Mount Rainier and North Cascades National Parks Vegetation Inventory Map Product Development

Project ID: P17AC01143

Federal Agency: National Park Service

Partner Institution: Portland State University

Fiscal Year: 2017

Initial Funding: $121,864

Total Funding: $423,566

Project Type: Research

Project Disciplines: Biological

National Park: North Coast and Cascades Network Inventory & Monitoring

Principal Investigator: Nielsen, Eric

Agreement Technical Representative: Coles, Janet

Abstract: A. Background

As part of a national effort to develop baseline information on the status of existing natural resources, the NPS is developing vegetation maps for all the national park units in the North Coast and Cascades Network (NCCN). The NCCN parks include large, diverse, and spectacular wilderness areas containing a wide range of vegetation types, from coastal bogs, through some of highest quality conifer forests remaining in the world, to spectacular alpine habitats.

Since 2008, the Institute for Natural Resources (INR) at PSU has been working with NPS through the Pacific Northwest CESU on highly collaborative projects to develop vegetation data for the large national parks in the NCCN. These multi-year cooperative projects includes vegetation mapping of Lewis and Clark National Historical Park, Mount Rainier National Park (MORA), Olympic National Park (OLYM), and North Cascades National Park (NOCA).

INR has developed a successful mapping strategy that relies on automated image segmentation of remotely sensed data and machine-learning classification algorithms to develop detailed vegetation maps. The mapping algorithms are “trained” with field data collected by NPS crews. The key to vegetation mapping is a determination in the field of the plant association (or plant community) sampled on the ground and a species list to support that determination. The plant association selection is then linked to the spectral signature of the imagery in that location, allowing extrapolation of the relatively limited field data to the entire extent of the mapping area using the classification models.

Challenges achieving target accuracy of the MORA map led to a delay in the original project timeline because the selections of associations made by various field crews over the multiple years of data collection were inconsistent. The inconsistencies made it difficult to generate clear relationships among the computer based models and the ground-collected data. Because each association is assigned to only one map class, the confused relationships led to confusion among the map classes particularly in some of the forested and subalpine classes, resulting in unacceptably low accuracies.

Working together, PSU and NPS took a comprehensive look at all field data collected from MORA, OLYM, and NOCA using the species information collected in the field to either confirm or update the association choices made by the field crews. During this Quality Assurance/Quality Control (QA/QC) process, confusion in the training data was resolved and the best possible data set for input into the models was created. The map classes were then updated to be as similar as possible among these large NCCN parks.

Most of the modeling methods and input datasets were developed for MORA and NOCA under previous CESU agreements. Upon completion of the agreements, modeling methods which evaluated how to best modify the random forests classification algorithm had been developed and refined. The models had been used to develop a draft map for MORA. All the training data had been collected and then carefully reviewed and QA/QC’d for both parks. In this current project, the next steps to complete the mapping project are to: finalize input data sets used in modeling, run models to create draft maps, review and refine the maps based on NPS input, create final maps, and create final project products for delivery to the NPS I&M Vegetation Inventory Program.

Expected products include: vegetation map and inventory products which meet the NPS I&M Vegetation Inventory Program Standards. This includes a polygon layer of the vegetation map at 1:24,000, with per class accuracy of 80% and map classes based on the National Vegetation Classification System. A table summarizing the accuracy assessment of the map, and a written report summarizing all project steps and methods, map class descriptions, and a key to the map classes will be produced for both NOCA and MORA.

B. Objectives

This project will be conducted in multiple phases. This agreement initially funds Phase 1 work, but may be modified to add future phases, subject to the availability of funding and satisfactory progress of project work. Investigators from PSU and NPS staff will collaborate to accomplish the following specific objectives.

1. Phase 1: NOCA Map Development and Refinement (Final Map), NOCA Accuracy Assessment and Map Classification Field Key

NPS and PSU will jointly determine how best to map disturbed areas of the parks. Following that, two stages of mapping will occur. The first stage is to generate single-class confidence-filtered maps. Training data plots that belong to map classes with lower model accuracies will be flagged and evaluated. This stage will shape the final association/map class crosswalk because during the evaluation process, the three layer relationship between the field data, association, and map classes can be peeled apart and reconfigured where needed in order to improve map functionality and/or accuracy. A small group of NPS users will then review draft map class prevalence and distribution. NPS input will be used to modify the mapping model and then the final modeled map for NOCA will be generated. The final map will be generated by manual editing of the final modeled map. Manual editing of the final model map is the last step in creating the final maps.

An accuracy assessment will be conducted on the final NOCA map. Some map classes may need to be modified so that the NOCA map can meet NPS accuracy standards. Typically this involves combining similar map classes that prove too difficult to model separately. Once target accuracy is met, the results of the accuracy assessment will be summarized for the project report and the final key to the map classes can be created. Creating and verifying a field key to the final map classes is the other major objective during Phase 1.

2. Phase 2: NOCA Final Report, MORA Map Development (subject to the availability of funding)

A primary objective of Phase 2 is to create the NOCA final project report. NPS will coordinate the development of the final report by obtaining the required NPS peer review, incorporating review comments in the report and delivering the NOCA product suite to the NPS I&M Vegetation Inventory Program (VIP). The other primary objective is to use the same techniques as described above in Phase 1 for developing the MORA map. The MORA single class confidence filtered map will be generated and reviewed for map class prevalence and distribution by NPS staff. NPS input will be used to modify the mapping model and then the final modeled map for MORA will be generated.

3. Phase 3: MORA Final Map, Accuracy Assessment, and Final Report (subject to the availability of funding)

An accuracy assessment will be conducted of the final MORA map products. Some map classes may need to be modified so that the maps can meet NPS accuracy standards. Typically this involves combining similar map classes that prove too difficult to model separately. The results of the accuracy assessment will be summarized for the project report. Phase 3 will lead to the development of final MORA mapping products, including NPS peer review, incorporating review comments in the final report and preparing the product suite for delivery to the NPS I&M Vegetation Inventory Program.

C. Public Purpose

This project will advance the public understanding about the vegetation resources of MORA and NOCA by providing current maps and data describing vegetation communities. These maps will directly inform the public about natural vegetation and biological diversity represented within the parks. The MORA vegetation map can be used to help develop a visual answer to that question, and to many other questions the public and research community pose regarding the unique vegetation found in these two national parks. The vegetation maps can be displayed in visitor centers and incorporated into interpretive presentations. The vegetation map and final report will be publicly available from a NPS web site, and can be used by a wide range of public entities for a variety of applications including assessing and managing natural resources, conducting research, identifying sensitive or threatened habitats, and creating educational programs. The products will also serve as baseline estimates of current vegetation useful for evaluating changes in these valuable public resources over time