|
|
2005 July — Vol. XIV, No. 5 |
 |
Current Issues |
|
|
EOM July 2005 > Features
Monitoring Canada's Forests: Earth Observation for Sustainable Development of Forests Land Cover
Mike Wulder, Jeff Dechka, Ron Hall, Joan Luther, Andre Beaudoin, Morgan Cranny, and Sarah McDonald
Canada is determined to be a good steward of its more than 400 million hectares of forested land — a renewable resource, contributing $37 billion dollars per year to the country's balance of trade. Ensuring effective resource management requires current and reliable forest information. In support of national and international reporting requirements, the Canadian Forest Service (CFS), in partnership with the Canadian Space Agency (CSA), is using space-based, Earth observation (EO) technologies to monitor the sustainable development of Canada's forests through an initiative called Earth Observation for Sustainable Development of Forests (EOSD). The initiative is producing a land cover map of the forested area of Canada using Landsat satellite data. The short-term goal of EOSD is to complete a land cover map representing circa year 2000 forested area conditions by 2006. Longer-term project goals include not only land cover maps, but maps of forest composition, change over time, and biomass.
The EOSD initiative was implemented in collaboration with the Provinces and Territories. Inputs from EOSD will be an important data source in the National Forest Carbon Accounting Framework and will also be used to enhance Canada's new plot-based National Forest Inventory">. The National Forest Information System will be used to integrate and synthesize applicable data and products. The delivery of EOSD data products and metadata is via SAFORAH. In this article we summarize the EOSD Land Cover program, from conception to planned completion.
Background
The initial goal of EOSD is to produce a national map of the forested land cover in Canada. Accounting for image overlap into non-forest areas, it will map more than 60 percent of the country. EOSD is working closely with provincial and territorial mapping agencies on a phased process to complete mapping the land cover of the forested area of Canada by the end of 2006. To meet this goal, the CFS and provincial and territorial mapping agencies have determined areas already being mapped, areas to be mapped, and areas not being mapped and have developed a plan to ensure complete coverage. CFS and CSA plan to integrate the methods and products produced by EOSD with provincial and territorial Landsat land cover map products. The specific goals of the EOSD project are to:
- Support, with space-based technology, the Government of Canada's priorities and international commitments for monitoring sustainable development of its forests (such as the Canadian Council of Forest Ministers' Criteria and Indicators, the Kyoto Protocol, and the United Nations Framework Convention on Climate Change). Meeting these commitments requires quantifying the composition, distribution, structure, and dynamics of forests over time.
- Use Earth observation datasets for deriving products that can be used both as input to national and global forest inventories and as geospatial information describing forest cover change, forest productivity, and forest carbon stocks.
- Provide international leadership and opportunities for Canadian value-added exports of remote sensing information systems and services for monitoring sustainable development of forests.
- Support Canadian economic development opportunities through technology transfer to the private sector.
Using single scenes of Landsat data to produce land cover information is a common practice. However, it is still relatively uncommon to develop a map covering a large area by combining several Landsat scenes, or even hundreds of them. Natural Resources Canada's Centre for Topographic Information-Sherbrooke (CTI-S) is leading a national consortium to collect and orthorectify Landsat-7 ETM+ imagery of Canada representing circa year 2000 conditions. The CTI-S Ortho7 project — a partnership of federal, provincial, and territorial agencies — will produce a complete set of orthorectified imagery for Canada, freely available to the public via GeoGratis. CTI-S Ortho 7 images are the primary source of data supporting the EOSD project.
The initial phase of EOSD will require the 2000 data to be classified in conjunction with CTI's acquisition and orthorectification schedule. Mapping was initiated shortly after the launch of Landsat-7, necessitating the mapping of images largely on a scene-by-scene basis. Figure 1 illustrates the steps in this classification. As indicated in the figure, masks are generated by thresholding an 8 bit NDVI channel into 4 broad classes (to represent water, non-vegetated, coniferous, and deciduous ground cover). The masks are not intended to be exhaustive but to provide a narrowing of the variability of digital numbers processed by the K-means classifier. To capture image spatial information, indicative of forest structural characteristics, a texture surface is computed from the 15x15 meter spatial resolution panchromatic channel. The "intra-pixel texture" is generated by computing the variance on the panchromatic data within a 3x3 meter window. The variance is resampled using a bilinear algorithm to correspond to the 30 meter spatial resolution of the other multispectral channels input to the classifier. The resampling yields an indication of the structural variability found within pixels of the Landsat optical channels. The 6 optical ETM+ channels and the intra-pixel texture channel are input to the K-means classifier (requesting 241 classes, a movement threshold of 0.1, 12 iterations, and a 50 percent sample of the pixels under the mask). Following the clustering, similar clusters are merged, then labelled as meaningful classes. Some steps in the classification process are completed through batch processing and / or specialized automation of tasks. The planned mapping approach is based on using proven, or new and tested, methods. This approach is intended to ensure delivery over the short production timelines identified.
Figure 1: General steps in the EOSD classification process. Click on image to see enlarged.
It was desired to have a classification legend that was appropriate to the level of information captured by Landsat-7, and to enable integration of satellite, forest inventory, and ground-based information. The level of detail captured by the NFI is greater than that available from Landsat imagery. Therefore, to integrate the Landsat data with the NFI and other provincial, territorial, and national programs, the EOSD developed a legend that was compatible with the cover type level of the NFI class structure. By using the NFI class structure as a base, EOSD is able to standardize classified image products and work in close association with provincial and territorial mapping agencies. The project also developed cross-walks between the EOSD legend and other provincial, national, and international legends.
Timelines, Products, and Delivery
EOSD land cover is scheduled to be completed during 2006 (Figure 2). The CTI-S Ortho7 project is to be completed in 2005, allowing sufficient time for the final orthorectified Landsat images to be classified prior to the EOSD deadline in 2006. The most efficient and cost-effective approach to date has been to partner with agencies engaged in these existing land cover mapping activities in order to avoid duplication of effort, combine financial resources, and technical expertise, and develop legend translation protocols that would yield both a provincial land cover and an EOSD land cover product at the same time.
Figure 2: Staged EOSD implementation; classification status by Worldwide Referencing System tile. Click on image to see enlarged.
EOSD land cover products are based upon the national topographic database (NTDB) national topographic system (NTS) map sheet framework. The EOSD land cover products are available for download on a 1:250,000 NTS map sheet basis (Figure 3). Each map sheet represents an area of approximately 14,850 square kilometers. A total of 610 1:250,000 tiles will be required to cover the forested area of Canada. The products will be available in a Paletted GeoTiff format, with a disabled tiff world file. Final products are re-sampled to a 25 meter spatial resolution. As a single tile may be composed of several images, ESRI shape files are provided to communicate source image information and actual mosaic lines (Figure 4). Federal Geographic Data Committee- compliant metadata will also be provided with the EOSD land cover products. Users of the land cover data may utilize the data as is or use the available data as source information to generate value-added products. As the EOSD land cover products are geolocated, it is possible to integrate them with a variety of other spatial datasets.
To date approximately 128 of the 610 NTS sheets to cover the forested area of Canada are complete and available for download. These materials are made available online as they are completed and the rate at which they are posted is increasing as mapping production activities near completion.
Figure 3: Sample EOSD land cover classification product, representing same area as NTS 1:250,000 map sheet 93F (Nechako River, B.C., Canada). Click on image to see enlarged.
Figure 2: Sample EOSD land cover product illustrating mosaic lines and source imagery (NTS 1:250,000 map sheet 93F, Nechako River, B.C., Canada) Click on image to see enlarged.
The primary vehicle for the delivery of EOSD projects is SAFORAH, an advanced networking infrastructure developed to manage, catalogue, store, and disseminate large volumes of Earth observation data. SAFORAH also enables the sharing of computing resources and facilitates efficient and collaborative research and automation over broadband communications. Registered users may access EOSD products, imagery, and associated metadata through the SAFORAH portal. Multiple items may be ordered concurrently and there is also a facility for user comment and EOSD product feedback. The catalogue system used by SAFORAH provides centralized metadata management and simplified access to large volumes of EOSD images and products. SAFORAH also aids EOSD users in tracking what data is available, where the data are stored, and how to access the data.
Accuracy Assessment
Independent validation is planned to ensure the integrity of EOSD products. Validation of each individual image, or NTS tile, is not possible for a project this large in scope. As a result, a protocol for addressing the accuracy of the national EOSD product based upon a stratified random sample has been proposed and is being tested.
The classification of land cover over large geographic areas using remotely sensed data is increasingly common due to the requirements of national inventory and monitoring programs, scientific modeling, and international environmental treaties. Standard operational protocols for validation of these large area land cover products do not exist. A protocol for the validation of the EOSD product has been designed in accordance with the following criteria:
- a logistically feasible and statistically valid sampling strategy;
- capability to assess the accuracy of the validation data; and
- stable and informative measures of accuracy.
The protocol addresses issues such as sample design and sample size, in order to facilitate reporting of accuracy at a national scale (with inclusion of appropriate confidence intervals). Identifying a source of cost-effective purpose-acquired data is one of the greatest challenges for large area land cover validation. Airborne video offers a potential source of relatively low-cost, context-rich data for validation. The general approach followed is also relevant for applications outside of forested areas. A classification accuracy estimate, with error bars, for each class (and overall) is planned.
Summary
EOSD will contribute to meeting Canada's national and international reporting requirements related to climate change and sustainable forest management by mapping the forested areas of Canada. This task will require the support and concerted effort of many partners to complete a circa year 2000 land cover map during 2006. Cooperation and communication both within and between various levels of government provide an opportunity to share resources and work toward common objectives. In addition to land cover mapping, research efforts are working towards automating procedures as well as methods for change and biomass mapping. Products generated from this project will be an integral component of Canada's new forest measuring and monitoring system and will assist the public and interested organizations in understanding the composition, distribution, and dynamics of Canada's forests. 
About the Authors
Mike Wulder is a research scientist with the Canadian Forest Service based out of the Pacific Forestry Centre in Victoria, British Columbia, and leads the Land Cover component of EOSD. He can be reached at [email protected]. Web: http://www.pfc.forestry.ca/profiles/wulder/
Jeff Dechka is Director of Forest Information at the Pacific Forestry Centre and is also the EOSD Project Manager. He can be reached at [email protected].
Ron Hall is a research scientist with the Canadian Forest Service, based out of the Northern Forestry Centre in Edmonton, Alberta. Ron is the Regional Land Cover Implementation Leader overseeing the mapping of Alberta, Saskatchewan, and Manitoba. He can be reached at [email protected].
Joan Luther is a research scientist with the Canadian Forest Service, based out of the Atlantic Forestry Centre in Corner Brook, Newfoundland and Labrador. Joan is the Regional Land Cover Implementation Leader overseeing the mapping of Newfoundland and Labrador, Nova Scotia, New Brunswick, and Prince Edward Island. She can be reached at [email protected].
Andre Beaudoin is a research scientist with the Canadian Forest Service, based out of the Laurentian Forestry Centre in Quebec City, Quebec. Andre is the Regional Land Cover Implementation Leader overseeing the mapping of Quebec. He can be reached at [email protected].
Morgan Cranny is the EOSD Project Coordinator. She is based out of the Pacific Forestry Centre and can be contacted at [email protected] for any data or project information. Morgan is overseeing the mapping of British Columbia, Yukon Territory, Northwest Territories, Nunuvut, and Ontario.
Sarah McDonald is a remote sensing analyst assisting with addressing EOSD research and implementation issues. She is based out of the Pacific Forestry Centre and can be reached at [email protected].
|
|
©Copyright 2005-2021 by GITC America, Inc. Articles may not be reproduced, in whole or in part, without prior authorization from GITC America, Inc. |