Land Cover Change Analysis

This work was done as partial requirements of a Master of Science in Applied Geomatics, a joint program at Acadia University and the Centre of Geographic Sciences [COGS, NSCC]. The project was sponsored by Clean Annapolis River Project [CARP]. Funding was provided by  Mitacs Accelerate , with contributions from Environment and Climate Change Canada, the Nova Scotia Department of Natural Resources and Renewables, the Canada Nature Fund and Clean Annapolis River Project.

Land cover maps are utilized in almost every industry: science and research, conservation, resource extraction, urban planning, agriculture, health services, and more. They are used for gathering all kinds of useful information about the landscape, such as:

• locating natural resources
• identifying environmental risks and hazards
• to determine the composition of an ecosystem
• to observe weather and seasonal patterns
• measure the frequency and severity of natural disasters
• to assess agricultural or forestry capacity and production

This information can then be used to:

• develop sustainable land-use practices and policies
• understand the impacts of climate change and improve adaptation strategies
• monitor ecosystems to facilitate research and study
• inform emergency preparedness strategies and community resilience

When land cover is mapped over a significant period of time we can determine how and why the landscape is changing.

Understanding the history and change of a landscape improves our ability to plan, manage and mitigate impacts to our communities and ecosystems.

Land cover maps were originally produced by conducting field surveys, measuring and drawing up the features by hand, and creating a final printed map product. As you can imagine, this was time consuming, could result in wide margins of error, and was not created in a format easily accessed, edited or shared with others. Now, we can utilize mapping software known as Geographic Information Systems (GIS) and remotely sensed data to more efficiently produce dynamic digital land cover maps over large areas. These maps can be easily edited, updated, and shared online for improved accessibility and data sharing.

The goal of this project was to map the land cover of the Annapolis Valley over time through air photo interpretation and determine what landscape changes may have impacted historic sand barrens. Said to have encompassed more than 200 km ²  in the past, the  Nova Scotia Department of Lands and Forestry  estimates that less than 3% of the original sand barrens remain. Urbanization, intensive agriculture and fire suppression resulting in dense reforestation are the most likely factors. Understanding the history of the Valley and evolution of the landscape will help us to develop targeted conservation strategies that benefit our communities while protecting a rare and unique Canadian ecosystem for generations to come.

To begin this work, a project area had to be determined to find the right data and potential scope for assessment. The following elements were considered when identifying the project area:

1. Regional Topography and Ecology
2. The Extent and Composition of Sandy Soils
3. Known Sand Barren Sites
4. Availability and Overlap of Air Photo Data
5. Processing Time Constraints

After establishing the project areas and selecting the appropriate data, the imagery was processed and integrated into the mapping software. Land cover was classified, mapped and measured to determine how change was occuring over time. Finally additional landscape data was incorporated to assess what changes were impacting sand barrens and suggest strategies for conservation and further assessment.

The individual air photos were processed into orthomosaics for land cover mapping.

An orthomosaic is a large, detailed image made up of smaller orthophotos -- air photos that have been corrected for any scene distortions and terrain relief to create a top-down view of an area with uniform scale throughout.

Orthomosaicking was done with  Agisoft Metashape Professional  software involving the following steps:

1. All of the air photos were resampled and resized where necessary to create a uniform set of a manageable size. Large files can slow down processing and interpretation.
2. Individual images are checked for visual quality and markings that can confuse the software processing. E.g. Air photo from 1931, pictured (right)
   1. Shadows cast by objects or clouds
   2. Clear surface detail for interpreting land cover
   3. Watermarks and fiducial marks, typical of film photos
3. Fiducial marks, watermarks and borders were manually masked out so that only ground features are processed by the software during alignment -- matching features between overlapping images and stitching them together.

4. The photos are then georeferenced, giving the scene real-world location information. - Ground Control Points were collected from the mapping software then imported and matched to the ground features in each air photo. - Newer cameras have location information built in to the sensors that speed up this process, but this remains a manual process with historic air photos. - Georeferencing is a necessary step to integrate the imagery with the mapping software.

5. The software completes the final steps of merging all of the photo details and geographic information together to produce an orthomosaic that can be imported to the mapping software for land cover mapping.

The land cover mapping was done in  Esri's   ArcGIS Pro .

The first step was to determine a system of classifying the land cover to be mapped. Classification often depends on:

• the amount of surface detail that can be identified from the image
• the type of change/phenomena being assessed
• the level of distinction required between land cover types
• other available data used

Two land classification datasets ( Cycle 1 and Cycle 3 ) produced by the Department of Natural Resources and Renewables' Forestry Division were used as the baseline for mapping land cover in 1992 and 2012, matching the photo year used for interpretation of the data. The land cover is divided into over 30 categories, many of which differentiate between types and compositions of tree stands. This level of classification was not necessary for this assessment, so the following schema was created to include and merge the Forestry categories into 6 distinct classes.

Agricultural areas are relatively easy to make out. Fields are typically rectangular with distinct edges and patterns from planting and harvesting can be seen from above. A subclass of Agriculture is Orchards. This crop type is also easy to differentiate from other ground features and are a staple crop of the Valley.

The Forest class represents areas of dense forest cover.

Wetland includes dry and wet bogs, treed bogs and any wet areas not included in the Water class, which represents open water bodies such as: lakes, ponds, and rivers.

The Urban class includes any buildings and industrial spaces. Subclasses include: Quarries and Roads.

The final Sparse Vegetation class is the best representation of what could be heathlands or sand barrens. It is made up of areas with low-lying, brush-like vegetation.

With all of the mosaics completed, and land cover delineated, the classified areas can be calculated to observe land use during each period and landscape change over time.

The large mosaics, covering a majority of the Project Extent, from 1931, 1977 and 2012 show the general trends of change over time across the Valley.

The Minor Area focuses on ~7,000 hectares of the Valley including the encompassing the communities of Aylesford, Greenwood and Kingston. Chosen mainly because of known existing sand barren sites in the area, but also to maximize the amount of air photo overlap. Interpretation and delineation of the land cover resulted in 5 fully classified maps used to calculate the landscape change over 81 years in roughly 20 year increments.

Although a portion of the 1931 and 1992 imagery does not cover the scene resulting in some No Data values (no classified land cover), the general trends observed throughout the Project Extent, can also be observed within the Minor Area.

When the No Data values are excluded, and the land cover results seen side by side (graphed below), the trends of urbanization and reforestation are clear. While dedicated agricultural space has somewhat decreased over time, the greatest land cover loss has been of the sparse vegetation class.

When viewed by photo year, the rate and extent of change is better understood.

Reforestation and urbanization appear to be the primary factors, each occupying roughly 11% more of the significantly sandy soils (≥70%) in 2012 than in 1955. In areas of greatest sand content (≥ 85%) urbanization increased by 21.5% in that 57 year period whereas forests only increased by roughly 9%. This indicates that the soils most likely to support sand barren ecosystems have been developed or are privately owned.

Involving communities and local stakeholders in sand barren restoration will be key to the success of these efforts.

The final aim of this project was to present the data in a dynamic format that would allow viewers to engage with the results and better understand how and why change was occuring. While the maps presented so far give a good sense of those trends, the beauty of a GIS and online sharing platform, is that data can be visualized in many different ways.

For this reason, a dashboard is used to compile the results in a single map frame and more easily intract with the various layers. In particular, the known sand barren sites.

This project provides important historical information about the landscape of the Annapolis Valley and the significant changes that have occured since the 1930s. It has demonstrated the value of air photo data for evaluating and mapping land cover. There is still more we can learn from the imagery and by utilising the maps for other analysis.

• Land cover classification could be completed to the full extent of the Project where data exists to see if the trends observed in the Minor Area are reflected across the Valley.
• Incorporate additional air photo data to observe change over smaller increments.
  • Air photo data of the valley exist for the 1940s, 60s, 80s, and early 2000s. These could be incorporated to examine the rates of change over time to possibly isolate factors or specific events.
• Incorporating other sand barren data. As understanding and study of sand barrens increases, related data or parameters could be used to better interpret the land change results.
  • Use an ecologically determined threshold of total sand by weight.
  • Determine if historic fire records exist that can be mapped to better examine the relationship of fire suppression and sand barren loss.
  • Locate and map sand barren-related species to determine if general reforestation is detrimental to the ecosystem or if specific species (e.g. Scots Pine, Pinus sylvestris) are inhibiting their growth.
  • Incorporate property boundary information to determine the stakeholders of historic heathlands. Because much of the valley is privately-owned, stakeholder engagement and involvement will be key to successfully protecting sand barrens in the long-term.

The completion of this project would not have been possible without the contributions of many people. Foremost thanks to my academic advisor David Colville (COGS, NSCC), who inspired my continued education and career in geospatial science. Your guidance and encouragement kept this project going through the many unexpected challenges, and I will consider your counsel in all my future work! A warm thanks to my academic advisor Dr. Ian Spooner (Acadia), who was always willing to advise, support, and engage in my process and learning despite his busy schedule. Final, and special thanks to CARP members, Katie McLean, Brittni Scott and Gillian Kerr, and the members of the Sand Barrens Committee for their enthusiasm and support. You made it all possible.

Thank you!