 240, and 360 degree). It depicts terrain displacement and portrays aspect and slope intensity.  Compound Topographic Index (CTI) – Used a customized model to generate a steady state wetness index. This index is a function of both the slope and the upstream contributing area per unit with orthogonal to the flow direction (Moore et al., 1993).  Heatload – Used a customized model to generate an estimate of the direct incident solar radiation based on slope, aspect, and latitude of the study area (McCune and Keon 2002).  Curvature – Used a customized model to generate a grid layer showing the physical characteristics of a drainage basin.  ValleyBottom – Used a customized model, stream order layers, and elevation information to delineate the shape of the valley bottoms. Valley shape assessment was used to identify where potential riparian zones may occur (Goetz, 2001).

Appendix B: 3

Ancillary Raster Data  NAIP imagery - was acquired for the Payette National Forest. The 1-meter DOQs were mosaicked for each mapping GA and resampled to 10m.  MTBS data – was acquired from the Remote Sensing Applications Center MTBS website (http://www.mtbs.gov/). MTBS identifies burn severity and fire perimeters for fires greater 1,000 acres in the western U.S. occurring from 1984 through 2010.  Vegetation Change Tracker (VCT) – was used to generate a thematic disturbance year product using Landsat imagery from the years 1984 through 2010. ( Li et al., 2009) describes VCT as “an automated forest change mapping algorithm based on the spectral and temporal properties of forest disturbance and post-disturbance recover process, and is designed for analyzing dense Landsat images”. A suite of indices are generated during the process, these are then used to assign the likelihood of being a “forest” throughout the entire time series. Abrupt decreases in the forest likelihood values indicate a change.  Texture - was derived using the resampled (5.25 meter) 1-foot resource photography. The 1-foot resource photography was first processed to a single layer through principal components. Then an adaptive texture model was applied in ERDAS using a 15x15 kernel. Then the texture imagery was resampled to 5.25 meter.  Daymet climate data - was downloaded from the from the Daymet website (http://www.daymet.org/). This data was available in 1-km pixel resolution and included information on temperature, precipitation, solar radiation, and growing days.  Raw Height data - was acquired from the National Biomass and Carbon Dataset (NBCD) website (http://www.whrc.org/mapping/nbcd/index.html). NBCD produced a one 16-bit signed raster layer at 30m resolution. Digital numbers represent the average basal area weighted height in meters * 10.  Airborne Interferometric Synthetic Aperture Radar (ifSAR) - was acquired from Intermap Technologies. These data were used to assist in the modeling of tree size for the Payette National Forest.  Landtype - was provided by the Forests.  Geology - was provided by the Forests.

Appendix B: 4

Summary of raster data layers generated/acquired: Spatial Data Layers Type Resolution Landsat Landsat TM - Spring Continuous 30-m Landsat TM - Spring NDVI Continuous 30-m Landsat TM - Spring Tasseled Cap Continuous 30-m Landsat TM - Spring PCA Continuous 30-m Landsat TM - Summer Continuous 30-m Landsat TM - Summer NDVI Continuous 30-m Landsat TM - Summer Tasseled Cap Continuous 30-m Landsat TM - Summer PCA Continuous 30-m Landsat TM - Fall Continuous 30-m Landsat TM - Fall NDVI Continuous 30-m Landsat TM - Fall Tasseled Cap Continuous 30-m Landsat TM - Fall PCA Continuous 30-m Topographic Elevation Continuous 10-m Slope Continuous/Thematic 10-m Aspect Continuous/Thematic 10-m Compound Topographic Index Continuous 10-m Heatload Continuous 10-m Fully Illuminated Hillshade Continuous 10-m Valley Bottom Thematic 10-m Other Imagery NAIP (4 bands) Continuous 10-m Resource Photography Texture Continuous 5.25-m ifSAR Continuous 5-m NBCD Continuous 30-m Ancillary Vegetation Change Tracker (VCT) Thematic 30-m MTBS Thematic 30-m Landtype Thematic 30-m Geology Thematic 30-m Precipitation Continuous 1-km Radiation Continuous 1-km Growing Days Continuous 1-km Temperature Continuous 1-km

Appendix B: 5

Appendix C: Training Data Protocols and Forms

Boise/Payette National Forests Existing Vegetation Mapping Project Field Data Collection Guide & Protocols – May 2010

This document will serve as a guide to data collection for the Boise and Payette National Forests Vegetation Mapping Project. Detailed instructions on how to fill out the datasheets are included in this document. These protocols have been established following the USFS Existing Vegetation Classification Mapping Technical Guide as well as guidelines from the Remote Sensing Applications Center (RSAC).

Overview

The Boise and Payette National Forests are responsible for managing vegetation to meet a variety of uses while maintaining the integrity of ecosystem components and processes. The 2003 Land and Resource Management Plans (Forest Plans) provide the strategic management framework that guides project planning and implementation. This framework is based in part on a vegetation classification, mapping, and quantitative inventory system current through 2000. An updated vegetation map is needed to account for significant changes that occurred since 2000, to ensure that Forest Plan objectives are based on the most accurate information, and to substantiate effects disclosures pertaining to National Forest Management Act (NFMA) obligations concerning vegetative and wildlife species diversity.

The data you collect will be used to create a 1:100,000 mid-level map of existing vegetation communities across the Boise and Payette National Forests for use in Forest planning, habitat assessment, and other regional analysis. Data gathered will include information on species composition, forest and shrub canopy cover, and forest size class. This data will be estimated from a “bird’s eye” or “satellite” view of the field plot, vegetation canopy overlap will not be considered. Estimations will be done using a combination of ocular estimates and transects.

Tools

You have been provided with several tools to assist in the field data collecting process. They include:

. Dominance type key . Field data collection forms . Field overview maps (1:160,000 scale) . Field travel maps (1: 20,000 scale) . Plot maps (1: 9,000 scale) . Habitat Type Manual . Guide to Indicator Species

Appendix C: 1

General Data Collection Procedures

Field information will be collected in three ways: . Pre-selected field plots . Field observation polygons . Opportunistic field plots

Pre-Selected Field Plots There are 500 field plots identified for each geographic area (A total of 5 geographic areas cover the entire project area). These plots were chosen using spectral information from Landsat Thematic Mapper satellite imagery, elevation, and slope, and aspect. They are not a random sample of the mapping area and have not been established along a sample grid, or other sampling procedure. Plots were selected in vegetative homogenous areas generally within a quarter mile of a road or along trails (FIA B-Grid travel routes). Some sites are behind closed roads or in roadless areas not associated with the B-Grid travel routes. These sites should provide a sample of the landcover communities that occur on the forests.

Field Observation Polygons Approximately 5 additional field observation sites will be collected with each of the pre- determined field plots. You will use the plot maps (NAIP imagery, 1-meter resolution) to identify areas of homogenous vegetation and estimate the dominance type, vegetation map unit, canopy closure, and tree dbh class. This provides an opportunity to quickly collect additional vegetation information.

Opportunistic Field Pots Opportunistic or new field plots can be established if it is determined an area or vegetation type has not been adequately sampled. These plots follow the same data collection protocols as the pre-selected sites.

Sampling Process and Data Collection Procedures

The sampling process contains three steps: planning, navigation, and data collection.

Step 1 - Planning Before leaving the office, each crew should know where they are going, what information is going to be collected, and have the appropriate gear to complete the task. Review the overview maps and travel maps to determine the best travel routes. Check with your supervisor and/or crew lead before leaving. Coordination with designated Forest personnel to ensure access should be completed before leaving for field.

Gear check list: - GPS unit - Dominance type key - Digital camera - Travel maps & plot maps - Batteries (GPS and Camera) - Pencils & sharpies - Data sheets - Clinometer

Appendix C: 2

- 100m tape - Pin Flags - 100ft tape - Whiteboard - DBH tape - Plot location markers for - Compass wilderness as determined at pre-work - Flagging meeting

Step 2 - Navigation You have been provided with the coordinates of the plot center, and navigation and plot maps with 2006 NAIP imagery in the background to help with navigating to the plot. The waypoints should be pre-loaded on the GPS unit. Plots have been located generally within a ¼ mile of a road, along backcountry FIA B-grid travel routes (GA 4 & 5 only), or along trails, to make them as accessible as possible. However, there is no guarantee that the plots will be accessible. If you cannot get to the plot, but can clearly see it from some vantage point, fill out as much information possible. If a plot is completely inaccessible and cannot be viewed, note that the plot is unobservable on the field plot form, and either go on to the next plot location or create a new plot in a nearby area with similar vegetation and topographic characteristics including vegetation type, aspect, and slope.

Appendix C: 3

Plot map showing roads (color-coded by type), contour lines, streams, and plot locations. Scale of imagery is 1:9000.

Step 3- Data Collection . Pre-selected field plots The size of each plot is a 15 meter radius circle. Pace or measure and flag the plot boundaries from the pin-flagged plot center. All vegetation data will be estimated in this area from a “bird’s eye view” or top-down perspective. It is important to walk through the entire plot before estimating species, canopy cover, and tree size class. It may also be helpful to mark out a 1.5 meter radius subplot representing 1 percent of the plot area to assist with your calibration estimates.

Image map showing plot center location and corresponding 15 meter radius plot boundary surrounded by homogeneous vegetation cover.

For the first 5 shrubland plots per observer, use the transect intercept method to determine the shrub canopy cover to calibrate your N ocular estimates. For every 3-5 shrubland - plots thereafter (per observer), use the transect S intercept method to maintain consistency of your ocular estimates. The intercept method involves laying out two perpendicular 100- foot transects through the plot center; one running north-south and one running east- west, using tapes and stakes. Do not allow the vegetation to deflect the alignment of the tape. Estimate and record the number of feet of live canopy cover intercepted for each species within each 10-foot transect increment. Gaps

Appendix C: 4 within a single plant, flowers, and flower stalks should be counted as part of the shrub. The total for each transect is the canopy percentage. The N/S transect and E/W transect percentages are then averaged to calculate the overall shrub canopy cover.

. Field Observation Polygons For each of the pre-selected field plots, we also ask that five field observation sites be collected using the same map used to navigate to the pre-selected plots (1:9000 scale with NAIP imagery as a backdrop). On the plot map with the NAIP imagery backdrop, draw a polygon around an area of homogenous vegetation, label it (A, B, C, D, or E), and fill in the appropriate information on the left side on the back of the field plot form. Here you will provide general information on the map group, dominance type, vegetation map unit, canopy closure, and tree dbh class. Where easily identifiable, target a variety of vegetation types and structure classes to capture the representative vegetation communities occurring in the project area. If you cannot correctly make a determination on all of these, just complete those that you have confidence in. Make sure the labels are legible and the polygons you draw identify areas of homogenous vegetation, including canopy cover and size class. Remember the minimum mapping unit is 5 acres for upland areas and 2 acres for riparian and aspen stands. If you cannot adequately identify the location on the plot map (i.e. heavily forested areas) then record the GPS location so that the precise location can be accurately located and used for the vegetation modeling aspects of the project.

. Opportunistic Plots While you are traveling from plot to plot and encounter underrepresented vegetation types, you can delineate new field plot locations and collect vegetation information in the same way as specified for the pre-selected plots. We will refer to these as opportunistic plots. Project personnel will provide information on what is considered to be underrepresented for the project area.

Three principals should guide your selection of opportunistic field plots: 1. Plots should be located in homogenous vegetation types that are at least 5 acres in size or 2 acres size for riparian and aspen communities. 2. The plot should represent a single vegetation life form. 3. The plot should not cross roads, major topographic breaks, major streams, etc.

Opportunistic plots must be given a completely new number; a previously assigned unused number cannot be used for an opportunistic plot. Various field crew members can be assigned a “set” of numbers so that no one will duplicate a number. A good suggestion would be to use numbers that would come after the pre-selected plot numbers, and give each crew member a set of 10-20 in that GA. The individual crew member would then be responsible for keeping track of which numbers have already been used for an opportunistic plot.

An opportunistic plot must be within the project boundary (i.e. on NFS lands designated for the project). It cannot be adjacent to lands of the project boundary. If a pre-selected

Appendix C: 5 plot is revisited, it cannot be labeled as “opportunistic” and given a second number. It is the responsibility of field crews to keep track of plots visited and who has been assigned to visit a particular plot. It is also the responsibility of the field crew to assure that an opportunistic plot is within the project boundary, within the appropriate Geographic Area, and that it represents vegetation types that are harder to encounter, as directed by project personnel.

Initial direction regarding what is considered underrepresented will be given at the start of the project. As field data sheets are received by project personnel, tracking and tallying of both the map units/dominance types being collected and their distribution, will assist with future selection of opportunistic plots. It is the responsibility of field crews to coordinate with Forest Service personnel in the appropriate collection of opportunistic plots, which can be modified as the field data collection progresses.

Data Collection Forms This section provides information on how to fill out the datasheets.

Field Plot Form

1. Plot ID — Record the 4-digit field plot number.

2. Names of collectors— Record the names of the personnel collecting the data. Initials can be used if they are unique to the entire team. However, names are preferred on the first few forms for each geographic area.

3. Month/Day

4. Level of Observation— Record the level of observation. “VI” stands for visited field plot, “VFD” stands for plot viewed from a distance, “NO” stands for not observable, and “NEW” stands for new opportunistic plot.

5. UTM E & N— Record the coordinates for the center of the plot. For new plots you should collect a minimum of 30-60 positions for non-forested plots and 60-90 positions for forested plots (or as many as possible if experiencing difficulty). It is important to collect positions from the plot center, so be at the center to start collection. Every plot should use a PDOP mask of 6 and elevation mask of 15. IF the GPS is not working (low satellites, etc.), then raise the PDOP, using the highest accuracy (i.e. the lowest number) possible.

GPS unit should be set to the following projection: UTM, Zone 11 NAD83 GRS1980

Appendix C: 6 6. Field Photograph— Take a single representative photo of the field site (more can be taken if necessary) and record the digital photo number. This photo number will need to be completely unique to all photos taken so that when it is transferred it does not get confused with other photos. The photos should be renamed at a later time to match the field plot number. A whiteboard or other marker with the field site number can also be used when taking the photo to help identify the site.

7. Geographic Area— Record the geographic area (GA) that the site is located in. This number should appear on the field plot list and image map.

8. Ocular Plot Composition— (Estimated from a “Top-down” perspective). List out each major species type (starting with tree species, then shrub, then herbaceous/non-veg) using the provided standard PLANTS code. If the code is not known, its name should be written out and the code looked up later. Estimate the total canopy cover for trees, shrubs, herbaceous, and non-vegetated. Determine canopy cover as if you were looking down on the stand from the air, allowing overlap to influence the counts. So smaller sized trees being overlapped by larger ones will be ignored and not counted in the canopy cover estimate. These sums of all life forms should add up to 100%. Then determine the canopy cover for each of the listed species. You may not be able to list out all the herbaceous or shrub species so record the dominant (most abundant) species only – we don’t need a complete plant list.

9. Tree DBH Size Class— (Estimated from a “Top-down” perspective). List out each tree species & cover as it is in #8. Determine the overstory tree size class (dbh) for each species and enter it in the size class columns. Determine size as if you were looking down on the stand from the air, allowing overlap to influence the counts. So smaller sized trees being overlapped by larger ones will be ignored and not counted in the size class estimate. Total each size class estimate for each species in the Cover column.

10. Shrub Canopy Cover by line intercept— (Only use if Lifeform is likely to be Shrubland – not for tree sites). List the Plant Codes for each major shrub species. Lay out two 100-foot transects perpendicular to and intersecting the plot center; one running north-south and one running east-west. Estimate and record the number of feet of live canopy cover intercepted for each species within each 10-foot transect increment. Gaps within a single plant, flowers, and flower stalks should be counted as part of the shrub. Total it on the right to get % cover of that species. Total all shrub species percents to get the actual shrub canopy cover for that transect. Calculate the overall shrub canopy cover by averaging the total shrub cc from both the north-south and east-west transects. A measured line intersect should be completed for every three shrubland sites visited to help maintain consistency for the ocular plot composition estimate (#8).

Plot Summary 11. Map Group— Based on the canopy cover from the ocular plot composition (#8) and classification key, determine the map group and record it in the “1st” column. If shrub canopy information from transects (#10) has been collected use the overall shrub cc to determine the map group. If a plot is near the borderline between map groups, record the

Appendix C: 7 secondary map group in the “2nd” column. (For example, if tree canopy cover totals 12 percent, record Conifer or Deciduous Forest as the first call, and Shrubland, or Herbland as the second call based on the cover of those maps groups). If you are in an area that was forested and is now burned with standing dead trees, and there is not 10% canopy cover of live trees within the plot, determine if at least 2/3 of the standing trees in your plot are burned . If this is the case, record the dominance type according to the vegetation that is alive on the site, and record the map group as B (burned area). Map group letter codes are as follows:

C – Conifer Forest D – Deciduous Forest S – Shrubland H – Herbaceous R – Riparian A - Alpine V – Sparse Vegetation B – Burned Area N – Non-Vegetated

12. Dominance Type— Based on the ocular plot composition (#8) and the dominance classification key, determine the dominance type and record it in the “1st” column. If shrub canopy information from transects (#10) has been collected use the transect data to determine the dominance type for the shrubland map group. Again, if a plot is near the borderline between dominance types, record the secondary dominance type in the “2nd” column. The full dominance type list can be found in the dominance key.

13. Vegetation Map Unit— Based on the dominance type classification, determine the vegetation map unit and record it in the “1st” column. Again, if a plot is near the borderline between map units, record the secondary type in the “2nd” column. For example, if PIPO is the most abundant tree with PSME a close second, record PIPO as the first call and PSME as the second call. The full vegetation map unit list can be found in the dominance type key.

14. Canopy Cover— Based on the predominant map group determine the canopy cover class for forested, shrubland, and riparian sites only and record it in the “1st” column. Shrub canopy cover information can be obtained from the ocular plot composition (#8) or the shrub canopy cover line intercept (#10). Forested riparian areas should be assigned a forest canopy class and shrubland riparian areas should be assigned a shrubland canopy class. Again, if a plot is near the borderline between canopy classes, record the secondary class in the “2nd” column. The secondary canopy class should be based on the secondary map group class if it is different from the primary map group. The canopy classes are as follows:

Forested TC1 – Low canopy tree cover 10 – 19% TC2 – Low-medium canopy tree cover 20 – 29% TC3 – Medium canopy tree cover 30 – 44%

Appendix C: 8 TC4 – Medium-high canopy tree cover 45 – 59% TC5 – High canopy tree cover > 60%

Shrubland SC1 – Low shrub canopy cover 10 – 24% SC2 – Medium shrub canopy cover 25 – 34% SC3 – High canopy shrub cover > 35%

15. DBH Size Class— Based on the tree dbh size class (#9) determine the most abundant size class and record it in the “1st” column. Again, if a plot is near the borderline between size classes, record the secondary size class in the “2nd” column. The dbh size classes are as follows:

TS1—Seedlings < 4.5ft tall TS2—Saplings 0.1-4.9" DBH TS3—Small Tree 5-9.9" DBH TS4—Medium Tree 10-19.9" DBH TS5—Large Tree 20-29.9" DBH TS6—Very Large Tree > 30" DBH

16. Notes— Record a description of the plot. Include information on the vegetation conditions, disturbances (fire, timber harvests, insect outbreaks, wind events, etc.), approximate age of the disturbance, and any other information that is not included in the field form. This description is often the most valuable piece of information we have about a plot and provides details that can have an effect on the mapping process.

17. Habitat Type —Using the habitat type manual determine the habitat type.

Observation Polygon Form Approximately 5 additional field observation sites will be collected for each of the given field plots. Using the image maps provided (NAIP imagery, 1-meter resolution), draw a polygon around an area of homogenous vegetation, label it (A, B, C, D, or E), and fill in the data on the left side of the field form. This data provides general information on the map group, dominance type, vegetation map unit, canopy closure, and tree dbh class. Make sure the labels are legible and the polygons you draw identify groups of homogenous vegetation, including canopy cover and size class. The canopy cover information on the right side of the field form (8-12) will be collected at a later time using photo-interpretation techniques. If you think it would be helpful, designate a symbol on the NAIP plot map to indicate where you were standing when you made the field observation.

1. Map Group— Ocular estimate of dominant map group for the polygon you delineated on the imagery

2. Dominance Type— Ocular estimate of the dominance type for the polygon you delineated on the imagery

Appendix C: 9 3. Vegetation Map Unit— Ocular estimate of the vegetation map unit for the polygon you delineated on the imagery

4. Canopy Cover— Ocular estimate of the canopy cover in 5% increments for the polygon you delineated on the imagery

5. DBH Class— Ocular estimate of the dbh class for the polygon you delineated on the imagery

6. Coordinates— If the site was hard to delineate on the imagery, and you had to walk into the site to determine the vegetation characteristics, take the center coordinates.

7. Notes— Record any information, such as site description or general vegetation conditions, that may be relevant to the site.

Appendix C: 10 Region 4 - Boise/Payette NF – FIELD PLOT FORM

1- PlotID# ______2- Names: ______3- Month/Day ____-____-10

4- Level of Observation: VI VFD NO NEW v v Jjjjjh 5- UTM E: ______N: ______(UTM, NAD83, GRS1980, Zone 11)

6- Field Photograph: ______7- Geographic Area: 1 2 3 4 5

8- “OCULAR” PLOT COMPOSITION Tree Cover Shrub Cover Herbaceous Cover Non-veg Cover

Total Total Total Total

(Lifeform totals must add up to 100%)

9- Tree DBH Size Class

Plant Code Cover TS1 TS2 TS3 TS4 TS5 TS6 Tree Size Classes: TS1 < 4.5ft Seedling TS2 0.1 - 4.9”dbh TS3 5 - 9.9”dbh TS4 10 - 19.9”dbh TS5 20 - 29.9”dbh

TS6 ≥ 30”dbh

Total

10– Shrub Canopy Cover – by line intercept Transect North/South Overall Shrub CC = Plant Code 0-10’ 10-20’ 20-30’ 30-40’ 40-50’ 50-60’ 60-70’ 70-80’ 80-90’ 90-100’ Total

Transect East/West Total SHRUB CC Plant Code 0-10’ 10-20’ 20-30’ 30-40’ 40-50’ 50-60’ 60-70’ 70-80’ 80-90’ 90-100’ Total

Total SHRUB CC

PLOT SUMMARY 1 6 - Notes: 1st 2nd 11- Map Group |______|______| 12- Dominance Type |______|______| 13- Veg Map Unit |______|______| 17– Habitat Type: 14- Canopy Cover |______|______| 15- DBH Class |______|______|

Region 4 – Boise/Payette NF – POLYGON FORM

Field Observation PI Canopy Cover

Polygon 1-Map Group |_____| 7-Notes: 8-Conifer Canopy Cover |______| |A| 2-Dominance Type |______| 9-Deciduous Canopy Cover |______| 3-Veg Map Unit |______| 10-Shrub Canopy Cover |______|

4-Canopy Cover ~5% |______| 11-Herbaceous Cover |______| 5-DBH Class |______| 12-Non-Vegetated Cover |______| 6-Coordinates: TOTAL COVER: |_100%_|

Polygon 1-Map Group |_____| 7-Notes: 8-Conifer Canopy Cover |______| |B| 2-Dominance Type |______| 9-Deciduous Canopy Cover |______| 3-Veg Map Unit |______| 10-Shrub Canopy Cover |______|

4-Canopy Cover ~5% |______| 11-Herbaceous Cover |______| 5-DBH Class |______| 12-Non-Vegetated Cover |______| 6-Coordinates: TOTAL COVER: |_100%_|

Polygon 1-Map Group |_____| 7-Notes: 8-Conifer Canopy Cover |______| |C| 2-Dominance Type |______| 9-Deciduous Canopy Cover |______| 3-Veg Map Unit |______| 10-Shrub Canopy Cover |______|

4-Canopy Cover ~5% |______| 11-Herbaceous Cover |______| 5-DBH Class |______| 12-Non-Vegetated Cover |______| 6-Coordinates: TOTAL COVER: |_100%_|

Polygon 1-Map Group |_____| 7-Notes: 8-Conifer Canopy Cover |______| |D| 2-Dominance Type |______| 9-Deciduous Canopy Cover |______| 3-Veg Map Unit |______| 10-Shrub Canopy Cover |______|

4-Canopy Cover ~5% |______| 11-Herbaceous Cover |______| 5-DBH Class |______| 12-Non-Vegetated Cover |______| 6-Coordinates: TOTAL COVER: |_100%_|

Polygon 1-Map Group |_____| 7-Notes: 8-Conifer Canopy Cover |______| |E| 2-Dominance Type |______| 9-Deciduous Canopy Cover |______| 3-Veg Map Unit |______| 10-Shrub Canopy Cover |______|

4-Canopy Cover ~5% |______| 11-Herbaceous Cover |______| 5-DBH Class |______| 12-Non-Vegetated Cover |______| 6-Coordinates: TOTAL COVER: |_100%_|

Map Group Code Vegetation Map Unit Code Vegetation Map Unit Code Vegetation Map Unit Code Conifer Forest C Aspen AS Forest Shrub FS Burned - Standing Dead Trees SBT Deciduous Forest D Aspen/Conifer Mix ASC Mountain Shrub MS Agriculture AGR Shrubland S Douglas Fir DF Bitterbrush BB Developed DEV Herbaceous H Douglas Fir/Lodgepole Pine DFL Low Sagebrush LS Barren/Rock BR Riparian R Douglas Fir/Ponderosa Pine DFP Mountain Big Sagebrush MB Water WA Alpine A Engelmann Spruce ES Mountain Mahogany MM Unknown UNK Sparse Vegetation V Grand Fir GF Tree Canopy Cover Code Burned Area B Grand Fir/Ponderosa Pine GFP Perennial Grasslands PG Low 10 - 19% TC1 Non-Vegetated N Grand Fir/Western Larch GFW Annual Grasslands AG Low-Medium 20 - 29% TC2 Lodgepole Pine LP Forblands FO Medium 30 - 44% TC3 Tree DBH Class Code Ponderosa Pine PP Alpine AL Medium-High 45 - 59% TC4 Seedlings < 4.5 ft TS1 Subalpine Fir SA Weedy Herbaceous WH High ≥ 60% TC5 Saplings 0.1-4.9" dbh TS2 Subalpine Fir/Douglas Fir SAD Small Tree 5-9.9" dbh TS3 Subalpine Fir/Western Larch SAW Wetlands WET Shrub Canopy Cover Code Medium Tree 10-19.9" dbh TS4 Subalpine Fir/Whitebark Pine SAB Riparian Deciduous Tree RDE Low 10 - 24% SC1 Large Tree 20-29.9 dbh TS5 Western Larch WL Riparian Shrub RSH Medium 25 - 34% SC2 Very Large Tree ≥ 30" dbh TS6 Whitebark Pine WB Riparian Herbaceous RHE High ≥ 35% SC3

Boise & Payette National Forests Existing Vegetation Mapping Project Aerial Photointerpretation Guide August 2010

1. Introduction

This guide outlines the procedures for collecting aerial photointerpretation (PI) data for the Boise and Payette National Forests Existing Vegetation Mapping Project. An integrated approach using field visits, field sampled data, and stereoscopic photointerpretation techniques will be used to characterize vegetation composition and structure from the 2008 4-band digital resource imagery. The resulting aerial photo-based data will be used to supplement field-based data for modeling the distribution and extent of existing vegetation.

2. Background

The project area comprising the Boise and Payette National Forests has been divided into 5 geographic areas (GA’s). A sample design has been developed for the collection of both field and aerial photo-based training or reference data. For each GA, approximately 500 field plots and 500 aerial photointerpretation plots have been proportionally distributed within the feature space of a spectral-topographic landscape stratification. In addition, for each field plot collected, about 5 observation polygons are collected in the field (~2,500 observation polygons for each GA).

Prior to collecting the aerial photo-based data, photointerpreters will conduct site visits and investigate field-sampled data to calibrate stereo photo interpretations of vegetation composition and structure. The photointerpretation work consists of two major components:

 PI Canopy Cover Assessment of Field Data  PI Plot Data Collection

The PI canopy cover assessment of field data involves estimating percent cover for field plots and observation polygons collected by crews on the ground. Percent canopy cover will be collected for conifer and deciduous trees, shrubland, herbaceous, and non-vegetated categories. The corresponding canopy cover map class will be estimated for forest and shrubland field sites as determined using the vegetation keys found in the project work plan, Attachment 4. Additional attributes will include the interpreter name, interpretation date, stereo pair image file names, and comments. PI canopy cover assessment data will be entered into an ESRI geodatabase polygon feature class containing the field site locations and associated plot ID’s.

The second component of photointerpretation involves collecting data at PI plots independent of the field sites described above. Data collected at PI plots will consist of plot summary map attributes including vegetation map group and map unit, percent canopy cover, corresponding canopy cover map class, and overstory tree diameter class. Vegetation map group and map unit classes will be determined using the vegetation keys found in the project work plan, Attachment 4. For vegetation map unit and overstory tree diameter calls, the level of confidence in the interpretations will also be recorded (i.e. high, medium, or low). Additional

Appendix C: 13

attributes will include the interpreter name, interpretation date, and stereo pair image file names, and comments.

3. Photointerpretation 3.1 PI Canopy Cover Assessment of Field Data

3.1.1 Sample Locations - A field plot and field observation feature class containing the 15 meter radius circle of ~500 field plots and ~2,500 estimated observation polygon point locations within each GA will be provided by the FS.

3.1.1.1 For Field Plots, verify the plot contains vegetation that is representative of the vegetation type map unit and overstory tree size class collected by crews in the field. Interpret percent canopy cover data for the 15 meter radius area representing the fixed- location field plot. In cases where the plot is not representative of the field data, make a note of it in the Comments field and do not interpret the canopy cover.

3.1.1.2 For Field Observation Polygons, reference the polygons that were hand- delineated by field crews on the Plot Maps to relate the observation polygons to individual segments. Verify the identified segment contains vegetation that is representative of the vegetation type map unit and overstory tree size class collected by crews in the field. Move the digital points in the feature class layer representing the estimated observation polygon location to the approximate center of the segment related to the field-indentified observation polygon. Interpret percent canopy cover data across the entire segment area representing the field observation polygon. In cases where a representative segment cannot be identified, make a note of it in the Comments field and do not interpret the canopy cover.

3.1.2 Field Data Attributes - For each field data plot, record the following attributes: - Interpreter name - Stereo pair image file names - Interpretation date - Comments (as needed)

3.1.3 Percent Canopy Cover - Estimate canopy cover and record the attributes below. Canopy cover estimates will total 100% for each plot. Percent canopy cover attributes include: - % Conifer CC - % Herbaceous Cover - % Deciduous CC - % Non-vegetation Cover - % Shrub CC

3.1.4 Canopy Cover Map Class - Based on interpretation results, determine the canopy cover map class for upland and riparian forest and shrubland types (as determined by the vegetation classification keys). Canopy cover map classes include: Forest Shrubland TC1 - Low tree cover 10 - 19% SC1 - Low shrub cover 10 - 24% TC2 - Low-medium tree cover 20 - 29% SC2 - Medium shrub cover 25 - 34% TC3 - Medium tree cover 30 - 44% SC3 - High shrub cover ≥ 35% TC4 - Medium-high tree cover 45 - 59% TC5 - High tree cover ≥ 60%

Appendix C: 14

3.2 PI Plot Data Collection

3.2.1 Sample Locations - A PI plot feature class containing ~500 PI plot point locations within each GA will be provided by the FS. Interpret PI plot data across the entire segment area represented by the PI plot point location.

3.2.2 Creating New and/or Relocating PI Plots - As PI plot interpretation work progresses, use existing vegetation composition/location information provided by the FS (i.e. field training data, stand exam, GPS points for aspen, larch aerial surveys) to aid in creating new and/or relocating PI plots to under-represented vegetation map units as identified in the Proposed Map Units by GA.docx.

3.2.3 PI Plot Attributes - For each PI plot, record the following attributes:

- Interpreter name - Stereo pair image file names - Interpretation date - Comments (as needed)

3.2.4 Percent Canopy Cover - Estimate canopy cover and record the attributes below. Canopy cover estimates will total 100% for each polygon. Percent canopy cover attributes include:

- % Conifer CC by species - % Herbaceous Cover - % Deciduous CC by species - % Non-vegetation Cover - % Shrub CC

3.2.5 Plot Summary Attributes - Determine and record the plot summary map attributes including vegetation map group and map unit, canopy cover class, and tree size class using methods consistent with field data collection protocols as described below. For vegetation map unit and overstory tree diameter calls, record the level of confidence in the interpretations (i.e. high, medium, low).

3.2.5.1 Vegetation Map Group - Determine and record the map group based on the canopy cover interpretations and vegetation classification keys. If interpreting an area that was forested and is now burned with standing dead trees, and there is not 10% canopy cover of live trees, determine if at least 2/3 of the standing trees are burned. If this is the case, record the map group as B (burned area). Map group letter codes are as follows:

C - Conifer Forest A - Alpine D - Deciduous Forest V - Sparse Vegetation S - Shrubland B - Burned Area H - Herbaceous N - Non-Vegetated R - Riparian

3.2.5.2 Vegetation Map Unit - Once the map group has been identified, determine and record the map unit based on the vegetation classification keys. Also record the interpretation confidence level.

Appendix C: 15

3.2.5.3 Canopy Cover Map Class - Determine and record the canopy cover map class for upland and riparian forest and shrubland types. Canopy cover map classes include:

Forest Shrubland TC1 - Low tree cover 10 - 19% SC1 - Low shrub cover 10 - 24% TC2 - Low-medium tree cover 20 - 29% SC2 - Medium shrub cover 25 - 34% TC3 - Medium tree cover 30 - 44% SC3 - High shrub cover ≥ 35% TC4 - Medium-high tree cover 45 - 59% TC5 - High tree cover ≥ 60%

3.2.5.4 Tree Size Map Class - Determine and record the most abundant tree size class. Also record the interpretation confidence level. DBH size classes include:

TS1 - Seedlings < 4.5 feet tall TS2 - Saplings 0.1 - 4.9” DBH TS3 - Small tree 5 - 9.9” DBH TS4 - Medium tree 10 - 19.9” DBH TS5 - Large tree 20 - 29.9” DBH TS6 - Very large tree ≥ 30” DBH

Appendix C: 16

Figure 1 – Crown coverage scale to guide photo interpretation canopy cover estimates.

Appendix C: 17

Boise/Payette National Forests Existing Vegetation Mapping Project Accuracy Assessment Field Reference Data Collection Guide & Protocols v 8/12/2011

Introduction

This document will serve as a guide to accuracy assessment reference data collection for the Boise and Payette National Forests Existing Vegetation Mapping Project. Detailed instructions on how to fill out the datasheets are included in this document. These protocols have been established following the USFS Existing Vegetation Classification and Mapping Technical Guide as well as guidelines from the Remote Sensing Applications Center.

Background

The Boise and Payette National Forests are responsible for managing vegetation to meet a variety of uses while sustaining and restoring the integrity, biodiversity, and productivity of ecosystem components and processes. In building the knowledgebase required to accomplish this mission, existing vegetation information is collected through an integrated classification, mapping, and quantitative inventory process. This information structure is essential for conducting landscape analyses and assessments, developing conservation and restoration strategies, and revising land management plans that guide project development and implementation.

The data you collect will be used to assess the accuracy of a mid-level (1:100,000 scale) map of current (existing) vegetation communities across the Boise and Payette National Forests. The mid-level map depicts vegetation patches or stands that are 5 acres or greater for upland and 2 acres or greater for riparian and aspen communities. Data gathered will include information on species composition, forest and shrub canopy cover, and tree size class. The data will be estimated from a “bird’s eye” or “satellite” view from above. Vegetation canopy overlap will not be considered. Estimations will be done using ocular estimates.

Tools

You have been provided several tools to assist in the field data collection process. They include: . Vegetation dominance type key . Field data collection forms . Forest overview maps (1:160,000 scale) (Supplied by Photo Science Inc.) . Ranger district travel maps (variable scale) (Supplied by PSI) . Locator maps (1:20000) and Plot maps (1:6,000 scale) on same page (Supplied by PSI)

Appendix C: 18

General Data Collection Procedures

Field information will be collected from pre-selected accuracy assessment polygons. These polygons were chosen using a stratified random sample design based on the draft mid-level vegetation maps. Selected polygons are within a quarter mile of a road or approved motorized trails Some plots may be further away from the road or trail, based on the availability of polygons for selection, but the number of these distant polygons should be limited. Check with local districts for information on road conditions, private property, etc.

Sampling Process and Data Collection Procedures

The sampling process contains three steps: planning, navigation, and data collection.

Step 1 - Planning Before leaving the office, each crew should know where they are going, what information is going to be collected, and have the appropriate gear to complete the task. Review the overview maps, Ranger District maps, and travel maps to determine the best travel routes. Check with your supervisor and/or crew lead before leaving. Coordination with designated Forest personnel to ensure access should be completed before leaving for the field.

Gear check list: - GPS unit - Clinometer - Digital camera - 100ft tape - Batteries (GPS and camera) - DBH tape - Data sheets - Compass - Dominance type key - Flagging - Travel maps & polygon maps - Zippered storage bags - Pencils & sharpies - Whiteboard

Step 2 - Navigation You have been provided with the centroid coordinates of the accuracy assessment field plot along with navigation and segment/field plot maps with 2009 NAIP aerial imagery in the background to help with navigating to the area (Figure 1). The waypoints should be pre-loaded on the GPS unit. Segments have been located randomly so may require cross-country hiking over variable terrain. There is no guarantee that the segments will be accessible. If you cannot access the area due to hazardous terrain or if the segment is completely inaccessible, note that the segment is not observable and go on to the next location.

As soon as possible, send a notification about each inaccessible segment to Photo Science ([email protected] and [email protected]) and a replacement site will be rapidly selected. These inaccessible segments will be submitted to the Forest by the field crew along with the completed field plot data on a weekly basis, in blocks of ten. A weekly summary report of all the field plots collected during that week will be submitted to Photo Science showing a summary of the Map Group (MG), Map Unit (MU), Canopy Cover (CC), and Tree Size (TS) plots collected. No coordinate or field plot numbers will be included in the summary.

Appendix C: 19

Figure 1: Location and Plot maps showing each accuracy assessment segment and field plot location. The top map shows site access. The lower map depicts the accuracy assessment segment, the accuracy assessment field plot, and nearby roads and streams.

Appendix C: 20

Step 3- Data Collection

The size of each segment to be assessed will range from about 2 to 10 acres for upland areas and 0.25 to 5 acres for riparian areas. Once you arrive at the accuracy assessment segment (site), refer to the high resolution image map to determine the orientation of the segment. Find the center of the field assessment plot using the GPS coordinates provided. Walk through the entire segment area including 200 to 300 feet around the center point. Using both an inspection of the aerial imagery plot and a traverse of the polygon, confirm that the accuracy assessment field plot is representative of the entire segment for MG, MU, CC, and TS (if applicable). If the field plot location is representative of the entire segment, then begin filling out the field field form to record the requested data while using the vegetation key. If the designated field plot does not appear to be representative of the segment, then the field plot can be moved to another area of the segment for data collection, provided that the field plot remains entirely within the segment.

Each field plot will have a radius of 83' (area of 0.50 acres). The field plots selected for the field crew are completely contained within each segment. After arriving at the center of the accuracy assessment field plot, record the GPS coordinates and flag the center point. Pace or measure and flag the plot boundaries (83' from the center) in each cardinal direction from the center of the plot. Estimate all vegetation data within the plot area from a “bird’s eye” view or top-down perspective. It is important to walk through the entire field site before summarizing the data for the field site. Record information on the data form about the vegetation found in the plot for map group, dominance type, map unit, canopy cover, and tree size using the instructions below.

Data Collection Forms

This section provides information on how to fill out the Field Polygon Form

1. Polygon ID — Record the field polygon number.

2. Names of collectors— Record the names of the personnel collecting the data. Initials can be used if they are unique to the entire team. However, names are preferred on the first few forms.

3. Month/Day/Year

4. Level of Observation— Record the level of observation. “VI” stands for visited field polygon, and “NO” stands for not observable. Polygons recorded as "NO" will be submitted as soon as possible, not exceeding a week, to both the Forest and PSI for reselection.

5. UTM E & N— Record the coordinates and flag the center of the accuracy assessment field plot. You should collect a minimum of 30-60 positions for non-forested polygons and 60-90 positions for forested polygons (or as many as possible if experiencing difficulty). It is important to collect positions from the plot center. Every plot should use a PDOP mask of 6 and elevation mask of 15. If the GPS is not working (low satellites, etc.), then raise the PDOP, using the highest accuracy (i.e. the lowest number) possible. In the Notes section, record changes to PDOP and elevation masks. If using a GPS unit where the PDOP and elevation masks cannot be set, verify a precision of ≤30 feet before collecting positions.GPS unit should be set to the following projection: UTM, Zone 11; NAD83; GRS1980

Appendix C: 21

6. Field Photographs— Take a single representative photo of the field site (more can be taken if necessary) and record the digital photo number. Take the photo from a location along the plot perimeter that has a side-hill view toward the plot center to capture the slope of the site. This photo number will need to be completely unique to all photos taken so that when it is transferred it does not get confused with other photos. The photos should be renamed at a later time to match the field plot number and cardinal direction taken (e.g. 1224W). A whiteboard or other marker with the field site number can also be used when taking the photo to help identify the site.

7. Ocular Polygon Composition— (Estimated from a “Top-down” perspective). Estimate and record the total canopy cover for trees, shrubs, herbaceous, and non-vegetated. Determine percent over-story canopy cover as if you were looking down on the stand from the air; do not double count overlapping layers that are not viewable from above. For example, smaller sized trees being overlapped by larger ones will be ignored and not counted in the canopy cover estimate. The sum of all lifeform and non-vegetation type totals must add up to 100%.

Based on the lifeform percent canopy estimates and classification key, determine the dominant lifeform or non-vegetation status for the polygon and record the most abundant species (or non- vegetation types) for that lifeform. Use the PLANTS codes or non-vegetation type codes from Table 1. If the code for any species is not known, its name should be written out and the code looked up later. If a plant can only be identified to the genus level, e.g. due to seasonal condition or disturbance, record only the plant genus and make a note of it on the form.

Table 1. Non-vegetation Type Codes BARE Bare soil - soil particles <2mm in diameter ROCK Rock >2mm in diameter LITT Plant litter and duff, including twigs <1/4 inch in diameter WOOD Dead wood material >1/4 inch in diameter, including bases of standing dead trees and shrubs SNOW Area covered by permanent ice and/or snow WATE Water that obscures other cover types

For each of the listed species/non-vegetation types, estimate and record the percent canopy cover as viewed from above. Record the combined percent cover of all “other” species/non-vegetation types that were not individually listed on the form in the previous step. The cover estimates for each species/other category/non-vegetation type must sum to the cover estimate for that lifeform. These estimates will be used to key out the map units.

8. Tree Size Class— (Estimated from a “Top-down” perspective). For forest sites only (≥10% tree), list out each tree species and cover as recorded in #7. For each species, determine the percent cover of each over-story tree size class and enter it in the size class columns. Determine percent cover of each over-story size class as if you were looking down on the stand from the air; do not double count overlapping layers that are not viewable from above. For example, smaller sized trees that are being overlapped by larger ones will be ignored and not counted in the size class estimate. Total the estimated percent cover for each size class. For the first 5 tree sites, measure DBH to calibrate subsequent ocular estimates. For every 3-5 polygons thereafter (per observer), measure DBH to maintain consistency of your ocular estimates.

Polygon Summary 9. Vegetation Group (Map Group -MG)— Based on the canopy cover from the ocular polygon composition (#7), determine the vegetation group from the classification key and record it as the first call (“1st” column). If a polygon is near the borderline between vegetation groups, record Appendix C: 22

the secondary group in the “2nd” column. For example, if tree canopy cover totals 12 percent, record Conifer or Deciduous Forest as the first call, and Shrubland, Herbaceous, or Non- vegetation as the second call based on the cover of those groups. As another example, if shrub canopy cover totals 12 percent on a polygon that is clearly not forested, record Shrubland as the first call and Herbaceous or Non-vegetation as the second call based on the cover of those groups.

10. Dominance Type— Based on the ocular polygon composition (#7), determine the dominance type from the classification key and record it in the “1st” column. The full dominance type list can be found in the dominance key. If a polygon is near the borderline between dominance types, record the secondary dominance type in the “2nd” column. If difficult Grassland and Forbland dominance types can't be identified in the field, take a sample of the dominate species, place in a zipper bag, label with the field plot number, and take to Supervisor's Office or Ranger District office for definitive identification.

11. Vegetation Type (Map Unit - MU)— Based on the dominance type classification, determine the vegetation type map unit and record it in the “1st” column. If a polygon is near the borderline between vegetation types, record the secondary type in the “2nd” column based on the secondary dominance type. For each Geographic Area, a list of applicable map units can be found on the accompanying GA-MU table.

12. Canopy Cover— Based on the predominant vegetation group, determine the canopy cover class for forest and shrubland sites and record it in the “1st” column. Upland and riparian forest sites should be assigned a forest canopy cover class. Upland, riparian, and alpine shrubland should be assigned a shrubland canopy cover class. For shrubland polygons, use the overall shrub cover to determine the canopy cover class. If a polygon is near the borderline between canopy classes, record the secondary class in the “2nd” column. The secondary canopy class should be based on the secondary vegetation group if it is different from the primary vegetation group.

13. Tree Size Class— Based on the tree size class (#8) determine the most abundant size class and record it in the “1st” column. In case of a tie, record the highest tree size class. If a polygon is near the borderline between classes, record the secondary class in the “2nd” column.

14. Disturbance Event— If there is evidence of a relatively recent disturbance event (fire, timber harvest, insect outbreak, wind event, etc.) within the last 5 years, check the appropriate box and include any relevant information such as whether the site was previously forested, contains standing dead trees, etc. in the notes section.

15. Notes— Record a description of the polygon. Include information on the vegetation conditions, disturbances, approximate age of the disturbance, and any other information that is not included in the field form.

Appendix C: 23

Boise/Payette NF – Accuracy Assessment FIELD PLOT FORM v8/12/2011

1- PlotID# ______2- Names: ______3 - M/D/YY ___-___-___

4- Level of Observation: VI NO

5- UTM E: ______N: ______(UTM, NAD83, GRS1980, Zone 11)

6- Field Photograph: ______

7- “OCULAR” Plot Composition Tree Cover Shrub Cover Herbaceous Cover Non-veg Cover

Other Other Other Other Total Total Total Total Lifeform & Non-Veg totals must add up to 100%

8- Tree DBH Size Class Plant Code Cover TS1 TS2 TS3 TS4 TS5 TS6 Tree Size Classes TS1 < 4.5 feet tall TS2 <5" TS3 5 - 9.9" TS4 10 - 19.9" Other TS5 20 - 29.9" Total TS6 ≥30"

PLOT SUMMARY 14- Disturbance: □ Burn □ Harvest □ Other 1st 2nd 15- Notes: 9- Map Group |______|______| 10- Dom Type |______|______| 11- Veg Map Unit|______|______| 12- Cnpy Cover |______|______| 13- Tree Size |______|______|

Tree Cover Classes: Shrub Cover Classes: TC1 10 - 19% SC1 10 - 24% TC2 20 - 29% SC2 25 - 34% TC3 30 - 44% SC3 ≥ 35% TC4 45 - 59% TC5 ≥ 60%

APPENDIX D: Segmentation and Mapping

Image Segmentation - Using eCognition

Introduction

A geographic object-based image analysis (GEOBIA) was performed to produce segments on approximately 5 million acres (2 million hectares) of the Boise and Payette National Forests of Idaho, USA, for the purpose of vegetation classification, mapping and quantitative inventory. Within GEOBIA, image segmentation techniques, the process of generating segments, were used to automatically generate stand-level delineations that eventually formed the perimeters of final vegetation polygons.

Over 18,000 very-high resolution aerial resource images collected in 2008 from Zeiss/Intergraph’s Digital Mapping Camera (DMC), with 1-foot (0.35m) resolution and 80% forward overlap and 20% side overlap, are the base for this mid-level existing vegetation map product.

Mosaic Creation

To produce a better mosaic through the use of most nadir imagery, a novel approach was employed for the base image mosaic generation using the most nadir portion of each resource image to eliminate distortion and object lean within each image. Chipping subsets out of orthorectified imagery into a ‘Sweet spot’ chip was achieved through creating thiessen polygons using flight centroids that were then edited for orthogonal rectangularity. These polygons were then buffered by 10% the shortest distance of the thiessen polygon to provide an overlap area for subsetting the imagery. Then ERDAS' Mosaic Pro weighted seam line generation created intelligent seam lines for a 5.25m resource imagery mosaic product.

ERDAS Imagine’s seam line generation (weighted or most nadir) is ineffective in collecting the best seam line for any one image. However the one good aspect of the weighted generation is the fact that at the micro level the seam line will adhere to a ‘fluid’ line of variance through the image, but at the macro level will fail. Figure 1 – ortho frame 1128 is shown, with its weighted seam line highlighted in yellow. This polygon extends to the edge of the image beyond the most nadir area for the image, leaving undesirable feature lean within the mosaic. Figure 2 – shows the same ortho frame with an ‘idealized’ sweet spot based off of thiessen polygons of the frame center points. The limitation of the use of this polygon for seam lines is that it does poorly at the micro level due to a pure straight line but handles the macro level well by constraining the image in the center for the most nadir viewing.

Appendix D: 1

Figure 1 Figure 2

The proposed solution is to merge the efficiency of the weighted seam line micro level detail with the thiessen polygon’s macro level efficiency. To achieve this, each input image must be subset to the sweet spot chip. Merely cropping the image using mosaic pro will not suffice. This subset of the image to the ‘sweet spot’ location of the image will help constrain the weighted seam line's meanderings away from the edge of images.

Examine and edit thiessen polygon layer of overlap areas to produce rectangular sweet spots. The shift in horizontal placement between two vertical rows will cause irregular shapes, these are to be avoided.

BAD GOOD

vs.

Mosaic Results: The output image chips will look similar to this, as from figures 1 & 2

Bad Good

vs.

Appendix D: 2

These subset images are now refined for a better result from mosaic pro’s weighted seam line generation. (Please note: that for each seam line generation, exporting into ArcGIS is needed to clean the topology from self-intersecting polygons that create holes in the image.)

a. Weighted seam line with entire image used:

b. Weighted seam line with subset image used:

Generating Segments

We used Trimble eCognition and its Cognition Network Language to create a Size-Constrained Region-Splitting Multi-Resolution Segmentation Routine which facilitated the automatic delineation of homogenous objects of interest across the varying forested and non-forested landscape scales. Through eCogntion’s iterative process of rule set design, object primitives became objects of interest through applied expert knowledge utilizing individual DMC bands, DMC band ratios, Zabud, NDVI ratios, and a tri-color illuminated hill shade DEM derivative based from NED elevation dataset, and Canny's edge layers. Initially objects were created to

Appendix D: 3

approximate a known USDA Forest Service product called TEUI (Terrestrial Ecological Unit Inventory), which creates large objects that mimic landscape and land unit scales at nominally the 5th level hydrographic unit through the use of large scale parameter multi-resolution segmentation. These TEUI-like objects maintain criteria for objects not crossing valley bottoms nor ridgelines, maintain homogeneity within segment, maintain heterogeneity outside of segment, and are within a range for minimum mapping unit size. Smoothing of objects occurred through pixel-based smoothing and open image object morphology for a cartographic result in object shapes. Small object removal was based on the minimum domain extrema for neighboring objects based on minimum mean difference of NDVI values below an acreage threshold. An additional smoothing routine for the elimination of persistent 'finger-like' shapes was performed through application of the open image object morphology.

Once these large watershed TEUI-like objects were created, a Parent Process Object routine was employed to find objects above a 5 acre threshold for additional region-splitting. This, large object removal allows for an iterative region-splitting of objects based on acreage and thusly size-constrains the objects. To achieve additional splitting a multi-resolution segmentation region grow was performed at iteratively smaller scale parameters until a desired mean acreage size was obtained. The region-splitting occurred for all polygons which were labeled as upland. Additional polygons that intersected a valley bottom DEM derived product were selected and identified as candidate polygons for additional region splitting at a an even smaller scale parameter for smaller acreage riparian segments that met a smaller MMU.

Using Trimble eCognition’s workspace work-flow, the segmentation product statistically approached the minimum-mapping unit range of two and five acres with a mean segment size of 2.43 acres and a standard deviation of 1.17. These segments are then exported to shapefiles and will be the basis of consistent and continuous mid-level existing vegetation map products based on a regionally adopted vegetation classification system to document the distribution, extent, and landscape patterns of vegetation composition and structural attributes.

The segments were created separately for each Geographic Area (GA). When joining the individual GAs together to form the complete National Forest, segments along the GA boundaries overlapped. These overlapping segments had to be edge-stitched. The overlapping segments in one GA were selected using the GA boundary. This overlap area extended into the adjacent GA and became the new GA boundary. This new boundary was then used to clip the segments from the adjacent GA to form natural segments that did not follow an administrative line.

When the Boise and Payette National Forests had to be edge-stitched together, a similar process was used. All of the segments from the Payette National Forest were selected with the Payette administrative boundary. These segments were used as a boundary file for a clip of the Boise segments that were adjacent to the Payette National Forest's border. A total of 346 Map Unit polygons (2392 acres) from the Payette map are present on the Boise map along the northern Boise NF boundary. These border MU polygons may have a different Map Unit Appendix D: 4

description than identically named MU polygons on the Boise NF map. These Payette polygons are denoted within the "FOREST" field of the BoiseNF_Map_Unit feature class where "PNF" signifies edge-stitched polygons from the Payette National Forest and "BNF" signifies Boise National Forest polygons. lmage Stack Creation - eCognition

An image stack was developed from 58 different geospatial data layers representing topographic, spectral, textural, climatic, and other ancillary information. These base layers were summarized against stand-level delineations (segments) generated earlier and the reference data, to produce an image stack containing 58 different data layers comprised of zonal means or zonal majorities of the original base layers. For each polygon segment and reference data plot, a mean/majority value of all pixel values within the segment boundary was calculated and recoded to the zonal mean value within eCognition. The base data layers that were used as summary layers in the mapping process are shown below.

Base Geospat¡al Data Layers included in the image Cube

Source Season Layer Single Band # Source Season Layer Single Band #

Band 1 1 Band 1 29 Band 2 2 Band 2 30 Band 3 3 Band 3 31 Mosaic Mosaic Band 4 4 Band 4 32 Band 5 5 Band 5 33 Band 6 6 Band 6 34

Spring

NDVI 7 Fall NDVI 35 Landsat

Principal Components 8 Principal Components 36 Band 1 (Brightness) 9 Band 1 (Brightness) 37 Tasseled Cap Tasseled Cap Band 2 (Greenness) 10 Band 2 (Greenness) 38 (sig16bit) (sig16bit) Band 3 (Wetness) 11 Band 3 (Wetness) 39 Band 1 (Brightness) 12 Band 1 (Brightness) 40 Tasseled Cap Tasseled Cap Band 2 (Greenness) 13 Band 2 (Greenness) 41 (un8bit) (un8bit) Band 3 (Wetness) 14 Band 3 (Wetness) 42 Landsat Band 1 15 Aspect 43 Band 2 16 Wetness Index 44 Band 3 17 Curvature 45 Mosaic Band 4 18 Elevation 46 Band 5 19 DEM Angle 120 47 Fully Illuminated

Summer Band 6 20 Angle 240 48 Hillshade NDVI 21 Angle 360 49 Principal Components 22 Heatload 50

Band 1 (Brightness) 23 Slope 51 Tasseled Cap Band 2 (Greenness) 24 Annual Temperature 52 (sig16bit) Band 3 (Wetness) 25 Annual Precipitation 53 Daymet Band 1 (Brightness) 26 Solar Radiation 54 Tasseled Cap Band 2 (Greenness) 27 Growdays 55 (un8bit) Band 3 (Wetness) 28 NBCD Raw_Height 56 Resource Imagery Texture IDP 57 MDAF - 2007 Canopy 58

Appendix D: 5

Grading Reference Data

All reference data (stored in the GIS file Boise_Refdata_Map_Intersect_Summary.shp) were evaluated for inclusion into the modeling process. However, not all of the reference data were of equal quality. The difference in the quality of the reference points was due to several factors including different field crew authorship, a 10% quality check by NF staff, different PI estimates vs. on ground estimates, and observation polygons going through a hand-drawn polygon to centroid point conversion. The reference data were evaluated by Geographic Area (GA1 and GA3) by two different analysts. This process was established to standardize the varying authorships of the reference data prior to use for modeling. Once graded, the 'good' subset of reference data was submitted for use in a separability analysis to determine if all of the planned map units could be modeled statistically. The separability results indicated that the planned map units could be modeled, at that time.

The grading gave a +1 call to any reference plot that exhibited one of the following characteristics:

 Canopy Cover differed from field estimate, indicated different Map Group. +1  Canopy Cover differed from field estimate, indicated different Map Unit.... +1  Not homogeneous - Plot (not necessarily cover) is not homogeneous...... +1  Observation polygon, characteristics don't carry over...... +1  Cover is <30%...... +1  Other (i.e. if the reference data had an unknown (UNK) value or burned)..+1  ------ Summed total, expressed as the field "QC_grade"...... +n

This grading essentially gave all reference sites a 'golf-like' score, where a score of 0 was preferred, and a low summed total score was considered a good reference site. Approximately 20% of plots with the highest scores were dropped from each Map Unit. However, if a Map Unit only had 20 to 30 plots, then all sites were kept. The retained reference sites were used for modeling the Map Units. All of the data, regardless of quality score, were used in the Map Group analysis.

Use of reference sites for modeling is denoted in the field "MODELED" in the reference data GIS layer. A reference site used for Map Group (MG), Map Unit (MU), Canopy Cover (CC) and Tree Size (TS) modeling would have the text "MG MU CC TS" in the "MODELED" field. If a two letter code is not present in the field, then the reference site was not used for modeling that condition.

Appendix D: 6

Using See5 to Model Vegetation

A customized GUI-interface for See5 produced by Photo Science

Once the segments and the reference data are combined into one shapefile, the combined shapefile can then be ingested into the customized See5 interface. Independent variables are chosen from the data stack of zonal statistics. Dependent variables are then set from the classification scheme that will be applied to segments. The resultant classification is attached to the same database file as the shapefile. A boost was applied which tries to fit the reference data into the same class when it is modeled. Also attached is a confidence value showing the highest ranking of the classification for any particular segment.

Appendix D: 7

Multiple Model Runs

Any modeling effort will have a certain amount of omission and commission from existing condition on the ground. With that in mind, between 15 and 19 model runs were done to create Map Group masks. Map units were hierarchically combined into different iterations for each Map Group model run. For example, one run may drop burned/disturbed sites from dependent variables, another may drop the land uses (Agriculture and Developed), another run grouped all Map Units into their Map Groups, two runs used separability data at Map Unit level, and another run used all reference data. Separate model runs were done for burned map group, constraining to fire perimeters and no constraint of fire perimeters.

Example of 14 map group model runs and the hierarchical grouping of map units. N NoVeg VEG NV V V N N V VEG V V NoVeg VEG W SV VEG A W VEG A W VEG A W A D C R S H B NV A W A D C R S H B NV A W DF L P SA RD RH RS B F M M A P F W DE AS DF DFL SA AGR SV A P P P D E E H B S B S L G O H V W DF L P SA RD RH RS B F M M A P F W DE AS DF DFL SA AGR SV A P P P D E E H B S B S L G O H V W D DF P S SA RH RS B F M M A P F W DE AS DFL LP RDE AGR SV A F P P A D E H B S B S L G O H V W D DF P S SA RH RS B F M M A P F W DE AS DFL LP RDE AGR SV A F P P A D E H B S B S L G O H V W D DF P S SA RH RS B F M M A P F W AG DE AS DFL LP RDE B SV A F P P A D E H B S B S L G O H R V W D DF P S SA RH RS B F M M A P F W AG DE AS DFL LP RDE B SV A F P P A D E H B S B S L G O H R V

Once all of the Map Group models were run, the next step was to create Map Group masks. The first mask created was Water, followed by Agriculture/Developed, Burned, Riparian, Herbaceous and Sparse Veg, Shrub, and lastly Tree for a total 7 masks. Each mask was chosen from the total number of Map Group model runs, when there was a positive identification for that mask. This mask was then manually edited. Since the mask was a creation from every positive identification from the entire suite of Map Group model runs, it created many commission calls but very few omission calls. The editing process was to delete commission calls. The masks were done iteratively in order, so that once the first 6 were completed, the remaining segments were all tree segments.

Once each Map Group mask was created and edited, the mask was then modeled in See5 again. This Map Unit model run used only those Map Units reference plots that the MapGroup mask contained. The reference plots were taken from the separability subset. A similar set of models was used to create canopy cover and tree size products.

Appendix D: 8

After the models for Map Unit, Canopy Cover and Tree Size were run for the Boise National Forest, the map layers were compared with the adjacent Payette National Forest map layers for edge-matching. Since the Boise NF and the Payette NF were mapped simultaneously using similar techniques, the results were very similar. Burned areas in the Boise matched with burned areas in the Payette and the Boise NF Tree Map Group matched with the Payette NF Tree Map Group. Therefore, it was determined that both maps were already sufficiently edge- matched due to similar modeling techniques so that no additional hand-editing was required.

Appendix D: 9

APPENDIX E: Draft Map Review & Revision

Boise National Forest Draft Map Review

The draft map review process was organized with Boise National Forest district personnel. While meeting with district forest staff members, decisions were made concerning vegetation calls in the draft maps in regions they are most familiar with. All district members were provided with a series of 3x3 quad maps covering their individual forest districts. Each district was able to provide comments following each review.

The following instructions were provided to the Forest:

Background: This review will focus on the draft vegetation type maps only. Meetings scheduled at the RO and the Ranger District offices are planned to solicit feedback from knowledgeable staff members who can evaluate the maps and help us improve the depiction of existing vegetation on the final maps that will be delivered to the forests in December 2011. Map revisions will be based almost entirely on the information provided from the review process.

A total of 16 maps were produced at a scale of 1:42,000, with nine USGS quadrangles represented on each plot. An index of the plots is shown in the following graphic. Some quadrangles are shown on multiple maps to insure complete coverage of the forest. Each lettered map (A through P) has 2 plots depicting the same area: one map shows the vegetation types and a corresponding map shows aerial imagery for the same area. The imagery is provided to allow reviewers to easily compare the draft maps against the imagery.

For the review, provide as much information about the draft map as possible. This includes feedback on what is correct and what is incorrect. Please focus your attention on the general vegetation patterns and distribution of vegetation types. You are requested to write notes, circle areas of concern, and document any other information on the maps and the fill in the review form provided. A digital version of the form as an Excel spreadsheet is also provided. It is preferred that the comments be entered into the spreadsheet with the corresponding comment numbers marked on the draft map.

You must follow the “Payette and Boise Vegetation Keys” when determining the vegetation type map unit. This ensures that everyone is assigning types based on the same rules and descriptions. There are also four burn classes: burned shrubland, burned grassland, burned forbland, and burned sparsely vegetated. To classify an area as burned it must have been

Appendix E: 1 forested and now has standing dead trees, and there must not be 10% canopy cover of live trees. If this is the case, and at least 2/3 of the standing trees in the site are burned, then one of the burned labels can be given to the site.

Review Process: For the review, provide as much information about the draft map as possible. You have been provided with both digital and hardcopy draft maps. Either form of review is acceptable… Overall, it is important to focus your attention on the general vegetation patterns and distribution of vegetation types. We need information on what is correct and what is incorrect. Please remember this is a mid-level map (1:100,000 scale) and not a site map. The minimum size of an area that will be depicted on the final map is 5 acres for upland and 2 acres for riparian and aspen stands. This is not project level mapping; fine scaled vegetation patches or stands will not be represented on the final map. However, the current draft maps at this point have not been filtered and do show patches below the minimum mapping unit. These will be aggregated in the final map.

For either the hard copy or digital map review you must follow the “Payette and Boise Vegetation Keys” when determining the vegetation type map unit. This ensures that everyone is assigning types based on the same rules and descriptions. There are also four burn classes: burned shrubland, burned grassland, burned forbland, and burned sparsely vegetated. To classify an area as burned it must have been forested and now has standing dead trees, and there must not be 10% canopy cover of live trees. If this is the case, and at least 2/3 of the standing trees in the site are burned, then one of the burned labels can be given to the site.

In general, the draft map review process includes the following phases:  Review the entire district you work on. Focus on general vegetation distribution and patterns and determine if the overall community types that you see are represented.  Next focus on specific areas that you are most familiar with. These include areas that you have done more detailed project work on or localized studies.  If necessary follow up with field visits to areas that are confused and correct labels cannot be easily determined.

Hardcopy paper draft map review procedures: Write notes, circle areas of concern, and document any other information on the hardcopy maps and the fill in the review form provided. Enter the map letter identified from the upper right corner of the map and the quad name on the form. Label each area marked on the map with a unique ID (number, letter, or combination) that corresponds to the comments entered on the form. It is also important to include your name on the form to allow the mapping specialists to follow up with any questions and/or further discussion. A digital version of the form as an Excel spreadsheet is also provided.

Digital draft map review procedures: Digital versions of the draft map are available through ArcServer and webmap. It is important to review the general distribution and extent of vegetation patterns at a scale that corresponds to the midlevel mapping scale, e.g. 1:50,000 to 1:100,000.

Appendix E: 2

Use of Comments from BNF staff: A final table of review edits was compiled from all of the ranger districts. These draft map review comments were stored as a feature class, giving exact spatial location to comments. The following are the review comments from district offices on the Boise National Forest.

Review Comment Comment Forest Solution PSI Response ID Source Rush skeletonweed more extensive Expand polygon and 1 MMiller Edit made. and dominant. label WH. Label WH. May be Edit made. Examined 2 EAlexander Rush skeletonweed dominant. more extensive than adjacent areas. drawn polygon.

Cheatgrass and rush skeletonweed 3 MMiller Label WH. Edit made. are dominant along highway corridor

Should be rush Rush skeletonweed more extensive 4 MMiller skeletonweed - label Edit made. and dominant. WH Should be rush 5 MMiller Heavy rush skeletonweed infestation skeletonweed - label Edit made. WH 6 MMiller Poa bulbosa, sweet clover Label WH. Edit made. 7 MMiller Poa bulbosa, noxious weeds Label WH. Edit made. 8 MMiller Poa bulbosa, noxious weeds Label WH. Edit made. Should be rush 9 MMiller Rush skeletonweed infestation skeletonweed - label Edit made. WH Should be rush 10 MMiller Rush skeletonweed infestation skeletonweed - label Edit made. WH Should be rush 11 MMiller Rush skeletonweed infestation skeletonweed - label Edit made. WH 12 MMiller Weeds are dominant. Label WH Edit made. 13 TRuffing not WH but forb (Wyethia) not MB and BB: seeded into 14 TRuffing increase agriculture Edit made. grassland has pockets of No riparian deciduous 15 TRuffing coded riparian shrub cottonwood along this MU. Remained in section riparian shrub. map team can use 16 LKennedy see photos # 4 attached photo to decide what No photo provided. cover type is 17 CLancaster OK 18 CLancaster not WH seeded into grassland Edit made. 19 CLancaster OK

Appendix E: 3

No grand fir anywhere on Mt Home Some areas should be GF map unit 20 LKennedy District; grand fir generally not south PP; some should be remodeled without GF of the Payette River DFP training sites.

No subalpine fir in the area of the Mt 21 LKennedy Home District in GA1. Too dry for Should be DF Edit made. subalpine even at higher elevations.

In GA1 and GA3, 3b No lodgepole pine in the area of Mt should be DF. The rest Home District in GA1 and where of the District in this designated in GA3 (3b). Too dry for map area should be PP 22 LKennedy LP. Upon further examination to the (or in some cases may Edit made. east--concluded that there should be get swallowed up by no lodgepole pine on Mt Home adjacent DF). Example District in this map area area Dunnigan Creek (marked on map). No Engelmann spruce in the area of 23 LKennedy Mt Home District in GA1. Too dry for Should be DF Edit made. ES

In GA1: Too much aspen; looks like being confused with mountain shrub. Better mix would be 25% AS and 75% Reduce the amount of 24 LKennedy MS. AS tends to be on more convex AS and increase the Edit made. areas with higher moisture than MS. amount of MS AS could be a different spectral color in the fall than MS. In GA3, only A

Too much aspen. Most aspen should be in the drainage bottoms. Midslope or ridgetop areas mapped Reduce the amount of as AS should be MS. Example pointed 25 LKennedy AS and increase the Edit made. out on map. In most cases adjacent amount of MS mapped areas might help (e.g. a riparian map group versus a shrub map group) Looks good where mapped (in 26 LKennedy No fix Cottonwood Creek ironically) Should be enough tree cover (e.g. greater than 10%) to show up in the 27 LKennedy PP or FS or BFS Edit made. conifer group? Should connect to the PP MU on the eastern edge. Within the fire area, WH (skeletonweed and cheatgrass) area Increase WH and 28 LKennedy Edit made. should be greater and BB area should shrink BB and MB be smaller

Appendix E: 4

In general, the relationship between conifer map group and shrub map 29 LKennedy group looks good. The main concern is the relationship within the shrub map group described in comment 4

Should not be any LP in Grape 30 LKennedy Should be PP or MS Edit made. Mountain Area Should not be any LP in Twin Springs 31 LKennedy area south of the Middle Fork Boise Should be PP or BFS Edit made. River DF/SAF underrepresented; most area Should show about SAD not final MU. 32 LKennedy showing as DF 50% more as SAD Edited to SA Most should be DF or 33 LKennedy PP is over represented in this area Edit made. DFP Plantations are mostly PP or LP/PP mix; need to use plantation layer 34 LKennedy Plantations show up as DF Edit made. to get plantation boundaries in this area. Corporate layer Areas shown should be 35 LKennedy Aspen is under represented Edit made. mainly AS DF should be LP or SA 36 LKennedy DF is over represented with minor amounts of Edit made. DF DF should be LP or SA 37 LKennedy DF is over represented with minor amounts of Edit made. DF Should be a mix of LP 38 LKennedy SA is over represented Edit made. and SA WB is either majority 39 LKennedy WB to show within the Trinity area Edit made. or a mix with SA or LP Aspen is under represented should Show boundary as a 40 LKennedy Edit made. show up as a mix mix of aspen/conifer 41 LKennedy Shows up as riparian and is aspen show as aspen Edit made. Shows up as DF, but was planted to PP in 1963, but parts burned in 2008 42 LKennedy Show as PP Edit made. as part of the South Barker Fire. Replanted in 2008 to PP. This shows as DF, but there riparian 43 LKennedy Show as ES Edit made. area has large ES. This shows as sagebrush, but was a DF/PP stand. This area burned in Show as PP or burned 44 LKennedy Edit made. 2008 as part of the South Barker Fire. forest shrub. Replanted to PP in 2010.

Appendix E: 5

Shows as DF, but know that area is 45 LKennedy change to PP Edit made. PP.

The vegetation type shown is DF/DFP 46 LKennedy mix, but know the area is large PP change to PP Edit made. with second growth PP underneath.

The PP shown within this area is mainly LP. There is a mix of LP/PP 47 LKennedy change to LP Edit made. within adjacent area, but this is a strong LP stand. This area has been burnt within the past 10 years and the aspen has 48 LKennedy change to AS Edit made. expanded its range and it should show over the forest shrub. This ponderosa pine within this area 49 LKennedy doesn't exist and it should change to Change to DF Edit made. DF

I believe that there is very little LP found up on Bennett Mtn. BLM said they have 2 small patches (less than 2 Change LP to DF and 50 LKennedy acres) and that he has not seen any Edit made. PP to DF PP either. Most is private and don't know specifics of veg except in conversation with BLM.

Rush skeletonweed more extensive Should be rush 51 MMiller Edit made. and dominant. skeletonweed Rush skeletonweed more extensive Should be rush 52 MMiller Edit made. and dominant. skeletonweed Rush skeletonweed more extensive Should be rush 53 MMiller Edit made. and dominant. skeletonweed Rush skeletonweed more extensive Should be rush 54 MMiller Edit made. and dominant. skeletonweed Rush skeletonweed more extensive Should be rush 55 MMiller Edit made. and dominant. skeletonweed Rush skeletonweed more extensive Should be rush 56 MMiller Edit made. and dominant. skeletonweed

It seems that there might be a problem with Whitebark Pine as it Believe it is seems to be overrepresented. 57 AS predominately Edit made. Suspect it might be from lumping the Subalpine in this area. Subalpine fir Whitebark Map Unit into the Whitebark Pine Map Unit.

Appendix E: 6

We have no known Western Larch Remodeled without 58 AS sites on the district. Western Larch as a MU We have no known Grand Fir sites on Remodeled without 59 AS the district. Grand Fir as a MU 60 AS Seems to be good 61 AS Seems Good 62 As Seems Good 63 AS All Seems Good Consider changing to 64 AS Reclaimed Mine Site. Edit made. Developed. Remodeled without 65 M.Dimmett WL mapped but not present. Label as DFL? Western Larch as a MU ES mapped but not the dominant 66 M.Dimmett Should be LP. Edit made. conifer. ES mapped but not the dominant 67 M.Dimmett Should be LP. Edit made. conifer. Mapped as GF. 69 M.Dimmett This WL is labeled correctly. No GF should be GFP. Edit made. change needed. 70 M.Dimmett Mapped as GF. Should be GFP. Edit made. 71 M.Dimmett Mapped as GF. Should be GFP. Edit made. The three polygons 72 M.Dimmett Mapped as several MUs. (pink highlighter) are Edit made. field identified WB. ES mapped as the dominant conifer Many of these but is actually limited to a narrow polygons are actually 73 M.Dimmett Edit made. corridor (i.e. polygons appear too dominated by LP, but large, too abundant). some are also DFL. The single polygon 74 N.Lange Mapped as several MUs. (pink highlighter) is Edit made. field identified WL. The single polygon 75 N.Lange Mapped as several MUs. (pink highlighter) is Edit made. field identified WL. Request Many polygons incorrectly mapped as 76 N.Lange harvest/plantation Edit made. WL, AS, or ES. layers.

Many polygons incorrectly mapped as 77 N.Lange Should be RHE. Edit made. ES but canopy cover less than 10%.

WL mapped but not the dominant Plot data indicates it Remodeled without 78 M.Leis conifer. should be GFP. Western Larch as a MU DF mapped but not the dominant Plot data indicates it 79 M.Leis Edit made. conifer. should be GFP.

Appendix E: 7

ES mapped but not the dominant Plot data indicates 80 M.Leis Edit made. conifer. thick WL. Polygons are dominated by DF. No Remodeled without 81 M.Leis Mapped as WL and LP. WL present, some LP Western Larch as a MU present, field verified. Only two plantation areas that are large enough to be western larch map unit. All modeled WL map unit 82 CW Remodel all WL Edit made. should be something else. None of the modeled areas match the known locations

Only a few known locations large enough to be grand fir and grand fir- ponderosa pine map unit. All Remodel all the GF and Remodeled without GF 83 PW modeled GF and GFP except those GFP and GFP MUs identified in comment 5 should be something else. 84 PW Should be GFP or GF Burn in as GFP or GF Edit made. 85 PW Should be GF Burn in as GF Edit made. Leave as GF and GFP that are large enough Modeled GF and GFP referenced in Remodeled without GF 86 PW (1 polygon). All the comment 2 and GFP MUs other GF and GFP to absorb Too much MB--verified by Nadine Hergenrider that there is no Should probably be BB sagebrush in the areas being and MS on southerly classified as MB. However, there is 87 PW, CW, NH aspects (drier) and FS Edit made. bitterbrush. She speculates that the on northerly (wetter) MB is being confused with the BB and aspects the shrub communities on the slopes are probably a c VanZile/Ros Too much MB, MB appears to be Probably should be 88 Edit made. eberry inaccurate call classified as MS This area should be classified as WH, VanZile/Ros Too much GR, GR appears to be majority of the 89 Edit made. eberry inaccurate call vegetation along this river corridor is Skeletonweed Should look closely at the elevation and These areas should be classified as VanZile/Ros photography for these 90 Alpine??? These area are high Alpine not a final MU. eberry areas. I looked with elevation meadow areas. Nate at these locations and it was pretty easy Appendix E: 8

to tell what was happening.

Wrong classification, there should be an elevation mask to These areas should be classified as prevent MB from Edit made to MB in this 91 VanZile Alpine, FO, or FS??? occurring at high area. elevations. This area might need an alpine classification type??? Is it possible to add a slope mask? Typical situation is that RSH would occur on slopes Appears to be an inaccurate call, less than 15% in slope. 92 VanZile Edit made. should be classified as FS??? Looks like that the classification is selecting areas with high forb density in FS areas. Unsure, might need WL is wrong call. While scattered some photo individual trees may or could be interpretation to 93 VanZile found in these areas i.e. 1 tree on 500 Edit made. determine next most acres, there should not be enough to likely classification is trigger classification. GF. Where the ES is Across these maps, it appears that ES classified ES does exist. is classified on more acreage than I Might be the way the would expect. Tends to be on the key works which 95 VanZile PVG 6 areas. I'm thinking that many Remodeled ES. selects for Spruce first of these areas are GF. The spruce on then Grand Fir. Unsure the Westside of the district tends to of corrective action or be restricted to RCA's and rare if one is even needed. Unsure, might need Typically WL is restricted to the areas some photo N of High Valley and W of Middlefork interpretation to Remodeled without 96 VanZile Payette River drainage a specific area determine next most WL as a MU of concern is identified under likely classification is comment #6. GF.

Appendix E: 9

This is more of an observation than a concern. Across these maps, it appears that AS is classified on more Should be plenty of acreage than I would expect. point data and AS examined. Edits 97 VanZile However, this might be highly reference information made. accurate due to the Aspen can be to verify classification. either tree form or shrub form. Shrub form is very co

After taking a hard look one common concern Roseberry and I have is the Edit made to MB in this 98 VanZile accuracy of map group shrublands. Verify with point data. area. Specific area have been identified under comment #1 & #4.

Kathy should be able to give valuable This is a general observation. The information on mid elevation appears to be fairly distribution of MU's by accurate with map group conifers. 99 VanZile elevation extent so Edit made. MG and MU of concern appear to be they could possible use those at low elevation and at high elevation as a filter or elevation locations. mask to assist in the classification process. ES mapped but not the dominant 100 M.Dimmett Should be LP. Edit made. conifer.

Appendix E: 10

Example of the hardcopy draft maps supplied to the Forest.

Appendix E: 11

Final Map Production

Three final map products were produced for delivery to the FS: 1) map groups and map units; 2) canopy cover for trees and shrubs; and 3) tree size. For the MG-MU map, segments were first dissolved to eliminate adjacent polygons having exactly the same labels. To achieve the MMU of 2 acres for riparian and aspen and 5 acres for all other MUs, segments below the MMU were merged based on a set of rules developed by the FS and project staff. These rules are included starting on the following page. For the CC and TS final maps, segments were only dissolved and no merging occurred. Therefore, polygons are present in the CC and TS GIS layers that are smaller than the MMU.

Appendix E: 12

Merge Rules for Segments less than MMU Size 11/1/2011

This document shows the hierarchies of vegetation map units used in the minimum map unit (MMU) merging process.

Forest = WL, WB, DF, DFL, DFP, ES, GF, GFP, LP, PP, SA, AS Shrubland = FS, MS, MB, BB, LS Herbland = GR, FO, WE Riparian = RSH, RHE Burned = BHE, BFS, BSV Nonveg = AGR, DEV (no minimum size, no filtering) SparseVeg = SV Water = WA (no minimum size, no filtering)

Forest DF-Doug-fir DFL – Doug-fir/lodgepole 1. DFP 1. DF 2. DFL 2. DFP 3. PP 3. GF 4. GF 4. GFP 5. GFP 5. Forest 6. Forest 6. Shrubland 7. Shrubland 7. Herbland 8. Herbland 8. Riparian 9. Riparian 9. Burned 10. Burned 10. SV 11. SV 11. Nonveg 12. Nonveg

DFP – Doug-fir/ponderosa pine ES – Engelmann spruce 1. DF 1. SA 2. DFL 2. GF 3. GFP 3. LP 4. GF 4. Forest 5. Forest 5. RSH 6. Shrubland 6. RHE 7. Herbland 7. Shrubland 8. Riparian 8. Herbland 9. Burned 9. Riparian 10. SV 10. Burned 11. Nonveg 11. SV 12. Nonveg

Appendix E: 13

GF – grand fir GFP – grand fir/ponderosa pine 1. GFP 1. GF 2. DF 2. DFP 3. DFP 3. DF 4. DFL 4. DFL 5. ES 5. Forest 6. Forest 6. Shrubland 7. Shrubland 7. Herbland 8. Herbland 8. Riparian 9. Riparian 9. Burned 10. Burned 10. SV 11. SV 11. Nonveg 12. Nonveg

LP – lodgepole pine PP –ponderosa pine 1. SA 1. DFP 2. DFL 2. GFP 3. ES 3. DF 4. GF 4. GF 5. Forest 5. DFL 6. Shrubland 6. Forest 7. Herbland 7. Shrubland 8. Riparian 8. Herbland 9. Burned 9. Riparian 10. SV 10. Burned 11. Nonveg 11. SV 12. Nonveg

SA – subalpine fir 1. ES WB - Whitebark 2. LP 1. SA 3. DFL 2. ES 4. GF 3. Forest 5. Forest 4. Shrubland 6. Shrubland 5. Herbland 7. Herbland 6. Riparian 8. Riparian 7. Burned 9. Burned 8. SV 10. SV 9. Nonveg 11. Nonveg

Appendix E: 14

WL – western larch AS – aspen < 2 acres 1. GF 1. MS 2. DF 2. FS 3. GFP 3. DF 4. DFP 4. DFL 5. Forest 5. GF 6. Shrubland 6. Forest 7. Herbland 7. Shrubland 8. Riparian 8. Herbland 9. Burned 9. Riparian 10. SV 10. Burned 11. Nonveg 11. SV 12. Nonveg Shrubland: FS – Forest Shrub MS - Mountain Shrub 1. MS 1. MB 2. BB 2. BB 3. MB 3. FS 4. Shrubland 4. Shrubland 5. AS 5. Herbland 6. Herbland 6. SV 7. Forest 7. Forest 8. Riparian 8. Riparian 9. SV 9. Burned 10. Burned 10. Nonveg 11. Nonveg

BB - Bitterbrush LS - Low Sagebrush 1. MB 1. MB 2. MS 2. BB 3. FS 3. MS 4. Shrubland 4. Shrubland 5. Herbland 5. Herbland 6. SV 6. SV 7. Forest 7. Forest 8. Riparian 8. Riparian 9. Burned 9. Burned 10. Nonveg 10. Nonveg

MB - Mtn Big Sagebrush 1. MS 2. BB 3. FS 4. LS 5. Shrubland 6. Herbland 7. SV 8. Forest 9. Riparian 10. Burned 11. Nonveg Appendix E: 15

Herbaceous Map Group: FO - Forbland 1. GR GR - Grassland 2. WH 1. FO 3. RHE 2. WH 4. LS 3. RHE 5. MB 4. LS 6. Shrubland 5. MB 7. SV 6. Shrubland 8. Burned 7. SV 9. Forest 8. Burned 10. RSH 9. Forest 11. Nonveg 10. RSH 11. Nonveg

Riparian: RSH – Riparian Shrubland < 2 acres RHE – Riparian Herb < 2 acres 1. RHE 1. RSH 2. FS 2. GR 3. MS 3. FO 4. AS 4. WH 5. Shrubland 5. Shrubland 6. Forest 6. SV 7. Herbland 7. Forest 8. SV 8. Burned 9. Burned 9. Nonveg 10. Nonveg

Burned Map Group: BFS – Burned Forest Shrub BSV – Burned Sparse Veg 1. BHE 1. BHE 2. BSV 2. SV 3. SV 3. BFS 4. FS 4. WH 5. Shrubland 5. Herbland 6. Herbland 6. Shrubland 7. Forest 7. Forest 8. Riparian 8. Riparian 9. Nonveg 9. Nonveg

BHE – Burned Herbaceous 1. BSV 2. BFS 3. SV 4. WH 5. Herbland 6. Shrubland 7. Forest 8. Riparian 9. Nonveg Appendix E: 16

Sparse Veg Map Group: SV – Sparse Veg 1. BSV 2. GR 3. FO 4. Burned 5. Herbland 6. Shrubland 7. Forest 8. Riparian 9. Nonveg

Appendix E: 17

Appendix F: Boise National Forest Accuracy Assessment Report

Prepared for:

USDA Forest Service, Region 4

May 2012

Prepared by: Andrew Brenner and Brad Weigle Photo Science

CONFIDENTIAL NOTICE

The information contained in this report is proprietary and confidential. This report and its contents may not be used, duplicated, communicated, or disclosed, in whole or in part without the express written permission of the United States Department of Agriculture Forest Service

Boise National Forest Accuracy Assessment Report

Appendix F: 2

Table of Contents

1 EXECUTIVE SUMMARY ...... 5

2 BACKGROUND TO ACCURACY ASSESSMENT OF MAPS ...... 6 2.1 DEFINITION OF TERMS FROM WO-67 ...... 6 2.2 UNIT OF ANALYSIS ...... 7 2.3 STRATIFICATION ...... 7 2.4 NUMBER OF POINTS ...... 8 2.5 EXTENT OF ASSESSMENT ...... 8 2.6 NUMBER OF CLASSES ...... 8 2.7 THE CONFUSION MATRIX AND DETERMINISTIC ACCURACY ASSESSMENT ...... 8 2.7.1 Area weighting of the results ...... 9 2.8 FUZZY ACCURACY ASSESSMENT ...... 10 3 TECHNICAL APPROACH ...... 11 3.1 OVERVIEW OF TASKS ...... 11 3.2 TASK 1: QUALITY CONTROL THE ABILITY TO PHOTO-INTERPRET MU ...... 11 3.3 TASK 2: CREATION OF REFERENCE DATABASES ...... 14 3.3.1 Reference datasets ...... 15 3.3.2 Resulting location of plots ...... 18 3.3.3 Summary of plot use in the Boise National Forest ...... 19 3.3.4 Issues with reference data resolved through review ...... 20 3.4 TASK 3: ACCURACY ASSESSMENT OF MAPS ...... 23 3.5 FUZZY ASSESSMENT ...... 24 3.5.1 Site Fuzzy Assessment ...... 24 3.5.2 Class Fuzzy Assessment ...... 24 4 RESULTS FOR BOISE NATIONAL FOREST ...... 26 4.1 COMPARISON OF PREDICTED AREAS ...... 26 4.1.1 Map Group ...... 26 4.1.2 Map Unit ...... 27 4.1.3 Canopy Closure ...... 29 4.1.4 Tree Size ...... 31 4.2 CLASS ACCURACIES ...... 32 4.2.1 Map Group ...... 32 4.2.2 Map Unit ...... 35 4.2.3 Canopy Closure ...... 39 4.2.4 Tree Size ...... 40 5 DISCUSSIONS ...... 42 5.1 DISCUSSION OF ERRORS ...... 42 5.1.1 Misclassification errors ...... 42

Boise National Forest Accuracy Assessment Report

Appendix F: 3

5.1.2 Segment delineation errors ...... 43 5.1.3 Segment merging errors ...... 43 5.2 DISCUSSION OF CLASSIFICATION SYSTEM ...... 43 5.2.1 % canopy threshold of classes ...... 43 5.2.2 Burned vs. unburned ...... 43 5.3 OVERALL ACCURACY ASSESSMENT ...... 45 6 REFERENCES ...... 47

Boise National Forest Accuracy Assessment Report

Appendix F: 4

1 Executive Summary This report details the accuracy assessment of the Boise National Forest Vegetation mapping project. The accuracy assessment followed standard procedures to assess the accuracy of the maps; the three steps involved (1) verifying whether photo-interpretation (PI) could reliably separate all the map units (MU), (2) developing the reference databases and (3) analyzing the data.

The conclusions from the first task were that all MUs could not be Photo Interpreted and so accuracy assessment sites would need to be visited in the field. The second task involved the integration of Forest Inventory and Analysis plots (FIA), B-Grid (an intensification of FIA), CDC (a vegetation inventory) and specific sites visited in 2011 for the purpose of collecting accuracy assessment plots. Each of the databases was quality controlled and non-representative plots were removed. The reference database was compared to their corresponding map segments.

Two main analyses were conducted. For the first analysis, the area for each class on the map was compared with area estimates from FIA. At the map group (MG) level the numbers agreed reasonably well. However, the numbers varied at the MU, CC and TS levels. Low area estimates for the large tree classes was particularly noticeable. The second set of analyses included the creation of confusion matrices that show which classes were confused. These analyses were conducted using three analytical assessment techniques: 1) deterministic, where only primary calls were considered correct; 2) site fuzzy, where acceptable calls were also considered correct; and 3) class fuzzy, where similar classes were considered correct. A summary of the overall accuracies of the maps is shown below. A detailed analysis of class accuracies is contained in the full report.

Boise National Forest Assessment Type Producers Users Map Group Deterministic 80.5% 73.2% Site Fuzzy 82.4% 76.0% Class Fuzzy 83.3% 79.6% Map Unit Deterministic 44.4% 40.5% Site Fuzzy 47.4% 44.8% Class Fuzzy 65.5% 60.5% Canopy Closure Shrubs Deterministic 28.5% 28.6% Class Fuzzy 61.6% 65.6% Trees Deterministic 25.2% 22.5% Class Fuzzy 70.8% 67.8% Tree Size Deterministic 39.5% 34.8% Class Fuzzy 76.9% 70.9%

Boise National Forest Accuracy Assessment Report

Appendix F: 5

2 Background to accuracy assessment of maps

The objective of this report is to assess the accuracy of the Boise National Forest Vegetation Maps. This section provides some background on the fundamentals of accuracy assessment of remotely sensed maps in the context of this project. For a more in depth study, please refer to Congalton and Green (2009).

The purpose of an accuracy assessment is to provide a measure of confidence in a map. This is helpful for the map user and should support the decision-making process. For example, if a mapped class is 50% accurate, the user of the map should know that every time an area of that class appears on the map it will be correct once out of every two times.

There are errors associated with every map, since a map is a simplification of reality. However, most maps do not have an accuracy statement and are often assumed to be correct even when this is not the case. For many decades it has been understood that for land and forest cover maps, the accuracy of the map needs to be assessed to ensure that the errors that exist are both quantified and understood. This is particularly important when making decisions associated with the information provided by the map. Taking, for example, the class that is 50% correct, if of the 50% of the time it is incorrect, it is classified as something similar 40% of the time, then the user will know that 90% of the time it is mapped it will be the correct class or something similar. This provides the user with information that can support data driven decision making.

Over the last 40 years, the approaches to assessing accuracy have been refined so that there are accepted protocols to analyzing accuracies that are now considered standard in the industry. The following paragraphs relate to designing and implementing an accuracy assessment of a land cover or vegetation map.

2.1 Definition of Terms from WO-67 For clarity, a number of key terms are defined in the Forest Service (FS) publication WO-67: Existing Vegetation Classification and Mapping Technical Guide, Version 1.0 (Brohman & Bryant, 2005): • MU: Map Unit Class: A class that contains vegetation that has a collection of features defined and named the same in terms of their vegetation characteristics (USDA Soil Survey Division Staff, 1993). • CC: Canopy Closure Class: A class where the vegetation is consistent in the proportion of the ground usually expressed as a % that is occupied by the perpendicular projection downward of the aerial parts of the vegetation of one or more species. • TS: Tree Diameter Class: Overstory tree diameter class is defined as the plurality of the dbh within a map unit or plot. It is important to note that this class may not coincide with the definition of size class as having greater than a specific % of a size class that would favor the classification of areas into a larger tree size classes.

Boise National Forest Accuracy Assessment Report

Appendix F: 6

2.2 Unit of Analysis There are three units of analysis in this project. 1. The field sites: These data are vegetation inventory observations and measurement on the ground. There are four sources for the field plots used in this analysis. Forest Inventory and Analysis plots (FIA), an intensification of the FIA grid on both forests (B-Grid), CDC upland and lowland plots (CDC), and field plots collected in 2011 for accuracy assessment (2011FS). Each one of these datasets will be discussed in detail later in Section 3. Each of these field plots represents an area on the ground defined by its plot design. In the case of an FIA plot this can be characterized as a 144 ft. radius around the plot center. For the B- Grid plots for the Boise this is also a 144 ft. radius around the plot center. For the CDC plots it has been approximated to a 0.25 acre area centered on the plot center. For the 2011FS plots it is a segment. 2. The segment: This is the basic mapping unit from which the map is created. The segment has a unique value of MUxCCxTS for forested, MUxCC for shrub types, and MU for non- woody types. 3. The Map Unit Feature (MUF): This is a collection of adjacent segments that are grouped together as a polygon and represented with a unique map label in the map. These will appear as map polygons in the final map.

2.3 Stratification Generally, the accuracy of classes in the map and the accuracy of the map as a whole are desired. If the accuracy of the map is of interest and the accuracies of classes are of less interest, either a random selection of points or a grid across the project area is suitable. This approach would result in the number of points per class being directly proportional to the extent of that land cover classes on the landscape resulting in many plots for the common classes and very few for the rarer classes. To address this problem, most assessments include some level of stratification. This stratification usually leads to sampling the rarer classes to a greater extent than they are represented in the map, but provides the user with a better understanding of the accuracy and confusion between classes.

For this project, we would like to achieve both the unbiased random or grid sample across the forest (similar to that from FIA) and a sample so that rarer classes can be measured. The objective is to collect a sufficient number of points for each class that can be used to create an unbiased estimate of the overall accuracy.

In an ideal situation, we would be able to use a grid sampling that was of sufficient intensity that there would be sufficient samples in the rare classes to meet the minimum number of observations required. In reality, the cost of doing this is prohibitive. Although the FIA and B- Grid samples are laid out on a grid and could be used exclusively for the estimate of overall map accuracy, we are not able to use all the FIA and B-Grid sites in the assessments because of questions of whether they are representative of the segment. Using an incomplete grid would in itself introduce biases.

Therefore, to ensure we get an unbiased estimate of the accuracy of the map, we will weight the class accuracies by the extent of that specific class on the map as determined by FIA plots.

Boise National Forest Accuracy Assessment Report

Appendix F: 7

Using this strategy, all the accuracy assessment sites can be used and an unbiased estimate of map accuracy can be calculated. These data are shown in Section 4.

2.4 Number of points The number of accuracy assessment points per class is subject of much discussion and statistical analysis. It ultimately comes down to a balance between budget and accuracy goals. Based on statistical analysis (Congalton and Green, 2009), collecting over 50 points per class provides a statistically significant sample. Increasing the sample size over this amount leads to an increase in precision of the error but the increase may not be justified by the resource costs to gather the information. Also, when the number of points decreases below 30 points per class, confidence in the error statistic starts to decrease, meaning that it is difficult for the user to say what the actual accuracy of that class is. Although the objective of this assessment was to reach a number of 30 sites for each class, this was not achieved for all classes because of the need to 1) visit each site in the field and 2) identify where the rare classes were located based on a draft map.

2.5 Extent of Assessment The objective of the project was to determine accuracy statistics for the Boise National Forest. This means that the accuracy statistics provided in the report are only appropriate at the level of the forest. It is probable that some areas of the forest are mapped more accurately than other areas. If an accuracy assessment on a smaller region is required, the analysis would need to be run only using the reference sites for that smaller area and the statistics recalculated.

2.6 Number of Classes The map for the Boise National Forest was comprised of four attributes: Map Group (MG); Map Unit (MU); Canopy Closure (CC), divided into tree canopy closure (TC) and shrub canopy closure (SC); and Tree Diameter Class (TS). If the goal was to analyze the error of each combination of the three layers, it would have been necessary to collect points for each combination of MUxCCxTS. Since the goal was to analyze the error for each layer independently, it was possible to develop separate analyses for each layer: MG, MU, CC, and TS.

2.7 The Confusion Matrix and Deterministic Accuracy The confusion matrix is the accepted format for presenting information on accuracy assessment of land cover maps. This matrix provides a full analysis of the confusion in the map, i.e. what classes are confused with which other classes and to what degree. The user can then determine whether these errors are a significant problem for the use of the map or not. In order to assess accuracy, the vegetation map has to be compared to a dataset that is considered to be correct, called the reference dataset. This could be a dataset that involves field measurements or estimates, information derived from forest inventory, or information derived from photo-interpretation. The reference dataset often includes some degree of uncertainty. An approach to dealing with uncertainty is described in the Section 2.8. The reference dataset is compared to the map dataset, i.e. the class labels that are in the map that correspond to the

Boise National Forest Accuracy Assessment Report

Appendix F: 8

same locations as the reference dataset. Each accuracy assessment site has a reference label and a map label that can be represented in the confusion matrix (Figure 1). The numbers that fall on the diagonal cells (bold boxes) indicate areas that are correctly mapped; the numbers that fall in the off-diagonal cells indicate confusion in the map. The % accuracy is calculated by dividing the correct calls by the total observations for that class. For each class, there is a users and producers accuracy. The users accuracy indicates errors of commission; this is where a class has been mapped in places where it does not exist. The producers accuracy indicates errors of omission; this is where a class has not been mapped but exists on the ground. Thus, the confusion matrix provides the user of the map a good basis for making a decision and understanding the uncertainty associated with it.

Reference Users % accuracy Class 1 Class 2 Class 3 Class 4 Total Class 1 M Class 2 a Class 3 p Class 4 Total Producers % accuracy Figure 1: Confusion matrix for a deterministic accuracy assessment

This approach to accuracy assessment is generally referred to as deterministic accuracy assessment.

2.7.1 Area weighting of the results Although class accuracies can be determined straight from the confusion matrix, the overall map accuracy depends on the number of observations taken in each class. For example, if 100% of the samples were taken from a class that was well classified, then the overall map accuracy would be high. Vice-versa, if a large number of observations were taken from a class that has low accuracy, then the overall map accuracy would be low. Therefore, the overall map accuracy will depend on the distribution of observations with respect to their class accuracy. An important question is how to correct for this potential bias in the data when the distribution of the accuracy assessment points does not correspond to their extent on the map. Since the overall accuracy of the map should be seen as an assessment of the probability of getting a correct call over the whole map, it would make sense that the more common classes on the landscape get a higher weight in the overall accuracy statistic. In a pure random or grid sampling scheme, the number of samples in any specific class would be in proportion to the extent of that class on the landscape. Since a degree of stratification was required for this project, the number of accuracy assessment sites did not exactly reflect the extent of each class on the map. To correct for this, the accuracy of each class was weighted by the extent of each class and the weighted accuracies were summed to provide the overall map accuracy. These numbers are reflected in the weighted accuracies provided in this report. The actual extent of the area for each class can be estimated two ways for this project: either from FIA data or from the mapped extent. Since

Boise National Forest Accuracy Assessment Report

Appendix F: 9

the FIA dataset is an independent measure of the area of each class, the FIA data were used for this project.

2.8 Fuzzy Accuracy Assessment An assumption of deterministic accuracy assessment is that the reference dataset is 100% correct and that an accuracy assessment point can only have one label. It is known that when two field staff are in the field, they can often label the same point as a different class. Likewise two photo-interpreters do not always label the same stand to the same class. This may not reflect on the staff member but on the fact that we are putting a line in a continuum and there is uncertainty associated with which label can be put on a specific polygon. Irrespective of the detail of the classification key, there will be sites that fall on the border between classes and should be considered realistically as being equally valid in both classes. The reason this type of analysis is important for field decision making is that it will provide a more realistic accuracy of the map than an assessment where it is assumed that vegetation classes are discrete entities that change clearly from one class to the next. If fuzzy assessment is not used, the map user might be less confident in decisions based on the map than the user should be.

The use of fuzzy sets (Gopal and Woodcock, 1994) is recognized as a way of dealing with the uncertainty associated with the reference dataset. This becomes increasingly important as classification systems become more complex, as in this project. The fine dividing lines between classes open up the possibility that exact reference calls cannot always be made, allowing for primary (best guess) and secondary (acceptable) calls to be made.

There are a number of ways to identify fuzzy sets in the reference dataset. Each one of these are used and explained in Section 4. 1) Site basis: For each PI or field point, an assessment is made of the primary and the secondary calls. This is made for each observation. If a point is definitely one type, only the primary call is made. If a point is on the edge of a class, the primary call is made and as many secondary calls as appropriate are made. 2) Class basis: Based on class definitions, there may be a reason that two classes cannot be differentiated and may be so easily confused. For example, classification may depend on soil type or classes are part of a continuum. The confusion may be an acceptable confusion from the management perspective so the fact that two classes are confused may not be considered a problem. 3) Range basis: For continuous variables such as canopy closure and height, it may be more appropriate to allow for measurement accuracy. If observations are made using field or PI points, it is often difficult to estimate the height of a stand and or canopy closure and measurements of individual trees may not characterize the conditions across a whole 20 acre stand. In these cases, there has to be an acceptance of a measurement error around the reference estimate. These errors can be estimated at the beginning of the project and then applied to the estimate; i.e. it is generally accepted that ocular estimates of canopy cover are ± 10% of actual values (Congalton and Green, 2009). Although the 10% rule is often used, it was decided at the beginning of the project that for both canopy closure and tree size, it would be considered acceptable if the mapped class was ± 1 class.

Boise National Forest Accuracy Assessment Report

Appendix F: 10

All of the above approaches to fuzzy accuracy assessment were used in this assessment and are discussed in Section 3.

In addition to the confusion matrix, an accuracy statistic the Kappa statistic can also be calculated. The Kappa statistic is used to determine statistical measure of the agreement, beyond chance, between two maps (i.e. classification and PI or field plots). This corrects for the fact that some observations may be assigned correctly by chance and not because the classifier was correct. The Kappa statistic shows the accuracy of the classification after this chance has been removed. This statistic ranges from 0 where there is no correlation to 1 where there is perfect agreement. For this project, the Kappa statistic was calculated only for the deterministic accuracy assessment.

3 Technical Approach

3.1 Overview of Tasks In the undertaking of this project, we went through a series of tasks to test assumptions and develop the reference data set. These tasks were.

Task 1: Quality control the ability to photo-interpret MU Task 2: Create and QC reference datasets Task 3: Conduct the accuracy assessment of the maps

3.2 Task 1: Quality control the ability to photo-interpret MU

The objective of this task was to assess if the photo interpretation (PI) of areas from the resource imagery compared well with the field collected data. This analysis compared PI calls for all the FIA, B-Grid, and CDC plots with the FS derived labels for these plots. The objective was to identify where there was confusion between PI calls and field calls for MU, CC and TS, although the focus of the analysis was mainly on MU.

An overview of the task is shown in Figure 2.

Boise National Forest Accuracy Assessment Report

Appendix F: 11

CDC-Riparian Rust – Upland FIA plot data B-Grid plot data plot data ( Boise plot data ( Boise (222 – Payette, (462 – Payette, – 238, Payette - 238, Payette - 213 - Boise) 249 - Boise) 200) 90)

Step 1.1

Plot locations Plot locations Plot locations Plot locations provided by USFS provided by USFS provided by USFS provided by USFS

Field plot Creation of (x,y of plot MUxCCxTS label for center) field plots (for location nonwoody only MU Area reference will be recorded) delineation database rules Step 1.1 FIA = 144ft

radius ~1.5 USFS Responsibility acres Boise B-grid Delineate PI areas same 144ft around plot centers radius Field reference Step Others label 1.4 including database Payette B-Grid = 0.25 acres PI for Plot areas Step (Veg) MUxCCxTS 1.2

USFS compare field PI database Provide PI database data labeled plots for plot areas with labels to USFS with PI labeled plot areas Step 1 . 3

USFS Assessment of MU provided Step that can be PIed and plot level those that can not reference 1.5 External inputs USFS Tasks database

Reference Photo Science Tasks Table of MU PI Databases accuracy

Figure 2: Overview of Task 1

Boise National Forest Accuracy Assessment Report

Appendix F: 12

Step 1.1: The areas for comparison between field and PI calls were defined: 1. Buffer FIA plots with 144 ft. around plot center 2. Buffer Boise B-Grid plots with 144 ft. around plot center 3. Buffer CDC with 59 ft. around plot center Step 1.2: The area delineated is PIed in the same way as the training sites were PIed with a MU x CC x TS call. These data were maintained in the database with a link to the site id from the original database from which the plot was extracted. Step1.3: The database was checked for completeness and accuracy by an independent QC analyst running automated and manual checks. Then the database was delivered to the FS for comparison with their assessment. Step 1.4: The field data from each plot was characterized by the FS into map labels that contain MU x CC x TS for forested plots, MU x CC for shrubs, and MU for non-woody vegetation. Step 1.5: The FS developed a confusion matrix by comparing the site database with the PI database. From this matrix, the FS assessed separately MU x CC x TS in three different tables.

The details of the comparison are not presented here because the reference data against which the PI calls were assessed were not all quality controlled to ensure accuracy. Based on subsequent analysis, around 13% of the sites on the Boise were dropped because they were either labeled incorrectly, non-representative, or too close to other sites. However, even with this level of uncertainty, the results of the PI vs. field assessment was not of sufficient accuracy to provide confidence in the MU calls if they were delineated by PI. Thus, it was concluded that additional accuracy assessment sites should be gathered only by field work.

Boise National Forest Accuracy Assessment Report

Appendix F: 13

3.3 Task 2: Creation of Reference Databases

The creation of the four reference databases is described in this section. An overview of the steps is shown in Figure 3.

USFS Use Draft MU provided Estimate number of Summarize number segments to identify Draft MU map plot level sites per MU of sites by MU areas to conduct unit segments reference required 2011 field sampling database

Select segments Create accuracy randomly from available assessment maps Field data collection segments That fulfill Step 2.1 and protocols for July – October 2011 certain accessibility field data collection parameters

Incorporation into Reference QC of data sheets database

Review plot data Review sites against Is reference site considered and imagery with aerial imagery usable No USFS team

Yes Step 2.2

Incorporate into reference Keep/Fuzzy Group decision Drop database drop/fuzzy/keep Drop

Combined segment reference database that has MUxCCxTS for the segment verified against the aerial imagery Step 2.3

Figure 3: Overview of Task 2

Boise National Forest Accuracy Assessment Report

Appendix F: 14

Step 2.1: Labels for the three existing databases were created and attached to each of the reference sites. These sites were then assessed to ensure that they did not fall too close to a training site. A distance of 300 m was chosen as the required separation between training site and accuracy assessment site with the same MU. Sites were then assessed with respect to their distribution of sites within each MU class. The MU classes that had less than 30 sites were identified as requiring additional collection of accuracy assessment sites. The classes were those that were rare in the landscape. Using the draft MU map that was delivered on in July 2011, sites that fell within the selected classes were selected. The site selection process used the following protocols. 1) It was decided that field sites needed to be within ¼ mile of a motorized road or trail to ensure that it did not take too long to get to any single site. 2) Sites were only taken on FS lands and other land ownerships were removed as possible sites. 3) All existing accuracy assessment points were buffered by 300 m to reduce the impact of spatial autocorrelation. In the accuracy assessment each observation is assumed to be independent. Therefore, two samples should not be collected from the same segment or, for this analysis, in the same vegetative MU unless separated by a minimum distance of 300 m. 4) From the available segments for analysis, the sites required for the accuracy assessment were selected using a random selection tool developed by Photo Science. The tool allowed the selection of a specified number of segments randomly from the available pool. More segments were selected than were considered to be needed because some would prove to be unsuitable for sampling. 5) Each site was reviewed against the imagery and was dropped if it met any of the following considerations a. The segment was < 1 acre b. The segment required crossing a water body c. Access to the segment meant going up a steep (> 35%) gradient d. Access was somehow otherwise restricted e. The sites were within 300 m of another accuracy assessment site or signature site 6) The MU labels for the selected sites were removed and a shapefile was provided to the field teams along with a packet of maps and forms to support data collection 7) The field teams collected sites between August and October 2011. The field data sheet was used to collected tree inventory data as well as making calls on the MU, CC, and TS. The field plots were centered on the centroid of the segment so that the plots were representative of the segment. 8) The field data sheets were QCed by the Forest Service and provided to Photo Science after the final maps were delivered.

3.3.1 Reference datasets This section reviews the field based datasets that were collected by the Forest Service prior to the creation of the vegetation map and that needed to be evaluated to determine whether each site was appropriate for use as an accuracy assessment site. In all cases, these datasets should be used with some caveats discussed below. The primary consideration has to be not whether the plot data represents what is on the ground, but whether the plot data represents the vegetated segment with which it is associated. The use of non-representative sites does not

Boise National Forest Accuracy Assessment Report

Appendix F: 15

support an assessment of the accuracy of the map and will reduce the numeric assessment inappropriately. Therefore, sites that do not represent the segment should be excluded from the analysis.

3.3.1.1 FIA Each FIA plot consists of four sub-plots that fall within a 144 ft. diameter circle based on the plot center. Plots are distributed on a fixed grid across the country and the number of plots for each vegetation type should be in proportion to the extent of a vegetation type on the landscape. For more details on FIA, please see http://www.fia.fs.fed.us/. The FIA data have detailed inventory information and photographs that support the labeling of the sites. The data used were collected between 2004 and 2009. In general, only forested FIA plots have field data collected on them. However in the case of the Boise National Forest, there are data on all FIA plot locations that allow labeling the map class, whether forested or non-forested. The issues with using the FIA data are that they can, and often do, cross segment boundaries, making it difficult to associate a specific FIA plot with a specific segment. An example of overlap of an FIA plot with several vegetation segments is shown in Figure 4.

Figure 4: An example of an FIA plot that crosses multiple MU, CC and TS class boundaries.

All plots that were determined to be non-representative or too mixed were dropped. In addition, secondary calls were made on plots where there were two possible labels for a single plot. These secondary calls were included in the site fuzzy accuracy assessment. Decisions regarding plots were made by consensus between Forest Service staff (RO and forest) and Photo Science accuracy assessment staff.

3.3.1.2 B-Grid The B-Grid plots are an intensification of the FIA grid and thus are distributed evenly over the forest landscape. These plots also have a distribution of plots that are in proportion to the extent of MUs on the landscape. The plot size differs depending on the forest. In the Boise National Forest , the plot design was the same as the standard FIA plot (i.e. plots fall within a 144 ft. diameter circle around plot center. The data were collected over forested and non-forested areas and the plots were collected between 2006 and 2008. As with the FIA plots, the B-grid

Boise National Forest Accuracy Assessment Report

Appendix F: 16

inventory data may sometimes be difficult to link to a specific segment since the plot may cross MU class boundaries.

3.3.1.3 CDC The CDC plots were collected for a previous ecological inventory of dominance types on the forests and were divided between two major datasets: Riparian and Upland. The objective of collecting these sites was very different from accuracy assessment and they were collected opportunistically along transects with little or no randomization. Plot areas are very small and have been nominally set as a diameter around plot center of 59 ft. Samples are clustered together and measure patches of vegetation that have a small extent, often significantly smaller than the 2 acre MMU for riparian or aspen and the 5 acre MMU for uplands. In addition to the sampling location, there was sometimes not enough information recorded to be able to label the plots using the MU key or add TS or CC class. However, the biggest problem was that often sites do not represent the map segment due to their small area when compared to the area of the corresponding segment. An example of this is shown in Figure 5 below.

Figure 5: Example of a CDC plot within a segment. The plot falls on a riparian shrub area while the segment as a whole has forest on it. Yellow lines indicate MUF boundaries and white lines show segment boundaries.

Boise National Forest Accuracy Assessment Report

Appendix F: 17

3.3.2 Resulting location of plots

Combining all the reference databases resulted in a total of 1038 accuracy assessment sites for the Boise NF. Not all of these sites had a complete set of attributes associated with them. When the maps were clipped to the final ownership boundaries, seven sites outside the Boise ownership were removed from the analysis.

The locations of the CDC and field plots that were used in the final analysis for the CDC and field plots were distributed as shown in Figures 6a and 6b. The Ranger District boundaries are shown in black. Locations of the FIA and B-Grid plots are not permitted to be shown.

Figure 6a: sites used from CDC plots Figure 6b: sites used from CDC (triangles) and 2011FS (circles)

Overall the distribution of sites across the forests is comprehensive, supporting the reliability of the overall assessment numbers.

Boise National Forest Accuracy Assessment Report

Appendix F: 18

3.3.3 Summary of plot use in the Boise National Forest The table below shows the breakdown of numbers of plots used in the accuracy assessment. The total sites in the FS database column indicates the number of sites that were in the existing database. These sites were filtered so that any sites that were within 300 m of a reference site and had the same MU call were eliminated from the database. This was done to avoid any possibility of having a training and accuracy assessment site come from the same vegetation patch. The "# sites provided by the FS" column represent the number of sites that were provided to Photo Science for analysis. The "# sites in AA database" column indicates the number of sites that were used for the analysis. There is a small discrepancy (plots unaccounted for) between the number of sites provided and the number of sites in AA database, but this is not thought to represent a problem in the accuracy assessment analysis. There were a series of reasons that sites were not used in the analysis that have been discussed previously. The number of sites that were dropped for a reason is noted and the breakdown of these sites and the reason from dropping them are detailed in the table below. The final column indicates the total number of sites used for the analysis.

# sites dropped Total from sites in # sites # sites in Plots analysis Data FS provided by AA unaccounted for # sites used Source database FS database for reason* for analysis FIA 213 204 203 1 1 202 B-Grid 249 238 238 3 235 CDC 238 197 196 1 29 167 Upland CDC 238 176 175 1 10 165 Lowland 2011 Field 272 272 10 262 plots Total 938 1087 1084 3 53 1031

Reason dropped Not Plot in Too Mixed Changed represent- wrong Outside Plot Type Close < MMU plot Condition ative location Boundary Total 1 1 FIA 1 2 3 B-grid 1 3 4 8 CDC upland CDC 12 3 1 10 3 2 29 lowland 2011 Field 5 5 5 plots Grand Total 1 15 3 2 19 6 7 53

Boise National Forest Accuracy Assessment Report

Appendix F: 19

3.3.4 Issues with reference data resolved through review As discussed in Section 3.3.1, most of the reference data were not collected with the objective of being used for accuracy assessment of a vegetation map. Therefore, all plots were reviewed against the NAIP imagery to remove the most obvious errors. Differences were easier to interpret at the MG level rather than the MU level. Since it was shown that MU differences were difficult to verify using PI within MG, errors of the reference sites were not determined and the inventory data were considered to be representative of the site. These errors may play into the accuracy figures. Many of the differences between the inventory derived label and what was observed from the imagery were reviewed first by Photo Science and then by a team of forest service and contractor staff. Decisions were made whether to drop the plot from the analysis, add a fuzzy call to the site, or accept the site as good. The team generally accepted the inventory derived label wherever there was not good reason to drop it. Some of the issues found are described in the figures below.

3.3.4.1 Burned - Non-burned One of the biggest questions was whether an area was classified as burned. This occurred because it was not just a case of whether there had previously been a fire but whether 1) the area was previously forested, as indicated by standing dead trees 2) the fire was relatively recent indicated by standing dead trees as opposed to fallen dead trees, 3) there were sufficient dead trees to indicate a previous tree canopy and a recent fire, resulting in over 2/3 of the plot covered by standing dead trees. Two examples of this issue are shown in Figure 7.

Figure 7a: The plot label indicates that the Figure 7b: The plot label indicates that the site is SV, site is RHE, however the area is burned. however the area is burned. Whether there are sufficient The riparian area is smaller than the MMU dead trees to justify the burned MG is unknown. The and the mapped MU (in yellow) of BHE is mapped MU (in yellow) of BSV is also acceptable. Action more correct. Action was to drop the plot. was to include a fuzzy label of BSV.

Boise National Forest Accuracy Assessment Report

Appendix F: 20

3.3.4.2 Minimum mapping unit The minimum mapping unit for the map is 2 acres for riparian or aspen and 5 acres for all other classes. For some of the sites, the inventory data were collected on MUs that were smaller than the MMU and thus would not be mapped in the final map. This was particularly true of the riparian MUs. Figure 8 shows some examples of this.

Figure 8a: The plot label indicates that the site Figure 8b: The plot label indicates that the site is RSH, is riparian shrub (RSH), however the area is however the area is forested. Although the riparian area forested. The riparian area is probably very is visible it is very small and the MU (in yellow) of DF is small and the area looks forested. The MU (in probably more correct. Action was to drop the plot. yellow) of ES is probably more correct. Action was to drop the plot.

Boise National Forest Accuracy Assessment Report

Appendix F: 21

3.3.4.3 Plot location issues Another issue was plot location. It was difficult to evaluate this issue except where the inventory data obviously did not match the aerial imagery. In these cases the plot center was often close to the class it matched but was not located within the correct segment. Two examples are shown in Figure 9.

Figure 9a: The plot label indicates that the site is Figure 9b: The plot label indicates that the DF; however the area is non-forested. It is adjacent site is SA; however the area is non-forested. It to a DF segment which is probably what it should be is adjacent to a SA segment which is probably associated with. Action was to drop the plot. what it should be associated with. Action was to drop the plot.

There were other reasons for dropping sites; however, these were the main issues identified.

When reviewing the sites, it became apparent that when a site should be associated with an adjacent segment, the record was linked to that segment and the site location was not moved. As a result, a review of the accuracy assessment data against the map will sometimes have cases where the map label was taken not from the exact location of the plot but from an adjacent segment. The decision to make this association with adjacent segments was reached during review of the plots with the Forest Service staff.

Boise National Forest Accuracy Assessment Report

Appendix F: 22

3.4 Task 3: Accuracy assessment of maps

Once the reference data were reviewed, the reference databases were overlaid on the maps and the class accuracies were calculated for MG, MU, CC and TS.

Two assessments of accuracy were made. The first looked at the extent of each class when compared to estimates made from FIA. The second looked at the accuracy of the classes when compared to the reference datasets. The overall analysis approach followed Figure 10.

Final MU map

FIA estimates of Comparison of FIA and map estimates of extent Final CC map areal extent of area Step 3.1

ANALYSIS 1: Comparison of areas between the map and inventory assessments of areal Final TS map extent of classes

Combined segment Overlay final maps on segment reference reference database database

Step 3.2

ANALYSIS 2: Comparison of MG, MU, CC, and TS between accuracy assessment sites and map

Figure 10: Overview of Task 3

Step 3.1: In this step, the area estimates for each class were made by RO staff using FIA data and provided to Photo Science. The class area estimates were compared to summaries of the area of each class made by summing all polygons with the same class in the forest map. Differences in the % extent were calculated and areas are provided as graphs in this report. Step 3.2: By overlaying the reference dataset and the map, a map label was assigned to each of the reference sites. The agreement and disagreement between these datasets is

Boise National Forest Accuracy Assessment Report

Appendix F: 23

contained within the confusion matrices for the forest. Included within the matrices are the fuzzy assessments. All results are provided in Section 4.

3.5 Fuzzy Assessment As discussed in Section 2, the use of fuzzy assessment is essential to the understanding of the usefulness of the map. It is acknowledged that the reference dataset is not flawless and it is important to take this into account when making an assessment. For this analysis, we have broken fuzzy assessment into two parts: site fuzzy assessment and class fuzzy assessment.

3.5.1 Site Fuzzy Assessment Site fuzzy assessment focusses on the uncertainty associated with an individual site. When reference sites are not collected specifically for accuracy assessment, a site may have a primary label but also have a secondary acceptable label. In many cases, the secondary label is as valid as the primary label. This uncertainty is plot specific and, in the case of this project, was recorded by the person assigning the label to the plot or during group review.

3.5.2 Class Fuzzy Assessment Class fuzzy assessment focuses on where there is either easy confusion between classes or where differences between classes are not significant. This accuracy is applied to all sites within a class. Class fuzzy matrices were determined on size and canopy closure classes at the beginning of the mapping part of the project and MU fuzzy matrices were determined in December 2011. The final MU fuzzy matrix is shown in the table below.

Map Unit Map Unit Name/Species Acceptable MU Call AS Aspen MS, FS MS Mountain shrub MB, FS, AS FS Forest shrub MS, AS, BFS BFS Burned forest shrub FS, BHE, BSV MB Mountain big-sagebrush MS, BB, LS BB Bitterbrush MB, MS LS Low sagebrushes MB

PP PIPO DFP, GFP DF PSME DFP, DFL DFP PSME-PIPO DF, GFP, PP DFL PSME-PICO DF, GF, LP GF ABGR GFP, WL WL LAOC GF, GFP GFP ABGR-PIPO DFP, GF, PP SA ABLA ES, WB ES PIEN SA, LP LP PICO DFL, SA WB PIAL SA, ES

RHE Riparian herb RSH

Boise National Forest Accuracy Assessment Report

Appendix F: 24

RSH Riparian shrub RHE

FO Forb GR, WH GR Grass FO, WH BHE Burned herbaceous GR, FO, BSF, BSV WH Weedy herb GR, FO SV Sparse veg BSV BSV Burned sparse veg SV, BHE, BFS

WA Water

These fuzzy calls were associated with the map label. They are not completely symmetrical and this is reflected in the matrices shown in Section 4.

Boise National Forest Accuracy Assessment Report

Appendix F: 25

4 Results for Boise National Forest The following section presents results from the analyses described in Section 3.3. First, the area estimates between FIA and the map are compared and, secondly, the class accuracies are compared. It should be understood that the FIA areas are also estimates and have errors associated with them.

4.1 Comparison of predicted areas

4.1.1 Map Group

Figure 11: Comparison of % area estimated by FIA and the map for the Boise National Forest by Map Group (MG).

Overall the agreement between the map and FIA was good with respect to total area occupied by different MGs. The biggest differences were between Conifer (5.7% lower) and Shrub (6.9% higher). The % distribution of MG across the map for the Boise National Forest is shown below. These differences will be discussed in the following sections.

% area estimated by % area estimated by the MG Class FIA map Difference (%) D 0.93 1.19 -0.26 C 65.88 60.22 5.66 S 13.75 20.69 -6.94 H 3.38 2.88 0.50 R 1.00 1.99 -0.99 N 5.51 1.84 3.67 B 9.55 11.18 -1.63

Boise National Forest Accuracy Assessment Report

Appendix F: 26

4.1.2 Map Unit

Figure 12: Comparison of % area estimated by FIA and the map for the Boise National Forest by Map Unit (MU). Numbers indicate FIA % estimates – Map % estimates.

Differences were observed in the extent of map units, although many of the differences constituted < 1% of the area. The biggest differences were observed in DF, being 7.8% lower than FIA estimates, and with lower areas on the map for shrubs MS and BFS. These were compensated for with PP (+6.1%) in the conifers and MB (+7.5%) in the shrub category.

A full table of the % areas is shown in the following table.

Boise National Forest Accuracy Assessment Report

Appendix F: 27

% area estimated % area estimated by the MU by FIA map Difference (%) AGR 0.00 0.34 0.34 AS 0.93 1.19 0.26 BB 2.47 1.97 -0.50 BFS 6.67 2.91 -3.76 BHE 2.88 5.22 2.34 BSV 2.86 3.04 0.18 DEV 0.00 0.30 0.30 DF 23.82 16.01 -7.81 DFL 1.49 0.40 -1.09 DFP 1.40 3.97 2.57 ES 2.69 0.59 -2.10 FO 1.50 0.53 -0.97 FS 3.23 6.43 3.20 GF 1.98 0.55 -1.43 GFP 0.51 0.17 -0.34 GR 1.88 1.69 -0.19 LP 5.86 6.76 0.90 MB 2.37 9.86 7.49 MS 5.68 2.44 -3.24 PP 17.92 24.01 6.09 RHE 0.50 0.53 0.03 RSH 0.50 1.46 0.96 SA 9.76 7.45 -2.31 SV 1.63 0.50 -1.13 WA 1.02 0.70 -0.32 WB 0.45 0.23 -0.22 WH 0.00 0.67 0.67 WL 0.00 0.06 0.06

Boise National Forest Accuracy Assessment Report

Appendix F: 28

4.1.3 Canopy Closure

Figure 13: Comparison of % area estimated by FIA and the map for the Boise National Forest by Tree Canopy Closure (TC).

For the tree canopy closure, the map had a lower estimate for TC3 (-10.2%) when compared with FIA. This was reflected in a higher map estimate for TC1 (+9.1%). This breakdown is shown in the table below.

% area estimated by FIA % area estimated by the map Difference (%) TC1 23.14 32.28 9.1 TC2 24.05 22.93 -1.1 TC3 31.46 21.26 -10.2 TC4 17.01 18.80 1.8 TC5 4.34 4.73 0.4

Boise National Forest Accuracy Assessment Report

Appendix F: 29

Figure 14: Comparison of % area estimated by FIA and the map for the Boise National Forest by Shrub Canopy Closure (SC).

For shrub canopy closure, lower map estimates of area were occupied by SC2 (-10.2%) when compared to FIA and were made up by higher estimates in SC3 (+8.7%). These data are also shown in the table below.

% area estimated by FIA % area estimated by the map Difference (%) SC1 45.12 46.60 1.5 SC2 25.24 15.02 -10.2 SC3 29.64 38.38 8.7

Boise National Forest Accuracy Assessment Report

Appendix F: 30

4.1.4 Tree Size

Figure 15: Comparison of % area estimated by FIA and the map for the Boise National Forest by Tree Size (TS).

One of the biggest concerns may be in the tree size class areas for the Boise National Forest. There are higher estimates of area in the map for classes TS2, TS3, and TS4 and substantially lower estimates for TS5 and TS6. TS4 occupies 64.8% of the map; however is only 51.3% of the forest. The data are also shown in the table below.

% area estimated by FIA % area estimated by the map Difference (%) TS1 0 0.10 0.1 TS2 3.19 6.38 3.2 TS3 13.07 21.30 8.2 TS4 51.33 64.76 13.4 TS5 24.53 7.32 -17.2 TS6 7.87 0.15 -7.7

Boise National Forest Accuracy Assessment Report

Appendix F: 31

4.2 Class accuracies In the tables below, the reader is presented with a large amount of information. The table formats are similar to aid the reader. The sections are presented for map group and map unit as three accuracy tables: deterministic, site fuzzy and class fuzzy. The deterministic tables show the accuracy assessment assuming no uncertainty exists in the labels assigned to the reference dataset. For example, a specific site could be acceptably called either of two classes but only one class is assigned. The site fuzzy tables show the accuracy of the dataset if two or more acceptable calls are included as correct in the analysis; this is done on a site by site basis. The class fuzzy tables show when calls associated with one class could also be acceptable when confused with another class. The acceptable confusions are provided in Section 3.5.2.

Within each table, the reader can see the users and producers accuracy. The users accuracy indicates errors of commission; this is where a class has been mapped in places where it does not exist. The producers accuracy indicates errors of omission; this is where a class has not been mapped but exists on the ground.

For individual class accuracies, the weighted value has no meaning. However, for the overall map accuracy, the weighted value provides a way of standardizing the overall map accuracy based on the proportion of the class in the forest rather than the proportion of the class in the accuracy assessment database. The proportion of the area in each class is provided in the Tables in Section 4.1. The weighted users and producers accuracies are the best measure of the likelihood of an area being mapped correctly.

4.2.1 Map Group

4.2.1.1 Deterministic Map Group Reference Dataset Users Grand B C D H N R S Total Correct Weighted Overall B 90 16 0 5 1 5 3 120 75.0% 7.16 C 10 418 4 18 7 20 30 507 82.4% 54.32 D 0 7 2 2 0 0 6 17 11.8% 0.11 H 2 0 0 35 0 2 29 68 51.5% 1.74 N 0 1 0 0 5 3 0 9 55.6% 3.06 Map Dataset R 1 7 2 12 0 74 4 100 74.0% 0.74 S 8 10 0 38 3 11 140 210 66.7% 6.05 Grand Total 111 459 8 110 16 115 212 1031 73.2% Producers Correct 81.1% 91.1% 25.0% 31.8% 31.3% 64.3% 66.0% 764 74.1% Weighted AA 7.74 60.00 0.23 1.08 1.72 0.64 9.08 80.5% Unweighted Kappa = 0.63

The unweighted overall accuracy is 74.1%, while the weighted accuracies for the producers and users are 80.5% and 73.2% respectively. The deterministic accuracy assessment does not include any level of fuzzy assessment either at the site or class level. In this analysis, we can see that the producers class accuracies range between 25% for deciduous and 91% for conifer. Likewise, the users accuracies range between 12% and 82% for deciduous and conifer,

Boise National Forest Accuracy Assessment Report

Appendix F: 32

respectively. However, the deciduous MG, with only 8 reference samples, has too few sites to make a statistically valid determination of accuracy.

Analyzing the biggest confusions between MG for each class, we can draw some general ideas of where and why the confusion occurs.

• Burned classes are confused with o Conifer: 7 out of 10 are in mapped TC1 o Shrub: 7 out of 8 are reference BFS • Conifers are confused with o Burned: 11 out of 16 are in reference TC1 or unknown o Shrub: 4 out of 10 are in reference TC1 • Deciduous confusion occurs with o Conifer 2: out of 4 are in TC1 or unknown • Herbaceous classes are confused with o Conifer: 18 out of 18 are in mapped TC1 o Riparian: 10 out of 12 are in mapped RHE o Shrub: 25 out of 38 are in mapped SC1 • Non-vegetated areas are confused with o Conifer: 6 out of 7 are in mapped TC1 • Riparian confusion occurs with o Conifer: 6 out of 20 are in mapped TC1 o Shrub: 11 out of 11 are mapped S • Shrubs are confused with o Conifer: 18 out of 30 are in mapped TC1 o Herbaceous: 12 out of 30 are in reference SC1

From this analysis, it is possible to see that the greatest source of confusion in the map occurs in the lowest canopy cover classes. In these low cover classes (10-19% cover), most of the classification signals (from imagery and ancillary data layers) are coming from the background (80-90% of a segment) rather than from the tree or shrub canopy and so confusion would be expected to be common. Other areas of confusion occur where there are composite MGs rather than vegetative definitions for classes as in the case of burned and riparian areas. Some of these issues are resolved with the use of class fuzzy rules. This is discussed in more detail in Section 5.

4.2.1.2 Site Fuzzy Assessment Using the acceptable site fuzzy (secondary) calls, there is an improvement in the overall accuracies with the unweighted accuracy at 77.1% vs. 74.1% for unweighted deterministic accuracy. The weighted producers and users accuracies increased to 82.4% and 76.0% respectively. Accuracies increased across all MGs apart from deciduous.

Boise National Forest Accuracy Assessment Report

Appendix F: 33

Map Group Reference Dataset Users Grand B C D H N R S Total Correct Weighted Overall B 92 16 0 3 1 5 3 120 76.7% 7.32 C 7 432 4 16 6 20 22 507 85.2% 56.13 D 0 7 2 2 0 0 6 17 11.8% 0.11 H 1 1 0 36 0 2 28 68 52.9% 1.79 N 0 1 0 0 5 3 0 9 55.6% 3.06 Map Dataset R 0 6 2 8 0 80 4 100 80.0% 0.80 S 8 7 0 34 2 11 148 210 70.5% 6.80 Grand Total 108 470 8 99 14 121 211 1031 76.0% Producers Correct 85.2% 91.9% 25.0% 36.4% 35.7% 66.1% 70.1% 795 77.1% Weighted AA 8.14 60.55 0.23 1.23 1.97 0.66 9.64 82.4% Unweighted Kappa = 0.68

4.2.1.3 Class Fuzzy Assessment Since most of the fuzzy decision rules at the class level are within MGs, there is little change in the overall accuracy with a class fuzzy assessment. The exception to this is between burned and unburned classes.

Map Group Reference Dataset Users Grand B C D H N R S Total Correct Weighted Overall B 97 16 0 1 0 5 1 120 81% 7.72 C 7 432 4 16 6 20 22 507 85% 56.13 D 0 7 7 2 0 0 1 17 41% 0.38 H 1 1 0 36 0 2 28 68 53% 1.79 N 0 1 0 0 5 3 0 9 56% 3.06 Map Dataset R 0 6 2 8 0 80 4 100 80% 0.80 S 8 7 0 34 2 11 148 210 70% 9.69 Grand Total 113 470 13 97 13 121 204 1031 79.6% Producers Correct 86% 92% 54% 37% 38% 66% 73% 805 78.1% Weighted AA 8.20 60.55 0.50 1.25 2.12 0.66 9.98 83.3% Unweighted Kappa = 0.69

The overall accuracy of the MGs is 78.1% for the unweighted overall increasing to 83.3% for the weighted producers accuracy. Much of the confusion occurred where the canopy cover was low and that confusion will remain even after a fuzzy assessment. The accuracy assessment of classes with fewer than 30 observations is generally considered unreliable which in this analysis includes both the D and N classes.

Boise National Forest Accuracy Assessment Report

Appendix F: 34

4.2.2 Map Unit

4.2.2.1 Deterministic The deterministic accuracy assessment of MU is shown in the matrix below

Map Unit Reference Dataset AS BB BFS BHE BSV DF DFL DFP ES FO FS GF GFP GR LP MB MS PP RHE RSH SA SV WA WB WH WL Total % correct Weighted AS 2 0 0 0 0 2 0 0 1 1 3 1 0 1 0 1 2 3 0 0 0 0 0 0 0 0 17 11.8% 0.109412 BB 0 7 0 0 0 0 0 0 0 0 1 0 0 5 0 6 2 1 1 3 0 1 0 0 1 0 28 25.0% 0.6175 BFS 0 0 17 5 0 1 0 0 0 1 2 0 0 0 1 0 1 0 0 1 0 1 0 0 0 0 30 56.7% 3.779667 BHE 0 0 9 19 8 0 0 0 3 2 0 0 0 1 3 0 0 0 1 1 2 0 0 1 0 2 52 36.5% 1.052308 BSV 0 0 7 12 13 0 0 0 0 0 0 0 0 1 1 0 0 0 2 0 2 0 0 0 0 0 38 34.2% 0.978421 DF 1 0 1 2 3 55 5 3 4 0 6 2 0 2 8 1 1 5 0 1 13 1 0 2 0 0 116 47.4% 11.29397 DFL 0 0 0 1 0 4 2 0 2 0 0 1 0 0 1 0 0 0 0 0 2 0 0 0 0 0 13 15.4% 0.020839 DFP 0 0 0 0 0 15 2 7 2 0 1 5 2 0 1 0 1 5 0 0 1 0 0 0 0 0 42 16.7% 0.233333 ES 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 1 1 0 0 0 0 0 5 20.0% 0.538 FO 0 0 0 0 0 0 0 0 0 9 1 0 0 4 0 6 0 0 1 0 0 0 0 0 0 0 21 42.9% 0.642857 FS 0 1 6 1 0 3 1 0 0 3 14 1 0 1 0 8 7 0 1 3 0 0 0 0 0 0 50 28.0% 0.9044 GF 0 0 0 0 0 0 0 0 0 0 0 11 2 0 0 0 0 0 0 0 0 0 0 0 0 1 14 78.6% 1.555714 GFP 0 0 0 0 0 0 0 0 2 0 0 3 0 0 0 0 0 3 0 0 0 0 0 0 0 0 8 0.0% 0 GR 0 2 0 1 1 0 0 0 0 6 1 0 0 2 0 5 4 0 1 0 0 0 0 0 2 0 25 8.0% 0.1504

Map Dataset LP 1 0 0 0 0 7 5 1 4 0 1 2 0 0 21 0 0 0 0 1 8 1 0 0 0 0 52 40.4% 2.366538 MB 0 3 1 0 0 1 0 1 0 7 2 0 0 10 0 50 12 1 0 2 0 2 0 0 4 0 96 52.1% 1.234375 MS 0 1 0 0 0 1 0 0 0 6 3 0 0 1 0 9 13 0 0 1 0 0 0 0 0 0 35 37.1% 2.109714 PP 2 1 2 0 0 25 4 20 9 0 11 4 0 7 2 2 3 74 1 14 1 4 0 0 0 1 187 39.6% 7.091337 RHE 0 0 0 1 0 0 0 0 1 1 0 0 0 9 2 2 0 0 17 5 0 0 0 0 0 0 38 44.7% 0.223684 RSH 2 0 0 0 0 0 0 0 3 1 2 0 0 1 0 0 0 1 18 34 0 0 0 0 0 0 62 54.8% 0.274194 SA 0 0 0 0 1 13 2 0 5 4 1 1 0 3 4 0 0 0 0 2 19 0 0 2 0 0 57 33.3% 3.253333 SV 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 3 100.0% 1.63 WA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 2 0 0 0 5 40.0% 0.408 WB 0 0 0 0 0 0 0 0 0 1 0 0 0 1 1 0 1 0 0 0 3 1 0 1 0 0 9 11.1% 0.05 WH 0 3 0 0 0 0 0 0 0 4 0 0 0 4 0 4 2 0 0 0 0 0 0 0 4 0 21 19.0% 4.16E-05 WL 0 0 0 0 0 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 4 0.0% 0 Total 8 18 43 42 26 127 22 32 38 46 49 33 4 53 46 94 49 93 46 69 53 14 2 6 11 4 1028 40.5% % Correct 25.0% 38.9% 39.5% 45.2% 50.0% 43.3% 9.1% 21.9% 2.6% 19.6% 28.6% 33.3% 0.0% 3.8% 45.7% 53.2% 26.5% 79.6% 37.0% 49.3% 35.8% 21.4% 100.0% 16.7% 36.4% 0.0% 38.6% Weighted % 0.23 0.96 2.64 1.30 1.43 10.32 0.14 0.31 0.07 0.29 0.92 0.66 0.00 0.07 2.68 1.26 1.51 14.26 0.18 0.25 3.50 0.35 1.02 0.08 0.00 0.00 44.4% Unweighted Kappa = 0.34

The overall deterministic unweighted accuracy of the map was 38.6%. The weighted producers and users accuracies were 44.4% and 40.5% respectively. As discussed before, class accuracies with fewer than 30 sites should be treated with caution, in particular the classes AS, GFP, WB, and WL. Water was not a focus of the assessment but has been included for completeness. Certain classes had very low accuracies such as AS, DFL, ES, GFP, GR, and WL and should be considered unreliable as distinct classes. However, it is important to consider the uncertainty of the reference data when dealing with some of these classes.

Many of the major confusions are addressed within the class fuzzy matrix but those that are not are described below. As can be seen, many of the sites that are confused correspond to the low canopy closure sites where the background signal in the analysis

Boise National Forest Accuracy Assessment Report

Appendix F: 35

probably dominates the classification of the site rather than the small amount of canopy (10%) used to define the lower limit of a class.

• BB o Confusion with WH: 2 out of 3 are reference SC1 • DF o Confusion with PP: 9 out of 25 are reference TC 1 o Confusion with SA: 3 out of 13 are reference TC1 • ES o Confusion with PP: 0 out of 9 are reference TC1 • FO o Confusion with MB: 5 out of 7 are mapped SC1 • FS o Confusion with PP: 6 out of 11 are mapped as TC1 • GF o Confusion with DFP: 0 out of 5 are mapped as TC 1 • GR o Confusion with RHE: 9 out of 9 mapped as RHE o Confusion with MB 7: out of 10 mapped as SC1 • LP o Confusion with DF: 1 out of 8 reference TC1 • MB o Confusion with FS: 0 out of 8 reference SC1 o Confusion with MS: 0 out of 9 reference SC1 • RSH o Confusion with PP: 5 out of 14 mapped as TC1 • WH o Confused with MB: 4 out of 4 mapped as SC1

From these results, we can see that most of the confusion between map units that result in a change between map groups results from low canopy closure. However, confusion between MUs within the conifer and shrub MGs is most likely the result of confusion between the spectral signatures.

Boise National Forest Accuracy Assessment Report

Appendix F: 36

4.2.2.2 Site Fuzzy Assessment The site fuzzy assessment increases the overall accuracy by 3 to 4%, with the unweighted overall accuracy at 42.4%. The weighted producers accuracy was 47.4% with a corresponding users accuracy of 44.8%.

Map Unit Reference Dataset AS BB BFS BHE BSV DF DFL DFP ES FO FS GF GFP GR LP MB MS PP RHE RSH SA SV WA WB WH WL Total % correct Weighted AS 2 0 0 0 0 2 0 1 1 2 4 1 0 0 0 1 1 2 0 0 0 0 0 0 0 0 17 11.8% 0.11 BB 0 8 0 0 0 0 0 0 0 0 2 0 0 5 0 5 2 0 1 3 0 1 0 0 1 0 28 28.6% 0.71 BFS 0 0 17 5 0 1 0 0 0 1 2 0 0 0 1 0 1 0 0 1 0 1 0 0 0 0 30 56.7% 3.78 BHE 0 0 5 21 6 1 0 0 3 1 0 0 0 1 3 0 0 1 1 1 5 1 0 1 0 1 52 40.4% 1.16 BSV 0 0 5 14 13 0 0 0 0 0 0 0 0 0 1 0 0 0 2 0 3 0 0 0 0 0 38 34.2% 0.98 DF 1 0 1 2 1 63 5 2 4 0 4 1 0 1 8 1 0 6 0 2 11 1 0 2 0 0 116 54.3% 12.94 DFL 0 0 1 0 0 4 2 0 2 0 0 1 0 0 1 0 0 0 0 0 2 0 0 0 0 0 13 15.4% 0.04 DFP 0 0 0 0 0 15 1 7 2 0 1 5 1 0 2 0 1 4 0 0 3 0 0 0 0 0 42 16.7% 0.23 ES 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 1 1 0 0 0 0 0 5 20.0% 0.54 FO 0 0 0 0 0 1 0 0 0 9 1 0 0 4 0 5 0 0 1 0 0 0 0 0 0 0 21 42.9% 0.64 FS 0 1 6 1 0 3 0 0 0 3 15 1 0 1 0 8 6 1 1 3 0 0 0 0 0 0 50 30.0% 0.97 GF 0 0 0 0 0 0 0 0 0 0 0 13 0 0 0 0 0 0 0 0 0 0 0 0 0 1 14 92.9% 1.84 GFP 0 0 0 0 0 0 0 0 2 0 0 3 0 0 0 0 0 3 0 0 0 0 0 0 0 0 8 0.0% 0.00 GR 0 3 0 0 1 0 0 0 0 5 1 0 0 2 0 5 5 0 1 0 0 0 0 0 2 0 25 8.0% 0.15

Map Dataset LP 1 0 0 1 0 5 2 1 5 0 1 0 1 0 26 0 0 0 0 0 8 1 0 0 0 0 52 50.0% 2.93 MB 0 3 1 0 0 2 0 1 0 8 2 0 0 8 0 53 10 0 0 2 0 2 0 0 4 0 96 55.2% 1.31 MS 0 0 0 0 0 1 0 0 0 5 3 0 0 0 0 10 14 1 0 1 0 0 0 0 0 0 35 40.0% 2.27 PP 2 1 1 0 0 22 1 16 9 2 8 4 1 3 5 2 3 85 1 15 1 4 0 0 0 1 187 45.5% 8.15 RHE 0 0 0 0 0 0 0 0 1 0 0 0 0 7 2 1 0 0 21 6 0 0 0 0 0 0 38 55.3% 0.28 RSH 2 0 0 0 0 0 0 0 3 1 3 0 0 0 0 0 0 0 19 34 0 0 0 0 0 0 62 54.8% 0.27 SA 0 0 0 0 1 12 1 0 4 4 1 1 0 3 6 0 0 0 0 2 20 0 0 2 0 0 57 35.1% 3.42 SV 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 3 100.0% 1.63 WA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 2 0 0 0 5 40.0% 0.41 WB 0 0 0 0 0 0 0 0 0 2 0 0 0 1 1 0 0 0 0 0 3 1 0 1 0 0 9 11.1% 0.05 WH 0 3 0 0 0 0 0 0 0 4 0 0 0 5 0 3 2 0 0 0 0 0 0 0 4 0 21 19.0% 0.00 WL 0 0 0 0 0 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 4 0.0% 0.00 Total 8 19 37 44 22 132 13 28 38 47 48 32 3 41 57 94 45 103 51 71 58 15 2 6 11 3 1028 44.8% % Correct 25.0% 42.1% 45.9% 47.7% 59.1% 47.7% 15.4% 25.0% 2.6% 19.1% 31.3% 40.6% 0.0% 4.9% 45.6% 56.4% 31.1% 82.5% 41.2% 47.9% 34.5% 20.0% 100.0% 16.7% 36.4% 0.0% 42.4% Weighted % 0.23 1.04 3.06 1.37 1.69 11.37 0.23 0.35 0.07 0.29 1.01 0.80 0.00 0.09 2.67 1.34 1.77 14.79 0.21 0.24 3.37 0.33 1.02 0.08 0.00 0.00 47.4% Unweighted Kappa = 0.38

Although the issues seen in the deterministic assessment remain, there is an improvement in accuracy across many of the classes.

Boise National Forest Accuracy Assessment Report

Appendix F: 37

4.2.2.3 Class Fuzzy Assessment In the table below, the class fuzzy matrix assessment was used. The orange shaded cells indicate fuzzy acceptable classes. The class fuzzy correct shows the number of observations that are considered correct based on the fuzzy class matrix. The class fuzzy correct, % correct, and weighted columns indicate what happens when the orange shaded boxes are considered to be correct, with both producers and users accuracies rising to over 60%.

Map Unit Reference Dataset Class Fuzzy

AS BB BFS BHE BSV DF DFL DFP ES FO FS GF GFP GR LP MB MS PP RHE RSH SA SV WA WB WH WL Total % Correct Weighted # Correct % Correct Weighted AS 2 0 0 0 0 2 0 1 1 2 4 1 0 0 0 1 1 2 0 0 0 0 0 0 0 0 17 11.8% 0.11 7 41.2% 0.38 BB 0 8 0 0 0 0 0 0 0 0 2 0 0 5 0 5 2 0 1 3 0 1 0 0 1 0 28 28.6% 0.71 15 53.6% 1.32 BFS 0 0 17 5 0 1 0 0 0 1 2 0 0 0 1 0 1 0 0 1 0 1 0 0 0 0 30 56.7% 3.78 24 80.0% 5.34 BHE 0 0 5 21 6 1 0 0 3 1 0 0 0 1 3 0 0 1 1 1 5 1 0 1 0 1 52 40.4% 1.16 34 65.4% 1.88 BSV 0 0 5 14 13 0 0 0 0 0 0 0 0 0 1 0 0 0 2 0 3 0 0 0 0 0 38 34.2% 0.98 32 84.2% 2.41 DF 1 0 1 2 1 63 5 2 4 0 4 1 0 1 8 1 0 6 0 2 11 1 0 2 0 0 116 54.3% 12.94 70 60.3% 14.37 DFL 0 0 1 0 0 4 2 0 2 0 0 1 0 0 1 0 0 0 0 0 2 0 0 0 0 0 13 15.4% 0.04 8 61.5% 0.92 DFP 0 0 0 0 0 15 1 7 2 0 1 5 1 0 2 0 1 4 0 0 3 0 0 0 0 0 42 16.7% 0.23 26 61.9% 0.87 ES 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 1 1 0 0 0 0 0 5 20.0% 0.54 3 60.0% 1.61 FO 0 0 0 0 0 1 0 0 0 9 1 0 0 4 0 5 0 0 1 0 0 0 0 0 0 0 21 42.9% 0.64 13 61.9% 0.93 FS 0 1 6 1 0 3 0 0 0 3 15 1 0 1 0 8 6 1 1 3 0 0 0 0 0 0 50 30.0% 0.97 27 54.0% 1.74 GF 0 0 0 0 0 0 0 0 0 0 0 13 0 0 0 0 0 0 0 0 0 0 0 0 0 1 14 92.9% 1.84 14 100.0% 1.98 GFP 0 0 0 0 0 0 0 0 2 0 0 3 0 0 0 0 0 3 0 0 0 0 0 0 0 0 8 0.0% 0.00 6 75.0% 0.38 GR 0 3 0 0 1 0 0 0 0 5 1 0 0 2 0 5 5 0 1 0 0 0 0 0 2 0 25 8.0% 0.15 9 36.0% 0.68

Map Dataset LP 1 0 0 1 0 5 2 1 5 0 1 0 1 0 26 0 0 0 0 0 8 1 0 0 0 0 52 50.0% 2.93 36 69.2% 4.06 MB 0 3 1 0 0 2 0 1 0 8 2 0 0 8 0 53 10 0 0 2 0 2 0 0 4 0 96 55.2% 1.31 66 68.8% 1.63 MS 0 0 0 0 0 1 0 0 0 5 3 0 0 0 0 10 14 1 0 1 0 0 0 0 0 0 35 40.0% 2.27 17 48.6% 2.76 PP 2 1 1 0 0 22 1 16 9 2 8 4 1 3 5 2 3 85 1 15 1 4 0 0 0 1 187 45.5% 8.15 102 54.5% 9.77 RHE 0 0 0 0 0 0 0 0 1 0 0 0 0 7 2 1 0 0 21 6 0 0 0 0 0 0 38 55.3% 0.28 27 71.1% 0.36 RSH 2 0 0 0 0 0 0 0 3 1 3 0 0 0 0 0 0 0 19 34 0 0 0 0 0 0 62 54.8% 0.27 53 85.5% 0.43 SA 0 0 0 0 1 12 1 0 4 4 1 1 0 3 6 0 0 0 0 2 20 0 0 2 0 0 57 35.1% 3.42 26 45.6% 4.45 SV 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 3 100.0% 1.63 3 100.0% 1.63 WA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 2 0 0 0 5 40.0% 0.41 2 40.0% 0.41 WB 0 0 0 0 0 0 0 0 0 2 0 0 0 1 1 0 0 0 0 0 3 1 0 1 0 0 9 11.1% 0.05 4 44.4% 0.20 WH 0 3 0 0 0 0 0 0 0 4 0 0 0 5 0 3 2 0 0 0 0 0 0 0 4 0 21 19.0% 0.00 4 19.0% 0.00 WL 0 0 0 0 0 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 4 0.0% 0.00 2 50.0% 0.00 Total 8 19 37 44 22 132 13 28 38 47 48 32 3 41 57 94 45 103 51 71 58 15 2 6 11 3 1028 44.8% 630 61.3% 60.5% % Correct 25.0% 42.1% 45.9% 47.7% 59.1% 47.7% 15.4% 25.0% 2.6% 19.1% 31.3% 40.6% 0.0% 4.9% 45.6% 56.4% 31.1% 82.5% 41.2% 47.9% 34.5% 20.0% 100.0% 16.7% 36.4% 0.0% 42.4% Weighted % 0.23 1.04 3.06 1.37 1.69 11.37 0.23 0.35 0.07 0.29 1.01 0.80 0.00 0.09 2.67 1.34 1.77 14.79 0.21 0.24 3.37 0.33 1.02 0.08 0.00 0.00 47.4% Class Fuzzy Correct 2 11 33 40 19 82 9 25 5 17 24 19 1 8 28 59 32 92 40 40 29 3 2 3 6 1 630 % Correct 25.0% 57.9% 89.2% 90.9% 86.4% 62.1% 69.2% 89.3% 13.2% 36.2% 50.0% 59.4% 33.3% 19.5% 49.1% 62.8% 71.1% 89.3% 78.4% 56.3% 50.0% 20.0% 100.0% 50.0% 54.5% 33.3% 61.3% Weighted % 0.23 1.43 5.95 2.62 2.47 14.80 1.03 1.25 0.35 0.54 1.62 1.18 0.17 0.37 2.88 1.49 4.04 16.01 0.39 0.28 4.88 0.33 1.02 0.23 0.00 0.00 65.5% Unweighted Kappa = 0.59

It should be noted that the confusion between MUs resulting from low canopy closures remains as confusion in this matrix. Overall accuracies are substantially improved giving an overall accuracy of 65.5% for producers and 60.5% for users accuracies. The classes that remain poorly classified (<40% accuracy) at this stage are AS, ES, FO, GR, GFP, SV and WL. The MUs AS, GFP, SV, and WL all have fewer than 30 reference sites.

Boise National Forest Accuracy Assessment Report

Appendix F: 38

4.2.3 Canopy Closure In this analysis, sites were dropped when the site was not representative of the segment. Site fuzzy calls were not made so the fuzzy assessment is based on whether a site call is within 1 canopy closure class of the deterministic call.

Analyzing woody (trees and shrubs) and non-woody cover using a deterministic approach resulted in an overall 82.8% correct classification see table below. The table shows that the woody classes are more accurately mapped (91.4% and 86.7%) than the non-woody (55.8% and 67.3%). This suggests that there is less map confusion when identifying forested areas than when identifying non-forested areas.

Non- Woody woody Total % correct Woody 689 65 754 91.4% Non-woody 106 134 240 55.8% Total 795 199 994 % correct 86.7% 67.3% 82.8%

Canopy Closure Reference Dataset Deterministic Fuzzy Weighted Weighted Fuzzy 1 Weighted Weighted NC SC1 SC2 SC3 TC1 TC2 TC3 TC4 TC5 Total % correct Shrub Tree class fuzzy % Shrub Tree NC 134 21 10 20 8 3 2 1 0 199 67.3% 163 81.9% SC1 37 19 17 41 3 0 2 0 0 119 16.0% 7.2% 73 61.3% 29% SC2 6 4 5 13 0 1 0 0 1 30 16.7% 4.2% 22 73.3% 11% SC3 28 10 14 83 3 3 0 2 0 143 58.0% 17.2% 97 67.8% 26% TC1 31 8 8 11 32 38 22 7 2 159 20.1% 4.7% 101 63.5% 14.70 TC2 3 0 2 10 19 15 38 15 8 110 13.6% 3.3% 72 65.5% 15.74

Map Dataset TC3 0 2 1 5 20 22 30 20 10 110 27.3% 8.6% 72 65.5% 20.59 TC4 0 1 0 5 10 9 25 22 20 92 23.9% 4.1% 67 72.8% 13.68 TC5 1 0 0 0 1 0 9 7 14 32 43.8% 1.9% 21 65.6% 3.10 Total 240 65 57 188 96 91 128 74 55 994 28.6% 22.5% 65.6% 67.8%

% Correct 55.8% 29.2% 8.8% 44.1% 33.3% 16.5% 23.4% 29.7% 25.5%

Weighted Shrub 13.2% 2.2% 13.1% 28.5%

Deterministic Weighted Tree 7.7% 4.0% 7.4% 5.1% 1.1% 25.2%

Fuzzy 1 class 202 44 36 96 59 75 93 49 34 688 % Correct 84.2% 67.7% 63.2% 51.1% 61.5% 82.4% 72.7% 66.2% 61.8% Fuzzy Weighted % shrub 30.54 15.94 15.14 61.6% Weighted % tree 14.22 19.82 22.86 11.26 2.68 70.8% Unweighted Kappa = 0.64

A summary of the weighted and unweighted average accuracies for trees and shrubs, excluding any non-woody plots, is shown in the table below.

Boise National Forest Accuracy Assessment Report

Appendix F: 39

Producers Accuracy Users Accuracy Shrub Tree Shrub Tree Deterministic 34.5% 25.5% 36.6% 22.5% Weighted Deterministic 28.5% 25.2% 28.6% 22.5% Fuzzy 56.8% 69.8% 65.8% 66.2% Weighted Fuzzy 61.6% 70.8% 65.6% 67.8%

As we can see from the table above, weighting does not always increase the overall accuracy of the dataset. In the case of weighted deterministic dataset for shrubs, lower classification accuracy was found in the more common class and, thus, the weighting decreased the overall accuracy. Deterministic accuracies of shrub and tree canopy closures classes were in the order of 25 – 28% and fuzzy accuracies (+/- 1 CC class) were between 61 and 70%. These analyses excluded the non-woody sites.

4.2.4 Tree Size

In this analysis, reference sites were dropped where the sites were not representative of the segment, but site fuzzy calls were not made. The class fuzzy calls indicate ± 1 tree size class. The analysis below includes all sites, both forested and non-forested. The table below that shows the overall accuracies excluded the non-forested sites by calculating only the accuracy of the size class in forested sites.

Tree Size Reference Dataset Deterministic Fuzzy

NS TS2 TS3 TS4 TS5 TS6 Total % correct Weighted Fuzzy % correct Weighted NS 469 5 6 18 2 1 501 93.6% 469 93.6% TS2 23 5 4 14 7 3 56 8.9% 0.3% 9 16.1% 1.0% TS3 23 6 13 42 11 6 101 12.9% 1.7% 61 60.4% 12.9% TS4 37 7 28 167 51 20 310 53.9% 27.7% 246 79.4% 51.4%

Map Dataset TS5 8 0 1 18 8 3 38 21.1% 5.2% 29 76.3% 5.6% TS6 0 0 0 0 0 0 0 0.0% 0.0% 0 0.0% 0.0% Total 560 23 52 259 79 33 1006 34.8% 814 80.9% 70.9%

% Correct 83.8% 21.7% 25.0% 64.5% 10.1% 0.0% 662

Determinsitic Weighted 0.7% 3.3% 33.1% 2.5% 0.0% 39.5% Fuzzy 469 11 45 227 59 3 814 % Correct 83.8% 47.8% 86.5% 87.6% 74.7% 9.1% 80.9%

Fuzzy Weighted 1.53 11.32 44.99 18.32 0.72 76.9% Unweighted Kappa = 0.70

Although the overall accuracy of 70 – 77% indicates a high level of accuracy for tree size class, the accuracy is primarily in the TS3, TS4 and TS5. TS1 did not exist in any of the reference sites. Although TS6 had 33 reference sites, none were mapped since there was almost no TS6 mapped across the entire map.

Boise National Forest Accuracy Assessment Report

Appendix F: 40

Producers Users Deterministic 43.3% 38.2% Weighted Deterministic 39.5% 34.8% Fuzzy 77.4% 68.3% Weighted Fuzzy 76.9% 70.9%

The overall accuracy of the size class differentiation within the forest category is between 34.8 and 43.3%. This means that, on average, around 3.9 out of every 10 observations the size category will be exactly correct. The fuzzy accuracies are between 70.9 and 77.4% correct, which means that, on average around 7.4 times out of every 10 observations, the size category will be within 1 size class of the correct value.

Boise National Forest Accuracy Assessment Report

Appendix F: 41

5 Discussions

5.1 Discussion of errors

5.1.1 Misclassification errors All maps have errors. This is also true of the Boise vegetation map. It is also important to recognize where the errors occur and whether they are important or not to the analysis under consideration. The most obvious errors are those that result from model misclassification. These are found in the map and are especially pronounced when comparing map units within a map group. Some of these errors are acceptable, such as the spectral confusion between DF and DFL or DFP. Other errors are not acceptable but are probably related to the spectral properties of the two species such as confusion between ES and LP or PP. It is important to note that it was not possible to verify the representativeness of sites where there was within-MG confusion. As a result, some of the confusion within MG could result from some of the issues discussed earlier. Misclassification also occurred between GR and FO and between the shrub and herbaceous MGs.

Some of the misclassification occurred as a result of the ability of the sensor to detect a signature at a very low canopy density. Section 4.2.2.1 shows that the largest proportion of the confusion between map groups occurred when the CC fell into TC1 (10 – 19% CC) or SC1 (10 – 24%), i.e. at low canopy closure. The remote sensing based mapping approach used in this project requires that there is a significant signal from the vegetation to map it. When up to 90% of the signal is made up of something other than the type of vegetation that is being mapped, significant error can occur since the classification will primarily be responding to a segment's background (herbaceous, shrub or barren) rather than the species being mapped. If you add to that an error in the estimate of the canopy closure, whether estimated from basal area or from PI, there are significant uncertainties surrounding the classification of those sites that fall into the TC1 and SC1 classes.

This impact can be seen in the table below using the reference data from the Boise National Forest. In this table, all the sites where the reference MG and the map MG did not agree were divided into those sites that fell into the reference NC, SC1 and TC1 classes, and those that fell into all other classes. This was compared to the whole reference dataset where there were CC labels. This table shows that if a site fell into one of the classes that would be most impacted by low canopy closure changing the MG, it was around twice as likely to be incorrect at the MG and therefore, at the MU level.

Site that fell into NC, SC1, All other classes Total TC1 classes # sites where MG 178 77 255 was incorrect All Sites 482 519 1001 % incorrect 36.9% 14.8% 25.5%

Boise National Forest Accuracy Assessment Report

Appendix F: 42

5.1.2 Segment delineation errors Errors were also introduced through the use of the segmentation routines. While the segmentation routines generally provided a good delineation of vegetation types, there were cases where the automated delineation would produce segments that combined a patch of vegetation of one MU within a segment where the majority of the segment was a different MU. When evaluating a site on the ground with respect to whether it should be mapped as a separate polygon or not, the approach taken was that if the area on the ground was larger than the minimum mapping unit it should be mapped with the appropriate class label. There were instances where a reference site that would be of sufficient size to be mapped would be divided across multiple segments, where each part of the site was small relative to the rest of the segment. So, although the mapped segment labels were correct with respect to the majority of the MU, the map label did not match the reference site label and therefore would be considered an error. Therefore, segment delineation on the map in some places did cause some errors in the final map.

5.1.3 Segment merging errors All segments were classified individually, then merged based on common MU labels. MU features (MUFs) that were less than 5 acres (or 2 acres for riparian or aspen) were identified and then merged into adjacent MUFs based on a series of decision rules. These rules were developed so that there was a certain degree of automation of the process and to ensure consistency. In some cases, the merging of segments into MUFs may not have resulted in the best merge. It is also possible that the MU at the segment was accuracy classified but that the merging of the segment resulted in the label changing to another MU.

5.2 Discussion of Classification System

5.2.1 % canopy threshold of classes In general, classification systems for image interpretation both using PI and other remote sensing approaches require there to be a significant signal to differentiate one vegetation type from another. Although the desire of the classification system designed for this project was to provide as much information as possible in differentiating vegetation types for the user, the key created did not play to the strengths of the classification approach. The issues that surround the 10% canopy closure for tree and shrub definition have been discussed above. The other issue is with the term abundant. This means that in a mixed conifer stand, which constitutes much of the forest, the proportion of the species being mapped for could be between 25% - 35% of the relative canopy that could in turn be 10% - 20% of the ground area. This type of scenario is very difficult to classify using any technique apart from field visits.

5.2.2 Burned vs. unburned Large areas of the Boise National Forest have been burned by fires in recent years and their mapping is important. The definition of burned areas was created to ensure that areas which were previously forested were identified as burned, and that the fire did not occur a long time

Boise National Forest Accuracy Assessment Report

Appendix F: 43

ago. Therefore, the extent of standing dead trees was used as a definition of burned. The ability to remotely sense standing dead trees is poor and so the sensors are better at picking up whether an area has been burned or not burned rather than the number of snags. Although the fuzzy rules eliminated some of this confusion, the definition of burned is something worth reviewing before future vegetation classification work plans are finalized.

Boise National Forest Accuracy Assessment Report

Appendix F: 44

5.3 Overall accuracy assessment

A summary of the results are shown below for the Boise National Forest. The table contains the weighted (unbiased) accuracy assessment for each of the analyses for producers (omission errors) and users (commission errors) accuracies.

Boise Assessment Type Producers Users Map Group Deterministic 80.5% 73.2% Site Fuzzy 82.4% 76.0% Class Fuzzy 83.3% 79.6% Map Unit Deterministic 44.4% 40.5% Site Fuzzy 47.4% 44.8% Class Fuzzy 65.5% 60.5% Canopy Closure Shrub Deterministic 28.5% 28.6% Class Fuzzy 61.6% 65.6% Tree Deterministic 25.2% 22.5% Class Fuzzy 70.8% 67.8% Tree Size Deterministic 39.5% 34.8% Class Fuzzy 76.9% 70.9%

The overall accuracies at the deterministic level are low across the analysis levels. This relates to the difficulty in both classifying the vegetation to the classification scheme and the potential uncertainty in labeling the reference segments (sites) to a single class using the data available. This analysis would suggest that expecting a high accuracy for a specific MUxCCxTS combination would be questionable. Specific classes have much higher accuracies than others. Generally, the more common classes are classified with a higher degree of accuracy than the rare classes and mixes are usually confused with the pure types. Some specific species such as ES and AS are poorly classified and these MUs should be used with caution. However, the map is with these MUs is better than having no idea of the extent and location of these types. Review of the accuracy sites often indicated that while the exact location of the reference site was incorrect the mapped MU was present in the adjacent segments. Ultimately, the use of the map will help the user determine the field accuracy of the dataset and its usefulness for forest planning and projects.

When the inherent confusion between classes is taken into account, indicated here by the inclusion of site and class fuzzy assessments, the accuracies of the classes are similar to expectations of other projects of this type, especially given the discussion on the classification system in Section 5.2. At the MG level an accuracy of 80 – 85% would be expected. At the MU

Boise National Forest Accuracy Assessment Report

Appendix F: 45

level, an accuracy of 60 – 65% would be expected using a fuzzy matrix as was done here. The canopy closure accuracy between 60 and 70% is reasonable, although a little lower than expected. The reason for this may be that the estimates were in many cases made directly from the inventory data which generally estimates lower CC than when CC is photo interpreted. The tree size accuracy overall is higher than would normally be expected with this type of technology when including the non-forest agreement. Using just the forested classes analysis, the accuracies of 35 – 40% for the deterministic and 70 – 75% for the fuzzy are as expected.

These average estimates mask the concern that the large tree accuracy is very low (0 – 10%) for the deterministic assessment for classes TS5 and TS6. The rarity of these classes within the training sites may be a partial cause for confusion and very few training sites were available for these classes because they probably exist in the more remote areas.. However, a more important cause is probably the inability for the imagery classification routines to accurately capture variation in height and then convert that information into discrete DBH size classes. Therefore, the classification tended towards the class that is the most common: TS4. We believe that the refinement of tree size classification techniques, especially at the large tree categories, may need a different approach.

Overall, the map represents the landscape well. From the users perspective, the biggest errors are in the rare classes in both MU and TS. The errors are largest where the canopy closure is low and will be the most accurate where there is significant cover of the MU being mapped.

When the class definitions are compiled, they will indicate the expected degree of mixing of the classes. Although the ability to predict the MUxCCxTS of any specific stand of trees may be low, the overall distribution of vegetation over the landscape within these mid-level scale map products will be at an accuracy that will support planning decisions.

Boise National Forest Accuracy Assessment Report

Appendix F: 46

6 References Brohman R.J. & Bryant, L.D. eds, 2005. Existing Vegetation Classification and Mapping Technical Guide, Version 1.0. General Technical Report WO-67. Washington DC: United States Department of Agriculture, Forest Service, Ecosystem Management Coordination Staff. 305 pp.

Congalton, R.G. and Green, K., 2009. Assessing the Accuracy of Remotely Sensed Data— Principles and Practices (Second edition), CRC Press, Taylor & Francis Group, Boca Raton, FL (2009) ISBN 978-1-4200-5512-2 183 pp.

Gopal, S. and C. Woodcock, 1994. Theory and methods for accuracy assessment of thematic maps using fuzzy sets. Photogrammetric Engineering and Remote Sensing, 60(2):181-188.

Boise National Forest Accuracy Assessment Report

Appendix F: 47

Appendix G

Directory to Boise and Payette Vegetation Mapping Data and Reports on the Final Delivery Hard Drive September 2012 All Directories (Folders) are highlighted with BOLD text

SWID_Vegetation_Final Base directory

contains following directories (folders) and this Word document

01_Digital_Maps_PDFs

BoiseNF_MXD_PDF contains the following files:

ArcGIS map document (.MXD) files for entire Boise NF and each of the 5 ranger districts for Map Units, Canopy Cover, and Tree Size. MXDs reference data layers in directories 02_Digital_GEOdb and 05_Supporting_Data. Moving MXD files or associated data to a different structure may break the data links.

Adobe PDF files for maps of entire forest and each of the 5 ranger districts for Map Units, Canopy Cover, and Tree Size.

ArcGIS layer files (.lyr) with symbology for Map Units, Canopy Cover, and Tree Size.

PayetteNF_MXD \ Final_maps contains the following files:

ArcGIS map document (.MXD) files for entire Payette NF and Council-Weiser, Krassel-McCall, and New Meadows ranger districts for Map Units, Canopy Cover, and Tree Size. MXD files reference data layers in this directory and subdirectories so a copy of the final Payette vegetation Geodatabase is located here as well as in 02_Digital_GEOdb. Moving MXD files or associated data to a different structure may break the data links.

02_Digital_GEOdb

VCMQ_2012_Boise_NF_final.gdb : Final Boise NF vegetation file geodatabase

VCMQ_2012_Payette_NF_final.gdb : Final Payette NF vegetation file geodatabase

03_Class_&_Field_Keys

Field forms for Training - Reference data collection and Accuracy Assessment data collection

Protocols for Training - Reference data collection, Accuracy Assessment data collection, and Photo Interpretation data collection

Boise-Payette Vegetation Key from the Regional Office (also in Report Appendix A)

Appendix G: 1

04_Field_Plots_&_Photos

BP_Digital_Field_Data

An ArcGIS personal geodatabase containing all the training-reference sites for both the Boise and Payette NFs (BP_Reference_Data_2011_0419_ALL_GAs.mdb)

Completed_Field_Forms_and_Photos

An Access database that contains all of the training-reference field data for both the Boise and the Payette NF (BP_VMap_Training_Data_20120307.accdb).

Ten additional directories with field data photos, forms, and maps: Two directories for each GA that contain Field Plot Photos (.jpg files) and Scanned Field Forms / Plot Maps (PDF files). PDF files are named based on the Field Plot number where field plots in GA1 have numbers in the 1000 range, plots in GA2 have numbers in the 2000 range, etc. Field plot photo numbers correspond to the field plot numbers with an additional cardinal direction at the end of the plot number signifying the direction in which the camera was pointed.

Training_Data_Analyses_Boise_NF

Shape file of training site points used and not used was intersected with the final MGs and MUs (Boise_Refdata_Map_Intersect_Summary.shp)

Excel spreadsheet showing intersected data and pivot tables of Boise Training sites vs. MUs and MGs along with deterministic accuracy (Boise_trainingdata_intersect_pivot_tables.xlsx)

Training_Data_Analyses_Payette_NF

Shape file of training site points used and not used was intersected with the final MGs and MUs (Payette_Refdata_Map_Intersect_Summary.shp)

Excel spreadsheet showing intersected data and pivot tables of Payette Training sites vs. MUs and MGs along with deterministic accuracy. (Payette_trainingdata_intersect_pivot_tables.xlsx)

05_Supporting_Data

BP_Ancillary_Veg_Data contains directories with vector data used for the project

ClimateData Fire_History Fire_Severity GA2_4_5_bnds Harvest_Boise MTBS Ownership Payette_Standexam_Tree_Data

Appendix G: 2

Roads Stand_Data Streams_and_Lakes.gdb Trails Valley Bottom (VB)

Raster_data contains raster data files used for the project

DEMs and derivatives Landsat imagery mosaics and derivatives for spring, summer, and fall Texture derived from imagery Climate Data Valley Bottom (VB)

06_Accuracy_Assessment_Data

An Access database that contains all of the accuracy assessment field data for both the Boise and the Payette NFs (BP_VegMap_AA_FieldReferenceData2011.accdb).

Boise _AA_Data This directory contains some sensitive FIA data that should not be released to the anyone who has not been granted access to the data by FIA.

Boise_AA_Field_Forms_&_Photos

Boise_Submission_Cover_Sheets Scanned cover sheets (PDF files) for field forms delivered to FS offices

Compromised_Boise_AA_files Boise AA sites (PDF files) and photos (jpg files) that were potentially compromised with field crews. All Boise Accuracy Assessment scanned field form files begin with "B_AA_XXX" where XXX is a numeric like 001. Photo file names are formatted like "B_AA_XXXD_PHOTO" where XXX is the numeric AA site and D is the cardinal direction signifying the direction in which the camera was pointed.

Scanned AA field forms (PDF files) and associated photos (jpg files). All Boise Accuracy Assessment scanned field form files begin with "B_AA_XXX" where XXX is a numeric from like 001. Photo file names are formatted like "B_AA_XXXD_PHOTO" where XXX is the numeric AA site and D is the cardinal direction signifying the direction in which the camera was pointed.

Payette_AA_Data This directory contains some sensitive FIA data that should not be released to the anyone who has not been granted access to the data by FIA.

Appendix G: 3

Payette_AA_Field_Forms_&_Photos

Payette_Submission_Cover_Sheets Scanned cover sheets (PDF files) for field forms delivered to FS offices

Scanned AA field forms (PDF files) and associated photos (jpg files). All Payette Accuracy Assessment scanned field form files begin with "P_AA_XXXX" where XXXX is a numeric like 1001. Photo file names are formatted like "P_AA_XXXXD_PHOTO" where XXXX is the numeric AA site and D is the cardinal direction signifying the direction in which the camera was pointed.

07_Presentations

Boise_PPTs_2011_December

PowerPoint presentations delivered to the Boise NF on December 6, 2011

Payette_PPTs_2011_December

PowerPoint presentations delivered to the Payette NF on December 7, 2011

08_Report

Boise_NF

Word documents and PDF files for Report and Appendices

Payette_NF

Word documents and PDF files for Report and Appendices

Comments

Comments and fixes from draft reports

09_Meeting Notes

Conference call notes

Word documents containing summaries of project conference calls

Final Map Unit decisions

Excel spreadsheets showing final Map Unit decisions

Appendix G: 4