A proposal for a

Landscape Relief Framework for Canada

                         

An open invitation to help produce open soil data

We invite all interested parties to join us in a voluntary activity to jointly create a new and improved landscape relief framework for Canada.  A new polygon-type framework has been proposed, and is available for review.  Details on how to access the framework and provide comments are indicated below.

We want the procedures used to produce this new framework to be transparent and open to all. We also want all data used to produce new maps and all new map products created to be free and open source.   We hope to find that crowdsourcing and voluntary contributions can succeed in this new era of reduced direct governmental activity and support for national soil map products.

Some context                

Traditional soil mapping has always been based on the identification of soil-landscape entities that could be recognized and described as discrete soil bodies. More recently, statistical models that use covariants in digital soil mapping have reinforced the observation that soils of similar properties do tend to occur as discrete soil bodies occupying more or less contiguous areas. Thus, we argue that  there is a clear need for a comprehensive spatial fabric that can provide a structure of recurring and recognizable objects within which the continuous variation in soil properties can be generated, summarized, validated, and used.

                        

Recent efforts to create a set of relief classes for all of Canada included applying automated landform classification procedures proposed by Hammond (Dikau et al., 1991) and SOTER (Dobos et al., 2005). Results from both of these classifications make sense and look reasonably good at smaller scales (1:4 Million to 1:8 million for Hammond, and 1:2 M to 1:4 M for SOTER), but they do not look quite so convincing when viewed at larger scales. At larger than 1:2 M the boundaries produced by these classifications tend to look arbitrary and out of sync with visible drainage lines and other topographic features.

                        

Finer resolution digital elevation models available for all of Canada now make it feasible to create an improved set of relief classes that can provide a consistent, high quality segmentation of the Canadian landscape. We contend that such a segmentation could well inform improvements to existing products like the Soil Landscapes of Canada, as well as assist in the creation of new digitally-generated estimates of soil classes and soil properties.

Objectives and intended uses                

Why would we want to produce a polygon-type map of landform classes for Canada? We offer the following explanation for how such a classified map might prove useful and beneficial.

  1. The polygons can act as initial proto-polygons that can be split, merged or modified by local experts to produce a final, and accepted polygonal spatial data fabric to be used as a basis for (manually or automatically) creating national (or regional) polygon type maps of soil associations, parent material complexes or any other phenomenon that displays a consistent relationship to landform position and landform type. .
  2. Each final polygon could be attributed with a list of classes of expected major soils and expected major parent materials using class codes defined by CanSIS and CSSC.
  1. Attribution could be manual by overlaying final polygons on top of any locally available and relevant maps of soils or parent materials and manually identifying major included soils or parent materials in each polygon or group of similar polygons
  2. Or attribution could be semi-automated or automated using data mining to detect patterns of association between groups of similar polygons and spatially registered point data sets or point data sampled from existing soil or surficial geology maps.
  1. A new polygonal national map of soil classes (or soil parent materials) could be used as a covariate in any subsequent automated Digital Soil Mapping (DSM) exercise to predict the detailed spatial distribution of gridded soil properties or soil classes nationally.
  2. The polygons can be used to define a comprehensive and consistent spatial fabric across the entire country. This spatial framework can be used for reporting, visualizing, organizing and standardizing raster maps of soil properties or soil classes.
  1. Many uses benefit from having a spatial reporting framework that is stable in time, consistent and comparable everywhere (in this case across Canada)
  2. Many uses benefit from defining spatial entities (objects) that have geomorphic and hydrological meaning and that can be recognized and described in the field as recognizable portions of the landscape (e.g. valley bottoms or valley sides, terraces, toe slopes, ridges or peaks, escarpments, and so on).
  3. A stable and consistent spatial fabric can help to put raster maps of individual properties or classes in context. It also provides a stable base for assessing and reporting on changes or differences in time within spatial objects that do not change (much) in time.

The proposed framework        

A new, and highly detailed, set of relief classes has recently been produced independently for all of  Canada at 250 m by Bob MacMillan, using a modification of the original approach of Hammond. These new relief classes (Table 1) are presented here for review and assessment.

The classification is very simple. It makes use of just two terrain derivatives computed here (for Canada) using a 250 m resolution DEM. These are absolute relief measured as elevation in m above a defined stream channel (Z2St) and relative relief (PctZ2St) defined as the ratio of elevation above a channel (Z2ST) to total vertical distance channel to divide (Z2ST+S2Cr).

The 250 m DEM was produced on contract for LandMapper Environmental Solutions Inc. by scilands GbmH of Gottingen, Germany  using as input the CDEM DEM provided by Natural Resources Canada on its Geogratis web site along with contour lines and hydrology data. This DEM was reprocessed at a resolution of 50 m to re-interpolate to continuous elevations expressed using real numbers to make use of, and respect, hydrological networks and contour line data.

Table 1. Legend and classification rules for the new simple  MacMillan landform classification

Grid Code

Cell Count

Areal Extent (%)

Rel Slope Position (PctZ2ST)

Abs Relief (Z2ST)

Description

10

5,512,003

4%

Slope Gradient = 0

Pos = 0

Z2St < 0.5

Lakes and Wetlands: Lakes and wetlands defined as having a slope of zero (0) and an elevation above base level of less than 0.5 m (to account for tilted base planes in the DEM).

11

25,272,297

16%

Dep2Low (< 10%)

< 1 m

Lake and Wetland Margins: Low-lying, level, Lake Margins, estuaries, shorelines, low-lying, level bowls and depressions, bogs and fens; a few low, level river channels and floodplains.

12

4,458,716

3%

Dep2Low (< 10%)

1-10 m

Sloping Valleys in Plains: Low Relief Depressions and Hollows: Low relief lake margins, low lake shores, sloping valley bottoms and river channels, raised floodplains and terraces.

13

915,850

1%

Dep2Low (< 10%)

10-30 m

Sloping Valleys in Hills: Lower to toe slopes, hollows and draws in areas of moderate relief hills. Sloping bottoms and sides in hills. Mid level terraces in valleys in hills.

14

1,238,424

1%

Dep2Low (< 10%)

30-91 m

Sloping Valleys in Mountains: Sloping valleys, elevated terraces and lower valley sides in areas of high hills to mountains.

22

12,485,787

8%

Low2Mid (10-30%)

1-10 m

Toe Slopes in Level Plains: Mid to lower slopes, hollows and draws in areas of low, level plains to low hills. Mostly in low relief plain areas. Also in lower parts of low hills. Drainage ways and lake margins in areas of low relief.

23

14,576,641

9%

Low2Mid (10-30%)

10-30 m

Toe Slopes in Hills: Mid to lower slopes, hollows and draws in areas of low to moderate hills. More often in upland areas than in basins or low plains. Lower and slightly moister portions of slopes below low to moderate hills. Some channel sides.

24

8,185,682

5%

Low2Mid (10-30%)

30-91 m

Toe Slopes in High Hills: Mid to lower slopes, hollows and draws in areas of moderate to high hills (some low mountains).

25

1,471,570

1%

Low2Mid (10-30%)

91-152 m

Toe Slopes in Mountains: Mid to lower slopes, hollows, draws and lower valley sides in areas of high hills to mountains.

26

4,557,861

3%

Low2Mid (10-30%)

152- 305 m

Toe Slopes in High Mountains: Lower to toe slopes, hollows, draws and lower valley sides in areas of mountains to high mountains. Steep mountain valleys.

32

6,510,041

4%

Mid2Crest (>30%)

1-10 m

Level Plains to Basins: Very flat mid to upper slopes of interfluves in very low relief basins and plains. Crests to crowns of interfluves and/or rises in very low relief landscapes. Basins, low plains, river valleys.

33

19,392,517

12%

Mid2Crest (>30%)

10-  30 m

Low Relief Undulating Plains: Mostly upper to mid slopes in gentle, low relief plains and basins. Crowns, interfluves, low ridges, some channel sides in low relief plains, basins and very low hills.

34

25,055,737

16%

Mid2Crest (>30%)

30-  91 m

Low Hills: Mid to upper slopes, crests and crowns of low hills (30-91 m), interfluves, ridges, valley sides. Drier upper slope positions. Some upper terraces in mountain valleys

35

6,917,312

4%

Mid2Crest (>30%)

91- 152 m

Moderate Hills: Upper to mid slopes, crests, crowns and interfluves in areas of moderate to high hills. Not in mountains except mountain valleys.Highest and driest parts of moderate to high hills.

36

7,519,975

5%

Mid2Crest (>30%)

152- 305 m

High Hills: Crests and crowns of high hills (maybe a few low mountains). Upper to mid slopes, crests, crowns and interfluves in areas with relief of 152-305 m.

37

10,245,523

7%

Mid2Crest (>30%)

305- 915 m

Mountains: Mid to upper slopes, crests to crowns of mountains under 915 m. Steeply sloping areas. No valleys or toe slopes.

38

1,056,332

1%

Mid2Crest (>30%)

915- 4000 m

High Mountains: Crests and crowns of high mountains over 915 m. Very steep, rocky, high. Highest parts of high mountains.

The proposed polygons can be seen and reviewed at http://isric.org/landforms, courtesy of ISRIC.

Web application

The proposed polygons can be seen and reviewed at http://isric.org/landforms, courtesy of ISRIC.

Backdrop images

                                

Georegistered JPG images with world files are provided to be used as backdrops for visual review and evaluation of shape files as vector overlays in a GIS.

MacMillan M38r100xL classification (jgw) (original, not filtered, minor classes removed, clusters less than 100 cells removed and a new lake class (10) cut in)

Georegistered JPG (zipped)

https://www.dropbox.com/s/0kp5swskxd2watt/c250_m38r100l.jpg?dl=0

SOTER level 3

Georegistered JPG (zipped)

Hammond level 2

Georegistered JPG (zipped)

Hammond level 3

Georegistered JPG (zipped)

Hillshade for Canada

Georegistered JPG (zipped)

slope gradient

Georegistered JPG (zipped)

Blue Marble satellite image mosaic

Georegistered JPG (zipped)

Scilands TCI-Low compound topographic index of moisture and landscape position

Georegistered JPG (zipped)

We welcome contributions of any other Canada-wide grid maps or images that depict variation in vegetation, land use, geology, ecology or any other spatially varying natural phenomenon. These additional images and maps can also help to evaluate the utility of the proposed classification.

We recommend reviewing how the MacMillan vectors outline areas of similar slope gradient, similar vegetation (as per Blue Marble), similar drainage (as per TCI-low) and similar relief (as per Z2St). We believe that the new polygons really do separate out areas of internally similar morphology, vegetation, slope, relief and drainage compared to areas outside each polygon. We welcome the contributions of others to assess the degree to which polygons of the proposed classification meaningfully differentiate natural phenomenon at local, to regional to national levels.        

GRIDS

                                

All in ArcView GRID format.          

MacMillan's classification of M38R100xL

ArcView GRID

Hammond level 2

ArcView GRID

Hammond level 3

ArcView GRID

SOTER level 3

ArcView GRID

SOTER level 4

ArcView GRID

input layer Z2St (change in elevation to stream)

ArcView GRID

input layer PctZ2St (percent change in elevation to stream)

ArcView GRID

SHAPEFILES

MacMillan M38_R100_XL

M38_R100_XL.zip

Hammond Level 2

Note: These files currently contain some geometry errors.

LEGENDS

A text legend in Excel format for as well as ArcView legends for assigning colors to the, Hammond and s,, and         

MacMillan's M38R100XL classification

Excel

M38 grid

ArcView

Hammond

ArcView

SOTER grid

ArcView

Slope Gradient

ArcView

Z2St

ArcView

PctZ2St

ArcView

Initial comments        

In Bob’s words:

                                                                         

This classification makes sense and looks best at scales of 1:500,000 to 1:1 M. The biggest improvement is the level of detail and drainage exactness - perhaps too much detail but at least correct detail. If you overlay Hammond as a vector shape file over top of raster or grid images of the current MacMillan classification (M38ZR100XL ) you can see how closely the MacMillan classes  correspond with the Hammond vectors. These new detailed classes are almost perfectly nested within equivalent Hammond generalized classes. This comes from using essentially the same rules for both just applying them to different entities at different scales.

                                        

The original MacMillan landform (relief) maps for Canada were reclassified to remove several initial classes that had a very small extent or were closely similar to, and confused with, other classes. There are presently 17 classes defined as per the attached legend. These classes closely follow the Hammond relief classification system, except that an extra 3 classes were added at the low end (0-0 m, 0-1 m and 1-10 m instead of the original single Hammond class of 0-30 m). Also, unlike Hammond, where upland interfluves are mixed with, and classified with, lowland basins and valleys (as in High Hills with Valleys), the current approach extracted and classified basins and valleys separately from upland ridges and interfluves. So, we now recognize valleys in flat basins, valleys in undulating plains, valleys in low hills and valleys in high mountains, and so on. The new procedures also cut out lower to toe slopes in each of these relief classes, so as to differentiate what we can colloquially refer to as “things that stick up” (mountains, hills, interfluves) from “things that stick down” (valleys, basins, lakes, wetlands, toe slopes and depressions). It is argued that this approach better reflects what human interpreters do when they tend to draw boundaries that outline different parts of the landscape starting with river valleys and channels or lakes at the lowest landscape positions and working progressively up slope until they outline upland ridges or interfluves.

                                        

My personal feeling is that this classification does provide a consistent and meaningful spatial data fabric that closely approximates the kind of soil-landscape spatial entities that a human interpreter would delineate at 1:500,000 scale if presented with air-photos or a DEM hillshade depicting the topography of an area (at 1:500,000 scale). These landform entities can be given names and descriptions (e.g. High Hills) that are meaningful and cognitive. The rules produce entities that are consistent and comparable everywhere, so no differences in detail or scale or concepts anywhere in the country. The entities identify and respect drainage courses that are identifiable on 250 m DEM data. If one accepts that a stable polygon type spatial data fabric that outlines identifiable landform based entities is beneficial, then I feel this classification presents a meaningful and interpretable implementation of such a spatial fabric. In fact, if I overlay Soil Landscape of Canada or AGRASID polygons on top of this classification I see a great deal of spatial correspondence between boundaries drawn by hand and boundaries evident on this classification (except that this classification produces orders of magnitude more spatial detail).

                                        

It is possible to envisage a second stage where we develop and apply fuzzy classification rules to cut out further sub-classes within each of these 17 main classes. For example rules could be produced to differentiate class 11 lakes and wetlands into lake shores and beaches, lacustrine or tidal flats, active flood plains, low terraces and so on. Similarly, mountains could be further subdivided into steep NE and Steep NW slopes, bare rock or vegetated, steep mountain hollows or draws, steep mountain spurs, gentle slopes or terraces on mountains and so on. It is premature to  undertake this second stage sub-classification until we obtain feedback from users and get some sense of whether this classification has any potential to be taken up and used.

                        

Comments and Feedback

We welcome comment and feedback. This page is an attempt to determine if this new work has such potential.  Please provide comments through the public access Google Group at

Terms of Use

                                

This data is licensed under the Open Data Commons Attribution License.                         

Collaborators

                                

Bob MacMillan (private sector consultant) - initial inspiration and initial classification development.         Jorge Mendes de Jesus (ISRIC) - web mapping application (CesiumJS)

Peter Schut (AAFC) - CDSDC website creation and administration, advice and support

Michael Bock - preparation of DEM input layer and TCI-low classification

Chuck Bulmer - input on classification criteria and utility, polygon reprocessing and cleanup

Xioyuan Geng (AAFC) - initial problem definition, advice and support

Scott Smith - input on classification criteria and utility

Joanna Zawadzka - application of classification rules in other countries

Cindy Shaw (NRCan/RNCan) - input on classification criteria and utility

John Simms - programming and analysis contributions, upgrade of LandMapR programs

Jean-Daniel Sylvain - evaluation of landform classification utility in Quebec

Mike Duncan - programming and application contributions

Tomislav Hengl (ISRIC) - provision of global 250 m DEM, data reformatting, programming

Rik van den Bosch (ISRIC) - support and encouragement

Final Thoughts

                                                                         

I think this new classification represents an improvement on Hammond and SOTER but in an evolutionary way not a revolutionary one. Especially Hammond. These new polygons nest almost perfectly within Hammond’s. They should. This classification is essentially Hammond applied to individual hillslopes from divide to channel. It is a spatially exact, hydrologically consistent Hammond classification. It addresses our desire to use the drainage network as a starting point for drawing polygons around drainage lines and lakes or depressions at the bottom end and ridges and peaks at the top end. Things that stick up and things that stick down. All we ever really did in manual mapping.

                                        

Both Hammond and SOTER are a little deficient at the top (MOUNTAINS) and bottom (valleys and level basins) end. This is just down to the fact that neither splits out any relief classes less than 30 m (hammond) or 50 m (SOTER) or over 1500 m. If they recognized a few more classes at both the bottom and top ends both would produce more complete results. The neighbourhood analysis window (NAW) derivatives produced by both Hammond and SOTER are actually quite effective. But both underestimate the true maximum relief. SOTER is the poorer. SOTER should use at least a 2 km radius window instead of 1 km. Hammond uses a 10 km NAW and so generalizes better but boundary location is only convincing when viewed at very small scales of 1:4 million or more. When viewed at even 1:1 to 1:2 million boundaries look poorly located.

                                        

I have noted some inconsistencies, where different polygons of the same class can outline areas of different conditions. But I think it should be possible to identify individual polygons that differ within the same relief class by attaching additional attributes such as dominant slope gradient, slope position, curvatures, wetness, etc to every polygon and then classifying polygons of the same initial class into two or more sub-classes, if they truly differ in ancillary attributes.

                                        

Also, I have also considered that these polygons could well provide the base for a top level framework in a hierarchical mapping approach. Each of these 15 classes could become a domain or zone, within which one could recognize and delineate refined subclasses. For example, mountains could be subdivided into steep SW and NE slopes, spurs, draws, terraces or more gentle upper slopes, cirques, and perhaps a few other entities I can't think of now. The class Lakes and Wetlands clearly needs to be subdivided into actual water (Z2St = 0 and perhaps some other data from imagery) and non water (shorelines and littoral zones), Active channel banks, channel bars and islands and maybe organic wetlands). The point is that the initial, top level classification, can provide context for subsequent second or even third level refinements.

                                        

Some caveats:

                                        

  1. This classification has not been produced, veted or approved yet by any government agency.
  2. The classification misses out on classifying a few small areas and so is not complete for Canada. The 50 m DEM used was missing a small area in southern Quebec.
  3. There are some small areas that are clearly classified incorrectly and need to be fixed. The two most obvious are small islands in lakes that got filtered out into the 11 wetland shore class because they were smaller than the minimum size of 100 cells. The other is areas of water where the source DEM has a tilted reference base creating edges, small slopes and minor relief. These confuse the classification rules. To fix I need to either get a new DEM without tilted tiles or manually reclassify polygons that are clearly water and not classified as such.

                                                         

         

================

https://www.dropbox.com/home/Canada_Landforms/RAM_2015_Products/4_Sharing