A Comparative Study Of Analytical Tools For Strategic & Tactical Forest Management Planning

2,047 views

Published on

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.

Published in: Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
2,047
On SlideShare
0
From Embeds
0
Number of Embeds
6
Actions
Shares
0
Downloads
35
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

A Comparative Study Of Analytical Tools For Strategic & Tactical Forest Management Planning

  1. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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. 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

×