The purpose of this project was to compare the current forest management planning process in New Brunswick with an alternative based largely on computer software tools. Two New Brunswick Crown Licenses were used as case studies: forest classification schemes, yield estimates and assumptions about forest dynamics used in the study were identical to those used by each of the participating Licensees in their respective forest planning models. However, Remsoft staff used Woodstock to develop a strategic forest management schedule, Crystal to generate potential harvest blocks and Block to develop a spatially feasible block harvest schedule.

1. A Comparative Study Of
Analytical Tools For
Strategic & Tactical Forest
Management Planning
Executive Summary
The purpose of this project was to compare the current forest management
planning process in New Brunswick with an alternative based largely on
computer software tools. Two New Brunswick Crown Licenses were used as
case studies: forest classification schemes, yield estimates and assumptions
about forest dynamics used in the study were identical to those used by each
of the participating Licensees in their respective forest planning models.
However, Remsoft staff used Woodstock to develop a strategic forest
management schedule, Crystal to generate potential harvest blocks and Block
to develop a spatially feasible block harvest schedule. All analyses were
Final Report conducted on a Gateway 2000 4DX2-66V microcomputer running MS-DOS
6.2, with 16MB of memory and a 425MB hard drive using Doublespace
compression.
To complete the analyses for both Licenses required six person-months of
submitted to
labor. This included time to obtain and process GIS map coverages for each
New Brunswick License, to develop models and determine alternative solutions for them, and
Forest Research to write the final report. The bulk of the work lay in writing custom software
Advisory to facilitate conversion of Licensee provided input files to, to update and
Committee obtain attribute and topological information from the GIS data files, to
automate data manipulation between planning models and to present mapped
April 1994
solutions. With the system of procedures currently in place, we estimate that a
Licensee familiar with the programs could complete the tasks undertaken in
this study in approximately one week.
Results of the Woodstock runs showed that linear programming has
significant advantages over simulation, in particular the ease with which
outputs and activities can be constrained and the ability to readily control
indirect outputs such as wildlife habitat. Overall, the Woodstock models
yielded 9% to 22% increases in strategic allowable cut estimations over the
baseline values provided by the Licensees. Moreover, mature conifer
furbearer habitat requirements were met in all planning periods of the
Woodstock analyses, unlike their baseline counterparts.
Despite a few shortcomings related to the ability to simultaneously address
multiple harvest actions, Crystal appeared to work well, and compared to a
manual approach, it was a vast improvement. Generating 10 alternative block
layouts for each of 10 different blocking parameter specifications on one
License required approximately 7 hours of processing using Crystal; on the
other License, the time required was just over 22 hours. Depending on the

2. License and the blocking parameters used, Crystal was able to generate blocks for 50% to 95% of the area
scheduled for harvest in the first 7 periods.
Block performed adequately but required a significant amount of effort and custom programming to be able to use
it efficiently. Despite its awkward input file structure and the inability to easily accommodate non-clearcut harvest
prescriptions, Block produced good results in this study. For one particular block layout, the schedule developed by
Block projected an average harvest just 2% lower than the harvest level of the Licensee's pre-block baseline
analysis. Adjacency delays and maximum opening size constraints were never violated in any of the solutions,
although License 4 was more prone to adjacency conflicts and thus a higher percentage of blocks remained
unharvested by Block than was the case for License 4.
Whether or not Woodstock, Crystal or Block are used operationally in New Brunswick, it seems inevitable that
software solutions will be adopted for harvest scheduling and blocking in the future. Several areas of potential
problems in the future are identified which may become an issue as technology makes it faster and easier to explore
more alternatives and include more constraints in forest planning models. Therefore, in addition to suggesting
future modifications to Crystal and Block, recommendations include:
• Conducting a benchmarking exercise for the two Licenses in this study using the approach taken by the
FORMAN 2000 group.
• Providing Licensees with detailed attribute and topological data from the provincial geographic
information database.
• Establishing consistent guidelines for planning procedures and articulating requirements and regulations
in terms that are not dependent on a particular frame of reference.
• Implementing linear programming techniques as part of a broad-based planning methodology.
• Implementing joint planning activities for Crown and Licensee freehold lands.

3. This three step process is implemented in New
Background Brunswick using FORMAN+1 and FORMAN+2 for
the strategic analysis and manual procedures for the
Problem statement remaining steps. Unfortunately, it is these steps
Comprehensive forest management has always been which are the most data and labor intensive and
difficult because of the magnitude of the problem. several problems arise:
Early attempts concentrated on attaining relatively
straightforward management goals such as forest • It is not uncommon for licenses to spend over 60
regulation, from which the related goals of sustained person-days to achieve an initial harvest block
yield and perpetual supply are realized by definition. schedule that meets all of the spatial, temporal
Over time, the notion of the regulated forest has and harvest flow constraints.
largely been discarded due to the dynamic nature of • Because of the high cost of finding and
forest ecosystems and the inability to rationalize evaluating each solution manually, exploring
methods such as area control with the need to adapt alternatives is rarely undertaken and the impact
changing social and economic demands. However, of increased spatial and temporal constraints on
the goals of sustained yield remain, although they are wood supply is not addressed.
much wider in scope than simply timber volumes.
Thus the problem of forest management has become • The decision criteria used to obtain a particular
one of deciding what actions to perform on what part schedule are usually not explicit enough to make
of the forest and when, to provide the desired the process repeatable.
benefits. Because some actions are incompatible with
the production of some products, trade-offs exist for Licenses used in the case study
virtually all combinations of actions. Two Licensees agreed to be participants in this study:
In New Brunswick, the Crown Lands and Forests Act Valley Forest Products (VFP) (License 8) and
requires licensees to produce an 80 year strategic Miramichi Pulp and Paper (MPP) (License 4). Each
plan, a 25 year management plan, and a 5 year Licensee agreed to provide us with information used
operational plan. The purpose of the strategic plan is in the preparation of their most recent Crown Land
to define ways to meet long term management Management Plan. This information included the
objectives, while the management and operational class and yield information used in their FORMAN
plans are location specific and details geographic analyses, along with lists of stands comprising each
locations of proposed activities. Currently, strategic class. The Department of Natural Resources and
plans are developed using a stratum or stand-type Energy (DNRE) provided us with 1988 vintage
based approach for determining periodic harvest Forest Development Survey (FDS) base maps and
levels and management prescriptions. watercourse buffer/deer wintering area overlay files
in ArcInfo export format.
However, since spatial factors (stand location,
minimum and maximum harvest block sizes, We decided to begin the study with License 8 since it
maximum opening size, adjacency delay was a more fragmented land base with a more
requirements) are not considered, following the heterogeneous forest classification and fairly
stratum based harvest schedule is unlikely to produce complex management objectives. The rationale for
a feasible management or operational plan. Instead, this was to test the worst case scenario – if
the stratum based harvest schedule is used as the procedures could be developed for converting data
basis for delineating sufficient numbers of harvest for this License, then it would be relatively simple to
blocks to generate a block harvest schedule for a 25 do so for other Licenses with less complicated
year planning horizon. In the process of generating planning problems.
blocks, deviations from the stratum based schedule Valley Forest Products subdivided the License 8
are necessary to comply with adjacency constraints forest area into different capability classes
and to even harvest flows. Once a feasible block (unrestricted versus restricted access, softwood
harvest schedule has been found, it must then be versus hardwood, even-aged versus uneven-aged),
validated by incorporating it into the strategic plan to resulting in six individual FORMAN+1 models, plus
ensure long term sustainability. If the resulting long individual models for deer wintering areas. Because
term harvest level is unacceptable, adjustments to the of the need for regular flows of softwood and
block harvest schedule must be made until long term hardwood products, and to avoid the negative
sustainability is ensured. allowable cut effect due to subdividing the forest, we
decided to build a single Woodstock model which

4. would encompass all the different capability classes Hardware and software tools used in
except the deer wintering areas. the case study
In contrast, License 4 is largely dominated by All analyses conducted in this study were performed
softwood forest with large tracts of contiguous on a Gateway 2000 personal computer with an Intel
Crown land. Both the Licensee and sub-Licensees are 486DX2-66 processor, 16MB of memory and a
primary softwood users and hardwood utilization is 425MB IDE hard disk. To perform the map import
fairly low hence the management objectives tend to and overlay procedures, we used pcArcInfo Version
be rather consistent for all parties. In addition, 3.4D; other database manipulations were performed
License 4 has fairly large mature conifer furbearer using FoxPro 2.0. In addition, numerous conversion
habitat (MCFH) requirements compared to License 8, programs and utilities were developed by Remsoft
which is in a different wildlife zone with more Inc. as a part of our own research and development
emphasis on deer wintering areas. Unlike Valley program, including a polygon adjacency scan and
Forest Products, Miramichi Pulp and Paper used a utilities to draw and color code map sheets by harvest
single FORMAN+1 model for the unrestricted land period.
base plus additional models for deer wintering areas.
Valley Forest Products is the wood procurement Woodstock
agency for the Ste. Anne-Nackawic pulp mill, a Woodstock is an MS-DOS based forest modeling
hardwood mill which uses minimal amounts of system developed by Remsoft Inc. to conduct forest
softwood during processing. However, License 8 planning analyses, including harvest scheduling.
must also supply a number of sub-licensees, the Woodstock models can be inventory projections,
majority of which are softwood users, primarily Monte-Carlo simulation models or linear
interested in spruce-pine-fir saw material and spruce- programming (LP) models. Because of the very
fir pulp. One of the major problems faced by VFP is powerful constraint capabilities of LP, we decided to
maintaining a balance between the hardwood needed formulate the strategic wood supply analyses of both
by the pulp mill, and the softwood fallout arising Licensees as linear programs. A brief overview of
from harvesting in mixed wood stands. Simple linear programming is given in Appendix 1.
maximization and/or constraining of a single product
output leads to unacceptable fluctuations in the flow
of other product outputs. Crystal
Crystal (Walters, 1991) is an MS-DOS computer
Miramichi Pulp and Paper manages two Crown program, developed at the University of New
Licenses in north-eastern New Brunswick. License 4 Brunswick which is designed to allocate harvest
is comprised largely of lands bordering the upper prescriptions from a stratum-based harvest schedule
Miramichi River basin. Unlike License 8, much of to individual stands thereby providing a spatial
the forest is comprised of softwood species, primarily configuration for part of a strategic management
spruce and fir. In general, softwood pulp and log plan. Crystal allocates prescriptions on a stand by
material is of primary importance with a much stand basis, and thus the blocks it generates are only
smaller demand for hardwood material. precursors to final operational blocks. Blocking
The two License boundaries encompass roughly the parameters such as block size and allowable
same area: License 8 is distributed over 135 Forest deviations from the strategic schedule are controlled
Development Survey (FDS) map sheets, License 4 by the user. A brief overview of the Crystal
over 132. However, the Crown land portion of algorithm is given in Appendix 2.
License 8 (126 157 ha) is substantially less than
License 4 (356 871 ha); on License 8, much of the
Block
Crown land base is made up of Crown woodlots and
small tracts, as opposed to License 4 which is Block (Dallain, 1989), also an MS-DOS computer
essentially one large tract of contiguous Crown land. program developed at the University of New
The average stand size on License 8, after overlaying Brunswick, determines spatially feasible block
watercourse and exclusion zone buffers, was harvest schedules under opening size, adjacency and
somewhat smaller than the average on License 4 (2.8 harvest flow constraints. Block uses a Monte-Carlo
ha and 3.2 ha respectively). integer programming (MCIP) algorithm to generate
many alternative solutions to the block harvest
scheduling problem. By retaining those feasible
solutions with the highest objective function values,

5. Block can generate very good, near optimal solutions would need to be modified; a new area file could
in a relatively short time. Maximum opening size, be produced in minutes and the linkage to
adjacency delay and harvest flow constraints can all component stands would necessarily be
be specified by the user on a global basis as well as maintained,
for individual management units and habitat zones. A
brief overview of Block is given in Appendix 3. • in the future when a new round of management
plans is implemented, the work done to associate
ages and yield
Methodology and Results NOTE: Both of the
curves is saved;
Licensees provided us
with forest class files. In rather than go
Development of strategic harvest order to be certain that through the
schedules using Woodstock every stand was
process of
accounted for, with no
The automated blocking procedures used in the possibility of duplication or individually
Crystal and Block programs require topological omission, we embedded assigning stands to
information about the arrangement of stands across the landscape themes forest classes, the
directly into the PAT files.
the forest: what forest class each stand belongs to, information used
what stands are adjacent to each stand, and the size of in the previous
each stand. Since these data are readily available planning cycle can
from GIS data files, we decided to combine the simply be updated.
stratified forest information embodied in the
Licensee's models with the stand level information Using a combination of visual inspection and
provided in the forest cover and exclusion zone programming, we devised a consistent classification
coverages from ARC/INFO. Since the ultimate goal scheme for both Licensees, where unique 4 or 5 part
of the study was to automatically produce pseudo- labels were assigned to each forest class; each part of
blocks for block harvest scheduling, we decided to the label was designated a landscape theme. A
begin the strategic planning process with a spatially- custom program was written to modify the polygon
referenced forest database to facilitate disaggregation attribute table (PAT) files in each coverage. New
later on. fields added to the PAT files included one for each
landscape theme used to classify the forest, one to
uniquely identify every polygon within the forest,
Building the Classification Schemes and fields to assign block numbers and harvest
We examined the model input data provided to us by periods later in the process. Once the block numbers
each of the Licensees to determine how they and harvest periods are incorporated into the GIS
stratified their forests. Both used similar database, it is trivial to produce maps of the block
classification schemes (working group, site, harvest schedules for visual inspection.
silvicultural status and management unit), however
Valley Forest Products divided the forest into several Accounting for watercourse buffers and
capability classes with a separate model devoted to
exclusion zones
each one. FORMAN+1 allows users to assign
descriptive names to yield curves and forest classes, Next, we overlaid the forest coverages with
but these names need not be unique, nor do they have coverages of watercourse buffer and wildlife
to correspond to one another. Instead, FORMAN+1 exclusion zones. The overlay process combined the
uses a numerical encoding format to match forest forest cover attributes with the buffer/wildlife
classes to yield curves. One disadvantage of this attributes to create a new set of maps. Because many
approach is that the codes themselves have little of the polygons in each coverage were not part of the
meaning, and the process of checking for errors in productive land base, we decided to use a re-select
meaning is difficult. Therefore, we decided to build operation to remove all of the ineligible stands to
a classification scheme for the Woodstock models reduce the disk space requirements to store all the
directly into the GIS database rather than simply maps. Using a batch process to conduct the initial
convert the numerical encoding structure of the overlays followed by the re-select operation, it took
baseline models. There are two major advantages to more than 20 hours of processing to complete each
this procedure: License. The resulting coverages included all Crown
land, with attributes from both the FDS and buffer
• should a change in the classification scheme be coverages.
necessary at the GIS level, none of the other
steps to produce an area file for Woodstock

6. Model formulation - dynamics Model formulation - LP constraints
After the new maps had been created, we merged all The most difficult task in formulating the
of the individual PAT files into a single attribute management problems of the two Licensees as linear
table. On the basis of landscape attributes and stand programs was establishing constraints. The
age, we used a database report writer to group the underlying principle of simulation models is trial and
individual stands into unique classes and create a error: you tell the model what to do and it reports the
Woodstock analysis area file. Then, using a custom results. The approach depends on the analyst's ability
program written for the task, we converted the to deduce the impacts of various changes and
baseline input files to Woodstock format: yield curves implement controls which produce a desired result.
were formatted in Woodstock format, harvest and With a linear programming approach, you tell the
silvicultural actions were defined using Woodstock model what kind of solution you want and it reports
syntax, and the baseline transition response file was the best means of accomplishing it. In effect, the
converted to Woodstock syntax using the new roles of analyst and model are reversed, with the
classification system. analyst providing the bounds for the solution space
and the model determining the course of action.
Once the major sections of the Woodstock model
were in place, we manually edited the files to remove Because of the long history of simulation modeling
redundancies and to structure the constraints and in New Brunswick, regulations and policy have come
objective functions for to reflect the modeling paradigm of FORMAN. For
the linear programming NOTE: A nondeclining example, we were told that the minimum requirement
formulation. The yield constraint sets up a for gross mature conifer furbearer habitat (MCFH)
conversion takes only series of linkages was based on the ratio of gross to net MCFH at the
between planning periods
minutes to complete, but where the output level of low point in the projected growing stock. We
the manual editing any period must be recognize that this determination is based on past
process can take a few greater than or equal to experience with FORMAN projections and is a
the output level of the
hours, depending on the reasonable approach for this type of model.
previous period.
amount of streamlining However, LP models require fixed quantities for
desired. Once the constraints, either single numbers (i.e. X ≥ 30) or
procedures were fixed proportions of
finalized, converting a FORMAN+1 data set to a another quantity (i.e.
NOTE: The perpetual
Woodstock model structure took roughly one day. timber harvest constraint X ≥ 30% of Y) and a
The value in being able to conduct a forest
assumes that if the ending specific period for
inventory is at least equal applying the
management scheduling analysis within a single to the average inventory
model framework should not be underestimated. over the entire planning constraint; the
Many of the difficulties associated with forest horizon, then a regime of requirement for
harvest and silviculture MCFH, as stated
management planning arise because of competing similar to the one used
resources and co-production of outputs. For example, during the planning
earlier, provides
it is difficult to produce hardwood pulp by horizon should be feasible neither piece of
clearcutting mixedwood stands without also
for all future periods. information.
generating softwood pulp. Conversely, the Furthermore, it is
production of mature conifer furbearer habitat possible to formulate a
competes with softwood volume production since the LP model where the
same development types furnish both outputs. The growing stock is at a minimum in any desired period,
only way around these difficulties is through trade- or one which does not exhibit a dip in the growing
offs – judicious selection of activities and their stock at all.
timing to best meet In order to produce harvest schedules which were
multiple objectives. By NOTE: An overview of reasonable approximations of those produced by the
separating the various linear programming Licensees, we constructed Woodstock models with
components of the forest harvest scheduling
models is given in the
constraints which we felt captured the intent of
into discrete planning appendix. provincial regulations and Licensee objectives. To
models, it is impossible ensure that silvicultural activities were maintained at
to make these types of required levels over the planning horizon, we
trade-offs. imposed a perpetual timber harvest constraint in the
final planning period. Without such a constraint, the

7. optimal solution will produce just the amount of harvest of total softwood volume only. Valley Forest
inventory in the last periods to sustain the required Products also projected significant harvest volumes
harvest level. However, such inventory levels would from uneven-aged management. Since these
not likely result in sustainable harvests beyond the projections originated in FORMAN+2, we simply
end of the planning horizon. took the results of the FORMAN+2 runs and coded
the outputs as time dependent yields in Woodstock.
By examining the solutions found by the Licensees,
Although the Woodstock model had the option of
we were able to determine a minimum ratio of gross-
implementing the unevenaged management
to-net MCFH area for a specific period. To
prescriptions, it could not change the harvest levels
approximate the wildlife habitat requirements on
arising from these prescriptions. Therefore, the
each License, we established two constraints. The
unevenaged volume components are exactly the same
first constraint guaranteed that the area of MCFH-
as those reported by Valley Forest Products.
eligible age classes within the specified zones did not
fall below initial values for the first seven planning While it would have been possible to maximize total
periods. In all subsequent planning periods the gross volume over the planning horizon, this would have
MCFH area was constrained to be at least a fixed placed as much emphasis on harvest volumes from
area: this minimum area was determined by the last planning period as the first. Furthermore, the
examining the gross MCFH area in the period where first period harvest may have been reduced so that
the growing stock was at a minimum in the baseline additional volumes could be harvested in later
analyses. periods. Neither of these outcomes reflects Crown or
company objectives for forest management planning
Valley Forest Products expressed a need to control
and so we limited the objective function to the first
the flow of both primary and secondary products.
period.
Since FORMAN+1 does not provide a means of
directly controlling secondary product flows, the In keeping with Provincial policy on silviculture, we
License 8 forest was subdivided into capability did not place any constraints on silvicultural
classes based on the predominant product harvested activities. DNRE regulations stipulate that the
from each forest type. This approach allows you to Licensee must perform the level of silviculture which
set the predominant output as the primary product will maximize the allowable cut effect. With an
and control it, however all other outputs remain as objective function to maximize first period harvest
fall-out products. The net result may be less variation and concurrent flow constraints on the outputs being
overall in periodic output levels, but there will still be maximized, the Woodstock models determine the
some variation due to fallout products. Furthermore, maximum allowable cut effect by default.
a negative allowable cut effect can be expected Furthermore, only the silviculture which contributes
because of the subdivision of the land base. to an increase in first period harvest is performed;
additional silviculture that could increase inventory
For the License 8 model, we implemented
but not increase the first period harvest is not done.
nondeclining yield constraints on total harvest
Although LP models are efficient in finding this type
volume, softwood pulpwood/logs, and mixed-
of solution, the marginal cost of producing this wood
hardwood pulpwood. Other product flows were not
may be very high.
directly constrained but because they were
components of total harvest volume the harvest levels
of these products were bounded by the non-declining Solving the Woodstock Models
yield constraints. For the License 4 model, Although the land base of License 4 is significantly
nondeclining yield constraints were placed on total larger than License 8, the complexity of the License
softwood volume and total hardwood volume. 8 model resulted in a LP matrix more than twice the
For both Licenses, we formulated objective functions size of the License 4 matrix. Furthermore, whereas
which represented the major product demands from the License 4 LP solved in about 30 minutes on our
the License. For License 8, the Licensee requires computer, the License 8 LP required nearly 5 hours
hardwood material but the sub-Licensees are to solve on the same machine. Much of the difference
primarily softwood users so the objective function in solution time between the two is due to the greater
maximized first period harvest of total softwood and number of constraints present in the License 8 model;
hardwood volume from even-aged and uneven-aged LP solution time is particularly sensitive to the
silvicultural prescriptions. For License 4, both the number of constraints.
Licensee and sub-Licensees are primarily softwood
users so the objective function maximized first period

8. Results of the Woodstock models Harvest Flows - FORMAN+1
Determining the model structure, developing the 3000000
conversion and utility programs, updating the GIS
2500000
coverages and producing
the final Woodstock
Harvest level (m3)
2000000
Although it is not visible in HW uneven
model required about the graph, small amounts SW uneven
1500000
four weeks time. of hardwood log volume HW uneven
SW even
However, now that the are produced in later 1000000
periods. Note also the
procedures have been shift toward softwood pulp 500000
developed, it should be production after period 5.
0
possible for users 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Planning period
familiar with both
modeling approaches to Figure 2. Projected harvest levels from Valley Forest
convert a FORMAN+1 analysis to Woodstock within Products' baseline models for License 8.
a day or two. Although evenaged softwood products exhibit
In the case of License 8, the Woodstock model relatively little variation period to period, evenaged
projected an allowable cut significantly higher than hardwood products vary a great deal. Furthermore,
the allowable cut reported by Valley Forest Products despite a trend toward increasing harvest levels in
using FORMAN+1. Linear programming models are later periods, there is a significant lapse in this trend
particularly adept at capitalizing on trade-offs among in the middle periods. In addition, the evenaged
different stand types and across planning periods, a softwood component does not exhibit the increases in
feature of particular value in the highly constrained allowable cut of the Woodstock model
Woodstock model for License 8. The Woodstock model reported an annual harvest in
Harvest Flows - WOODSTOCK
the first period of 279,000 m3 from the evenaged
capability classes, whereas the baseline models
3000000
projected annual harvests in the first period of
2500000 255,000 m3. A comparison of the inventory profiles
of the Woodstock and baseline models showed a
Harvest level (m3)
2000000
HW uneven general decline in inventory levels over time in the
SW uneven
1500000
HW even
baseline runs, while the Woodstock model maintained
1000000
SW even more than double the level of inventory of the
baseline models, despite harvesting more wood.
500000
Inventory Profiles for Evenaged Capability Classes
Growing stock (m3)
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 8000000
Planning period
7000000
Figure 1. Projected harvest levels from Woodstock 6000000
strategic model for License 8. 5000000
Baseline Softwood
Baseline Hardwood
4000000
The evenaged hardwood component includes birch 3000000
WOODSTOCK Softwood
WOODSTOCK Hardwood
and poplar products which were not subject to flow 2000000
constraints. These products are the cause of the minor 1000000
variation in the harvest flows of evenaged hardwood. 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
However, the harvest profile reflects a general Planning Period
increasing harvest level over time, particularly for
evenaged softwood products. The unevenaged Figure 3. License 8 inventory profiles projected by
component harvest flows of the Woodstock model are Woodstock and baseline models. What is particularly
identical to those of the License 8 baseline models. striking about this figure is that the Woodstock model
was able to retain more than twice the inventory of
the License 8 baseline models, while harvesting more
wood.
The harvest profile for License 4 was very different
from License 8. Although the same harvest flow
constraints were used, the flow of total softwood and

9. total hardwood NOTE: An overview of the requiring treatment. Since eligibility for treatment in
components were strictly algorithm used in Crystal the strategic model was based on a forest-wide
even, although there was is given in the appendix. sample rather than stand-level attributes, any stand
a shift toward increasing within an eligible development type may or may not
softwood pulp and decreasing softwood logs in later actually require treatment. Therefore, we assumed
planning periods. that Licensees would implement treatment where
needed.
The allowable cut projected by the Woodstock model
was approximately 787,000 m3 annually; as The Crystal algorithm was designed only for single
compared to an AAC of 647,000 m3 using the entry harvest prescriptions. Although commercial
baseline strategy reported by Miramichi Pulp and thinning is not a single entry harvest, none of the
Paper. The MCFH requirement was satisfied in all treated development types were scheduled for second
planning periods for the Woodstock model, whereas entries during the seven period planning horizon and
it was not met in periods 14 through 16 in the thus the commercial thins could be accommodated.
License 4 baseline projections. Two-pass harvests, however, could not have been
easily accommodated in Crystal or Block. In the
initial runs of the License 8 Woodstock model, a
Harvest Flows - WOODSTOCK
limited amount of two-pass harvesting was also
selected. However, two-pass harvests did not
4000000
contribute a large amount of volume, and because
3500000
Valley Forest Products did not implement two-pass
3000000
harvests in their baseline runs, and because of the
Harvest level (m3)
2500000 HW logs complex workarounds that would have been required
HW pulp
2000000
SW logs
to use Crystal and block, we modified the Woodstock
1500000 SW pulp model to exclude two-pass harvests for License 8. In
1000000 the License 4 model, two-pass harvesting was never
500000 selected.
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Planning period
Adjacency tables
Figure 4. Projected harvest levels from Woodstock
The pcArcInfo topology structure can provide
strategic model for License 4.
information on stand adjacencies within a map
For both Licenses, the Woodstock models yielded coverage, but cannot provide adjacency information
higher allowable cuts than the corresponding baseline across map boundaries. Also, the map sheets
analyses performed by the Licensees. Furthermore, provided by DNRE had not undergone edge-
the optimal solutions found using Woodstock met all matching, a process which guarantees common
planning requirements that were formulated as boundaries between adjacent map coverages. Since
constraints; the baseline models of both Licensees pcArcInfo provides no librarian functions available
appeared to project shortfalls in one or more outputs in the workstation versions of ArcInfo, we developed
during the planning horizon. a custom program to determine stand adjacencies
within and across map boundaries. The output of this
Developing harvest blocks using Crystal program was imported into a xBASE file, duplicate
records were removed and then the file was
restructured as a double entry list. The process of
Harvest treatment tables
generating the adjacency table had to be done only
The harvest schedules developed with Woodstock for once for each License, and required less than an hour
each License were the basis for blocking with to complete on our computer.
Crystal. The report writing capabilities of Woodstock
were used to write an analysis area report for the first
7 periods of the planning horizon. These ASCII files Eligibility tables
were imported into xBASE format data files, one for The eligibility table was simply the common attribute
each License. Only harvest prescriptions (commercial table generated earlier when the map overlays were
thinning or clearcutting) were maintained in the data processed. The only modification required for
file and all other actions were deleted (planting, Crystal was to sort the file on the basis of map and
spacing, senescence). Silvicultural prescriptions were stand number. Preparation of a Crystal input data set
not blocked because of the inability to predict sites

10. required no more than a couple of hours, including to allocate first would have required more thought.
time to generate the adjacency table. We used a 10 ha minimum block size for all
commercial thins and did not allow any timing
deviations whatsoever for either License. Still,
Setting blocking parameters
Crystal was able to allocate virtually all of the area
One of the objectives of the study was to determine scheduled for commercial thin prescriptions. Because
how well Crystal worked under different planning there were no timing choice deviations and only one
conditions. The two Licenses in this study had very minimum block size used for commercial thins, we
different forest structures only ran Crystal once for each License, retaining the
Each block generated by
and management Crystal is shaded using a highest scoring block layout.
objectives. To retain a random color; unallocated
degree of comparability, areas are white. However, The time required to generate 10 alternative clearcut
we decided to apply the adjacent blocks may block layouts for License 8 (35 - 52 minutes) was far
represent identical less than that required for License 4 (123 - 145
same sets of blocking harvest prescriptions and
parameters to both timing choices (see Figure minutes), but the variation between runs was much
Licenses. Although we 9). higher for License 8 than License 4. In total, to
did not know what generate 100 different block layouts for License 8 on
minimum block size our computer required 7 hours, 16 minutes; the 100
would be acceptable to different block layouts for License 4 required 22
each company, we tested minimum block sizes hours, 2 minutes.
ranging from 5 up to 25 hectares in size with target
block sizes double the minimum. To determine the Results of the Crystal block allocation
impacts of timing choice deviations within blocks,
Crystal was much more successful at allocating
we established allowable deviations in timing choices
larger blocks (20 or 25 ha minimum) on License 4
were ±2 periods for type 1 stands , ±1 periods for
than on License 8; for the small blocks, there was
type 2 stands and ±3 periods for use in the cleanup
little difference. In both cases, allowing more timing
routine for the first set of runs. The second set of 5
choice deviations enabled Crystal to allocate more of
runs used the same range of minimum block sizes,
the area schedule for harvest. Furthermore, as the
but allowed ±4 period deviations for type 1 stands,
minimum block size increased, the proportion of
±2 period deviations for type 2 stands, and ±5 period
scheduled area Crystal was able to successfully
deviations in the cleanup routine. In all, 100 block
allocate fell, but at a faster rate on License 8 than
configurations were generated for each License.
License 4. Overall, on License 8 delineating harvest
blocks much larger than 10 ha is problematic because
Stand Eligibility Harvest
significant amounts of area scheduled for harvest
Adjacency Table
Table Treatment Table remain unallocated.
Adjacent Map & Block layouts as a function of size and timing choice deviations
Map & Polygon ID Analysis Area ID
Polygon ID Number of blocks generated
14000
Lic 8 High
Lic 8 Low
12000
Map & Polygon ID Analysis Area ID Treatment Period Lic 4 High
Lic 4 Low
10000
8000
Stand Area Treatment Area
6000
4000
Other fields
2000
Figure 5. Relational structure of Crystal input files. 0
5 10 15 20 25
Minimum block size (ha)
Allocating blocks Figure 6. Area successfully allocated by Crystal for
To accommodate both commercial thinning and each License under various blocking parameters.
clear-cut prescriptions in Crystal, we allocated
commercial thinning to blocks first. Because the area
to be allocated to commercial thins was far less than
clearcuts, we did not see this as a problem; had the
area of commercial thins been comparable to the
clearcut area, the decision as to which prescriptions

11. Allocation Success
Proportion of scheduled area allocated (%)
95
90
85
80
NOTE: An overview of
75
Lic 8 High the algorithm used in
Lic 8 Low
70 Lic 4 High
Block is given in the
Lic 4 Low
appendix.
65
60
55
50
5 10 15 20 25
Minimum block size (ha)
Figure 7. Number of blocks generated by Crystal for
each License under various blocking parameters.
There was a great deal of variation in solutions across Figure 9. Preferred harvest times for individual
block runs (different minimum block sizes or blocks on License 4. The various shadings on this
deviations permitted), but little variation within runs. figure represent the final harvest periods for blocks.
Typically, the overall score values for individual Where two or more blocks may be assigned the same
solutions (a measurement used to penalize large harvest period, they will appear as a uniformly
timing choice deviations) and the number of blocks shaded opening. Stands not eligible for harvest are
allocated were very similar. For example, the number white.
of blocks allocated on License 8 with a minimum
block size of 5 ha using low deviations ranged from
5050 to 5073 with an average of 5061 blocks.
Developing block harvest schedules
using Block
Preparing the Block input files
Four different block layouts for each License were
selected for scheduling. Each time Crystal was run,
the best solution found thus far was saved, as well as
information on the number of blocks generated, the
overall score values, the proportion of area allocated
using a specific timing choice deviation, area
impossible to allocate and area left unallocated. The
solution files are stored as dBASE IV files and detail
Figure 8. License 4 map sheet showing individual the component stands for each block, size of each
Crystal blocks. block, and block adjacencies.
Although Crystal provides most of the information
required by Block, it is not in an appropriate format
to be used directly. Furthermore, Block requires
block volume estimates rather than stand type
estimates of volume. To assist in producing a
properly formatted Block input file, we wrote two
custom programs. The first program reads the
Woodstock input files to obtain yield and analysis
area information. It then produces an intermediate
file, which details per hectare estimates of previously
defined outputs for each analysis area defined in
model. The second program uses this intermediate
file, along with the solution files produced by Crystal
to calculate block volume estimates and write out a

12. properly formatted Block input file. Finally, the block Results of the Block runs
information for the commercial thin blocks is For License 8, the 10 ha minimum block layouts
manually added to the input file. With the assistance yielded about 330 blocks as opposed to about 115 for
of the conversion programs, developing a Block input the 20 ha minimums. Unlike License 4, only one or
file takes just minutes. two blocks at most were left unharvested, regardless
Like Crystal, Block was designed only for single of minimum block size. Again, the high deviation
entry harvest prescriptions. However, the maximum layouts yielded higher average harvests than the low
opening size and adjacency delay parameters can be deviation layouts.
different for different management units or habitat Block Harvest Levels - License 4
zones. Because the commercial thins are not Periodic Harvest Volume (m3)
3500000
considered openings and the final harvest of these
3000000
areas does not occur during the planning horizon, we
2500000
separated the two types of blocks using the
2000000
management zone option. This allowed us to apply a
maximum opening size of 100 ha and a 10 year 1500000
harvest delay for clearcut blocks without restricting 1000000
commercial thin blocks whatsoever. Also, volume 500000
obtained from both harvest prescriptions contribute 0
1 2 3 4 5
to the volume objective, which would not be possible Planning Period
with separate runs for each. For License 8, the 10 ha
minimum block layouts Figure 10. Block
yielded about 330 blocks developed spatially
as opposed to about 115
Block runs for the 20 ha minimums.
feasible harvest
For each run, we restricted the availability of Unlike License 4, only one schedules for License 4.
or two blocks at most
commercial thin blocks to the periods in which they were left unharvested, A comparison of the
were originally scheduled by Woodstock. Clearcut regardless of minimum results from the various
blocks could be scheduled during any of the 7 block size. Again, the high runs showed that
planning periods. To obtain relatively good solutions, deviation layouts yielded
higher average harvests smaller minimum block
an iterative approach was followed. For the first run, than the low deviation sizes in Crystal allow
we applied no limits on individual product flows and layouts. more of the schedule
generated 100 feasible solutions. Then, we examined area to be allocated to
the best solution found, and noted what the lowest blocks than larger
harvest level was for each product over the planning minimum block sizes,
horizon. We then ran Block again with lower limits with concomitant increases in average harvest levels
on each product set to the minimum values found in in the corresponding Block runs.
the previous run. By applying the same procedure 3
or 4 times, we quickly found appropriate lower limits Periodic Harvest Volume (m3)
Block Harvest Levels - License 8
which would yield approximately one feasible 800000
solution for every 100 attempts. Then, we ran Block 700000
once more, using the final lower limits for each 600000
500000
product, to generate 100 feasible solutions. The best Min 10, low
Min 10, high
400000
3 solutions from each run were retained. In most Min 20, low
Min 20, high
300000
cases, generating a final solution set for a particular
200000
block layout required about an hour. 100000
The final step in the process was to match up the 0
1 2 3 4 5
final harvest periods for each block from the Block- Planning Period
generated harvest schedule with the individual stands Figure 11. Block developed spatially feasible harvest
in the master polygon attribute table. A custom schedules for License 8.
program was written to perform this function, which
simply updated the period field with the harvest Mapped solutions quickly illustrate the differences in
period selected by Block. Blocks left unharvested by allocation success between the two Licenses. In
Block were assigned a harvest period of zero. particular, note the fragmentation in the land base,
and the number of watercourse or wildlife buffers
present on the map sheets from the two Licenses.

13. License 4 appears to be more prone to adjacency
conflicts than License 8. In all cases, the number of
blocks left unharvested by Block was proportionally
higher on License 4 than License 8. We presume that
this is so because License 4 is far less fragmented
than License 8, necessarily increasing the likelihood
of adjacency conflicts.
Periodic variations in block harvest schedule
# of blocks Avg block size (ha)
1200 30
1000 25
800 20
Figure 12. Scheduled block layout for a single map 600 15
sheet from License 8.
400 10
200 5
0 0
1 2 3 4 5 6 7
Planning period
# of blocks harvested Average size of blocks
Figure 14. Variation in average block size and
number of blocks harvested for a 10 ha minimum
layout on License 4.
Based on the results found using Woodstock, Crystal
and Block, and the ongoing research into related
approaches, it seems inevitable that software
solutions will be adopted for harvest scheduling and
Figure 13. Scheduled block layout for a single map blocking. Even as technology makes it possible to
sheet from License 4. explore more alternatives and consider more
variables in forest planning models, those same
Presuming that the blocks could be harvested as
capabilities can give rise to several areas of potential
scheduled, the block harvest schedules for each
problems. These issues are discussed in this section.
License resulted in substantial decreases in AAC as
compared to the optimal forecasts from Woodstock.
For License 4, the decreases ranged from 19% to Issues
30% whereas the decreases for License 8 were
between 36% and 56% depending on the minimum Strategic harvest scheduling issues
block size used and the degree of timing choice
deviations allowed. Planning horizons
Using a 10 ha minimum block layout generated by In many jurisdictions, the convention for setting the
Crystal on License 4, the spatially feasible AAC planning horizon is to at least double the average
produced by Block was 635 500 m3 per year; rotation length. The rationale for this is to ensure that
compared to the pre-blocked AAC from the preferred by the end of the planning horizon only wood from
strategy developed by Miramichi Pulp and Paper regenerated stands is contributing to the allowable
staff which was 647 000 m3 annually (a difference of cut. Since long term sustained yield (LTSY) by
2%). definition is based solely on expected regeneration
volumes, the final period harvest is usually a good
The variation in block size period to period was indication of LTSY.
relatively constant for every Block schedule: for
example, using a 10 ha minimum layout from The data presented in the figure comes from a Forest
License 4, the average block size for the seven Management Area in northern Alberta where average
periods ranged from 23.4 ha to 25.3 ha with no rotations range from 80 to 110 years. The objective
violations of the adjacency constraint (see Figure 14). function maximized first period harvest subject under

14. non-declining yield constraints. With a sufficiently
long planning horizon, the harvest level and the
↖ overestimating
sustainable harvest
The data presented in the
LTSY would be equal but the differences shown here figure comes from a Forest levels (refer to Figure
are due primarily to surplus inventory – the models Management Area in 16).
with shorter planning horizons liquidate the surplus northern Alberta where
average rotations range Effect of ending inventory constraints on AAC estimates
at a faster rate thereby increasing the cut. from 80 to 110 years. The
Nondeclining harvest
level (m3/period)
objective function 2250000
maximized first period 2000000
In New Brunswick, the required planning horizon is harvest subject under non-
1750000
80 years, which is less than two average rotations on declining yield constraints.
With a sufficiently long 1500000
License 4 and License 8. Short planning horizons planning horizon, the
generally exhibit higher AAC and lower LTSY harvest level and the LTSY
1250000
values than longer planning horizons (see Figure 15). would be equal but the 1000000
Because the existing inventory can be liquidated in a differences shown here are 750000
due primarily to surplus
shorter time, allowable cuts are usually higher for inventory – the models with
500000
short planning horizons; as the planning horizon shorter planning horizons 250000
lengthens, the existing inventory must last longer, liquidate the surplus at a 0
faster rate thereby 8 periods 12 periods 16 periods 24 periods
until finally regeneration volumes are sufficient to increasing the cut.
sustain the harvest. Figure 16. Allowable
cut estimates for various planning horizon lengths
Volume (m3)
Impacts of planning horizon lengths with and without ending inventory constraints.
60
Harvest Although it is true that a new wood supply analysis
LTSY
50 every 5 years will correct for overestimates in harvest
40
level, there will likely be more variation in allowable
cuts by doing so. One advantage of using longer
30 planning horizons and ending inventory constraints
20
to estimate AAC's linked to long term sustained yield
is for evaluating Licensee management performance.
10
For example, a License which demonstrated
0
maintenance or an increase in long term sustained
5 decade 10 decade
Length of planning horizon
20 decade
yield would generally be considered in compliance
with provincial management objectives; on the other
Figure 15. Changes in AAC and LTSY due to hand, falling LTSY estimates would indicate
planning horizon length. potential problems.
For the hypothetical forest depicted in Figure 15, an
arbitrary ending inventory of 7 million m3 Allowable cut effect
(approximately 50% of initial inventory) was Current New Brunswick policy requires all Licensees
required in the last planning period. As the planning to perform basic silviculture at levels which will
horizon increases in length, more regeneration maximize the allowable cut effect (ACE) – the
volume contributes both to the allowable cut and to immediate increase in harvest due to changed
the inventory. Beyond 24 periods increasing the assumptions about future productivity or utilization
planning horizon makes little difference – in other standards. In attempting to comply, Licensees using
words, the allowable cut is essentially the same as the FORMAN+1 have tried various silvicultural regimes
long term sustained yield. to find the combination which yields the highest
The effects of shorter planning horizons can be offset AAC. However, simulation models are rather poor at
somewhat by imposing an ending inventory finding marginal increases in output and except in
requirement. The perpetual timber harvest constraint relatively simple models, linear programming models
works to counter inventory liquidation and thus are better able to capitalize on silvicultural treatments
ensure harvests beyond the end of the planning and report substantially higher allowable cut effects
horizon. Although it is not a perfect substitute for than corresponding simulation models (Jamnick,
longer planning horizons, it does tend to lower the 1990).
estimated AAC closer the LTSY for the forest. Using a linear programming formulation for License
Otherwise, there is a very real possibility of 8, we were able to find determine a silvicultural
regime which maximized allowable cut effect.

15. However, the cost of this regime is substantially contiguity issue in future periods. Setting aside areas
higher than the one proposed by Valley Forest for the present excludes them from harvesting, but no
Products and yielded only a 9% higher harvest in the attempts are made to locate harvests in specific areas
first planning period. Despite the fact that LP models to create contiguous areas of forest with similar age
nearly always yield higher allowable cut effects, the and species composition. Without some form of
question remains whether such gains are zone-based spatial constraints it is doubtful that
economically viable. suitable habitat areas will be available at the
appropriate times in the future. Moreover, the current
Not only does the
policies on maximum opening size and adjacency
optimal silvicultural Although the Woodstock
model consistently constraints promotes even further fragmentation of
regime result in far
projected less MCFH area the forest.
larger treatment areas than the baseline model
and associated costs, but up to period 11, it always An automated blocking algorithm like Crystal
the fluctuations in met the minimum depends on a strategic harvest schedule to determine
requirement, which the
treatment period to baseline projections failed
eligible stands for harvest in each period. Crystal is
period are likely to do in the last four only able to work within parameters established by
unacceptable from an planning periods. strategic harvest schedule and if that schedule reflects
operational standpoint. dispersed harvesting and fragmentation, so will the
Although constraints on blocking strategy generated by Crystal. The only way
treatment could smooth to counter this and concentrate harvesting would be
out these fluctuations, they do not address the root to deviate from the strategic harvest schedule, the
problem of the ACE policy itself – that it is not exact opposite of what Crystal was designed to do.
economically justifiable, at least for basic
There are aspects of the current planning procedures
silviculture. A more justifiable policy might be to set
related to habitat management which have strong
basic silviculture budgets at the point of diminishing
implications for harvest scheduling. First, the
returns, where further investments no longer increase
eligibility windows for MCFH typically encompass
at the rate of investment. Additional silvicultural
the point of the yield curve where mean annual
investment to improve product quality or to increase
increment (MAI) is culminated. A stand which could
future productivity would be the decision of the
be applied to the MCFH requirements can only be
Licensee.
harvested after it is in decline to maximize its
Silviculture Regime for License 8 membership in the eligibility window; in many cases,
Area Treated (ha)
14000
PCT HW - baseline
PCT SW - baseline
the eligibility window extends beyond the usual
Plant WS - baseline
PCT HW - WOODSTOCK
operability window resulting in the complete loss of
12000
PCT SW - WOODSTOCK
Plant JP - WOODSTOCK that stand for harvesting purposes. The result is that
Plant WS - WOODSTOCK
10000
Plant BS - WOODSTOCK the objective of volume maximization is directly at
8000 odds with fulfillment of the habitat objective. Since
6000
both cannot be simultaneously attained, some form of
4000
trade-off is needed and the analyst needs to
determine its magnitude.
2000
0
Although the Woodstock model consistently
1 2 3 4 5 6 7 8 9
Planning period
10 11 12 13 14 15 16
projected less MCFH area than the baseline model up
to period 11, it always met the minimum
Figure 17. Silviculture regimes for License 8 using requirement, which the baseline projections failed to
Woodstock and baseline planning models. do in the last four planning periods.
Because the Woodstock models were able to make
Wildlife habitat trade-offs across planning periods and among
The mature conifer furbearer habitat (MCFH) silvicultural treatments, the reductions in AAC due to
objectives require contiguous areas of mature MCFH requirements could be minimized. In general,
softwood types. The current policy is to identify such the LP solver selects an appropriate harvest and
areas and preserve them for as long as possible. silvicultural regime to just meet the MCFH
Thereafter, new areas will need to be identified to requirements and nothing more. In contrast,
replace those that are no longer suitable. The problem FORMAN+1 models are rule-based and cannot make
with the current modeling approach used on Crown trade-offs across planning periods. Overall, the
land is that no attempt is made to address the