Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
Quesnel PEM Input Data Quality Report for Manually Derived Terrain
Attributes
Extracted From
Quesnel PEM Predictive Ecosystem Mapping
Knowledge Base and Attribute Summary
Submitted to:
Dr. David Moon,
Technical Monitor, Quesnel PEM Project
CDT Core Decision Technologies Inc.
42-11391 7th Avenue
Richmond, B.C.
V7E 4J4
Ray Coupé,
Regional Research Ecologist
Ministry of Forests,
Southern Interior Forest Region,
200-640 Borland Street
Williams Lake, B.C.
V2G 4T1
By:
Robert A. MacMillan., Ph.D., P. Ag.
LandMapper Environmental Solutions Inc.
7415 118 A Street NW,
Edmonton, Alberta,
T6G 1V4
780-435-4531
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
1
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
1.0 Introduction
This material is extracted from MacMillan R.A., 2005. Quesnel PEM Predictive Ecosystem Mapping
Knowledge Base and Attribute Summary. Available from the MSRM Land and Resource Data Warehouse
under the Quesnel PEM as file pem_4073_Kbase-pem_4145-kb01.pdf.
2.0 Base Maps
All input data layers used TRIM I or TRIM II base maps. The respective PEM and TEM standards define
TRIM I and TRIM II as the standard base map for all PEM and TEM products and exempt it from IDQ
reporting.
3.0 Description of the LMES DSS input layers
As previously indicated, a major distinction can be made between those LMES DSS input layers that were
prepared using manual visual interpretation and manual digitizing and those prepared using automated
processing of available digital input data (principally the TRIM II 25 m DEM data). The following
discussion maintains this distinction between manually and automatically prepared input layers though, in
several cases, there was overlap with automated modeling contributing to the preparation of manually
created input layers and manual intervention contributing to the preparation of automated layers.
3.1 Manually prepared input layers
The philosophy behind utilizing manually prepared input layers is basically one that recognizes that it is
better and more economical to directly map those inputs that are more easily recognized and delineated
using manual visual interpretation than by automated modeling.
It does not make sense, from the point of view of either cost or scientific validity, to attempt to model those
spatially distributed features that are more clearly, unambiguously and easily identifiable using manual
visual interpretation. Therefore, for example, a decision was made to have human interpreters manually
delineate easily recognized spatial entities such as open water, non-forested wetlands (e.g. swamps,
marshes and fens), bare rock, disturbed or urban areas and similar features that are clearly and easily
distinguished on available imagery. It is, of course, possible to automatically extract such entities by
applying image analysis and classification techniques to available digital imagery (both air photo and
satellite). However, automated recognition of these easily recognizable features was recognized to possess
some drawbacks that were not as manifest in manual approaches. Firstly, some spatial entities, such as
lakes and wetlands, have very distinct, hard boundaries. These boundaries are easily perceived and located
by humans viewing appropriate imagery. Automated techniques for modeling or classifying these entities
will inevitably result in delineations whose boundaries do not coincide exactly with boundaries that are
perceived using human vision. This leads to an undesirable spatial disconnect between the hard boundaries
of discrete objects, as extracted using automated techniques and the hard boundaries as perceived by human
vision and interpretation. This spatial disconnect can reduce the confidence that users of any map will
place in the map, if clear obvious boundaries, such as lake shores or wetland margins are not located where
they can be clearly seen to occur. Additionally, specifications for PEM mapping in B.C. clearly state that
the boundaries of all permanent water bodies, as delineated by a PEM, are to conform exactly to the
boundaries of water bodies as they appear on the TRIM digital base maps. This makes it imperative that at
least the mapped boundaries of lakes conform exactly to the TRIM digital base maps. If we are going to
require boundaries of permanent water to conform to boundaries identified by manual visual interpretation
it is just as well to adopt a similar approach for delineating non-forested wetlands and other clearly visible
surface features.
From a cost standpoint, it has so far proven more efficient and economical to generate manually interpreted
maps of readily visible surface features than to attempt to model and classify these features from digital
imagery. The maps of manually interpreted surface features prepared by JMJ Holdings Inc. were produced
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
2
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
for less than 3½ cents per hectare. It has, so far, not been possible to produce maps of equivalent content,
spatial precision and thematic accuracy using automated analysis and classification of available digital
imagery.
Similar arguments about cost and reliability apply to the other manually prepared input layers used in the
LMES DSS procedures. It has not yet proven possible to model the physiographic regions, zones of
elevated frost hazard, seepage areas and areas of different textures of parent material in a manner that is
more correct, efficient or cost effective than can be accomplished using manual interpretation and manual
digitizing. It is hoped that, in the future, some or all of these input layers may prove amenable to reliable
and cost-effective preparation using automated modeling or image classification techniques. In the
meantime, LMES continues to advocate use of manual interpretation and digitizing as the most efficient
and cost-effective alternative.
3.1.1 Localized Biogeoclimatic Ecosystem Classification (BEC) Input Layer1
The localized BEC maps (referred to as Big BEC) were prepared by the Regional Research Ecologist for
use in the Quesnel PEM project. Because the BEC maps were supplied by the regional ecologist, there is
no requirement that they be subjected to an input data quality assessment. They are assumed to be correct
and to present an accurate depiction of the spatial extent of each of the BEC sub-zones or variants
recognized to occur within the Quesnel PEM study area.
LMES simply received the localized BEC information as a polygonal map in ArcView shape file vector
format and UTM Zone 10 projection, NAD83 datum. LMES then arbitrarily assigned a unique integer ID
number to each of the BEC sub-zones or variants identified on the ArcView shape file vector map (Table
1). The unique integer ID number associated with each of the BEC sub-zones or variants was subsequently
used in the construction of integer ID numbers for unique “classification regions” defined by the
intersection of the BEC polygons with polygons representing physiographic, frost hazard and parent
material texture classes.
Table 1. Integer ID numbers used by LMES to identify BEC sub-zones and variants
BEC_ID
21
22
23
24
25
26
27
28
29
30
33
40
41
42
50
52
Bgc_Zone
MS
SBPS
SBPS
SBPS
SBPS
SBS
SBS
SBS
SBS
SBS
SBS
At
ESSF
ESSF
IDF
IDF
Bgc_Subzone
Bgc_Vrt
xv
dc
mc
mk
xc
dw
dw
mc
mc
mh
dk
mv
xv
dk
xm
1
2
2
3
1
1
3
Area (ha)
455377.42
353858.06
64073.87
232863.40
108027.88
40158.71
154168.19
110430.67
19988.62
21972.99
719.80
13222.21
1052.65
22999.77
9559.34
2936.49
1
A full report is presented by Coupé R. and Steen, 2004. Large Scale Biogeoclimatic Mapping in the
Quesnel and Canim Lake PEM Project Areas. Available from the MSRM Land and Resource Data
Warehouse under the Quesnel PEM as file pem_4073_BEC-pem_4145-bgc.pdf.
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
3
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
The unique integer ID numbers listed under the heading BEC_ID in Table 1 are purely arbitrary and carry
no underlying meaning. One may perceive that the first 10 BEC zones are arranged in alphabetic order and
numbered consecutively from 21 through to 30. The first 9 BEC zones were identified as “high priority”
zones for which LMES was contracted to produce knowledge base (KB) classification rules on a priority
basis. Development of KBs for the remaining BEC zones was not a requirement of the initial LMES
contract and so these other zones were not originally included in the list and were not initially assigned
unique ID numbers. LMES received a request late in the project to produce an additional set of KB rules
for the IDFdk3 variant. At this time, LMES decided to produce initial, interim rules for ALL BEC subzones and variants that lay within the study area defined for the Quesnel PEM project, regardless of
whether they had been requested, or paid for, as part of the LMES contract. The gaps in the numbering of
these extra BEC sub-zones and variants reflect the fact that they represent a subset of a larger list of zones
that LMES developed early in the project that identified all zones that occurred within a larger rectangular
area covered by the original DEM that was provided to LMES. Many of these originally numbered BEC
sub-zones and variants ended up being outside the final boundary of the Quesnel PEM project area, so they
do not appear in Table 1 and there are some missing numbers in the sequence.
3.1.2 The JMJ prepared map of parent material texture, depth and exception classes2
LMES was provided with a map of parent material texture, depth and exception classes prepared by JMJ
Holdings Inc following specifications originally developed by LMES. JMJ was responsible for preparing
this map and for preparing and submitting separately the input data quality report associated with it.
LMES received this map from JMJ in ArcView shape file format, UTM Zone 10 projection and NAD 83
datum. Each JMJ polygon was associated with a data base record that stored several pieces of information
about that polygon (Tables 2 & 3).
Table 2. Example of the data base records linked to each polygon on the JMJ-prepared map
poly_id
1
2
3
4
5
6
7
8
9
10
geocode
1
2
3
4
5
6
7
8
9
10
depth
100
100
100
100
20
100
20
20
100
100
texture
1
1
1
1
50
1
50
50
1
1
non_forest
seepage
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Each polygon on the JMJ-prepared map of parent material depth, texture and exceptions contained an entry
for each of the fields of information identified in Table 2. The poly-id field provided the link to the JMJ
polygon map. The field named geocode is an exact copy of the poly_id field and was used by a custom
LMES program that created a GEOFILE DBF table for use in the LMES DSS procedures in which the
geocode attached to each and every grid cell in an area of interest was used to look up and enter values for
2
A full report is presented in Ketcheson M., J. Shypitika, M. Kalmakov, K. Misurak, and V. Lipinski.
2004. Quesnel PEM Materials Mapping Component FRS#4233005. Available from the MSRM Land and
Resource Data Warehouse under the Quesnel PEM as file pem_4073_Materials-pem_4145_idq.pdf
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
4
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
depth, texture, non-forest exception type and seepage into the GEOFILE. By this mechanism, the JMJ
polygon map was converted into a grid-based DBF file that contained data for each of the fields in Table 2
for every 25 m grid cell in the study area.
The numbers in that appear in the columns labeled depth, texture, non-forest and seepage in Table 2
represent codes (see Table 3) for estimates of parent material depth and texture, non-forest class and
presence or absence of visible seepage that were made by air photo interpreters employed by JMJ Holdings
Inc. The JMJ interpreters were instructed to visually seek out, and delineate, all readily visible surface
features that represented “exceptions” to normal, modal conditions.
Thus, for example, all parts of the study area were assumed to have a default depth to bedrock of 100 m or
greater (code 100) and JMJ interpreters would only delineate and attribute areas that were conspicuously
and clearly shallow to bedrock (code 20).
Similarly, JMJ interpreters were instructed to seek out areas that appeared to be associated with parent
material textures that were significantly coarser (code 70) or finer (code 20) than normal or were obviously
organic (code 1). All other areas were considered to possess normal, medium textured parent materials
(code 50). Only the “exception areas” of clearly coarser, finer or organic parent materials were identified
and delineated. Recognition of areas of coarser or finer textured materials was aided by visual observation
and interpretation of patterns of vegetation and land use that were observable on geo-registered digital
orthoimagery. The visual interpretation of available digital orthoimagery was supplemented by visual
interpretation of the DEM (as a hillshade) and of a number of terrain derivatives computed from the DEM
that were judged to be useful in visually identifying different types of landforms that might be associated
with different textures of parent material.
Table 3. Codes used on the JMJ-prepared map to identify different classes of “exception” features
Field
Name
Poly_ID
Field Description
Geocode
Depth
Empty (copy of Poly_ID used by LMES)
Depth to bedrock
Texture
General material/texture of the dominant
parent material
Code
The ArcView or ArcInfo polygon ID
number
A provincially acceptable
polygon ID
20
0
1
Non-forest
Identifies areas of shallow
depth to bedrock (all other
areas are coded as 100)
All areas are assumed to be
medium texture by default
(coded as 50)
Exposed bedrock
Organic Wetland
20
50
Fine Textured
Medium (default)
70
Coarse Textured
Identifies “exception areas”
consisting of lakes and nonforested wetlands, and nonforested uplands
Lakes and open water
Areas that do not support forested
ecosystems
10
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
Code Description
5
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
Seepage
11
Non-forested wetlands
21
22
23
24
meadow
pasture
non productive brush
Avalanche track (none
mapped)
31
Wetter than expected
Identifies wetter “seepage” areas
With respect to the identification of areas of coarser parent materials, JMJ interpreters basically looked for
areas that were occupied by thinner stands of tree species that preferred drier conditions (e.g. pine or
Douglas fir) and that were located in landscape settings that could be reasonably assumed to be associated
with coarser textured materials (e.g. flood plains, terraces, fans). Similarly areas of organic parent
materials were identified on the basis of stand density of tree species such as Black Spruce that are often
associated with wetter areas of organic soils. This rapid seeking out of “exception areas” proved to be both
cost-effective and acceptably accurate. It provided the maximum amount of usable information for input
into a predictive PEM process at the lowest possible cost in terms of both time and expense.
Non-forested areas represented a different sort of “exception area” than areas of shallow soils or coarser or
finer parent materials. Areas of different texture or depth were still expected to be occupied by forested
ecological entities (Site Series). The main difference in these exception areas of different texture or depth is
that the Site Series expected to occur would likely differ from the Site Series normally expected to occupy
the same location in areas of deep, medium textured materials.
Non-forested “exception areas” are not viewed as capable of supporting forested ecosystems according to
the JMJ interpreters. There is therefore no expectation that any defined forested ecological class can occur
in these locations. The approach adopted by LMES has been to predict an expected forested Site Series for
each and every grid cell in non-forested areas, as if the areas had not been identified as non-forested. The
non-forested “exception areas” are maintained in a separate grid coverage which is “cookie cut” into the
final LMES PEM map at the last possible moment, after all operations related to predicting, correcting,
smoothing, filtering or otherwise improving the LMES predictions of forested ecological entities have been
completed. This is done in order to make it possible to smooth and filter the LMES DSS classification grid
maps to reduce local noise without blurring or moving the hard boundaries that are typically associated
with the non-forested “exception areas” mapped by JMJ. Most of these non-forested “exception areas”
represent surface features such as lakes, wetlands, pastures, meadows or areas of non-productive brush that
are clearly identifiable on available imagery and that have clearly defined, hard boundaries. Any filtering
that might be applied to a grid map that already portrayed the spatial distribution of these surface features
with hard boundaries would only serve to corrupt obviously correct boundaries. As LMES wished to
maintain all hard boundaries in their original locations, the non-forested “exception area” classes were not
mapped out as part of the initial LMES DSS predictions but were reserved for “cookie cutting” into the
final PEM map at the last possible moment.
The final field in the JMJ-produced map of material texture, depth and exceptions is the column labeled as
“SEEPAGE”. The seepage field was added at the suggestion of the Regional Ecologist. It is an attempt to
provide a mechanism for manually recognizing areas that appear to the interpreter to be wetter than
expected for any given location in the landscape. This field allows manual interpreters to alert the LMES
DSS procedures to the presence of wetter than normal conditions. Such conditions might be associated with
seepage from bedrock sources on hillslopes that is not modeled adequately by any of the terrain derivatives
or they might be associated with areas of shallow depth to groundwater table that are also not well modeled
by any of the available terrain derivatives. Typically, such “seepage areas” represent areas that would
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
6
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
normally have been expected to exhibit normal, mesic Site Series but that, for whatever reason, exhibit a
Site Series associated with somewhat moister conditions. The “Seepage” code is therefore used to alert the
LMES DSS procedures to the presence of wetter than normal conditions. This code can serve as a flag that
causes the LMES DSS procedures to utilize a different rule that will identify a wetter than normal Site
Series at a location at which a wetter Site Series would not normally be predicted. The code 31 is simply
used as a binary flat to identify these wetter conditions and flag them for the LMES DSS procedures.
The JMJ-prepared map of material texture, depth and exception classes is a key input to the LMES DSS
procedures. It provides the opportunity to make the most effective possible use of human visual
capabilities and human interpretation. This map represents an attempt to directly map what is easily and
readily visible and that can be mapped more rapidly, accurately and cost-effectively through direct manual
interpretation than through more complex and costly modeling. The LMES philosophy is to directly map
what is easily mappable and to only model what is too difficult, tedious or time consuming to map directly
by manual visual interpretation.
3.1.3 The LMES prepared map of physiographic classes
Experience in previous PEM projects in the Cariboo region had led to the realization that a single set of
LMES DSS rules was often not able to be applied successfully to different landscapes within the same BEC
subzone or variant if the landscapes exhibited significantly different expressions of relief, slope gradient,
slope length and wave length (short range complexity). KB rules that produced acceptable results when
applied to normal or modal landscapes that exhibited the most common range of slope lengths, gradients
and relief often performed unsatisfactorily when applied to landscapes within the same BEC subzone that
exhibited markedly different ranges of these physiographic characteristics. It was determined that better
results could be obtained if the LMES procedures were set up to permit slightly different KB rule bases to
be applied to areas that exhibited markedly different relief and physiography within any given BEC
subzone. Often, modifications consisted simply of adjusting the threshold values applied to input variables
to establish spatial locations for boundaries for such concepts as relative moisture status (e.g. slightly dry or
very wet) or relative landform position (e.g. toe slope or depression). The LMES procedures were
therefore adapted to permit creation and use of maps that distinguished regions of different physiography
and relief within BEC subzones and variants and that then applied different versions of KB rule tables for
any given subzone or variant within these different physiographic regions.
LMES prepared a map depicting the location and extent of a number of different classes of physiography
and relief within the Quesnel PEM study area (Table 4). This map was prepared using a combination of
automated modeling of DEM data and manual interpretation and digitizing.
Table 4. Codes used by LMES to identify classes of landforms with different physiography and relief
Zone ID
20
30
40
50
60
Description of the topography and physiography of the LMES defined zones
Areas of relatively low relief (< 100 m ridge to channel) and relatively short slopes.
Areas of relatively high relief (> 100 m ridge to channel) and relatively long slopes.
Areas of high predicted frost hazard – usually also exhibited relatively low relief.
Areas of very gentle to flat topography the nonetheless had long gentle slopes.
Areas of high relief that exhibited unusually long continuous slopes.
The list of zones defined by LMES for use in the Quesnel PEM (Table 4) is not fixed or formalized. The
zones represent an ad-hoc response to challenges to effective classification as they are encountered. LMES
began by defining only two different physiographic classes for the Quesnel PEM map area; namely areas of
low (20) versus high (30) relief. During a modeling workshop held in Williams Lake February 22-26, 2004
the Regional Ecologist reviewed classification results produced by the initial LMES KB rule bases and
identified deficiencies and concerns. One of the results of this review by the Regional Ecologist was a
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
7
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
decision to recognize and map the spatial extent of three additional classes of landforms. These new,
additional, landform classes were defined for low-relief, bowl shaped areas that were deemed to have a
high likelihood of experiencing elevated frost hazard (40), areas that had very gently sloping to almost level
terrain but that nonetheless exhibited long gentle continuous slopes (50) and areas of high relief that
exhibited unusually long continuous slopes, usually of high gradient (60).
The initial classification of physiography into areas of high versus low relief was accomplished by a
modeling effort that applied a threshold value of 100 m to a terrain derivative named Zpit2Peak computed
by the LMES program FormMapR. This variable is computed as the sum of two other variables named
Z2Pit and Z2Peak. As their names imply, Z2Pit is the vertical difference in elevation between any given
cell in a DEM and the depression (pit) cell to which it is connected by a path of simulated surface flow and
Z2Peak is the equivalent vertical difference in elevation between a cell and the peak cell to which it is
connected by a flow path. Add the two measures together and one gets a value for the total vertical change
in elevation from pit to peak for all cells that drain to the same depression and are below the same local
peak. This is a reasonably useful measure of local relief that applies to large groups of cells that are all part
of the same local catchment. LMES experimented with using other measures of local relief (Stream to
divide elevation change-ZSt2Div) and slope length (Slope length stream to divide (LSt2Div) or pit to peak
slope length (LPit2Peak)), but settled on using the measure Zpit2Peak with a threshold value of 100 m to
differentiate areas of low relief (< 100 m pit to peak) from areas of high relief (> 100 m pit to peak). Areas
of high relief tended to exhibit long continuous slopes that developed high values for diffuse upslope
contributing area and wetness index, two terrain derivatives that were used extensively in the LMES DSS
knowledge base rule tables. Conversely, areas of low relief tended to develop and exhibit significantly
lower values for diffuse upslope accumulation area and wetness index. These variables were used as
surrogates for approximating relative landform position and relative moisture status in the LMES DSS rule
bases. They were being used in a relative sense, but their absolute values were only comparable within
areas that had landscapes with similar relief and slope lengths. Therefore, it was necessary to be able to
apply different threshold values to these variables in areas with different landscapes in order to achieve
comparable results in recognizing relative landform positions or relative degrees of wetness. The automated
modeling did an acceptable job of partitioning the entire Quesnel PEM map area into regions that could be
thought of as exhibiting high versus low relief for which different LMES KB rule bases could be developed
and applied. The amount of effort required to prepare the initial map of high versus low relief areas was
less than two days, including time spent investigating alternative measures to use to accomplish this
differentiation. The cost, at less than 0.1 cent per hectare, was so low that this task was not identified and
charged for separately in the budgeting for the project.
It was necessary to define and map three new physiographic/landscape classes after the modeling workshop
in February, 2004. Maps for these three new classes had to be created quickly in response to suggestions
from the Regional Ecologist. There was not sufficient time to experiment with different alternatives by
which these areas might be extracted automatically from available digital data sets. The only feasible
approach, given the extremely tight time lines imposed by the need to complete all project mapping by mid
March of 2004, was to identify the location and extent of all three of these new areas visually and to
digitize polygons depicting their extent manually. This manual recognition and digitizing was done over
part of a single day toward the end of the week during which the modeling workshop was held. With input
and advice from the Regional Ecologist, LMES identified and digitized polygons that outlined all areas that
the Regional Ecologist considered to exhibit landscape patterns consistent with any of a) high likelihood of
elevated risk of frost hazard, b) very level areas with long slopes or c) very steep areas with exceptionally
long slopes. Polygons for these three new classes of physiography/landform were overlaid on the original
polygons of high versus low relief to replace these original classes wherever one of the three new classes
was defined. The new, revised map of physiography/landform classes portrayed the spatial distribution of
all 5 classes listed in Table 4.
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
8
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
3.1.4 The LMES prepared map of areas of elevated frost hazard
This map, that depicted the location and extent of areas deemed likely to have an elevated risk of frost, was
digitized manually following input and suggestions made by the Regional Ecologist. The Regional
Ecologist had background knowledge about those locations where he was aware that frost was more likely
to present a hazard and of areas where increased concentrations of Black Spruce were indicative of
increased incidence of frost. Once the Regional Ecologist had identified any particular region where he felt
there was a reason to expect elevated levels of frost hazard, it was found that visual review of several of the
terrain derivatives that had been computed from the 25 m DEM for the study area proved helpful in
deciding where to digitize polygon boundaries for areas of elevated frost hazard. Graphical backdrops
consisting of colored and hillshaded depictions of the variables wetness index (Qweti), log of diffuse
upslope area (LnQarea) and slope gradient (Slope) helped to reveal the most likely geographic locations of
areas of elevated risk of frost. These areas were seen to occur mainly within the lowest portions of large
structural basins in areas with low slope gradients and low relief that exhibited complex topography with
relatively short slopes and frequent reversals of slope (e.g. low relief hummocky to undulating). These
conditions were highlighted in the backdrop images of Qweti and LnQarea which helped to guide the
manual placement of polygon boundaries outlining areas of elevated frost hazard. The Regional Ecologist
reviewed the polygon boundaries as they were being digitized and offered suggestions for changes or
confirmed the correctness of boundaries that seemed reasonable to him.
Once digitized, the LMES prepared map of areas of elevated frost hazard was simply treated as another
kind of physiography/landform class and used to define unique regions within BEC subzones or variants
within which separate knowledge bases (KBs) and LMES DSS rules could apply. The separate rules
permitted the LMES DSS procedures to assign different Site Series to particular landform positions than
would be assigned to the same landform positions in areas that did not experience elevated risk of frost.
3.1.5 The LMES prepared map of “classification regions”
The spatial information contained in the manually prepared maps of BEC subzones, physiographytopography, frost hazard and parent material texture described above was mainly used to produce a final
grid map that depicted “classification regions” as sub-divisions of BEC subzones and variants.
The grid map of “classification regions” treated the various input layers as Boolean constraints that
determined whether a particular set of classification rules would apply within any given “classification
region” or would not apply. These Boolean constraints imposed the hierarchical classification logic
described in section 2.0. This logic required BEC subzones and variants, as depicted on the updated
localized “Big BEC” map, to be further sub-divided according to considerations of physiography, landform
type and local climate variation (the physiographic class map including areas of elevated frost hazard) and
then to be further sub-divided according to considerations of parent material texture (here only coarse
versus medium to fine textured as mapped by JMJ). Computationally, the overlaying and merging of the
three different source maps was accomplished as follows.
Firstly, the polygon map of unique integer ID numbers assigned to each of the BEC subzones and variants
(Table 1) was converted into a grid map with a grid size of 25 m and dimensions (rows and columns)
identical to the 25 m raster DEM grid used to establish the analysis extent for the Quesnel PEM study area.
This grid map of unique integer ID numbers was multiplied by 100 to convert numbers such as 21 to 2100
and so on to produce a temporary map named MAP 1.
Secondly, the polygon map of physiographic classes (including the frost hazard zone class) was also
converted to a temporary grid map (MAP 2) with a grid size of 25 m and dimensions identical to the DEM
and BEC zone grid (MAP 1) maps. The integer values associated with each of the physiographic classes
(e.g. 20, 30, 40, 50 & 60) were maintained and assigned to every grid cell located within a polygon labeled
with a given physiographic class ID number.
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
9
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
Thirdly, the JMJ produced polygon map of parent material texture, depth and exception classes was used to
create a temporary grid map of parent material texture classes (MAP 3), also with a grid size of 25 m and
dimensions identical to the other grid maps described above. For this grid map, all areas mapped by JMJ as
being expected to contain coarse textured parent materials received an integer value of 1 while all other
areas were assigned an integer value of 0 (e.g. were assumed to be medium textured).
Finally, the three grid maps of identical grid size and dimensions were simply added together to create a
single new grid map of integer values according to ZONE MAP = MAP 1 + MAP 2 + MAP 3.
This new ZONE MAP assigned a unique integer number to every grid cell. The unique integer number
identified a specific combination of BEC subzone or variant, physiographic class and parent material
texture for every cell. For example, 2241 = BEC zone 22 (SBPSdc) + physiographic class 40 (frosty and
low relief) + texture class 1 (coarse textured).
A complete list of all unique integer ID numbers generated by this process and used to define
“classification regions” for the Quesnel PEM is presented in Table 5. The grid map of unique ID numbers
for “classification regions” was exported from ArcView and reformatted into a DBF table with the name
Q01_Zone2 that was used as an input file in the LMES DSS classification program FacetMapR. The ZONE
file is used to tell FacetMapR the file name of the KB rule file to consult to read and apply classification
rules for any particular grid cell in the data base. The FacetMapR program requires the existence of a pair
of KB rule files named Arulennnn and Crulennnn for each and every unique “classification region” integer
ID number, where nnnn is the unique integer ID number. LMES prepared a separate set of KB rule files
for each and every unique ID number listed in Table 5. Each pair of rule files, for each unique
“classification region”, contained fuzzy logic definitions for the specific set of ecological classes (Site
Series) defined to occur in a particular classification region (Crulennnn) and contained definitions for the
fuzzy attributes that were used in the fuzzy classification of each specified Site Series (Arulennnn).
Table 5. Complete list of all unique ID numbers for "classification regions" for the Quesnel PEM
ID
BEC zone
Physiography
Texture
BEC zone
Physiography
Texture
2120
MSxv
Low Relief
Medium
ID
2720
SBSdw2
Low Relief
Medium
2121
MSxv
Low Relief
Coarse
2721
SBSdw2
Low Relief
Coarse
2130
MSxv
High Relief
Medium
2730
SBSdw2
High Relief
Medium
2131
MSxv
High Relief
Coarse
2731
SBSdw2
High Relief
Coarse
2150
MSxv
Long gentle slopes
Medium
2740
SBSdw2
Frosty - low relief
Medium
2151
MSxv
Long gentle slopes
Coarse
2741
SBSdw2
Frosty - low relief
Coarse
2160
MSxv
Long steep slopes
Medium
2750
SBSdw2
Long gentle slopes
Medium
2220
SBPSdc
Low Relief
Medium
2760
SBSdw2
Long steep slopes
Medium
2221
SBPSdc
Low Relief
Coarse
2820
SBSmc2
Low Relief
Medium
2230
SBPSdc
High Relief
Medium
2821
SBSmc2
Low Relief
Coarse
2231
SBPSdc
High Relief
Coarse
2830
SBSmc2
High Relief
Medium
2240
SBPSdc
High Relief
Medium
2831
SBSmc2
High Relief
Coarse
2241
SBPSdc
High Relief
Coarse
2840
SBSmc2
Frosty - low relief
Medium
2250
SBPSdc
Long gentle slopes
Medium
2850
SBSmc2
Long gentle slopes
Medium
2260
SBPSdc
Long steep slopes
Medium
2920
SBSmc3
Low Relief
Medium
Coarse
2320
SBPSmc
Low Relief
Medium
2921
SBSmc3
Low Relief
2321
SBPSmc
Low Relief
Coarse
2930
SBSmc3
High Relief
Medium
2330
SBPSmc
High Relief
Medium
2931
SBSmc3
High Relief
Coarse
2331
SBPSmc
High Relief
Coarse
2940
SBSmc3
Frosty - low relief
Medium
2340
SBPSmc
Frosty - low relief
Medium
2950
SBSmc3
Long gentle slopes
Medium
2341
SBPSmc
Frosty - low relief
Coarse
3020
SBSmh
Low Relief
Medium
2350
SBPSmc
Long gentle slopes
Medium
3021
SBSmh
Low Relief
Coarse
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
10
Quesnel PEM Predictive Ecosystem Mapping Knowledge Base and Attribute Summary
2420
SBPSmk
Low Relief
Medium
3030
SBSmh
High Relief
Medium
2421
SBPSmk
Low Relief
Coarse
3031
SBSmh
High Relief
Coarse
2430
SBPSmk
High Relief
Medium
3050
SBSmh
Long gentle slopes
Medium
2431
SBPSmk
High Relief
Coarse
3051
SBSmh
Long gentle slopes
Coarse
Medium
2450
SBPSmk
Long gentle slopes
Medium
3320
SBSdk
Low Relief
2451
SBPSmk
Long gentle slopes
Coarse
3330
SBSdk
High Relief
Medium
2460
SBPSmk
Long steep slopes
Medium
4020
At
Low Relief
Medium
2461
SBPSmk
Long steep slopes
Coarse
4030
At
High Relief
Medium
2520
SBPSxc
Low Relief
Medium
4130
ESSFmv1
High Relief
Medium
2521
SBPSxc
Low Relief
Coarse
4220
ESSFxv1
Low Relief
Medium
2530
SBPSxc
High Relief
Medium
4230
ESSFxv1
High Relief
Medium
2531
SBPSxc
High Relief
Coarse
4250
ESSFxv1
Long gentle slopes
Medium
Medium
2550
SBPSxc
Long gentle slopes
Medium
4260
ESSFxv1
Long steep slopes
2551
SBPSxc
Long gentle slopes
Coarse
5020
IDFdk3
Low Relief
Medium
2620
SBSdw1
Low Relief
Medium
5021
IDFdk3
Low Relief
Coarse
2621
SBSdw1
Low Relief
Coarse
5030
IDFdk3
High Relief
Medium
2630
SBSdw1
High Relief
Medium
5220
IDFxm
Low Relief
Medium
2631
SBSdw1
High Relief
Coarse
5221
IDFxm
Low Relief
Coarse
5230
IDFxm
High Relief
Medium
5231
IDFxm
High Relief
Coarse
Future revisions to the Quesnel PEM could result in additions or deletions to the list of unique ID numbers
for “classification regions” presented in Table 5. The list is not fixed and varies in response to ad-hoc
decisions to create new sub-divisions based on different physiographic classes or parent material textures.
LandMapper Environmental Solutions Inc.
7415 118 a Street NW, Edmonton, AB, T6G 1V4
(780) 435-4531
email: [email protected]
11