ALCES Land Use Optimization Tool
Right-click here to download a Word version of this help topic.
Introduction
Historically, the prevailing manner in which ALCES has been deployed on Alberta’s regional landscapes is one of forecasting “defined” future landuse trajectories provided by each of the major industrial sectors (forestry, energy, agriculture, residential, transportation, tourism). The Alberta Land-use Framework has provided clear intent to move to a “management by objective” strategy whereby ecological, economic, and social thresholds will be defined for each region. As such, ALCES must be able to define combinations of land uses that are best able to collectively achieve the stated threshold goals for the region.
“Management by objective” presents some challenges with respect to scenario analysis. If multiple land uses are involved, the range of possible land use combinations (or scenarios) will be very large, likely reaching into the thousands. Manually exploring individual land use scenarios is therefore impractical. Also problematic is that the land use options must be compared with respect to their capacity to achieve ecological and socioeconomic objectives. Given the numerous land use options and multiple objectives typically involved, this can present a substantial analytical challenge. With these challenges in mind, it was determined that additional capacity should be added to ALCES to assist users with the task of identifying land use options best suited to achieve ecological and socioeconomic objectives for their study area. The result is the ALCES Land Use Optimization Tool.
The Land Use Optimization Tool builds upon ALCES’s capabilities to efficiently explore a range of possible land use combinations to identify land use strategies that best balance a suite of potentially conflicting ecological and socioeconomic targets. Principles that guided the design of the ALCES Optimizer were:
a) Maintain ALCES strengths of speed and versatility. This required an efficient strategy to evaluate the effects of a wide range of land use options. It also required that all major land uses and a range of ecological and socioeconomic indicators be included in the optimizer.
b) Provide capacity to identify optimal land use strategies that balance objectives even in situations where the objectives are competing and potentially unrealistic. This required a tool that can not only consider multiple objectives, but also intelligently adjust management objectives if necessary.
c) Be intuitive to use. The tool was developed in Stella and Excel to seamlessly link with ALCES. As with other components of ALCES, efforts were made to build an intuitive user interface that guides the user through a series of logical, sequential steps to apply the land use optimization tool.
The ALCES Land Use Optimization Tool’s analytical strategy is a factorial simulation experiment with optimization. A factorial simulation experiment is run in ALCES to efficiently probe the range of possible land use combinations. Output from the simulation experiment is then analyzed using multiple linear regression to estimate statistical models of the relationship between land use parameters (development rate, best practices) and ecological and socioeconomic indicators. An optimization algorithm is then applied to the regression models to identify the land use option that is best able to achieve multiple and potentially conflicting user defined management objectives. To accommodate unrealistic management objectives, the optimization algorithm is applied iteratively while adjusting the management objectives until a feasible set of management objectives is identified.
The analytical approach is now described in more detail, followed by a description of how to use the land use optimization tool.
Analytical approach
The first step in the land use optimization process is to run a factorial simulation experiment in an ALCES model that has been parameterized for the study area. The factorial simulation experiment is customized to explore the range of land use options within user-defined maximum and minimum growth rates for up to five land uses (forestry, oil, gas, agriculture, settlements). Simulation experiment output is then exported to an excel file where it is analyzed to fit linear regression models that describe the relationship between land use parameters and up to ten ecological and socioeconomic indicators. An optimization algorithm is then applied to the regression models in Excel to identify the land use strategy that best achieves user defined management objectives for each indicator. The optimization component is linked to but distinct from the factorial simulation experiment component such that multiple optimizations can be carried out to compare land use options that are best suited to a range of possible management objectives. Repeating the optimization process without having to repeat the time consuming simulation experiment is amenable to workshop situations where it may be desirable to evaluate the implications of alternative management objectives “on the fly”.
Factorial Simulation Experiment
A factorial experiment systematically evaluates the effects of multiple factors on a variable of interest as well as the effects of interactions between factors. In a factorial experiment, the influence of two or more factors, each with discrete possible values or “levels”, is evaluated by testing the effect of all possible combinations of these levels across all factors. Factorial experiment outcomes are analyzed using ANOVA or regression analysis to estimate the size and direction of each factor’s effect on the response variable.
In the context of the land use optimization tool, the factors refer to land uses and the levels refer to the amount of a given land use. A factorial experiment provides an efficient strategy to explore the effects of the wide range of land use options that are possible when multiple land uses are involved. Up to five land uses (i.e., factors) can be included in the factorial experiment: forestry, oil, gas, agriculture, and settlements. The resource production and population types that are included in each factor are as follows:
1. The forestry factor includes softwood and hardwood timber production. Associated footprints are cutblocks and inblock roads.
2. The oil factor includes the production of conventional oil, insitu oil, and oil sand mining. Associated footprints are wells, surface mines, well site access roads, seismic lines, and pipelines.
3. The gas factor includes the production of natural gas and coal bed methane. Associated footprints are wells, well site access roads, seismic lines, and pipelines.
4. The agriculture factor includes the amount of crop production and the size of livestock populations. Associated footprints are cropland, pasture, and other footprints such as feedlots.
5. The settlement factor includes the size of human populations. Associated footprint is settlements.
The transportation network is not included explicitly as a factor because roads are typically developed to serve other land uses. Instead, growth in the major and minor road networks in a given simulation is a derivative of the levels of the other land uses. The creation of inblock forestry roads and well site access roads is dependent on the rate of forestry and oil/gas development, as specified in ALCES panels x and y. To include the development of other roads in the simulation experiment, the user must specify the length of major and minor roads that need be built each year to maintain current resource production, the proportion of major and minor road development that is associated with each land use, and the length of a fully developed major and minor road network. Road development is then assumed to change in proportion to changes in the amounts of the other land uses in the simulation experiment. If the fully developed road network length is surpassed during a simulation, road network expansion stops.
The factorial experiments implemented by the land use optimization tool simulates each factor across three levels: the lowest permissible resource production or population level, the highest permissible resource production or population level, and the mid-point between the lowest and highest levels. All simulations begin at the current resource production and population levels (as set in ALCES), and change in a linear and gradual fashion to the targeted level over the course of the simulation. Experimenting with each land use across three levels requires 3n land use scenarios, where n refers to the number of land uses included. For example, a factorial experiment with five land uses consists of 243 land use scenarios to be simulated in ALCES. In addition, the user has the option of evaluating the effect of a suite of best practices in the experiment, which doubles the number of scenarios (i.e., all scenarios are simulated with best practices and without best practices).
Including three levels for each land use in the experiment balances the competing experiment objectives of efficiency and completeness. The maximum number of land use scenarios (486) is sufficiently small to be simulated within 24 hours using the sensitivity analysis (i.e., batch file) feature in ALCES. At the same time, simulating across three levels provides enough detail to explore across the range of possible land use options (i.e., between the land use minimums and maximums) and test whether the response surface between a land use and an indicator is linear, as is assumed by statistical modeling component of the land use optimization tool (as described in the next section). Including nonlinear relationships between land use and indicators will result in poor statistical models and therefore unreliable optimization outcomes.
Linear response surfaces between land use and ecological and socioeconomic indicators are typical in ALCES, with the exception of sudden reductions in resource production that occur when resource production (e.g., timber harvest, hydrocarbon extraction, cropland expansion, etc.) exceeds resource availability. To exclude sudden reductions in resource production from the simulation experiment, the user should ensure that the maximum land use options included in the simulation experiment do not exceed resource availability. To minimize the occurrence of resource production fall-downs, ALCES will automatically switch compatible resource production types if resource availability is exceeded. This feature exists for the forestry, oil, and gas factors.
1. Forestry: if during a simulation the harvest level can not be achieved for either softwood or hardwood, ALCES tries to make up for the timber shortfall by increasing the harvest level for the other timber type such that the targeted forestry GDP is achieved.
2. Oil: if during a simulation the production target can not be achieved for one of the types of oil, ALCES tries to make up for the shortfall by increasing one of the other types of oil production. The hierarchy applied when selecting oil production types to make up for a shortfall is that conventional oil is preferred, followed by insitu and then oil sands mining. When increasing a type of oil production to make up for a shortfall, ALCES adjusts production in order to achieve the targeted oil GDP (i.e., the oil production increase is adjusted to account for differences in GDP/m3 between oil types).
3. Gas: if during a simulation the production target can not be achieved for one of the types of gas, ALCES tries to make up for the shortfall by increasing the other type of gas production. When increasing a type of gas production to make up for a shortfall, ALCES adjusts production in order to achieve the targeted gas GDP target (i.e., the gas production increase is adjusted to account for differences in GDP/m3 between gas types).
Prior to running the simulation experiment, the user specifies up to 10 indicators to include as response variables. The indicators can be any of the various ecological and socioeconomic indicators that ALCES is designed to track. The output from the simulation experiment consists of the status of the indicators at the end of each simulation. This output is exported to excel where it is used to fit statistical models of the relationships between the indicators and land use, as described in the next section.
To avoid the time consuming process of running each of an experiment’s simulations manually, the land use optimization tool uses the sensitivity analysis feature in ALCES. The sensitivity analysis feature allows the user to specify the land use levels for each of the experiment’s scenarios ahead of time and then simulate the complete set of scenarios in sequence.
Fit Statistical Models of the Relationship between Land Use and Indicators
The results from the simulation experiment are analyzed using multiple linear regression. For each indicator, regression is applied to model the affect of land use amount and best practices. These models provide a succinct representation of the impact of land use on indicators. More importantly, from the perspective of optimization, they provide equations that can be used as objective and constraint functions by an optimization algorithm to identify land use options that are best able to balance indicator objectives. The models are of the following form:
y = b0 + b1•A + b2•O + b3•G + b4•F + b5•H + b6•B + b7•A•B + b8•O•B + b9•G•B + b10•F•B + b11•H•B (Equation 1)
where
y = indicator status at the end of the simulation period
A, O, G, F, H = the proportion of the initial level of agriculture production, oil production, gas production, forestry production, and human population size, respectively, at the end of the simulation. For example, 0.5 refers to 50% drop in resource production or population, 1.5 refers to a 50% increase in resource production, and 1 refers to no change in resource production.
B = best practice implementation rate, where 1 represents 100% implementation of the suite of best practices
b0 = intercept term, which can be interpreted as indicator status at the end of the simulation period if all land uses are stopped
b1-5 = land use main effect coefficients, which can be interpreted as the effect to indicator status at the end of the simulation period of maintaining resource production or population size at its initial level.
b6 = best practice main effect coefficient, which can be interpreted as the effect to indicator status at the end of the simulation period of implementing best practices if all land uses are turned off
b7-11 = best practices/land use interaction coefficients, which can be interpreted as the effect to indicator status at the end of the simulation of implementing best practices if a land use is maintained at its initial level.
Linear regression is used to fit the indicator models because exploratory simulations indicated linear relationships between the amount of a land use and indicator performance, and between the degree of best practice implementation and indicator performance. Exploratory simulations were run for two landscapes in Alberta: the southern east slopes and the LMA4 management unit in northeastern Alberta. The interaction terms (b7-11) were included because they substantially improved model fit, indicating that the influence of best practices on indicator status is dependent on the amount of land use that is occurring in the region. Exploratory analysis indicated that including interaction terms between individual land uses (i.e., A, O, G, F, H) did not improve model fit, and therefore these interaction terms are not included in the models fit by the optimization tool. For the two study areas used during exploratory analyses (southern east slopes and LMA4), the coefficients of determination (R2) for the multiple linear regression models exceeded 98% and often reached 100%. These high R2 values indicate that the models explain a very high proportion of the variation in indicator results across the simulations from the simulation experiment.
Because linear regression is used, it is essential that nonlinear relationships between indicators and land use parameters are not included in the simulation experiments. Doing so will result in models that poorly represent the effects of land use on indicators, and will ultimately recommend land use options that are poorly suited for balancing indicator objectives. As discussed in the previous section, relationships between indicators and land use are generally linear in ALCES with the exception of resource production fall-downs that occur if resource availability is exhausted. For this reason, it is necessary that maximum land use options included in simulation experiments do not exceed the resource production capacity of a study area.
Multiple linear regression models are fit by minimizing the residual sum of squares using the LINEST function in excel. The analysis is completed automatically when the user exports the results from the simulation experiment from ALCES (see the Using the Tool section for details).
Optimization with Adjusted Constraints
The final step of the land use optimization tool is to apply an optimization algorithm to the indicator models to identify the land use option that is best able to balance indicator objectives. Optimization problems have three components: a) the objective function which is to be minimized or maximized; b) the variables which influence the objective function; and c) a set of constraints that limit the allowable values of the variables. The optimization algorithm seeks to identify the variable values that are within the set of constraints that minimize or maximize the objective function.
In the context of the land use optimization tool, the variables are the land use and best practice factors. The indicator models (equation 1) are the objective functions. Although optimization algorithms can only maximize or minimize a single objective function, multiple objectives can be included by specifying them as constraint functions instead of objective functions. For example, the optimization solution must be such that the value of an indicator is above a user defined threshold according to its regression model (equation 1). Additional constraints that are relevant to the land use optimization tool are the minimum and maximum land use amounts specified in the simulation experiment. The task of the optimization algorithm is to identify the land use combination that is within the user-specified minimum and maximum land use amounts that maximizes or minimizes a given indicator’s value (according to the indicator’s regression model) while obeying the user-defined objectives for the other indicators (according to those indicator’s regression models).
A limitation of optimization algorithms is that no solution is found if two or more of the constraints are incompatible. Incompatible constraints are likely to be encountered by the land use optimization tool because the constraints represent ecological and socioeconomic objectives. Management objectives set at the start of land use planning exercises are often unrealistic by not adequately accounting for tradeoffs between ecological and economic indicators. One of the roles of scenario analysis during land use planning is to identify these incompatibilities and inform how indicator objectives need to be adjusted to improve their feasibility. Due to the likelihood of conflicting and perhaps incompatible management objectives, the land use optimization tool was developed to accommodate incompatible constraints.
Initially, two optimization strategies were considered as possible approaches for accommodating incompatible constraints: optimization with hierarchical constraints and pareto optimization. In optimization with hierarchical constraints, an optimization algorithm is applied and, if a solution is not feasible, a constraint is dropped and the algorithm is applied again. The dropping of constraints continues, following a user-defined hierarchy of constraint priorities, until a solution is feasible. This approach performed poorly because, in many cases, a solution will not be found until all ecological or all economic constraints are dropped. The result is an “optimal” land use option that either maximizes ecological objectives without attempting to accommodate economic objectives, or vice versa. This is inconsistent with a central goal of land use planning, which is identify land use options that balance ecological and economic objectives. As such, optimization with hierarchical constraints was determined to be poorly suited to the needs of the land use optimization tool. Pareto optimization was found to be poorly suited for the same reason. Pareto optimal is defined as a solution where an indicator can not be improved anymore without having a negative effect on one or more of the other indicators. Depending on the response surfaces of the indicators, multiple pareto optimal solutions may exist. A limitation of pareto optimization is that, if the objectives are sufficiently contradictory (as is often the case with ecological and economic objectives), the pareto optimal solutions maximize the contradictory objectives but do not balance them.
Due to the unsuitability of optimization with hierarchical constraints and pareto optimization, an alternative approach was developed to accommodate incompatible objectives whereby the optimization algorithm is applied iteratively while adjusting the constraints. If a solution to the land use optimization problem is not found due to incompatible constraints, the management objectives are relaxed and the optimization algorithm is applied again. This sequence continues until the management objectives have been relaxed to a point where they are compatible and a solution exists. The adjusting constraints approach is therefore able to identify “optimal” land use options that balance competing management objectives such as resource production and wildlife habitat.
The optimization with adjusting constraints approach was designed such that the default adjustment size for indicator objectives is 0.1% of the initial size of the objective. For example, if a resource production objective is 1000 units, each adjustment would decrease the objective by 1 unit. After exploration analysis, 0.1% was deemed to be an adjustment size that balanced precision with efficiency. Larger adjustment sizes may lead to inexact optimal land use solutions, and smaller adjustment sizes may require large numbers of iterations before a solution is reached and therefore are time intensive. To minimize the solution time, it is important to close other software when applying the optimization tool in order to free up RAM.
All indicator objectives are adjusted after each application of the optimization algorithm until a solution is found. However, the user has the option to assign weights to the indicator objectives so that the indicator adjustments are smaller for indicators that are deemed to be more important. The result is that indicators with higher weights are adjusted less in the process of identifying a feasible land use option. The weighted adjustment is calculated such that the indicator with the smallest weight is adjusted by 0.1%, and other indicators are adjusted by (ws/wi)*0.1%, where ws is the smallest weight and wi is the weight assigned to other indicators. For example, if an indicator is assigned a weight that is twice as large as the smallest weight, the indicator will be adjusted by 0.05% after each application of the optimization algorithm. The capacity to assign weights to indicators allows the land use optimization tool to evaluate the land use implications of prioritizing some management objectives above others.
The optimization algorithm with adjusting constraints is implemented in excel. A function was written in visual basic that adds adjusting constraints capacity to the Excel Solver tool. The Solver tool applies the Generalized Reduced Gradient method to identify local optimal solutions to minimization and maximization problems. An advantage of the Generalized Reduced Gradient method is that it can accommodate nonlinear objective functions. A nonlinear objective function occurs in the land use optimization tool when best practices are included in the model due to the interaction terms between best practices and land uses (see equation 1). See http://www.utexas.edu/courses/lasdon/design3.htm for details on the Solver tool.
Output provided by the land use optimization tool includes the information about the optimal land use option and its impact on the indicators. Optimal land use option parameters are the percent change in each land use over the next 50 years relative to today. For example, -50% change for a land use indicates that a land use should decrease by 50% relative to year 0 in order to achieve the optimal balance of indicator objectives. A +50% change indicates that a land use should increase by 50% relative to year 0 to achieve the optimal balance of indicator objectives. Indicator information provided by the optimization tool is the status of each indicator at the end of the simulation period under the optimal land use option and the percent of each indicator’s management objective that is achieved under the optimal land use option. The percent management objective achieved information is useful for determining how much the indicator objective had to be relaxed in order to make them realistic (i.e., compatible).
The entire optimization with adjusted constraints component of the land use optimization tool is implemented in excel, including the setting of indicator objectives and weights, calculating the optimal land use option, and presenting output. As a result, it is possible to run the optimization tool in isolation of ALCES (and the Stella platform) once the simulation experiment has been completed. This is advantageous because users can experiment with alternative indicator objectives and weights without having to rerun simulation experiments. Simulation experiments can take hours of computing time to complete whereas implementation of the optimization with adjusted constraints algorithm takes minutes. Experimenting with alternative indicator objectives and weights is instructive for developing an understanding of key tradeoffs among management objectives and the influence of these tradeoffs on optimal land use combinations.
It is important to understand the meaning of “optimal” in reference to the land use options identified by the land use optimization tool. The solution identified by the optimization tool is dependent on the settings of the ALCES model used in the simulation experiment, the range of land use options allowed in the simulation experiment, the indicators included and their objectives and weights. As such, the identified land use option is optimal only in the very narrow sense of being best able to balance the user defined indicator objectives given the land use system represented by the ALCES parameterization. Including different indicators or objectives will result in new “optimal” land use options. Modifying the assumed relationships between land use and indicators as represented by the ALCES parameterization will also result in new “optimal” land use options.
Using the Land Use Optimization Tool
Necessary Files
In addition to the ALCES model, the land use optimization tool requires that the following Excel files be located in the “C:/ALCES optimize” folder:
1. Optimizer.xls: this workbook contains the user interface for steps 6 to 8 of the land use optimization tool. The key roles of the workshop are to fit indicator regression models using simulation experiment results and to implement an optimization algorithm to identify the optimal land use options.
2. sensi specs.xls: this workbook generates “sensi specs” that are needed by ALCES to run a simulation experiment
3. datastore.xls: this workbook contains data that is exported from the ALCES simulation experiment. It is essential that this workbook not be manipulated by the user.
Parameterize ALCES for the Study Area
Prior to running the land use optimization tool, ALCES must be parameterized for the study area. The land use trajectories that are included in the simulation experiment are based on the initial size of each land use as defined in ALCES. For example, resource production either increases, decreases, or stays the same relative to the initial amount of land use defined in ALCES. In addition, the optimization tool will use the footprint intensities, sizes, and life-spans defined in ALCES for the simulation experiment. It is therefore essential that the initial status of resource production and footprint relationships are properly defined in ALCES. A list of key variables for each land use is as follows:
• Forestry: harvest goals and constraints as well as inblock road size and lifespan as defined in panel 8.1.2.
• Oil and Gas: the production types of interest must be turned on (table 1 in panel 8.2). Initial area of wells and surface mines (table 2 in panel 4) and historical percent contribution of different well types (table 4 in panel 8.2) must be defined as this is used during the optimization experiment to calculate the initial rate of oil production. The initial hydrocarbon volumes must be defined (table 3 in panel 8.2) as this is used to determine whether resource availability is sufficient to supply development trajectories included in the simulation experiment. Exploratory well requirements must be defined (tables 5 and 6 in panel 8.2). Various energy sector footprint intensities, sizes and life-spans must be defined using inputs in panels 8.2.2 (seismic), 8.2.3 (wells and access roads), and 8.2.4 (pipelines and surface mines).
• Agriculture: the initial amount of agriculture land (table 1 in panel 4) and agricultural footprint (table 2 in panel 4) must be defined as this is used during the optimization experiment to calculate the initial amount of agricultural land and crop production. The simulation experiment uses LT_to_Ag_rules (table 15 in panel 4) and FT_classification_code (table 10 in panel 4) to determine which LT’s and FT’s are agricultural. If agricultural footprints are to be reclaimed during negative growth scenarios, their footprint lifespan strategy must be switched to user-directed (option 2 in table 5, panel 8.11) and the destination of reclaimed agricultural footprints must be defined in tables 3 and 7 in panel 8.11. The initial size of livestock populations must also be defined (table 3 in panel 8.3).
• Settlements: the initial area of settlements (table 2 in panel 4) must be defined as this is used during the optimization experiment to calculate that initial size of settlements. FT_classification_code (table 10 in panel 4) is used to determine which FT’s are residential. If residential footprints are to be reclaimed during negative growth scenarios, their footprint lifespan strategy must be switched to user-directed (option 2 in table 5, panel 8.11) and the destination of reclaimed residential footprints must be defined in tables 3 and 7 in panel 8.11. The initial size of the human population must also be defined (table 2 in panel 8.5).
It is also necessary to define relationships required to track any indicators that are to be included as response variables in the simulation experiment. For example, GDP and employment relationships as defined in panel 8.14, water use and nutrient runoff relationships as defined in panel 8.13, and wildlife habitat relationships as defined in panel 9.
Using the Land Use Optimization Tool
To successfully complete a land use optimization, the user must implement a series of eight steps in sequence. The tool’s interface is organized into the eight steps to help ensure successful use. The first five steps use the Landuse Optimization panel in ALCES to run the simulation experiment. The final three steps use a customized Excel spreadsheet to fit the indicator regression models and run the optimization algorithm.
Step 1: Set land use parameters for the simulation experiment
In this step the following parameters are entered into ALCES to set-up the simulation experiment:
a) Lowerbound must be set for each land use that is to be included in the simulation experiment using the "Lower and upper bounds for land use" input table. It defines the lowest permissible level of a given land use. More specifically, it defines the largest reduction in resource production (timber, oil, gas, crops) or population (humans and their settlements, livestock) that can occur over the simulation period relative to today. For example, a lowerbound=0.3 for forestry indicates that timber production can decrease to no less than 30% of initial timber production over the course of the simulation. Declines in resource production and population occur gradually in a linear fashion over the course of a simulation.
b) Upperbound must be set for each land use that is to be included in the simulation experiment using the "Lower and upper bounds for land use" input table. To exclude a land use from the simulation experiment, upperbound must be set to zero. Upperbound defines the largest increase in resource production or population that can occur over the simulation period relative to today. For example, an upperbound=1.3 for forestry indicates that timber production can increase to no more than 130% of initial timber production. Increases in resource production occur gradually in a linear fashion over the course of a simulation.
c) The "include best practices?" switch must be turned on (i.e., up) if best practices are to be included in the simulation experiment. If best practices are included, the simulation experiment will include scenarios without best practices and scenarios with best practices. This provides an assessment of the degree to which best practices influence the indicators of interest. It is important to note, however, that including best practices will double the time needed to run the simulation experiment. The best practices that are manipulated by the "include best practices?" switch are those that are identified on the "Best Practices Switches and Levers" panel (8.16). Revising the set of best practices requires the assistance of an ALCES programmer.
d) Destination of reclaimed agricultural land. In scenarios where agricultural production declines, cropland and pasture no longer required for production are converted to other cover types (i.e., reclaimed). If the set of possible land use options includes scenarios where agriculture declines (i.e., lowerbound[agriculture]<1), the user must define which cover types agricultural land should be reclaimed to. This is done by setting the "ag reclaim destination" variables on the "Destination of reclaimed agricultural land" input table. "ag reclaim destination" for a given cover type specifies the proportion of reclaimed agricultural land that should become that cover type. The sum of "ag reclaim destination" across all cover types must equal 1.
e) Source of new agricultural land. In scenarios where agricultural production increases, cropland and pasture must increase in order to achieve growth in agricultural production. If the set of possible land use options includes scenarios where agriculture increases (i.e., upperbound[agriculture]>1), the user must define which cover types should be converted to agricultural land. This is done by setting the "ag growth destination" variables on the "Source of new agricultural land" input table. "ag growth destination" for a giver cover type specifies the proportion of new agricultural land that should be expanded onto that cover type. The sum of "ag growth destination" across all cover types must equal 1.
f) Road expansion. Unlike other land use sectors, transportation is not directly manipulated in the simulation experiment but instead varies depending on the levels of the other land uses. To define the relationship between road growth and the other land use sectors, the user must set the following variables in the "Road expansion" input table.
i. "A infrastructure to maintain production". The user defines the annual major and minor road growth (km's) required to maintain the current (i.e. initial) level of resource production. Major and minor road refers to all road types except for wellsite access roads, inblock forestry roads, and acreage access roads.
ii. "max infrastructure". The user defines the length of major and minor roads in a fully grown road network for the study area.
iii. "road_optimization_weights". The user specifies the proportion of road network expansion that is associated with each type of land use. Road expansion (relative to "A infrastructure to maintain production") in a simulation is then the weighted average (based on "road_optimization_weights") of the rates of change of the other land use sectors. If the "max infrastructure" length is exceeded during a simulation, road development stops.
Step 2: Test the simulation experiment settings
Prior to running the simulation experiment, it is necessary to test the settings to ensure that sufficient natural resources (timber, hydrocarbons, potential agricultural land) exist to achieve the user-defined upper bound for each land use. After running a test simulation by clicking the test button, you can check the viability of the user-defined upper bounds by viewing the "optimize check" graphs (double click on "optimize check" to open the graphs. The graphs compare user-defined upper bounds for natural resource production to the realized natural resource production during the test simulation. If the realized natural resource production is less than the upper bound, the upper bound exceeds the availability of natural resources in the study area. When this occurs, the upper bound must be reduced until a test indicates that it is within the availability of natural resources for the study area.
In order to minimize the occurrence of situations where resource availability is insufficient, ALCES automatically switches production to compatible resource types when a resource is used up. This function exists for timber production (hardwood and softwood timber harvest are interchanged), oil production (conventional oil, insitu bitumen, and surface bitumen are interchanged), and gas production (natural gas and coal bed methance are interchanged). When switching resource production, ALCES modifies development rates to maintain the targeted GDP goal.
NOTE: Do not skip the test step!!! Simulation experiment settings (as defined in Step 1) must be tested before proceeding to Step 3. Otherwise, incorrect sensi specs will be generated for the simulation experiment.
Step 3: Enter “sensi specs” for the simulation experiment
Simulation experiments in ALCES are run by entering "sensi specs" that define the suite of land use scenarios to include. To guide the user through the process of entering "sensi specs" for the simulation experiment, ALCES generates an excel spreadsheet that provides the necessary "sensi spec" settings. The user clicks the "Generate sensi specs" to export the simulation experiment settings to the Excel workbook “datastore.xls” and then clicks "Input sensi specs" to open the Excel workbook “sensi specs.xls” that provides the sensi spec settings and an explanation of how to enter them into ALCES.
Step 4: Prepare output tables
The ALCES land use optimization tool can identify land use strategies capable of balancing targets for up to 10 ecological and socioeconomic indicators. To select which indicators to include, click on the "optimize output" table pad. The table pad includes 10 tables, one for each indicator. To select assign an indicator to a table, double click on the table. This will open the "Define table" dialogue box. Select the indicator of interest from the Allowable list. Do not change any of the other settings in the table as it will interfere with the functioning on the optimization tool.
Prior to running a simulation experiment, it is essential that "optimize output" tables are cleared of any previous simulation results. This is achieved by clicking the "Clear output tables" button. It may take a couple of minutes for the action to be completed.
Step 5: Run the simulation experiment
Once the simulation experiment is set-up, the user clicks the “Run” button in “Run simulation experiment” box. The ALCES simulation experiment is a factorial experiment consisting of 3 levels for each land use treatment included (as specified in Step 1). The 3 levels for each land use are its lower bound, upper bound, and mid point between the lower and upper bounds. A simulation experiment with 1 land use requires 3 simulations, 2 land uses require 9 simulation, 3 land uses requires 27 simulations, 4 land uses requires 81 simulations, and 5 land uses requires 243 simulations. Including best practices doubles the number of required simulations. A simulation experiment will take approximately 100 seconds for every 50-year simulation.
Once the simulation experiment is complete, the results are automatically exported to the excel file "datastore.xls". The user then clicks the "ALCES Optimizer" button to open the Excel-based land use optimization tool (“Optimize.xls”). The tool analyzes the simulation experiment results to fit statistical relationships between the land uses and indicators (Step 6). These relationships are then evaluated by an optimization algorithm to explore which land use options are best able to balance user-defined indicator targets (Steps 7 and 8).
Step 6: Review the indicator models
When the user clicks the “ALCES Optimizer” button, an Excel spreadsheet called “Optimizer” is opened. The “Optimizer” spreadsheet automatically uses multiple linear regression to fit models of the relationships between indicators and land use parameters (Equation 1). The primary purpose for modeling these relationships is to provide input for the optimization algorithm (steps 7 and 8). Coefficients of determination are provided in the “Step 6” worksheet in the “Optimizer” spreadsheet (Figure 6). It is important to review the coefficients of determination to ensure that the models account for most of the variation in indicator response from the simulation experiment. As a rough guideline, indicators for which the regression model’s coefficient of determination is below 95% should not be included the optimization (steps 7 and 8). An indicator model with a coefficient of determination below this level likely indicates a nonlinear relationship between land use and the indicator (see the section “Fit statistical models of the relationship between land use and indicators” for details).
In addition to informing the optimization, the indicator models are also of interest as a succinct representation of the influence of land use on indicators of interest. The model coefficients are therefore presented in the “Step 6” worksheet in the “Optimizer” spreadsheet. See the section “Fit statistical models of the relationship between land use and indicators” for information about how to interpret the model coefficients.
Step 7: Run the land use optimizer
Once the user has ensured that the indicator models achieve sufficient fit, targets and weights are set using table 7.1 in the “Step 7” worksheet in “Optimizer.xls”. Including an indicator instructs the optimizer to consider the effects of land use on that indicator when selecting an optimal land use option. For each indicator included in the optimization, the user sets the target and states whether the target is a minimum (e.g., the minimum number of jobs) or a maximum (e.g., the maximum density of anthropogenic disturbance). The user can also assign weights to indicators. A high weight indicates that achieving an indicator’s target is of greater importance (see the section “Optimization with adjusted constraints” for details).
Once the indicator information is entered, the user clicks the “Run Optimization” button to run an optimization. The optimizer will identify a land use strategy that best balances the indicator targets, placing greater importance on those indicator targets with higher weights. An optimization typically requires hundreds of iterations (as specified by the “Number of indicator adjustments” counter) and takes a few minutes to complete. During that time, the user will not be able to manipulate the “Optimizer” spreadsheet.
Step 8: View Optimization Results
Once the optimization is complete, the user navigates to the “Step 8” worksheet to view the results (Figure 8). Output includes the optimal land use option (Table 8.1) and its impact on the indicators (Table 8.2).
The optimal land use option is described as the percent change in each land use over the next 50 years relative to today. For example, -50% change for a land use indicates that a land use should decrease by 50% relative to year 0 in order to achieve the optimal balance of indicator objectives. A +50% change indicates that a land use should increase by 50% relative to year 0 to achieve the optimal balance of indicator objectives.
Indicator results are presented as status at the end of the simulation period under the optimal land use option and the percent of each indicator’s management objective that is achieved under the optimal land use option. The percent management objective achieved information is useful for determining how much the indicator objectives had to be relaxed in order to make them realistic (i.e., compatible).
Experimenting with alternative indicator objectives and weights is instructive for developing an understanding of key tradeoffs among management objectives and the influence of these tradeoffs on optimal land use combinations. To evaluate the optimal land use option under alternative indicator objectives or weights, the user returns to Step 7 (Run the land use optimizer) and revises the indicator information prior to running the optimizer again.