Comments
Description
Transcript
Document 2662445
Vermont Renewable Electricity Transition: Spatial Simulation of Scenarios of Renewable Energy Development Christopher Clement1,2,3 Corresponding Author: Christopher Clement Address: 617 Main Street, Burlington, VT 05401 Email: [email protected] Phone: 828.273.4417 1 University of Vermont Rubenstein School of Environment & Natural Resources Gund Institute for Ecological Economics 3 NSF IGERT: Smart Grid: Technology, Human Behavior, & Policy 2 Abstract Vermont’s transition from a fossil fuel and nuclear-‐based electricity system to one primarily constituted of renewables has large implications not only for the shift in technology, but also in land cover change. This analysis considers scenarios under which electricity demand is met with resources from within the state. Using Dinamica EGO, a model was developed to simulate the spatial patterns of renewable energy development under different scenarios of achieving this renewable electricity transition. Based on this analysis, Vermont could meet this goal; however, this would be achieved only through developing thousands of hectares of land for solar and wind. To some degree, all scenarios would require development on environmentally sensitive land. Based on this analysis, it is recommended that energy policy and planning focus on strategically developing low-‐ impact regions with medium-‐ to large-‐scale wind developments, and fill remaining gaps with medium-‐scale ground-‐mounted solar developments closer to developed areas. 1 Table of Contents ABSTRACT 1 TABLE OF CONTENTS 2 FIGURES 3 TABLES 3 1 TRANSITIONING TO RENEWABLE ELECTRICITY USING VERMONT RESOURCES 4 2 STUDY AREA AND CONTEXT 5 2.1 STUDY AREA 2.2 EXISTING RENEWABLE ENERGY DEVELOPMENTS 5 5 3 METHODS: SIMULATION OF RENEWABLE ENERGY DEVELOPMENT SCENARIOS 6 3.1 AN OVERVIEW OF SPATIO-‐TEMPORAL MODELING WITH DINAMICA EGO 6 4 SUITABILITY ANALYSIS FOR RENEWABLE ENERGY DEVELOPMENT 7 4.1 MULTI-‐CRITERIA EVALUATION AND FRICTION COST SURFACE 9 5 LUCC MODEL OF FUTURE RENEWABLE ENERGY DEVELOPMENT 11 5.1 RENEWABLE ENERGY DEVELOPMENT LUCC MODEL PARAMETERIZATION 5.1.1 TRANSITION MATRICES 5.1.2 WEIGHTS OF EVIDENCE 5.1.3 LUCC MODEL CALIBRATION 5.2 RENEWABLE ENERGY DEVELOPMENT LUCC MODEL SIMULATION 5.2.1 RECREATING THE SPATIAL EXTENT OF EXISTING RENEWABLE ENERGY DEVELOPMENTS 5.2.2 SIMULATING NEW RENEWABLE ENERGY DEVELOPMENTS 5.3 SCENARIO MODELING 13 13 14 16 17 18 19 19 6 RESULTS 20 6.1 SUITABILITY ANALYSIS FOR RENEWABLE ENERGY DEVELOPMENT 6.2 RENEWABLE ENERGY DEVELOPMENT LUCC MODEL 6.2.1 SCENARIOS 20 21 21 7 DISCUSSION 24 8 CONCLUSIONS 25 ACKNOWLEDGMENTS 26 APPENDIX 1: MAPS 27 REFERENCES 44 2 Figures Figure 1: Flow Diagram of Multi-‐Criteria Evaluation and Friction Cost Surface Processes to Create a Transition Probability Map ________________________________________________________________________________________________________ 11 Figure 2: Renewable Energy Development LUCC Model Process ____________________________________________________ 13 Figure 3: Weights for Static Variables for the Transition from Developed Land to Ground-‐Mounted Solar ______ 16 Figure 4: Ground-‐Mounted Solar -‐ Constant Decay Multiple-‐Window Similarity of Differences Analysis _________ 17 Figure 5: Wind -‐ Constant Decay Multiple-‐Window Similarity of Differences _______________________________________ 17 Figure 6: Land Cover Impact for each Renewable Development Scenario __________________________________________ 23 Figure 7: Renewable Energy Development on Land Tiered by Contribution to Biodiversity _______________________ 24 Figure 8: Renewable Energy Development on Land Tiered by Protected Lands Status ____________________________ 24 Figure A-‐9: 2011 Vermont Land Cover Reclassified into Four Categories ___________________________________________ 27 Figure A-‐10: 2011 Vermont Land Cover and Existing Ground-‐Mounted Solar Installations _______________________ 28 Figure A-‐11: 2011 Vermont Land Cover and Existing Wind Solar Installations ____________________________________ 29 Figure A-‐12: 2011 Observed Land Cover with Ground-‐Mounted Solar Developments (Inset Map) ________________ 30 Figure A-‐13: 2011 Observed Land Cover with Wind Development (Inset Map) _____________________________________ 31 Figure A-‐14: Inverse Friction Cost Surface with Multi-‐Criteria Evaluation Filter -‐ Ground-‐Mounted Solar _______ 32 Figure A-‐15: Inverse Friction Cost Surface with Multi-‐Criteria Evaluation Filter -‐ Wind Solar ____________________ 33 Figure A-‐16: Inverse Friction Cost Surface with MCE Filter: Ground-‐Mounted with Existing Developments Overlaid (scaled to installed capacity (kW)) __________________________________________________________________________ 34 Figure A-‐17: Inverse Friction Cost Surface with MCE Filter: Wind with Existing Developments Overlaid (scaled to installed capacity (kW)) _______________________________________________________________________________________________ 35 Figure A-‐18: Simulated Land Cover Map for Scenario 1: All Ground-‐Mounted Solar _______________________________ 36 Figure A-‐19: Transition Probability Map for Scenario 1: All Ground-‐Mounted Solar _______________________________ 37 Figure A-‐20: Simulated Land Cover Map for Scenario 2: All Medium-‐ to Large-‐Scale Wind _______________________ 38 Figure A-‐21: Transition Probability Map for Scenario 2: Medium-‐ to Large-‐Scale Wind __________________________ 39 Figure A-‐22: Simulated Land Cover Map for Scenario 3: 50/50 Ground-‐Mounted Solar and Wind _______________ 40 Figure A-‐23: Scenario 1 -‐ Land Cover Impacts Percentage Distribution ____________________________________________ 41 Figure A-‐24: Scenario 2 -‐ Land Cover Impacts Percentage Distribution ____________________________________________ 42 Figure A-‐25: Scenario 3 -‐ Land Cover Impacts Percentage Distribution ____________________________________________ 43 Tables Table 1: Vermont Land Cover (2001-‐2011) ____________________________________________________________________________ 5 Table 2: Existing Renewable Energy Developments (REAVT, 2014) __________________________________________________ 5 Table 3: Energy Density and Generation Characteristics _____________________________________________________________ 6 Table 4: Criteria and Data Sources for Suitability Analysis ___________________________________________________________ 8 Table 5: Multi-‐Criteria Evaluation and Friction Cost Surface Process for Developing Suitability Analysis ________ 10 Table 6: Process for Developing Land Use Change Model for Renewable Energy Development Scenarios ________ 12 Table 7: Historical Transition Matrices _______________________________________________________________________________ 14 Table 8: Average Cramer Correlation Coefficients for Static Variables for Both Transitions ______________________ 15 Table 9: Summary Statistics for Existing Renewable Developments Across Land Cover Types ____________________ 18 Table 10: Parameters for Using Patcher for Future Simulations ____________________________________________________ 19 Table 11: Renewable Energy Supply Calculations for each Scenario ________________________________________________ 20 Table 12: Historical and Projected Transition Rates for Each Scenario _____________________________________________ 20 Table 13: Land Cover Impacts of Renewable Energy Development Scenarios ______________________________________ 22 Table 14: Environmentally Sensitive Land Impacts for Each Renewable Development Scenario __________________ 23 3 1 Transitioning to Renewable Electricity using Vermont Resources With the ratification of the Vermont Comprehensive Energy Plan (CEP) in 2011, Governor Shumlin’s administration set forth an ambitious vision for the transformation of the Vermont energy system. Among its many goals, the plan calls for 90 percent renewable energy generation, as compared to the current energy portfolio that contains 23 percent from renewable sources, and 75 percent reduction in greenhouse (GHG) emissions from the 1990 baseline by 2050 (DPS, 2011). Act 170, the Vermont Energy Act of 2011-‐2012, further codified portions of the plan by establishing a statutory total renewable energy goal, that require that 75 percent of electricity be derived from renewable sources by 2032 (Vermont General Assembly, 2012). The achievement of these goals will require a massive coordinated effort of state, regional, and local policy, technology development, and information systems, as well as the cooperation of private actors and independent organizations/industry. Precisely how this transformation occurs is a matter of significant debate among many stakeholders, with many differing views of what long-‐term policy, planning, and future scenarios are most feasible and desirable for the state of Vermont. There are policy design and modeling processes being employed to develop and evaluate policy and planning scenarios. However, none of these analytical efforts explicitly addresses the land-‐use implications of scaling renewable energy development to the degree necessary to achieve the CEP goals. There is an implicit assumption that Vermont will remain reliant on the large-‐scale hydroelectricity from Hydro Québec on which it has depended for decades. This study explores what the Vermont landscape-‐level impacts would be if current electrical demand were to be met exclusively with in-‐state renewable generation. Vermont’s reliance on large-‐scale hydroelectric and nuclear energy for much of its electricity and fossil fuel imports for its transportation fuels has disconnected many Vermont communities from the resource base on which their energy depends. There are some notable exceptions, such as Burlington, which in 2014 achieved sourcing 100 percent of its electricity from renewable sources, primarily biomass and local hydro dams. Using in-‐ state energy resources to meet in-‐state energy demand places a natural check on the scale of Vermont’s energy consumption. A key feature of this approach is putting in place a negative feedback mechanism whereby growth in energy demand is met with an increase in efficiency and the remainder provided within the bounds of what is available with local renewable energy resources. When Vermonters confront the scale of their consumption as evidenced through shifting land uses toward electricity generation, there will be a difficult set of trade-‐offs through which to navigate concerning land use, particularly agricultural and forest land, and renewable energy generation. The following study develops a spatial analysis and simulation of scenarios of meeting current electricity demand with current in-‐ state renewable resources. 4 2 Study Area and Context 2.1 Study Area The study area for this analysis includes the entire state of Vermont. The Multi-‐Resolution Land Characteristics (MRLC) National Land Cover Database (NLCD) was used to develop a baseline for land cover types in Vermont (Jin et al., 2013). The 15 categories of land cover at a spatial resolution of 30m-‐30m provided within the NLCD were reclassified into four categories to aggregate the land cover classes according to their similarities and relevance in terms of being potential renewable energy sites. Figure A-‐9 in Appendix 1: Maps shows these land cover classes for 2011, the most recent dataset. This was done to maintain some spatial resolution with the land cover data, while reducing the computational burden of simulating transitions across the full complement of 15 land cover classes. These aggregated land cover classes were used as the basis of the LUCC modeling to be described in subsequent sections. All other raster datasets developed for the following analysis were converted to 30m-‐30m grids using the same projection at the NLCD – Albers Conical Equal Area. Summary statistics are provided in Table 1. Table 1: Vermont Land Cover (2001-2011) Land Cover Class (1) Developed land (2) Forest and Wetlands (3) Herbaceous or barren land (4) Farmland and pasture Color Red Dark Green Light Green Yellow 2001 Area (ha) 134,267 1,875,600 49,612 332,999 2011 Area (ha) 135,843 1,861,321 63,018 332,248 Percent Change 1.17% -‐0.76% 27.02% -‐0.23% 2.2 Existing Renewable Energy Developments Table 2: Existing Renewable Energy Developments (REAVT, 2014) Data on existing renewable Renewable Installation Type kW Installed developments were obtained from Wind Site 120,981 the Renewable Energy Atlas of Methane Digester Site 4,195 Vermont (REAVT), which provided Methane Digesters 625 data on the location, installed size Woody Biomass Electric Site 70,000 (kW), and owner/operator of the 1,100 installation (see Table 2 for installed Woody Combined Heat and Power capacity per renewable Landfill Methane Site 11,560 development type). These data were Wastewater Treatment Biogas Site 1,000 geocoded in ArcMap using the Solar PV Ground Site 32,273 Vermont Center for Geographic Solar PV Roof Site 11,371 Information (VCGI) E911 data on Solar PV & Hot Water Site 382 building and road locations (VCGI, Total Renewable Installed Capacity 253,487 2011). Over 93 percent matches were obtained using the geocoding tool in ArcMap, and the remainder were hand-‐checked and manually located in the land cover map. The geocoded existing renewable energy developments were organized in a point shape file, which was then converted into a raster 5 file with cell size 30m-‐30m to correspond with the NLCD land cover data. This was done only for Solar PV Ground Site and Wind Sites. The general energy density and generation characteristic for ground-‐mounted solar and wind generation (70m) is shown in Table 3 (Denholm, 2009; Ong et al., 2013; Smil, 2013). The fact that these different technologies have different energy density factors in terms of hectares per MW will have large ramifications for the land cover trade-‐offs associated with pursuing a dominantly solar or wind strategy. It takes on average 2.74 hectares for each MW of ground-‐mounted solar, whereas only 0.87 hectares are needed for each MW of wind. There are also differences in potential generation, as a result of the capacity factors for each technology, or the ratio of actual versus potential electricity production. Wind’s comparatively higher capacity factor results in nearly twice the annual generation capacity compared to ground-‐mounted solar. These factors will have ramifications for the spatial modeling of renewable energy developments to meet future demand. Table 3: Energy Density and Generation Characteristics Generation Type Ground-Mounted Solar Wind Nameplate capacity (MW) 1 1 Average area per MW (ha per MW) 2.74 0.87 Capacity Factor 0.15 0.25 Annual Generation (MWh) 1,314 2,190 3 Methods: Simulation of Renewable Energy Development Scenarios Research Questions 1. How much of the Vermont landscape is suitable for renewable energy development? 2. What are the factors that drive the location, size, and rate of renewable energy development? 3. How much of Vermont’s landscape would need to be dedicated to renewable energy development to meet current electricity demand? 4. What are the spatial patterns of development that would likely be seen if Vermont were to meet it electricity demand with some combination of ground-‐mounted solar and large-‐scale wind development? 3.1 An Overview of Spatio-‐Temporal Modeling with Dinamica EGO Changes in land use and land cover can be analyzed and simulated using a range of spatial or geographic information systems approaches and environmental modeling platforms. The regional energy shed analysis, modeling, and simulation will be built with Dinamica Environment for Geoprocessing Objects (EGO) (hereafter Dinamica). Dinamica was originally designed as a land use/cover change (LUCC) modeling software to analyze land use change in the Amazon, but has in its more recent iterations evolved into a very flexible and powerful modeling platform for a variety of spatial analyses (Soares-‐Filho et al., 2002; Soares-‐Filho et al., 2003). Dinamica has been applied to a range of spatial analyses, 6 modeling, and simulations relevant to sustainable land management, planning, and policy (Kolb et al., 2013; Maeda et al., 2011; Soares-‐Filho et al., 2006; Sonter et al., 2014). Dinamica is considered among the more comprehensive modeling platforms that allow for the design of complex spatio-‐temporal models (Mas et al., 2014). The foundational approach in Dinamica is based on the use of Markov transition probability matrices to estimate state changes across spatial variables over time. Projections based purely on Markov matrices work on the assumption that the rates of change observed during the calibration period will remain the same during the simulation period. To account for the situations in which this assumption is erroneous, Dinamica computes the probability of each transition occurring, and calculates a probability map using the weights of evidence method. This method enables the application of statistical, data-‐driven or expert knowledge-‐driven approaches (Bonham-‐Carter, 1994). To address fluctuations in change rates, Dinamica allows for the replacement of the Markov matrix at specific steps of the simulation, which incorporates deterministic transitions between distinct states. Dinamica allocates change spatially by normalizing the probability maps of concurrent transitions, and uses two cellular automata-‐based transition functions (Expander and Patcher) that employ a stochastic selection algorithm in which pixels are ranked according to their change potential from greatest to lowest potential. As time progresses in discrete steps, all cells change their state simultaneously as a function of these transition probabilities. This process is mediated by the use of a pruning factor, which is multiplied by the expected number of cells to be changed and selects the cells that will take part in the selection mechanism based on their spatial probability. To evaluate the model, Dinamica computes either a decay or fuzzy similarity index, in which coincidence is not restricted to a strict, cell-‐by-‐cell overlay but also includes the cells in a neighborhood based on different sets of rules. The Dinamica spatial modeling platform was used to conduct the suitability analysis and create the land use and cover change (LUCC) model to simulate renewable energy development spatial patterns described in the following sections. A major advantage of the analytical platform that Dinamica uses is that it enables the integration of biophysical, environmental, and socioeconomic variables of different data types in order to parameterize the model. There are also various ways of incorporating dynamic feedbacks to vary transition rates as a function of landscape-‐level changes. These functionalities made Dinamica an ideal platform to model and simulate future renewable energy development patterns. 4 Suitability Analysis for Renewable Energy Development In renewable energy planning contexts, suitability analyses are a common quantitative approach to evaluate the relative viability of land parcels for the development of energy generation (Collins et al., 2001). There is a complex array of trade-‐offs in such analyses, as there is not only a variety of resource availability, infrastructural, and other physical factors of which to take account, but also social factors related to the preferences of local communities and landowners. Thus, the process whereby suitability is determined requires a multi-‐criteria analysis that addresses both objective and subjective criteria. 7 The objective criteria take into account renewable resource availability, proximity to infrastructure (i.e., roads and transmission lines), proximity to sources of demand (i.e., high-‐, medium-‐, or low-‐intensity development), and avoidance of protected or environmentally sensitive areas (i.e., high threat to habitat, high biodiversity value, proximity to wetlands, etc.). Land use governance and other social preference variables were not taken into account explicitly in this analysis, but could be done so in future research. Table 4 describes the variables, data sources, and GIS data types and projections. Table 4: Criteria and Data Sources for Suitability Analysis Variable Data Source Data Type Projection Existing Renewable Generation Facilities Solar (1) Existing PV Sites; (2) Existing Solar Ground-‐based Sites (REAV, 2014) Excel spreadsheet geocoded using ArcMap to create a point shapefile, which was converted to 30m-‐30m raster Albers Conical Equal Area Wind Existing Wind Developments (REAV, 2014) Excel spreadsheet geocoded using ArcMap to create a point shapefile, which was converted to 30m-‐30m raster Albers Conical Equal Area Potential Renewable Resources Wind Potential (1) Large Commercial (70-‐m); (2) Small Commercial (50-‐m); (3) Small Residential (30-‐m) (REAV, 2010) Polygon shapefiles which were combined and converted to 30m-‐30m raster; interpolated using inverse distance weighted function in ArcMap NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Solar Potential (1) Potential PV Sites; (2) Potential Solar Ground-‐based Sites (REAV, 2010) Point and polygon shapefiles merged and converted to 30m-‐ 30m raster; interpolated using inverse distance weighted function in ArcMap NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Line shapefile converted to 30m-‐30m raster NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Line shapefile converted to 30m-‐30m raster NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Electric Utility Electric Utility Franchise Boundaries (VCGI, 2012) Line shapefile converted to 30m-‐30m raster NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Slope and Aspect DEM converted into a 30m-‐ 30m raster and slope and aspect calculated using Slope and Aspect functions in GCS WGS 1984 converted to Albers Conical Equal Area Infrastructure and Physical Characteristics Roads Roads (VCGI, 2010) Transmission Lines Transmission Lines (VCGI, 2010) Digital Elevation Models (DEM) from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) 8 Global Digital Elevation Model (GDEM) (GDEM, 2011) ArcMap Development High, Medium, MRLC NLCD (Jin et al., 2013) and Low Intensity Developed Population and Housing 30m-‐30m raster Census Block Group Population Polygon shapefile converted to and Housing (U.S. Census, 2012) 30m-‐30m raster; interpolated using inverse distance weighted function in ArcMap Albers Conical Equal Area NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Environmental Constraints Threats to Habitat and Biodiversity BioFinder (VANR, 2013) 10m-‐10m raster converted into a 30m-‐30m raster using the nearest neighbor function in ArcMap NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area Protected Lands Protected Lands (VCGI, 2010) Polygon shapefile converted to 30m-‐30m raster; interpolated using inverse distance weighted function in ArcMap NAD 1983 State Plane Vermont FIPS 4400 converted to Albers Conical Equal Area 4.1 Multi-‐Criteria Evaluation and Friction Cost Surface A multi-‐criteria evaluation (MCE) was used to determine the areas deemed suitable for renewable energy development, based on renewable energy potential, land cover type, infrastructural requirements, and environmental constraints. The MCE analysis was constructed in Dinamica using a series of negative screens based areas deemed unsuitable for renewable energy development. The resulting map is a simple Boolean representation of suitable and unsuitable areas without any consideration for the relative cost (later referred to as a friction cost surface), of development on suitable lands. A friction cost surface was developed to estimate the relative cost or difficulty of developing within the suitable areas determined by the MCE. The friction cost surface analysis was constructed in Dinamica as a weighted multiplication of factors that describe the spectrum of costs or difficulty of developing renewable energy projects. For instance, lands with a higher slope would be given a subjectively heavier weighting than lands with lower slopes. The resulting map reflects a composite friction cost surface determined through combining renewable resource potential, environmental factors, and other land cover factors scaled to their contribution to the cost or difficulty of developing a renewable energy project. The inverse of the friction cost surface map was used as a proxy for transition probabilities, as it scales all values between 0 and 1, with the higher values representing a greater probability of transitioning to a renewable energy development site. These transition probability maps were then used in the LUCC modeling process as a point of comparison with the default statistical approach that Dinamica uses for developing transition probability maps. The steps of this analysis are described in Table 5 and illustrated in Figure 1. 9 Table 5: Multi-Criteria Evaluation and Friction Cost Surface Process for Developing Suitability Analysis Establish Development and Unmet Demand Baseline Step 1: Map existing renewable energy facilities. Step 2: Estimate the gap in energy production needed to meet demand. Create Multi-‐Criteria Evaluation of Suitable Areas for New Large-‐Scale Solar and Wind Facilities Step 4: Renewable potential: Include areas for which there is sufficient renewable resource potential. Step 5: Environmental constraints: Exclude protected areas, areas within a buffer from wetlands, and areas within a buffer from existing high-‐ and medium-‐density development. Step 6: Infrastructural requirements: Include areas within proximity to transmission lines and roads. Exclude areas within a buffer of existing renewable projects. Create Friction Cost Surface for Potential New Large-‐Scale Solar and Wind Facilities within Suitable Areas Defined by Multi-‐Criteria Evaluation Step 7: Renewable potential: Assign areas with lowest renewable resource potential to highest friction factor, indicating a preference for high potential areas. Step 8: Land cover factors: Assign areas that contain some land cover impediment the highest friction factor. Lands cover types such as barren land, agricultural fields, and grasslands were assigned the lowest friction factor given the relative ease of development in these areas. Land cover types such as forests, mixed forests, and low-‐intensity development or open space were assigned higher friction factors due to the higher cost and difficulty of development. Step 9: Environmental factors: Areas with the highest friction factor were lands exposed to the highest risk to habitat, lands with some degree of protection, and areas that contribute most to biodiversity. Create Suitability Map using Friction Cost Surface and Transition Probability Maps Step 10: Create friction cost surface map: Multiply friction factors and create a composite friction cost surface. Exclude all areas deemed unsuitable for renewable energy development in the MCE analysis. Step 11: Create transition probability map: Calculate the inverse of the friction cost surface map in order to generate a transition probability map for all suitable areas for renewable energy development. 10 Figure 1: Flow Diagram of Multi-Criteria Evaluation and Friction Cost Surface Processes to Create a Transition Probability Map 5 LUCC Model of Future Renewable Energy Development The suitability analysis was used as a basis for comparison with the outputs from the renewable energy development land use and cover change (LUCC) modeling process. The LUCC modeling process in Dinamica uses a “weights of evidence” statistical analysis to construct a probability map for future transitions. The LUCC model then utilizes Dinamica EGO’s Markov transition probability approach to estimate the magnitude of spatio-‐ temporal change for, in this case, transitions to renewable energy generation. This change in then distributed across the landscape based on the weights generated from the weights of evidence statistical analysis (Mas et al., 2014). The model can be calibrated by comparing simulated maps with actual landscape maps using a fuzzy similarity index and a constant decay multiple-‐window analyses. The calibrated model is then used to simulate future scenarios. The steps in this process are described in Table 6 and illustrated in Figure 2. 11 Table 6: Process for Developing Land Use Change Model for Renewable Energy Development Scenarios Generate Markov Transition Probability Matrices Step 1: Generate a historical transition matrix based on NLCD land cover types overlaid with existing renewable energy generation facilities. Step 2: Create discrete transition probability matrices to represent patterns that are not representative by historical trends (i.e., acceleration of renewable energy development). Apply Statistically-‐ or Expert-‐Driven Approaches to Generate Transition Probability Maps Step 3: Determine weights of evidence – the statistical relationships between static variables and likelihood that change occurs over specific ranges of values (how favorable an area is to change). Select variables that are significantly correlated with the desired transition, but are not cross-‐correlated with other static variables, in order to produce a map of spatial transition probability. If statistical relationships are not possible or desirable, expert-‐ driven estimate can be used, as well. These weights will be used to generate a transition probability map. Step 4: Use the suitability analysis to define the areas in which future renewable energy development can occur. Use the inverse friction cost surface map to specify a subjective or expert-‐drive approach to generate a transition probability map. Validate and Run LUCC Simulation Model Step 5: Run Expander to simulate spatial extent of existing patches of renewable energy developments. This calibrates the number of raster cells representing renewable energy development to the existing installed capacity. Step 6: Simulate historical transitions (2001-‐2011) using weights of evidence statistical relationships. Compare with simulations based on the expert-‐driven transition probability maps derived from the inverse friction cost surface maps. Step 7: Validate simulations on historical transitions (2001-‐2011) using exponential or constant decay distributions for adjacent cells. Run LUCC Simulation with Patch Formation and Expansion of Existing Patches Step 8: Run the simulation with the patch formation process, which is designed to generate new patches through a seeding mechanism that randomly selects viable areas for new patches with a specific geometry. Patch size and geometry was determined with a priori knowledge of renewable energy development historical patterns. Run LUCC Model under Different Demand Scenarios Step 9: Run LUCC model under different scenarios. 12 Figure 2: Renewable Energy Development LUCC Model Process 5.1 Renewable Energy Development LUCC Model Parameterization 5.1.1 Transition Matrices Dinamica uses a Markov transition probability (also known as Markov chain) approach to generate transition matrices to describe the probability of transitioning between different states. The probability that a cell will be in a given state at a given time1 is derived from the knowledge of its state at any earlier time0, and does not depend on the history of the system before time1. The Markov chain functionality in Dinamica was used in this study to determine the direction and magnitude of change in terms of renewable energy development land cover changes from the period of 2001 to 2011. The transition matrices generated using the Markov chain approach, however, reflect historical probabilities that were used as a baseline. Future rates of renewable energy development will have to significantly exceed historical rates in order meet Vermont’s renewable energy targets. Table 7 shows the historical transition rates. 13 Table 7: Historical Transition Matrices Transition from: 1 -‐ developed land 2 -‐ shrub, herbaceous, or barren land 3 -‐ forest or wetlands 4 -‐ farmland or pasture Ground Mounted Solar Wind 5 -‐ renewable energy development 0.00010999 0.00003578 0.00003142 0.00000571 0.00000725 0.00000242 0.00005960 0.00001842 5.1.2 Weights of Evidence Spatial data often violate the assumptions of statistical parametric methods such as the commonly used logistic regression technique. The weights of evidence Bayesian statistical technique is a more robust method with which to analyze spatial data because it allows for the relationship between the transition probability and the static variable to be flexible. Logistic regressions and other parametric methods, in contrast, assume that the relationship between the transition probability and the static variable is a sigmoid function, which is not always the case. Thus, utilizing the weights of evidence approach reduces the bias and subjectivity that commonly undermines multi-‐criteria evaluation techniques. In this study, the weights of evidence technique was used to compute transition potential maps based on infrastructural, biophysical, and socioeconomic data such as renewable resource potential, distance to various land cover types, and distance to roads and other infrastructure, while the Markov chains were used to generate the transition probabilities from the renewable generation cover maps. Transition potential maps represent the likelihood or the probability that the landscape would change from one land cover type to a renewable generation site, whereas the transition probabilities represent the temporal changes among the renewable energy cover classes. The cellular automata model functions were employed to simulate future renewable energy generation development based on the observed land cover maps, transition potential maps, and transition probabilities. 5.1.2.1 Correlations between Static Variables Since the basic assumption of the weights of evidence technique is that static variables must be independent, the correlation between static variables was tested using the Cramer coefficient (V). The weight of evidence correlations indicated that certain variables were cross-‐correlated. The Utility, RPC, and County static variables were all cross-‐correlated, as well as, “distance to roads” and “distance to development,” indicated by the Cramer V of greater than 0.3. As such, Utility, RPC, and “distance to roads” were removed from the weights of evidence analysis. Table 8 shows the Cramer correlation coefficients for the remaining static variables that were not cross-‐correlated. 14 Table 8: Average Cramer Correlation Coefficients for Static Variables for Both Transitions A B C D E F G H I J K L A B C D E F G H I J K L X 0.072 X N/A N/A X N/A N/A 0.004 X 0.161 N/A 0.191 0.049 X N/A N/A N/A N/A N/A X N/A N/A N/A N/A N/A N/A X 0.232 N/A 0.017 0.074 N/A 0.110 0.114 X 0.148 N/A 0.053 0.055 N/A 0.089 0.099 N/A X 0.253 0.036 0.010 0.218 0.068 0.159 0.172 0.190 0.159 X 0.099 N/A 0.070 0.023 N/A 0.065 0.031 0.052 0.144 N/A X 0.178 N/A 0.023 0.024 0.052 0.073 0.078 0.146 0.096 N/A 0.039 X A = County; B = Slope; C = Aspect; D = Building Density; E = NLCD; F = Renewable Potential; G = Distance to Potential Renewable Site; H = Distance to Transmission; I = Distance to Existing Development; J = Threat to Habitat; K = Distance to Wetlands; L = Protected Lands 5.1.2.2 Weights and Coefficients The weights of evidence coefficients estimate which weights are statistically significant and the range in which weights will be applied to generate transition probability map. Figure 3 displays graphs of the distribution of weights across different ranges for each static variable for the transition from developed land (land cover class 1) to ground-‐mounted solar (land cover class 2). These same weights were also calculated for the transitions from the three other specified land cover classes (2 – herbaceous and scrub land, 3 – forest land, and 4 – cropland and pasture) to a renewable energy development (ground-‐mounted solar or wind). Many of the weights generated for these static variables shown in Figure 3 were not statistically significant, largely due to the lack of transitions generated from the existing renewable development data from the Renewable Energy Atlas for Vermont. 15 Figure 3: Weights for Static Variables for the Transition from Developed Land to Ground-Mounted Solar 5.1.3 LUCC Model Calibration Once simulations were run using the statistical weights of evidence approach to produce a simulated map of future renewable energy development, a constant decay multiple-‐widow similarity of differences analysis was performed to calibrate the model to the observed final landscape map of renewable energy development. The constant decay multiple-‐window analysis provides another means of validating the simulated map. This analysis shows that the upper bound of the similarity of differences approaches 60 percent for ground-‐mounted solar at window size of 101, whereas the similarity of differences for wind plateaus at approximately 35 percent at a window size of 73. This result is expected given the differences between the comparatively denser and more developed ground-‐mounted solar and the sparsely developed wind. Consequently, there were significantly more data points on which to train the ground-‐mounted solar LUCC model compared to the wind LUCC model. Figure 4 (ground-‐mounted solar) and Figure 5 (wind) show how the similarity between the observed and simulated maps increases with window size. Given the limitations of the existing renewable energy development data sets and the resulting weakness of many of the weights of evidence, this level of similarity was deemed sufficient. 16 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101 Similarity Ground-‐Mounted Solar: Constant Decay Multiple-‐Window -‐ Similarity of Differences Window Size Figure 4: Ground-Mounted Solar - Constant Decay Multiple-Window Similarity of Differences Analysis Wind: Constant Decay Multiple-‐Window -‐ Similarity of Differences 0.3 0.2 0.1 0 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101 Similarity 0.4 Window Size Figure 5: Wind - Constant Decay Multiple-Window Similarity of Differences 5.2 Renewable Energy Development LUCC Model Simulation This cellular automata approach that Dinamica employs uses an Expander transition function to expand or contract previous land cover class patches, and a Patcher transition function used to form new patches. The Expander and Patcher transition functions both employ an allocation mechanism that identifies cells with the highest transition probabilities for each transition (Almeida et al., 2003). Assuming that all future renewable energy developments will be new developments, not expansions of previous developments, the Expander function was not utilized, only the Patcher function. The Patcher function performs transitions from state i (not a renewable energy development) to state j (renewable energy development) only in the neighboring cells with states other than j (e.g., any other land cover type). This transition function uses a stochastic selecting mechanism, in which an algorithm scans the initial renewable energy development land cover map to 17 identify cells with the highest transition probabilities and randomly selects from this data array cells to transition to new renewable energy developments (Soares-‐Filho et al., 2002). 5.2.1 Recreating the Spatial Extent of Existing Renewable Energy Developments As described in Section 2.2, the geocoded existing renewable energy developments were organized in a point shape file, which was then converted into a raster file to correspond with the NLCD land cover data. This meant that each point was converted into one 30m-‐ 30m cell, irrespective of the actual installed size or land area required by each installation. When accounting for the energy density of development in terms of hectares per MW installed, this resulted in a significant discrepancy between the spatial extent of existing renewable developments as represented in the raster data and the actual spatial extent. To better reflect the spatial extent of each installation, the Expander function in Dinamica was used to expand existing patches (the renewable energy developments). The Expander function was parameterized according to a lognormal probability function, whose parameters are defined by the mean patch size, patch size variance, and isometry. Statistics on the average size and standard deviation for installations within each land cover class was used for this purpose, as shown in Table 9. It was assumed that the isometry in all cases was 1.5, representing approximately uniform patch geometry. The statistical weight of evidence described in Section 5.1 was used to generate the transition probability map used to drive the transitions generated by the Expander function. The resulting simulated land cover map includes “expanded” patches of existing renewable energy developments that capture both the location and spatial extent of development as a function of the existing installed capacity. Figure A-‐12 and Figure A-‐13 in Appendix 1: Maps show observed land cover for ground-‐mounted solar and wind developments, respectively. These adjusted existing renewable developments map were used as the basis for recalculating the weights of evidence to simulate future renewable developments through patch creation. Table 9: Summary Statistics for Existing Renewable Developments Across Land Cover Types Land Cover Class Ground-‐Mounted Solar (1) Developed land (2) Herbaceous or barren land (3) Forest and Wetlands (4) Farmland and pasture Total Wind (1) Developed land (2) Herbaceous or barren land (3) Forest and Wetlands (4) Farmland and pasture Total Total Installed (kW) 13,874 1,052 13,203 4,143 32,273 1,076 69,255 42 50,609 120,981 Average Size (kW) 105.91 12.99 73.35 23.02 56.42 19 1,413 10 733 680 18 Standard Deviation (kW) 365.04 22.66 361.56 164.03 284.77 31.65 9,022.10 7.56 4,945.13 5,636 Average Size (ha) 0.291 0.036 0.201 0.063 0.155 0.02 1.24 0.01 0.64 0.59 Standard Deviation (ha) 1.002 0.062 0.992 0.450 0.781 0.03 7.89 0.01 4.32 4.93 5.2.2 Simulating New Renewable Energy Developments The simulated maps of existing renewable developments produced in the previous step were used as the baseline for simulating future patch creation, or new renewable energy developments. This was achieved using the Patcher function, which was parameterized using assumptions shown in Table 10. These parameters were informed by historical renewable energy development patterns and assumptions around the energy density of development, which were then applied to the Patcher transition function. Table 10: Parameters for Using Patcher for Future Simulations Land Cover Class Ground-‐Mounted Solar (1) Developed land (2) Herbaceous or barren land (3) Forest and Wetlands (4) Farmland and pasture Wind (1) Developed land (2) Herbaceous or barren land (3) Forest and Wetlands (4) Farmland and pasture Average Size (ha) 1.37 0.68 1.37 2.74 0.87 8.7 0.87 8.7 Standard Deviation (ha) 0.68 0.34 0.68 1.37 0.44 1.09 0.44 1.09 Isometry 1.5 1.5 1.5 1.5 1.5 1.5 1.5 1.5 Average Size (MW) 0.50 0.25 0.50 1.00 1.00 10.00 1.00 10.00 Standard Deviation (MW) 0.25 0.12 0.25 0.50 0.50 1.25 0.50 1.25 5.3 Scenario Modeling Following calibration and validation, a scenario-‐driven modeling approach was then used to simulate future renewable energy development patterns under three scenarios: (1) All Solar (all ground-‐mounted solar), (2) All Wind (medium and large-‐scale wind), (3) 50/50 Solar and Wind. Each scenario is designed to meet all current electricity demand, a total of 5,500,000 MWh per year, using in-‐state renewable energy resources (EIA, 2013). A such, these are hypothetical scenarios that assume no out of state imports of hydroelectricity and a 100 percent shift to renewable sources of electricity. It should be noted that electricity generated by biomass, such as from the Joseph C. McNeil Generating Station in Burlington, VT, would be included in any actionable plan for meeting future electricity demand with in-‐ state renewable resources. However, this analysis focuses just on ground-‐mounted solar and medium-‐ and large-‐scale wind developments as a proof of concept, which can be adjusted to accommodate other research questions. The renewable energy supply calculations are shown in Table 11, which indicate the renewable capacity needed to meet this current demand, and its translation into hectares and 30m-‐30m cells. 19 Table 11: Renewable Energy Supply Calculations for each Scenario Scenario Parameters Scenario 1: All Solar Scenario 2: All Wind Total MW for Current Demand Total Land Area (ha) Cell Transitions 3,970 10,893 121,039 2,281 1,994 22,152 Scenario 3: 50/50 Solar and Wind Solar Wind 1,985 1,140 5,447 997 60,519 11,076 Table 12 shows the historical and future projected transition rates for each scenario. The projected rates of transition were calculated based on a 1-‐step transition to the level of renewable energy development specified in the scenario, assuming that the historical proportion of renewable energy development across each land cover type will remain constant. Table 12: Historical and Projected Transition Rates for Each Scenario Transition from: Historical 1 -‐ developed land 2 -‐ shrub, herbaceous, or barren land 3 -‐ forest or wetlands 4 -‐ farmland or pasture Scenario 1: All Solar 1 -‐ developed land 2 -‐ shrub, herbaceous, or barren land 3 -‐ forest or wetlands 4 -‐ farmland or pasture Scenario 2: All Wind 1 -‐ developed land 2 -‐ shrub, herbaceous, or barren land 3 -‐ forest or wetlands 4 -‐ farmland or pasture Scenario 3: 50/50 Solar and Wind 1 -‐ developed land 2 -‐ shrub, herbaceous, or barren land 3 -‐ forest or wetlands 4 -‐ farmland or pasture Ground Mounted Solar Large-‐Scale Wind 5 -‐ renewable energy development 0.00011999 0.00003578 0.00003142 0.00000571 0.00000725 0.00000242 0.00005960 0.00001842 0.02386 0 0.00682 0 0.00157 0 0.01293 0 0 0.00445 0 0.00071 0 0.00030 0 0.00229 0.01193 0.00223 0.00341 0.00036 0.00079 0.00015 0.00646 0.00115 6 Results 6.1 Suitability Analysis for Renewable Energy Development Figure A-‐14 and Figure A-‐15 in Appendix 1: Maps show the results for the suitability analysis, represented as the inverse friction cost surface with an MCE filter for ground-‐ mounted solar and wind, respectively. The MCE map was produced as a result of a combination negative filters that excluded areas deemed inappropriate for ground-‐ mounted solar development. Figure A-‐16 (ground-‐mounted solar) and Figure A-‐17 (wind) 20 shows the suitability analysis with existing renewable developments overlaid. There is a clear coincidence between the suitability analysis and the location of existing renewable developments, however some existing sites fall outside the explicit boundaries demarcated in the analysis. This is largely a product of the fact that the MCE analysis defines viable development sites as those that fall in an area with sufficient renewable resource potential (REAVT, 2014). It appears as though historical development patterns did not perfectly adhere to the minimum renewable resource potential specification, which indicates that some development will occur in suboptimal conditions. This renders the suitability analysis less useful as a deterministic screen. The friction cost surface maps in Figure A-‐14 and Figure A-‐15 represent an estimate of the relative desirability of sites for renewable development, by incorporating subjective valuation of characteristics of desirable locations. This map was used as a comparison to the transition probability map produced as part of the application of the weights of evidence approach in the LUCC model described in Section Renewable Energy Development LUCC Model6.2. Ultimately, this suitability analysis was used just as a point of comparison with the transition probability maps produced in the LUCC modeling of renewable energy developments. The suitability analysis proved to be too coarse of a tool to supplement the weights of evidence statistical analysis. 6.2 Renewable Energy Development LUCC Model The renewable energy development LUCC model was designed to simulate future renewable energy development patterns. The model was built in Dinamica following the basic structure and framework provided in the Dinamica Guidebook (Soares-‐Filho et al., 2009). 6.2.1 Scenarios Three scenarios were modeled: (1) All Ground-‐Mounted (Figure A-‐18), (2) All Medium-‐ to Large-‐Scale Wind (Figure A-‐20), (3) 50/50 Solar and Wind (Figure A-‐22). Transition probability maps can be found in Appendix 1: Maps for Scenario 1 (Figure A-‐19) and Scenario 2 (Figure A-‐21). Simulation of Scenario 3 leverages the transition probability maps from Scenarios 1 and 2, so it was not provided separately. Table 13 shows descriptive statistics of the land use impacts for each scenario, while Figure 6 represents these data graphically. 21 Table 13: Land Cover Impacts of Renewable Energy Development Scenarios Land Cover Type Observed: Solar Ha MW Ha MW Ha MW Ha MW Ha MW Dev., Open Space Dev., Low Intensity Dev., Med. Intensity Dev., High Intensity 13.0 4.7 0.6 0.7 920 335 114 42 448 163 8.3 3.0 8.3 9.5 1,192 434 202 74 682 248 6.4 2.3 10.4 11.9 824 300 219 80 601 219 2.3 0.8 7.7 8.8 300 109 118 43 271 99 0.0 0.0 5.6 6.5 33 12 0 0 13 5 Barren Land Observed: Wind Scenario 1: All Solar Scenario 2: All Wind Scenario 3: 50/50 Solar & Wind Deciduous Forest 14.7 5.3 0.3 0.3 1,214 442 308 112 784 286 Evergreen Forest 4.7 1.7 19.6 22.4 849 309 137 50 527 192 Mixed Forest 7.0 2.6 3.6 4.1 818 298 145 53 478 174 Shrub/Scrub 3.7 1.3 4.8 5.4 344 125 36 13 194 71 57 21 15 6 41 15 3,855 1,405 720 262 2,479 904 Herbaceous 0.3 0.1 2.1 2.4 Hay/Pasture 36.2 13.2 37.7 43.2 Cultivated Crops 3.4 1.2 2.7 3.1 480 175 155 57 202 74 Woody Wetlands Emergent Wetlands 0.5 0.2 1.5 1.7 60 22 12 4 24 9 0.1 0.0 0.3 0.3 10 3 5 2 4 1 Total % VT Land 100.4 36.6 105.1 120.3 10,953 0.4% 3,992 2,186 0.1% 797 6,747 2,459 0.3% The land cover trade-‐offs between scenarios are clearly illustrated in Figure 6, with percentage distributions shown in Figure A-‐23, Figure A-‐24, and Figure A-‐25. Scenario 1 has significantly higher land cover impacts than other scenarios (10,953 hectares developed), with Scenario 2 the least impactful (2,186 hectares developed). Overall cropland and pasture and forestland are projected to absorb a large percentage of future renewable development under all scenarios. Due to the comparatively smaller contiguous land requirements of ground-‐mounted solar, more developed land is projected for this purpose. 22 Developed Area (hectares) Developed Area under each Renewable Energy Scenario 4,500 4,000 3,500 3,000 2,500 2,000 1,500 1,000 500 -‐ Current Landscape Scenario 1: All Solar Scenario 2: All Wind Scenario 3: 50/50 Solar and Wind Figure 6: Land Cover Impact for each Renewable Development Scenario There are likewise trade-‐offs related to land development on environmentally sensitive land, defined in terms of contribution to biodiversity and protected lands status. Table 14 and Figure 7 and Figure 8 represent the distribution of land impacts on environmentally sensitive lands. Not surprisingly, approximately 50 percent of renewable development in all scenarios falls in the category of the lowest contribution to biodiversity (6) (VANR, 2013). However, 12 to 20 percent of development is simulated to occur in the top three tiers (1-‐3), thus posing a potential threat to important species. Though the majority of renewable energy development was simulated to occur in unprotected land (1), there remains some pressure on lands with incomplete protected status, indicated by class 2 and 3 (VCGI, 2011). Taken together, scaling renewable energy development may have negative consequences for environmentally sensitive lands. Table 14: Environmentally Sensitive Land Impacts for Each Renewable Development Scenario Environmentally Sensitive Land Impacts (Hectares) Scenario 1: All Solar Scenario 2: All Wind Contribution to Biodiversity Classification (VANR, 2013) 1 -‐ Highest 217 2 530 3 752 4 3,196 5 913 6 – Lowest 5,344 23 Scenario 3: 50/50 Solar & Wind 166 110 132 539 262 977 191 250 361 1,743 695 3,508 Protected Lands Classification (VCGI, 2011) 1 – Unprotected Land 9,267 2 957 3 654 4 – Protected Land 72 1,814 75 269 28 5,845 464 398 40 Renewable Energy Development on Land Tiered by Contribution to Biodiversity Hectares 6,000 4,000 2,000 -‐ 1 2 3 4 5 6 Tiered Contribution to Biodiversity Scenario 1: All Solar Scenario 2: All Wind Scenario 3: 50/50 Solar and Wind Figure 7: Renewable Energy Development on Land Tiered by Contribution to Biodiversity Hectares Renewable Energy Development on Land Tiered by Protected Lands Status 10,000 8,000 6,000 4,000 2,000 -‐ 1 2 3 4 Protected Lands Status Scenario 1: All Solar Scenario 2: All Wind Scenario 3: 50/50 Solar and Wind Figure 8: Renewable Energy Development on Land Tiered by Protected Lands Status 7 Discussion Many studies of the transition to renewable focus on the aggregate, state-‐level impacts of renewable energy development, largely focusing on the benefits of solar and wind. Ingerson (2013) developed a compelling analysis of what the landscape-‐level impacts of renewable energy development in Vermont would be under very conservative scenarios of renewable penetration into the electricity and thermal sectors. This analysis builds on the 24 basic thesis that the myriad benefits of renewable energy development come with an often overlooked trade-‐off with land use, especially when considering the scale at which renewable energy is envisioned to meet future energy needs. The modeling and simulation approach taken in this study allows for a spatially explicit analysis of renewable energy development and diffusion patterns under different scenarios. The results of the modeling and simulation indicate that both ground-‐mounted solar and medium-‐ to large-‐scale wind would displace a significant amount of current land cover types, largely forest and cropland, if these sources are used to provide a substantial percentage of future energy needs. Over 10,000 hectares of ground-‐mounted solar would be needed to provide current electricity demand, given the current state of technology. Even if solar technology were to markedly improve, it is not likely to have a dramatic effect on the land intensity of development due to the need for supporting infrastructure and roads. In comparison, medium-‐ to large-‐scale wind seems a more appealing option, as approximately 1/5 of the area would need to be developed in order to meet current electricity demand. Much of this development, however, would need to take place on larger contiguous tracts of forest and cropland at higher elevations. To some degree, all scenarios would require development on environmentally sensitive land. All of these scenarios do not account for the dispatchable generation and storage that would necessary to smooth the generation profile of the stochastic supply profiles of solar and wind. These results were obtained by leveraging transition probability maps generated based on a “weights of evidence” statistical analysis of historical spatial patterns of development. If land use policies or other directives change the way in which renewable energy is developed in the future, it is possible that other land cover types would be subject to more pressure from development of renewable generation. These are hypothetical scenarios, as Vermont is currently heavily dependent on imports of electricity from large-‐scale hydro. If Vermont were to wean itself from these imports and rely more on in-‐state renewable resources, large tracts of primarily forest and cropland would need to be dedicated to renewable energy generation. As the Vermont Department of Public Service and many partnering organizations implement the 2011 Vermont Comprehensive Energy Plan’s goal of 90 percent renewable energy by 2050, further consideration needs to be taken of how best to govern land use decisions around how much and where to development renewable energy generation. Based on this analysis, it is recommended that energy policy and planning focus on strategically developing low-‐impact regions with medium-‐ to large-‐scale wind developments, and fill remaining gaps with medium-‐scale ground-‐mounted solar developments closer to developed areas. 8 Conclusions As Vermont considers how best to plan and strategically guide the transition to renewable energy, there are serious trade-‐offs to consider related to land use and cover. Renewable energy is generally a much more diffuse energy source, thus requiring larger areas of land than convention fossil fuel, nuclear, or large-‐scale hydro sources. When scaling renewable energy sources to supplant non-‐renewable sources at the level to which Vermont aspires, there will necessarily be more land dedicated to in-‐state electricity production than at any 25 point in history. This analysis and simulation was developed in Dinamica and explores what the land cover impacts would be if Vermont were to shift 100 percent of its electrical demand to in-‐state ground-‐mounted solar or medium-‐ to large-‐scale wind generation. In order to achieve this goal, three scenarios were modeled and simulation, Scenario 1 in which current demand would be met only with ground-‐mounted solar, Scenario 2 with medium-‐ to large-‐scale wind, and Scenario 3 with half coming from solar and the other half from wind. Scenario 1 results in the largest land cover impacts, with 10,952 hectares developed, 35 percent of which would take place in pastureland. Scenario 2 has the lowest impacts, with only 2,186 hectares developed, again, largely concentrated in pastureland and forestland. The result is fairly intuitive given the fact that solar requires over 250 percent the land per MW of installed capacity, and on average has a lower capacity factor than wind. Though solar performs more poorly in terms of its land intensity, it can be developed on less environmentally sensitive lands. Wind, in contrast, is best developed on Vermont’s ridge tops, which are both critical habitat for many species and also delicate environments for what amounts to industrial development. Even if Vermont elects to meet a fraction of future demand with in-‐state renewable resources, there will be large trade-‐ offs in terms of current land uses being repurposed for electricity generation. Acknowledgments C. Clement is supported by a fellowship from the National Science Foundation Integrative Graduate Education and Research Traineeship (IGERT) program on Smart Grid: Technology, Human Behavior, and Policy. He acknowledges the valuable contributions of data on existing renewable energy installations from the Renewable Energy Atlas of Vermont developed by the Vermont Sustainable Jobs Fund, assistance from the Dinamica EGO collaboration team based in the Federal University of Minas Gerais in Brazil, and analytical guidance and review from Research Assistant Professor Gillian Galford at the University of Vermont Gund Institute for Ecological Economics. 26 Appendix 1: Maps Figure A-9: 2011 Vermont Land Cover Reclassified into Four Categories 27 Figure A-10: 2011 Vermont Land Cover and Existing Ground-Mounted Solar Installations 28 Figure A-11: 2011 Vermont Land Cover and Existing Wind Solar Installations 29 Figure A-12: 2011 Observed Land Cover with Ground-Mounted Solar Developments (Inset Map) 30 Figure A-13: 2011 Observed Land Cover with Wind Development (Inset Map) 31 Figure A-14: Inverse Friction Cost Surface with Multi-Criteria Evaluation Filter - Ground-Mounted Solar 32 Figure A-15: Inverse Friction Cost Surface with Multi-Criteria Evaluation Filter - Wind Solar 33 Figure A-16: Inverse Friction Cost Surface with MCE Filter: Ground-Mounted with Existing Developments Overlaid (scaled to installed capacity (kW)) 34 Figure A-17: Inverse Friction Cost Surface with MCE Filter: Wind with Existing Developments Overlaid (scaled to installed capacity (kW)) 35 Figure A-18: Simulated Land Cover Map for Scenario 1: All Ground-Mounted Solar 36 Figure A-19: Transition Probability Map for Scenario 1: All Ground-Mounted Solar 37 Figure A-20: Simulated Land Cover Map for Scenario 2: All Medium- to Large-Scale Wind 38 Figure A-21: Transition Probability Map for Scenario 2: Medium- to Large-Scale Wind 39 Figure A-22: Simulated Land Cover Map for Scenario 3: 50/50 Ground-Mounted Solar and Wind 40 Scenario 1: All Solar Developed, Open Space 1% 4% 0% Developed, Low Intensity 8% Developed, Medium Intensity Developed, High Intensity 11% Barren Land Deciduous Forest 35% 8% Evergreen Forest Mixed Forest 3% 0% 11% Shrub/Scrub Herbaceous Hay/Pasture 1% 3% 7% Cultivated Crops 8% Woody Wetlands Emergent Herbaceous Wetlands Figure A-23: Scenario 1 - Land Cover Impacts Percentage Distribution 41 Scenario 2: All Wind Developed, Open Space 1% 0% Developed, Low Intensity 7% 5% Developed, Medium Intensity 9% Developed, High Intensity Barren Land 10% Deciduous Forest Evergreen Forest 33% 5% Mixed Forest Shrub/Scrub 0% 14% Herbaceous Hay/Pasture Cultivated Crops 1% 2% 7% 6% Woody Wetlands Emergent Herbaceous Wetlands Figure A-24: Scenario 2 - Land Cover Impacts Percentage Distribution 42 Scenario 3: 50/50 Solar & Wind 0% 3% Developed, Open Space 0% Developed, Low Intensity 7% Developed, Medium Intensity 10% Developed, High Intensity Barren Land 9% 37% Deciduous Forest Evergreen Forest 4% Mixed Forest Shrub/Scrub 0% 12% Herbaceous Hay/Pasture 0% 3% 7% Cultivated Crops 8% Woody Wetlands Emergent Herbaceous Wetlands Figure A-25: Scenario 3 - Land Cover Impacts Percentage Distribution 43 References 1. Almeida, C.M.; Batty, M.; Monteiro, A.M.V.; Camara, G.; Soares-Filho, B.S.; Cerqueira, G.C.; Pennachin, C.L. Stochastic cellular automata modelling of urban land use dynamics: Empirical development and estimation. Comput. Environ. Urban Syst. 2003, 27, 481-509. 2. Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM). (2011). Vermont Digital Elevation Models. (accessed on November, 14 2014). 3. Collins, Michael G., Steiner, Frederick R., & Rushman, Michael J. (2001). Land-Use Suitability Analysis in the United States: Historical Development and Promising Technological Achievements. Environmental Management, 28(5), 611-621. 4. Denholm, P.; Hand, M.; Jackson, M.; Ong, S. (2009). Land Use Requirements of Modern Wind Power Plants in the United States. 46 pp.; NREL Report No. TP-6A2-45834. 5. Hatziargyriou, N., Asano, H., Iravani, R., & Marnay, C. (2007). Microgrids. Power and Energy Magazine, IEEE, 5(4), 78-94. 6. Ingerson, A. (2013). Cumulative Landscape Impacts of Renewable Energy Alternatives for Northern New England. The Wilderness Society. Accessed at: http://sitingcommission.vermont.gov/sites/cep/files/Siting_Commission/Publications/Meeting022013/Cumulati ve%20Landscape%20Impacts%20of%20RE%20in%20NE%201-15-13%20rev%20(1).pdf 7. Jin, Suming, Yang, Limin, Danielson, Patrick, Homer, Collin, Fry, Joyce, & Xian, George. (2013). A comprehensive change detection method for updating the National Land Cover Database to circa 2011. Remote Sensing of Environment, 132, 159. 8. Mas, Jean-François, Kolb, Melanie, Paegelow, Martin, Camacho Olmedo, María Teresa, & Houet, Thomas. (2014). Inductive pattern-based land use/cover change models: A comparison of four software packages. Environmental Modeling & Software, 51, 94-111. 9. Moreno, N.; Wang, F.; Marceau, D.J. A geographic object-based approach in cellular automata modeling. Photogramm. Eng. Rem. Sens. 2010, 76, 183-191. 10. Ong, S.; Campbell, C.; Denholm, P.; Margolis, R.; Heath, G. (2013). Land-Use Requirements for Solar Power Plants in the United States. 47 pp.; NREL Report No. TP-6A20-56290. 11. Renewable Energy Atlas of Vermont. (REAV) (2014). Vermont Sustainable Jobs Fund. Retrieved September 19, 2014, from www.vsjf.org/resources/renewable-energy-atlas 12. Smil, V. (2013). Understanding the Spatial Dimension of the Unfolding Transition to Renewable Electricity Generation. www.vaclavsmil.com/wp-content/uploads/docs/smil-article-power-density-primer.pdf 13. Soares-Filho, B.S.; Cerqueira, G.C.; Pennachin, C.L. DINAMICA—A stochastic cellular automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier. Ecol. Model. 2002, 154, 217-235. 14. Soares-Filho, B. S.; Rodrigues, H. O.; Costa, W. L. Modeling environmental dynamics with Dinamica EGO (Guidebook). (2009). (Rev. Ed., July, 2013, Davis, J. L.). http://csr.ufmg.br/dinamica/dokuwiki/doku.php?id=tutorial:start#guidebook (accessed September 2, 2014). 15. UFMG (Universidade Federal de Minas Gerais). Dinamica EGO; Centro de Sensoriamento Remoto/ UFMG: Belo Horizonte, Brazil, 2009. Available online: www.csr.ufmg.br/dinamica/ (accessed on November, 14 2014). 16. U.S. Census Bureau. (2012). Vermont: 2010. Population and housing unit counts (2010 Census of population and housing, CPH-2-47). Washington DC. 17. Vermont Agency of Natural Resources (VANR). (2013). Biofinder. http://biofinder.vt.gov. 18. Vermont Center for Geographic Information (VCGI). http://vcgi.vermont.gov/ 19. Vermont Comprehensive Energy Plan (CEP). (2011). http://publicservice.vermont.gov/publications/energy_plan 44