Main article

View the most recent version.

1. Introduction
2. Overview: The settlements project
3. The methodology: research component
4. The methodology: processing the data
5. Initial research results and future directions

1   Introduction

The size, structure and form of settlements over time can have a variety of socio-economic and environmental implications. The expansion of cities, for instance, requires infrastructure spending for new roads, sewers and water supply lines. Potential environmental impacts include loss of wildlife habitat and/or high quality agricultural land, increased air pollution and greenhouse gas emissions, and the contamination of rivers, lakes and aquifers. The type or form of expansion is also significant. Low density expansion consumes more agricultural and forested land and results in more infrastructure costs per capita than high density expansion.

The nature and growth of settlements are of particular importance for policy research related to cities, particularly metropolitan areas. The new methodology introduced in this paper, addresses data gaps and limitations with respect to current measures of Canada's built-up or settled areas. This methodology also aims to produce spatial data sets that are coherent through time and space.

2   Overview: The settlements project

Coherent national analyses of the physical form and growth patterns of settlements and the resulting socio-economic and environmental impacts requires data sets that are comparable over time and space. At the same time, these data sets must track changes in the physical form of settlements as precisely as possible.

The research phase of this project included an in depth analysis of existing settlements-related data sets and a review of the relevant literature. While many useful national data sets on settlements were found, not all were produced at regular intervals and most, owing to their underlying methodology, yielded settlement boundary results that were not comparable over time and space. For example, in the past some boundaries were created using coarse data that did not necessarily reflect the true physical form of settlements; oftentimes this resulted in the inclusion of other land uses, such as agriculture, within the boundaries of settlements.

Datasets are produced for different reasons. For instance, census geographic boundaries at Statistics Canada were not developed to provide precise measures of settled areas or to conduct detailed time series analyses on the physical form of Canada's cities, towns and villages. These 'census geographies' are often designed according to administrative information. As a result, they tend to have limited utility when examining issues such as land use change because the boundaries do not always reflect areas that have been physically altered for settlement purposes.

The new approach presented in this research paper addresses current data gaps and allows the coherent mapping of settlements across the country. ¹  The new methodology takes advantage of more refined census boundaries and satellite-based data sets to help address the challenges noted above.

There is strong policy and research interest for this research project. For example, Infrastructure Canada will use the results for research and analysis activities related to of Canada's infrastructure needs and programs. The methodology also fills an important data gap in Statistics Canada's Land Accounts. It allows the creation of a new land use/land cover change matrix, which is used to track changes in land use and land cover over time.

2.1  Key project goals

The primary goal is to develop a new methodology and produce corresponding data sets that more specifically delineate or map the physical form of settlements across the country and through time. The resulting spatial data sets will reflect more accurately where people live and work. They will also ensure that the amount of unsettled land ²  within the new settlements boundaries kept to minimum and is easily measurable; the existence of these areas can impact calculations related to settlement growth and form.

The methodology described in this study is based on Geographic Information System (GIS) technology. ³  The application of the methodology results in a data set that contains digital boundaries depicting the limits of settlements.

2.2  What are settlements?

Settlements reflect areas where humans live and work. They are tracts or blocks ⁴  of land where humans have altered the physical environment by constructing residential, commercial, industrial, institutional and other installations/buildings. They include cities, towns, villages and other concentrations of human populations that inhabit a given area of the environment. ⁵ 

2.3  The need for a new data set

While some detailed data sets on settlements exist for Canada's largest cities, they are not necessarily comparable with one another and do not provide for a complete national picture of settlements. Extensive research and analysis found that national data sets do exist but have certain limitations.

For instance, the National Topographic Database (NTDB) includes a built-up layer, but it has not been consistently or regularly updated. Using the NTDB to accurately track land use/cover change over time is therefore quite a challenge. As well, various types of satellite data are available, but the time, expertise and financial resources required to develop a national coverage based only on high-resolution data make such an approach unrealistic. Many of these national datasets do not include attribute or other supplementary data useful for research.

The most common data set of national data from which national and regional settlement-type analysis or indicators are derived is Statistics Canada's Census of Population. Although used for planning and other purposes, census geographies were not developed to provide precise measures of settled areas or to conduct detailed time series analyses on the physical form of Canada's cities, towns and villages. That is, these boundaries were often designed according to administrative information. As a result, these boundaries tend to have limited utility when examining issues such as land use change, because the boundaries do not always reflect areas that have been physically altered for settlement purposes. ⁶  For instance, these boundaries may include agriculture or forest land. As an illustration of this issue, consider the CMA and settlement boundaries (based on the methodology outlined in this report) of Regina in comparison to satellite imagery outlined in Figure 1. ⁷ 

The development of a new data set is particularly timely given recent advances in Geographic Information Systems, satellite technology and the creation of the census geographical unit—the dissemination block.

Figure 1: Regina: census metropolitan area and settlement boundaries

3   The methodology: research component

Methodology development for the project began with a preliminary research and analysis phase. This included the identification and examination of available data sets, followed by in-depth research into settlements accomplished through the creation of the Settlements Earth Observation Inventory (SEOI). The SEOI was instrumental in the following key research tasks: identifying settled features and their location (for example, residential, employment, and recreational areas, etc.); analyzing spatial structure and relationships; and establishing rules and thresholds based on spatial and statistical analysis.

3.1  Available data sets: identification and assessment

Several geographic data sets ⁸  were considered as potential foundations for the project, including:

1. Statistics Canada geographic boundaries (dissemination blocks, block face, urban areas, designated places, census tracts, census subdivisions and localities);
2. the National Topographic Database (NTDB);
3. National Land and Water Information System (NLWIS) land cover (2000);
4. Natural Resources Canada's National Road Network (NRN);
5. Canada Centre for Remote Sensing MODIS (moderate resolution imaging spectrometer) product ('built-up area'); and
6. various types of satellite data, for example, Landsat, Google Earth.

Each of these data sets had certain limitations for use in the settlements project. For instance, some of the data were produced only for a given year or did not follow a regular update cycle and therefore did not permit time series analysis. Other considerations included issues related to scale or resolution, cost of product, differing representation or definition of settlements, inclusion of unsettled land, lack of attribute data and others.

Upon consideration of these existing data sources, the dissemination block geography from the Census of Population was chosen as the mapping unit for the settlements project. This data set was chosen due to the small size of the geographic units ⁹  and the variety of attribute data available such as dwelling, population and employed labour force (ELF) counts (see Text box: Key census variables and density measures for definitions). The data are also regularly updated every five years; this supports the need for consistency through time and space.

3.2  The dissemination block and census variables

The mapping unit for this methodology is the 'block'. Known more formally as the 'dissemination block', the block is an area equivalent to a city block bounded by intersecting streets. More precisely, blocks may be bounded on all sides by roads and/or boundaries of standard geographic areas. ¹⁰  Blocks cover all of Canada. In less densely populated areas, blocks are typically larger than in cities and towns due to the sparse road network.

The block is the smallest geographic unit available for creating census geographies. As such, it is possible to aggregate dissemination block data to other standard census geographies. Since the block structure was created in 2001, the time series for this study is limited to 2001 and onwards.

While the block is a suitable mapping unit for this project, the following issues related to the unit and certain associated data are present:

1. As the mapping unit of the settlements methodology, blocks are not split or otherwise changed. Since blocks can contain mixed uses, some blocks may be classified as "settled" due to high population density counts, but may nonetheless include portions of agriculture, forest, wetlands or other "unsettled" land. ¹¹ 
2. Because dissemination blocks are a function of the road network, the number of blocks created is a function of the timeliness of updates of the road network files prior to a given census. Highway medians, ramp areas and other irregular shapes may also form blocks on their own due to road networks. There are also changes or updates in block boundaries between censuses that are not due to actual change, but rather, to improvements in the block geography. A synchronization process in the settlement methodology is used to reduce the impact of these changes to blocks. In the data preparation phase of this study (see Section 4.1), some of these issues were addressed.
3. The Census of Population is distributed to all households in the country; information for all households is obtained regarding dwelling and population characteristics. However, more detailed questions (for example, the long form) were sent to 20% of Canadian households. These questions included information about one's place of work and employment. This sampled approach has implications on data quality. ¹² 
4. Certain census information are geo-coded (that is, located physically on a map) by Statistics Canada. Depending on the census variable, geo-coding is not always conducted at the block-face ¹³  or block level. For example, the workplace location of people working outside CMAs and census agglomerations (CAs) are coded to census subdivisions (CSDs), and not the block.

3.3  The Settlements Earth Observation Inventory (SEOI)

The SEOI is an inventory of geographic information and codes based on an overlay and analysis of the blocks in comparison to satellite imagery. The comparison includes, for example, estimates of the proportion of land that is covered by settlement features in each individual block. To produce the SEOI, imagery for about 260,000 blocks were visually assessed and coded. As a result, in addition to data from the census for each individual block, extra information is now available for each individual block based on satellite imagery.

The creation of the SEOI by Statistics Canada fulfilled a number of important needs for this project, including the following:

1. Characteristics of settled blocks
   Information and improved understanding about the characteristics of settled blocks was needed in order to develop the new methodology. This included information about densities, physical form, land uses, etc. For example, it helped answer the question: at what population density is a block considered settled?
2. Analysis of spatial relationships leading to standardized rules for the methodology
   The SEOI provided data for analyzing and understanding the spatial structure, distribution and relationships of settled blocks. The analysis of the SEOI led to the concepts and threshold-setting for rules related to population, employed labour force (ELF) and dwelling densities, as well as buffers. A 'buffer' is a zone around a block or cluster of blocks; for example, a block cluster with a buffer of 1 km would involve a zone 1 km wide around the periphery of the block cluster boundary. (see Figure 5 for an illustration of the buffering exercise).
   These concepts and rules were then used to establish an automated or rule-based process for delineating settlement boundaries across the country.
3. Data enhancement/improvement
   (i) SEOI data were also used to improve overall estimates of total settlement area. Even with well-established rules, some blocks that should have been included were initially omitted, due to limitations of the ELF data, geocoding, ¹⁴  geographic variability or other reasons. The SEOI allowed settled blocks that would have otherwise been missed to be added. The reverse is also true: some unsettled blocks that were initially included were excluded based on SEOI analysis.
   (ii) The supplementary data in the SEOI allows for improved calculations and flexibility. Limitations related to mixed-use settlement blocks, where settlements are not the dominant feature in a block, can be addressed through use of the SEOI. For example, estimates for settlement area can be adjusted to exclude blocks with limited settlement activity.
4. Data quality and accuracy assessment activities
   The SEOI provided an avenue for conducting data accuracy assessments for the rule-based standardized process (see Section 4). This product can also be used to assess the data quality of other products.

3.3.1  Creating the SEOI

Using satellite data, analysts verified whether or not a given block was settled and determined the proportion of the block that was settled. In doing so, the assessed blocks were given specific codes based on a visual assessment of the block data in comparison to the satellite imagery.

The satellite data consisted of either high-resolution satellite data (QuickBird) or medium-resolution data (Landsat), usually from Google Earth Pro. ¹⁵  Approximately 90% of the imagery used was high resolution, while the remaining 10% consisted of medium resolution data. All census subdivisions (CSDs) with a population of 1,000 or more were analysed, representing an overall total of about 260,000 blocks.

During the coding process, blocks were assigned the following codes:

1. Code 1—over 50% settled ¹⁶ 
2. Code 2—between 25% and 50% settled
3. Code 3—between 10% and 25% settled
4. Code 4—under 10% settled
5. Code 0—unsettled.

Other information was also tracked, including the year of the satellite imagery, the type of imagery (high or medium resolution), and general notes including observations on dominant land use.

Figure 2 outlines a sample block of the SEOI in Google Earth Pro as well as the accompanying data for a single block.

Since the SEOI contains satellite imagery from about 2005 to 2008, it is partnered with data for the 2006 Census. To benefit further from the work of the SEOI, rules were created to include information from the 2006 SEOI in the 2001 settlements data set. The rules were primarily based on census and land cover data.

Figure 2: Illustration of Settlements Earth Observation Inventory (SEOI)

3.4  Results of SEOI analysis: spatial structure and relationships

Analysis of the SEOI shows that there is great variety in settlement patterns across Canada. While some settlements are isolated and remote, others are located in regions with a higher concentration of settlements, such as in southern Ontario. Settlements themselves vary in size and scale, from small groupings of houses to large metropolitan areas with millions of inhabitants.

There is also much variation within settlements, with some areas densely populated and other areas more sparsely populated. Within a given settlement, there may be areas with high employment counts coexisting with areas of no employment activity whatsoever.

The use of the SEOI and GIS technology permitted the spatial analysis of these complex patterns and relationships. Given this complexity, it was clear that one simple rule or variable was insufficient to classify settlement area with a high degree of certainty.

The spatial analysis was conducted using CSDs as a frame. The CSDs were divided into seven categories based on their population. ¹⁷  This analysis of the spatial nature of settlements formed the basis of the thresholds and resulting rules established for the methodology of this project. Rules were initially developed and tested; testing frequently required the modification of rules or the addition of new rules. The final set of rules is presented in Section 4 "The methodology: processing the data."

3.4.1  Core areas and neighbours

Overall, settlements vary considerably in their form and size as a result of a number of factors including geographic location, historical evolution, socio-economic bases, etc. However, there are similar patterns with regards to their use of space.

Settlements contain different, distinct areas or zones. Relationships exist between these zones. Most settlements have concentrated areas or 'core areas' ¹⁸  where the use of land appears to be more intense. ¹⁹  Core areas can be located throughout a given settlement, not just in areas that would casually be referred to as "downtown". These core areas are usually surrounded by or neighboured by areas where land use is less intense, but that are still settled or partially settled. Typically, there is a dependent relationship between the more intensely used core zones and their less intensely used neighbours.

Analysis using the SEOI found that approximately 80% of blocks coded 1, 2 and 3 had a population and/or employment density of at least 400 people per square kilometre. Similarly it was discovered that over 85% of blocks with a population density of 400 people or employees per square kilometre were at least 50% settled (Code 1). Core areas also typically had a more concentrated road network due to the high degree of settlement activity. As a result, core area blocks were typically smaller in size compared to less intensely used zones. These statistics helped formulate the rules for core areas based on density and block size as outlined in Section 4 "The methodology: processing the data."

Less intensely settled blocks were located adjacent to core blocks. Intense settlement activity appeared to encourage additional settlement activity around its borders, but at less intense levels. For instance, over 80% of the less intensely settled blocks were located completely within 500 m of core areas. These blocks were considered 'first order neighbours' (Figure 3). A key attribute of these first order neighbour blocks is their relatively small size and the high frequency of certain specific land uses such as parks, golf courses, parking and areas of growth. This analysis helped formulate rules related to buffers and block size for first order neighbours (see Section 4 "The methodology: processing the data").

Figure 3: Distinct zones within settlements

Another distinct zone was discovered beyond the first order neighbours. These 'second order neighbours' (Figure 3), typically consisting of blocks larger than the first order neighbours, were often characterized by less intense settlement activity. About 94% ²⁰  of these lower density settled blocks had their centre of gravity or centroid ²¹  within 1,000 m of a core block. This statistic, and the testing activities noted above, led to the development of rules related to buffers, as well as densities and block size for second order neighbours (see Section 4 "The methodology: processing the data").

4   The methodology: processing the data

There are four key stages to the methodology (Figure 4):

1. data preparation;
2. the rule-based selection process, which identified core areas and first and second order neighbours;
3. data quality and accuracy assessment activities; and
4. data enhancement activities.

Key concepts used in this methodology include the use of census-based densities measures (see Text box: Key census variables and density measures), buffers ²⁹  (see Figure 5), and block size limitations (see Text box: Thresholds and rules).

Figure 4: Key steps of the methodology

4.1  Data preparation

Data preparation involved a number of different activities. Geospatial data preparation activities included map projection changes, ³⁰  the integration of a new hydrology layer, ³¹  the creation of a central geodatabase, and other activities. ³²  If this research were to be updated, the SEOI would be updated and analyzed in the data preparation phase.

The geodatabase consists of many attribute tables as well as spatial features (for example, census geographies) and raster data (for example, thematic layers and mosaics derived from satellite data). The attribute tables include information such as block identification codes, census counts (population, dwelling and employment), area calculations, linear and surface densities of census variables, correspondence codes, data codes and information from the SEOI and geospatial statistics based on the analysis by block of various national land cover products. This geodatabase is the key connecting foundation of this project; it connects census data with other geospatial datasets.

Another component of the data preparation activity was a synchronization process. Because of some changes in blocks between 2001 and 2006, a synchronization process was needed to ensure data integrity as well as temporal and spatial consistency. Essentially, population, dwelling and employment data from 2006 blocks were transferred to 2001 blocks if certain criteria were met. ³³ 

4.2  The rule-based selection process

An automated and standardized rule-based selection process was responsible for classifying the greatest number of blocks. Overall, 98% of blocks that were classified as settled were classified during the rule-based process.

The process applied rules to all blocks in Canada in an automated fashion within ArcGIS. The key elements of the rule-based process included the selection of core areas, the creation of the distance buffers around core areas and the application of thresholds or rules related to population, dwelling and employment densities of blocks for each of the two buffers. These rules were a result of analysis of the SEOI. Please refer to Text Box: Thresholds and rules, for more details regarding thresholds and rules selection.

An automated process is desirable for two principle reasons. Firstly, it is more cost effective than a process that requires a great deal of manual intervention. Secondly, an automated process helps ensure consistency over time and across the country. This is important for ensuring comparability, particularly for analytical purposes.

4.2.1  Selection of core areas

The first step of the rule-based process was to select blocks that had a population or employment density of at least 400 people per square kilometre and that did not exceed a certain size limitation; that is, in terms of population-related blocks, only blocks with a maximum area of 0.3 km² were selected, while for employment-related blocks, the blocks had to have a maximum area of 0.7 km². This first rule accounted for almost 80% of the total number of the blocks classified as settlements.

Figure 5: Illustration of the buffering exercise

4.2.2  Selection of first order neighbours

Once the 'core blocks' (Figure 5) were selected, a 500 m buffer ³⁴  was generated around these blocks (Figure 5). The use of buffers reflects the spatial relationship (proximity, adjacency, etc.) between the different spatial features of settled blocks. The effectiveness (that is, accuracy) of the buffer was evaluated and changes in buffer width were made in order to reduce the incidence of unsettled blocks being classified as settled. ³⁵ 

Only blocks less than 0.5 km² that fall completely within this buffer were also classified as settled (Figure 5). This method ensured that only small blocks were included (for more information, please see Text box: Thresholds and rules).

4.2.3  Selection of second order neighbours

The selection of the second order neighbours was more complex and required a range of rules or thresholds based on settlement structure, distribution of densities and block size. These thresholds ensured that only the most appropriate blocks were classified as settled. For instance, the rules ensured that large blocks with limited settlement activity were not included.

A 1,000 metre buffer was generated around all core blocks (Figure 5). All blocks within the 1,000 metre buffer that met the following criteria were selected as settlement blocks:

1. The block centroid ³⁶  must be found within the 1,000 m buffer (the entire block does not have to be within the buffer). If the centroid is outside the buffer, the block is not classified as settled at this time.
2. All blocks with a centroid falling within the 1,000 m buffer were examined for the following criteria; if the following criteria were met, they were classified as settled:
   • a minimum population density of 150 people/km² and minimum dwelling density of 50 dwellings/km² and a maximum area of 0.9 km² or
   • a minimum dwelling density of 55 dwellings/km² and a maximum area of 0.9 km²
   • a minimum ELF density of 100 employees/km² and a maximum area of 2.9 km²

4.3  Quality and accuracy activities

Various verification activities were undertaken to ensure quality and accuracy. These activities were limited primarily to thematic accuracy checks, for example, determining if the settlements are actually settled.

The settlement boundaries were converted into a format that could be viewed in Google Earth Pro. GIS specialists then compared all settlement boundaries with satellite imagery. All areas that appeared settled, but that were not included by the rule-based selection process were flagged and a follow-up was conducted. Similarly, areas that were identified as settled but that did not appear settled were also flagged for follow-up. Over 250 areas received follow-up examination and all were found to follow the rules created in this methodology and no changes were made.

A data accuracy assessment was conducted ³⁷  by comparing the results of the rule-based selection to the SEOI for 2006. If the rules classified the block as settled and the SEOI results also showed that the block was settled it was deemed a success.

Three types of accuracy are considered:

1. overall accuracy;
2. producer accuracy (or omission error); and
3. user accuracy (or commission error).

In terms of overall accuracy, there was a success rate of 97%. ³⁸  When the results of the rule-based selection process were combined with additions made from the SEOI, a success rate of around 99% was achieved.

Producer accuracy, or omission error, examines the proportion of settled blocks from the SEOI that should have been captured by the rules-based selection process, but were not. The number of blocks omitted was 7,443, for an omission error rate of 2.8%. ³⁹ 

User accuracy, also known as commission error, is the portion of unsettled blocks that were captured by the methodology but that, according to the SEOI, were not settled. The number of unsettled blocks captured by the rule-based was 2,141 blocks for a commission error of 0.8% or a user accuracy of 99.2%. ⁴⁰ 

4.4  Data enhancement activities

The settlements data were primarily improved by the inclusion of the SEOI data and by the cleaning and merging processes.

4.4.1  Inclusion of the Settlements Earth Observation Inventory (SEOI)

The flexibility of the SEOI allows users to include or exclude specific settlement blocks. In order to address various limitations and improve data quality, some SEOI data were included.

4.4.2  Cleaning and merging processes

The final stage of the methodology involved cleaning the settlements data selected by the rule-based process using the SEOI and merging settlement blocks into larger groupings.

The purpose of the cleaning process was to exclude areas that had been classified in error during the rule-based process, using the SEOI information. If a block was coded as forest or agricultural land in the SEOI, for example, but classified as settlement by the rules-based process, it was removed from the database.

The merging process involved dissolving the polygons; that is, merging individual settlement blocks together to create actual settlements.

5   Initial research results and future directions

5.1  Initial research results

Approximately 20,000 square kilometres of Canada's land was considered to be settled in 2006. ⁴¹  Map 1 illustrates the location of all settlements found through this project. Maps 2 to 4 provide a closer look at settlements for specific areas of the country. Concentrations of settlements occur in certain regions, particularly in central Canada.

5.2  Future directions

Using the methodology and data sources described in this study, a series of indicators are planned to help illuminate settlement-related issues in Canada. Several key themes have been proposed for further study. Proposed initial indicators may cover:

1. land converted to settlements;
2. density of settled areas;
3. compactness/dispersion; and
4. diversity of settlements.

The indicators will be based on two broad categories: settlements with a population over 500 and Canada's largest settlements.