Forest Stewardship Potential Pilot Overlay Study – Methodology Massachusetts Dept. of Conservation & Recreation Outline • Factors of Influence and Datalayer development Fire Protection Assessment: Risk of Insects/Pests: Risk of Development (change in census block households Private Forested Lands Wetlands: 1. Extraction guides 2. Data extraction 3. Creating the composite Forest Patches: Riparian Corridors: Natural Heritage Priority Habitats (Threatened & Endangered Species): Proximity to Publicly Protected Lands: Slopes Public Water Supply Areas: Analysis Mask Private Forest Mask Non-forest – Non-developed (NFND) Mask **All MA watersheds are High Priority, thus excluded from the analysis • The Overlay Model Model Overview Data Aggregation • Weighting and Quantitative Analysis 1. The Weighting schemes - Interval Scale Weighting Scheme - Rank Scale Weighting Scheme 2. Quantitative analysis 3. Determination of final Weighting scheme • Generating statistics from the overlay analysis Maps, Tables & Diagrams Map 1: Wetlands Composite Creation Map 2: Forest Fragmentation issues Map 3: Continuous grid output from overlay Map 4: Discrete grid with 3 classes Table 1: Natural Breaks for Data Aggregation Table 2: Interval Scale Weighting Scheme Table 3: Rank Scale Weighting Scheme Table 4: Comparison of Test Results Table 5: Tests on Continuous vs. Aggregated Results Diagram 1: Wetlands Composite Overlay Model Diagram 2: Composite Mask Model Diagram 3: Stewardship Potential Overlay Model Diagram 4: Results of aggregation 3/30/2006 (MA_Methodology.doc) Pg. 1 of 15 Factors of Influence and Datalayer development The four pilot states decided upon 11 factors that play a key role in influencing the suitability of land for Forests Stewardship. They were evaluated on 2 basic qualities; Threat to Resources and Resource Potential The Factors posing Threats to Resource Conservation: • Fire Protection Assessment • Risk of Insects/Pests • Risk of Development (change in census block households) Factors with Resource Potential: • Private Forested Lands • Wetlands • Forest Patches • Riparian corridors • Natural Heritage Priority Habitats (Threatened & Endangered Species) • Proximity to Publicly Protected Lands • Slopes • Public Water Supply Areas In addition, an analysis mask was developed to exclude from the analysis, areas that are not eligible for Stewardship for various reasons discussed further on. Fire Protection Assessment: This datalayer is composed of Massachusetts towns that have had a moderate to high incidence of fires. The original dataset was developed by the State of Maine GIS Team for the U.S. Forest Service North East Compact from tabular data supplied by each state about the number of fires in each town. It is a preliminary but dynamic dataset that is being improved and developed continuously. The original data was queried to produce a Raster layer that shows towns with "Firevalue" of 2 or 3. Risk of Insects/Pests: This datalayer shows areas of Massachusetts that have had insect infestations for 3 or more years in the past 10 years. It is derived from vector datalayers developed annually and maintained by the Department of Conservation Recreation. Risk of Development (change in census block households): This raster layer shows areas of Massachusetts where change in households per square mile between 1990 and 2000 has been less than or equal to 20. In other words, where the increase in the number of households has been 0 to 20 in the time period mentioned. The data are derived from Census block group data that were linked to a shapefile of Massachusetts towns. Private Forested Lands: This datalayer describes forested land in Massachusetts that are not within the extent of the Analysis Mask. The Mask covers Developed areas derived from the MRLC dataset, Surface Water from MRLC, Public non-CR land from MassGIS Open Space data and area covered by the MA Military Reservation). The Forested land extent was derived by querying MRLC's National Land Cover Datalayer (NLCD) datalayer for Forested Uplands, Shrubland and Woody Wetlands. Areas in common with the Analysis Mask were determined by adding the two grids and were then excluded using the Reclassification tool in Spatial Analyst. Wetlands: This layer shows wetland areas in Massachusetts as of May 2003. It was derived from three wetlands datalayers of varying resolutions and coverage of the state. They were pieced together 3/30/2006 (MA_Methodology.doc) Pg. 2 of 15 so that the best data was used in its entirety for areas where it was available and the remainder of the state was covered by data of the next best resolution and so on. The datalayers in order of decreasing resolution are: Orthophoto Wetlands and streams at 1:5000 (developed by DEP GIS), National Wetland Inventory or NWI data (developed by the US Fish & Wildlife Service) and Hydrography at 1:25,000 (developed by MassGIS). The tripartite process for creating this composite wetlands layer is: 1. creating ‘cookie-cutter’ extraction guides for areas of data availability at each level 2. extracting the data desired for each of those extents 3. putting together the three different extents for the State composite 1. Extraction guides: Each of these wetlands datalayers have Index shapefiles showing the current availability of data over the State of Massachusetts. These were clipped, where necessary to make ‘cookie-cutters’ for the actual data. To illustrate, the highest resolution, 1:5000 datalayer was used in its entirety but did not provide complete, statewide coverage at the time. So the next best data, all available NWI data, was used for the areas not covered by 0 20 Miles 1:5000. The exact extent of this data to be 'Cookie-Cutters' used had to be derived by excluding areas that were covered by the higher resolution data. Thus the shapefiles indicating availability Data Extracted for the NWI data was clipped to exclude the higher resolution (1:5000) data in order to create a sort of data ‘cookie-cutter’ that was Best resolution (5k) 2nd best resolution (NW I) Lowest resolution (25k) used to extract the required data from Map 1: Wetlands Composite Creation the NWI layer. The NWI layer however also has incomplete statewide coverage. The remainder of the state not covered by the NWI data was filled with wetlands from the 1:25,000 layer. Tools used to query and clip these vector files included ArcView’s GeoProcessing extension. 2. Data extraction: For each of the datalayers, features that fell within their respective extraction guides were queried using standard ArcView Theme Selection tools and then converted directly to grids using Spatial Analyst. Each grid was further formatted so that cells with wetlands had a value of 1 while the rest of the extent had cell values of 0. The extraction guides as well as data extracted are illustrated in Map 1 above. 3. Creating the composite: The data used for wetlands is periodically updated as more data becomes available for the state. This sort of update occurred twice in the duration of this pilot study, and the final wetlands datalayer was updated each time in order to fully avail of the best data at hand. In order to efficiently incorporate these updates into the wetlands datalayer, the final step of combining the different wetlands extents was modeled as an Arithmetic overlay in the ModelBuilder Extension. Each layer was added as a categorical grid with no transformation or weight attached. The resultant grid was reclassified with values of 1 to clearly indicate wetland areas for the entire state of Massachusetts to the best accuracy available at the time. 3/30/2006 (MA_Methodology.doc) Pg. 3 of 15 Diagram 1: Wetlands Composite Overlay Model Forest Patches: The source for Forest related information in this study has been MRLC data. The original land cover data set was produced as part of a cooperative project between the USGS and the USEPA to produce a consistent, land cover data layer for the conterminous US based on 30-meter Landsat Thematic Mapper (TM) data. National Land Cover Datalayer (NLCD) was developed from TM data acquired by the Multi-Resolution Land Characterization (MRLC) Consortium. Our Forest Cover layer was derived from the NLCD layer for Massachusetts by agglomerating areas of the following NLCD Land Cover Classes: Barren (Transitional), Forested Uplands (Deciduous, Evergreen, and Mixed), Wetlands (Woody, Emergent Herbaceous). All this forested land however is not truly contiguous land, but rather fragmented into discrete ‘patches’ by roads and highways (amongst other factors) that has a direct effect on the feasibility, or lack thereof, of forest Stewardship and management of such areas. A further consideration, once patches have been identified, is their size as smaller areas may poses difficulties in terms of Stewardship planning and management. Given this, it becomes necessary to identify not only Forested areas that are actually contiguous (‘patches’), but those that are also of a size that allows effective Stewardship. Roads were decided upon as the primary guide for patch identification and statewide forested areas fragmented into smaller patches using buffered roads. Orthophotos suggest that cleared shoulders off roads are on average roughly 15 meters (50ft), making that the optimal buffer distance. MA Highway Dept. Roads, maintained as vector files by MassGIS were used as base information for roads. The Spatial Analyst ‘Straight Line Distance’ tool created the desired buffer with an extra allowance to account for inaccuracies created in the vector-raster conversion process. The buffer thus varies from 20 to 60m depending on road direction. The final vector file used for fragmentation contains the roads as well as the buffers. Of the resultant forest ‘patches’, those below 1000 acres in size were excluded in the interest of identifying realistic Stewardship potential. Calculation of patch size posed some difficulty as the grid would visually indicate forest areas distinctly separated from others, but the grid was homogenous. This is due to the way the source grid (NLCD Forested area) was developed; it was the results of a query so all cell values were 1. It was not possible to calculate area of any discrete cluster of cells directly from this grid. The solution was to convert the grid to a shapefile, thus giving each patch a discrete shape and area that could be calculated. 3/30/2006 (MA_Methodology.doc) Pg. 4 of 15 Map 2: Forest Fragmentation issues A second problem was that the visually distinct patches also very often touched at diagonally opposite corners, functionally uniting the two patches when in reality the connection was a combination of the raster effect and resolution. This had the additional advantage of removing the misleading effect of cells from different patches that touch at diagonally opposite corners to visually appear to consolidate the 2 patches as illustrated above in Map 2. Riparian Corridors: This raster datalayer shows 100 meter riparian corridors encompassing perennial stream and river features from the MassGIS 1:25,000 hydro datalayer. Rivers and Streams were buffered by 100 meters by MassGIS to derive this data. The MassGIS vector file was queried for the field ‘Inside’ equal to 100 and converted to Grid format. Natural Heritage Priority Habitats (Threatened & Endangered Species): This raster data layer depicts estimated priority habitats of rare species in Massachusetts. It is a composite of two NHESP vector files: 'Priority Habitats' describing rare species priority habitats, and 'Estimated Habitats' including similar habitats that are in 'Resource Areas', as defined in the Wetlands Protection Act Regulations (310 CMR 10.02(1)). The two shapefiles were combined using ArcView’s GeoProcessing extension and ‘Merge’ tool. Proximity to Publicly Protected Lands: This raster layer shows areas of proximity to Public lands. The 'Protected and Recreational Open Space' vector datalayer maintained by MassGIS was queried to derive protected Open Space. Further queries distinguished areas under Conservation Restrictions from owned and managed Public access property. Quarter mile buffers were calculated on both of these extents and merged before conversion to raster format. Areas under Conservation Restriction were included in the data layer too as there is potential for greater Stewardship on those lands through acquisition and ownership. Slopes: This raster data layer describes areas of Massachusetts where slope is more than 15% and less than 30%. Percentage slope was derived from MassGIS 30 meter Digital Elevation Model (DEM) raster files using the Spatial Analyst ‘Surface Analysis’ tool. The DEM files are tiled so the slope rasters had to be merged together for coverage of the entire State. The resulting slope layer had a number of very small data gaps of Null cells corresponding in the most part to seams where individual tiles come together. A number of them however indicated locations of stone quarries (as indicated by topographical quads) for which information was presumably unknown or uncertain. Null cells that were not quarry pits were filled in using a Nearest Neighbor type function using ArcINFO’s GRID command ‘focalmean’. Each null cell was replaced with a mean value calculated from a rectangle measuring 3 cells by 3 cells centered on it. The final slope layer was then queried to the desired slope range of 15% to 30% slope. Public Water Supply Areas: This raster data layer is a composite derived from the following 3 DEP vector data layers: Wellhead Protection Areas - Zone II, Wellhead Protection Areas - IWPA (Interim Wellhead Protection Area) and Surface Water Supply Protection Areas - Zone C. This grid describes 3/30/2006 (MA_Methodology.doc) Pg. 5 of 15 areas that fall into the above categories with a value of 1 and the rest of the state of Massachusetts with the value 0. ArcView’s GeoProcessing Extension was used and the ‘Merge’ and ‘Dissolve’ functions within it. Analysis Mask: The analysis mask is a grid formatted such that all cells meeting the criteria for the Mask hold no value (are Null cells) while the rest of the Grid extent is holds values of 0. As a background mask within the Spatial Analyst environment as well as when treated as a datalayer, this has the effect of excluding from analysis cells in any datalayer that are coincident with the Mask’s null cells. The mask is composed of Surface Water, Urbanized areas, Public Protected Lands, bounds of the Massachusetts Military Reservation and a grid showing the extent of the state of MA. Surface Water was queried from the MRLC NLCD dataset category of Open Water (11) to produce a grid which was then converted to a vector format to allow calculation of discrete patches or water bodies. A Massachusetts Water Bodies Act that designates water bodies over 10 acres to be Public Property lead to only areas of Open Water over 10 acres to be included as Public Property and thus part of the Analysis mask. The vector layer was queried for such areas and converted back to raster format to be included in the mask. Urbanized areas were also queried from the NLCD’s categories 21, 22 and 23. (respectively, ‘Low Intensity Residential’, ‘High Intensity Residential’ & ‘Commercial/Industrial/Transportation’). The Public Protected lands are vector to raster conversions of queries of the MassGIS Openspace layer. They only include lands that are publicly owned and managed (i.e. by MA DCR) and not the Conservation Restrictions (which do actually have Stewardship potential). The MMR layer is simply a raster grid of the facility area. Finally, the MA Mask layer serves to exclude from Diagram 2: Composite Mask Model analysis cells within the rectangular bound or extent of the MA State grid, but outside the actual State’s geographical boundary. This discards from analysis those cells placed over the ocean or Rhode Island. All these layers are formatted such that cells meeting the criteria of each layer are Null, while the rest of the extent has a cell value of 1. The exception is the MA layer, which holds 0 in cells within the state and Null outside of it. An overlay model was created to combine these and create a composite mask given that at least one of the components of the mask, Publicly Protected Mask, is dynamic and sees frequent updates. The mask has been updated at least twice within the duration of this study. All layers were added as ‘categorical’ themes except the MA State layer, which was added as a ‘Numerical’ theme. A final Reclassification then converts all the non-null cells to 0 so that the Mask is ready to be used in the overlay. Private Forest Mask: The Private Forest Mask is a grid layer created using the Overlay Analysis result (the High-Medium-Low Composite) and represents Private Forest areas of High, Medium and Low Stewardship Potential. A Private Forest layer was created from the following categories of land use of the MRLC NLCD datalayer: Forested Uplands (41, 42 & 43), Shrubland (51), Non-natural Woody (61) and Woody Wetlands (91). All cells in these categories were coded 0 while the rest of the grid cells were given Null values, in effect creating a mask of Private Forests. When combined with the HML Composite grid (through raster addition), this has the effect of masking out or removing all cells in the HML Composite that are NOT Private Forest and showing only cells that 3/30/2006 (MA_Methodology.doc) Pg. 6 of 15 are. In addition the Private Forest cells now also have High, Medium or Low values showing Forest Stewardship Potential. Non-forest – Non-developed Mask (NFND) Mask: The NFND Mask is complementary to the Private Forest and Analysis masks. It is a grid layer created also with the Overlay Analysis result and represents Non-forest – Non-developed areas of High, Medium and Low Stewardship Potential. The actual NFND layer was created from the following categories of land use of the MRLC NLCD datalayer: Barren – Bare Rock/Sand/Clay (31), Barren –Transitional (33), Herbaceous Upland Grasslands (71), Herbaceous – Planted/Cultivated (81, 82, 83, 84, 85) and Emergent Herbaceous Wetlands 92. To ensure that this layer in addition to the Private Forest Mask and the Analysis Mask exhaustively covers all land-use categories in the MRLC NLCD, Open Water (11) patches less than 10 acres were also included. The rest of the Open Water category (i.e. areas over 10 acres) is included in the Analysis Mask. As with the Private Forest Mask, all cells in these categories were coded 0 and the rest of the grid coded as Null. Raster addition with the HML Composite followed, resulting in the masking out or removal of all cells in the HML Composite that are NOT on NFND Land, showing only those that are. The Overlay Model Model Overview The Overlay model was created using the Modelbuilder extension of ArcView. The Model spatially combines by addition the 11 different datalayers, each weighted according to their importance to Stewardship Potential. An Arithmetic Overlay was used to do this where each of the 13 datalayers were added as Numerical themes, with a Multiplier corresponding to predetermined Relative Weights (see section on Weighting and Quantitative Analysis). An analysis ‘mask’, created to exclude certain areas from consideration, was added to the overlay as a ‘Categorical’ theme with no particular weight assigned (its multiplier was 1). Diagram 3: Stewardship Potential Overlay Model The format followed to prepare each datalayer for this Arithmetic overlay was to code each input grid cell that meets the datalayer’s definition or criteria to cell values of 1, and the rest of the State of Massachusetts to cell values of 0. For example, for the Private Forest datalayer, all cells that 3/30/2006 (MA_Methodology.doc) Pg. 7 of 15 meet the predefined conditions for being Private Forests are reclassified to 1, while the remainders of grid cells that lie within the border of the State are classified to 0. The data development process for most datalayers resulted in values of 1 for cells meeting the criteria, and in cases where it didn’t but rather held other values resulting from querying or vector-raster conversion, they were simply reclassified to a value of 1. However the initial querying to develop the datalayers resulted in cells that did not meet the query having ‘No Data’ or Null cells. These cells have the effect of canceling out any other valid cell value within any sort of analysis that combines cells in 2 or more datalayers. In other words, if a cell with a value of 2 in a grid was added (in a Spatial Calculation) to a Null cell in a second grid, the corresponding cell in the output grid from the process would be a Null cell. In the Overlay process, this would mean that the only cells from all the datalayers that would appear in the end result would be ones that had values of 1 coinciding with values of 1 in the very first datalayer in the overlay. The rest of the data in other layers would be lost. All null cells within the datalayers were thus replaced with a value of 0 using the ‘IsNull’ and ‘Con’ functions in the Spatial Analyst ‘Map Calculator’ Tool. The result of this overlay is a continuous grid with values ranging from 0 to just under 1 where increasing cell value represents increasing degree of Stewardship suitability. Cells with a value of 0 indicate areas where none of the factors exert any influence and Stewardship Potential is the lowest, but not necessarily absent. Cells of ‘No Data’ in this grid correspond to the Analysis Mask – which does represent areas where Stewardship Potential is absent. Map 3 below shows a section of Massachusetts from this grid. Map 3: Continuous grid output from overlay Data Aggregation The continuous grid provides a good general impression of Forest Stewardship Potential throughout the state but is not conducive to a functional identification of discrete areas showing different levels of suitability. The distribution of cells within this rather large range of values results in some very small extents that are not easy to visually identify. The underlying trend of Forest Stewardship Potential can be better revealed by grouping together these raw values into classes, or in other words, aggregating the data. There is also compelling statistical support for aggregating this data, discussed in a following section on ‘Weighting and Quantitative Testing’. In order to do this the Continuous grid had to be converted to a Discrete Grid using ArcView’s Spatial Analyst (Map Calculator with a [Grid].Int expression). As a precursor to this step, the fractional cell values (below 1) in the continuous grid had to be converted to values over 1 in 3/30/2006 (MA_Methodology.doc) Pg. 8 of 15 order to preserve the data by multiplying by 10,000. The discrete grid was then reclassified to 3 classes of Natural Breaks. Since the actual cell values are amalgams of various datalayer weights, they have no direct meaning except in relation to one another to indicate trend. So for the sake of simplicity each class was assigned a value from 1 to 3 to indicate the level of Stewardship Potential. The three classes defined by the Natural Breaks in the statewide data, the values they were reassigned and what they represent is summarized below: Natural Breaks Reclassified to: Representing: 0 - 2718 1 Low Stewardship Potential 2719 - 4528 2 Medium Stewardship Potential 4529 - 9611 3 High Stewardship Potential Table 1: Natural Breaks for Data Aggregation Map 4 below shows this aggregated grid for the same region of the state as was shown in Map 3 on the Continuous grid. Visual comparison shows that the aggregated grid serves much better to highlight the trend while at the same time enabling identification of more useful discrete areas of specific Stewardship Potential. Map 4: Discrete grid with 3 classes Weighting and Quantitative Analysis There are various ways of identifying areas with potential for Forest Stewardship based on a given number of factors. The most straightforward one would be a basic Overlay analysis. Such a model would simply combine datalayers spatially representing the different factors to produce a map of areas of Stewardship potential. However the fact is that neither the factors influencing the Stewardship Potential of land, nor any possible area qualifying for Stewardship is likely to have homogenous attributes. There is bound to be varying degrees of influence of the different 3/30/2006 (MA_Methodology.doc) Pg. 9 of 15 Stewardship factors where some criteria are more important to conservation goals than others. Similarly, Stewardship Potential is very unlikely to be even throughout the state. Some areas will meet more Stewardship criteria and be better suited to stewardship than others. Given this kind of intrinsic and inevitable variation on both sides of the analysis, some sort of evaluation seems necessary to attempt to quantify or otherwise assess the relative importance of the different factors. A Weighted overlay model allows for just this by rating each input layer relative to the others. The process of calculating and devising an appropriate weighting scheme and then calculating the weights was a tripartite one: 1. Two different weighting schemes were designed to measure the relative importance of the datalayers. Each scheme employed a slightly different approach to evaluating each datalayer. A group of Forestry experts were asked to assign weights to each datalayer according to both schemes. 2. The overlays were run and Quantitative analysis performed on the results to check for any significant difference between weighted and non-weighted analysis results and between the two weighting schemes. 3. The applicability of weighting was determined and then one weighting scheme decided upon based on the quantitative analysis as well as detailed observations. Each of these steps is discussed further below. 1. The Weighting schemes Ten Massachusetts State Foresters evaluated the factors influencing Forest Stewardship Potential and attributed weights to the respective datalayers. The Foresters were chosen by virtue of their professional knowledge of factors influencing forest health and management, as well as their familiarity with more practical Stewardship issues like feasibility. The process of deriving weights for datalayers involved summarizing the ten essentially qualitative, subjective assessments to extract valid and meaningful weights. A predetermined weighting scheme was the only way for each Forester to evaluate each factor on a common scale shared by all Foresters. Two different scales were tested to try and allow for qualitative and quantitative opinions with the intention of picking the one that worked best through statistical testing. The ‘Interval Scale’ scheme allowed an exact numerical input within a defined range so each individual could offer a quantitative assessment of each datalayer’s importance. The ‘Rank Scale’ scheme on the other hand, called for ordinal numbers that allow for a more qualitative, ‘relative’ assessment. Both schemes ensured that the same scale is shared by all ten Foresters so that their Weight assessments are comparable. Interval Scale Weighting Scheme: This scheme comprises values from 0 to 2 in increments of 0.25, where 1 indicates a normal or medium degree of influence. Weights from all the foresters were averaged to find a ‘Mean weight’ for each datalayer. The importance of each datalayer in relation to the others was represented by ‘Relative weights’ calculated as each datalayer’s Mean weight as a proportion of the Total Mean weights (see Table 1 below). These proportions were utilized to weight each input datalayer in the overlay model in order to preserve their varying degrees of influence on Stewardship Potential, relative to one another. 3/30/2006 (MA_Methodology.doc) Pg. 10 of 15 Overlay Weighting Scheme 2: Interval Scale Decimal increments of .25 in a range of 0-2, where 1 is normal Datalayer Mean Weight As a % Relative Weight Fire Protection Assessments 0.5 0.5 0.75 0.3 1.5 0.5 0 0.25 1 0.25 0 0.5 4% 0.0389 Forest Patch 1.5 1 1.75 1.5 1.5 0.75 1 2 2 1.25 2 1.5 11% 0.1149 Natural Heritage Priority Habitats 1.5 1.25 1.5 1 1 1.5 1 1.75 0.5 1.75 1.5 1.3 10% 0.1007 Risk of Development 1.5 1 1.5 1.5 1.5 1.2 2 2 2 2 1 1.6 12% 0.1216 Insects/Pest Risk 0.5 0.25 0.25 1 0.25 1 2 0.75 1 0.25 0 0.7 5% 0.0513 2 1.75 1.25 1.5 0.75 1.2 2 1.5 1 1.25 1.5 1.4 11% 0.1110 Riparian Areas Public Water Supply Buffers 1 1.25 1.25 2 0.75 1 2 1.5 2 1.25 1.5 1.4 11% 0.1096 0.8 0.25 0.75 1 0.25 0.75 0 0.25 0.25 0.5 1 0.5 4% 0.0407 2 1 0.5 2 1.5 1 1 1 1 0.75 2 1.3 10% 0.0972 2 2 1.25 2 1.5 1 1 1.75 1 1.5 1 1.5 11% 0.1131 1 2 2 ** This is used as a surrogate for Wetlands 0.5 1 1.5 1 1 2 1 0.8 Total: 1.3 12.86 10% 90% 0.1011 1.00 Slope **Proximity to Successional Lands Proximity to Protected Openspace Private Forests Table 2: Interval Scale Weighting Scheme Rank Scale Weighting Scheme: This weighting scheme comprised an Ordinal or Rank type of score. Foresters assigned each datalayer a number ranging from 1 to 13, 1 being the highest, 13 the lowest. Weights were averaged and converted to ‘Mean weights’ similar to the previous scheme. The scale was constructed with 1 as the highest value to aid the weight assessment process. The analysis however, requires datalayer importance and magnitude of weight to advance or decrease in similar directions. In other words, datalayers of greater importance should have larger weights than those of relatively less importance. So each Mean was subtracted from the highest possible weight (that is, 13) to derive the ‘Inverse Weight’. These were then converted to ‘Relative Weights’ as a proportion of the Total Inverse Weights. Score Fire Protection Assessments Overlay Weighting Scheme 1: Ordinal or Rank Scale Values from 1 to 13 (1is the highest weight, 13 the lowest) Mean Weight Inverse Weight Relative Weight (proportion) 13 13 13 4 8 10 12 8 11 10 12 10.4 2.6 0.0373 Forest Patch Size 5 10 2 1 1 3 2 1 1 4 4 3.1 9.9 0.1403 Natural Heritage Priority Habitats 6 6 9 6 2 4 3 9 5 2 5 5.2 7.8 0.1107 Risk of Development 4 9 4 12 4 1 4 4 6 1 10 5.4 7.6 0.1081 Insects/Pest Risk 12 11 11 13 13 3 11 7 11 13 11 10.5 2.5 0.0348 Riparian Areas 3 4 6 9 7 2 5 10 8 8 6 6.2 6.8 0.0965 Public Water Supply Buffers 7 3 5 10 6 4 7 3 7 7 3 5.6 7.4 0.1043 11 12 12 7 12 10 13 12 12 12 9 11.1 1.9 0.0270 2 5 7 2 11 4 8 11 4 11 2 6.1 6.9 0.0978 1 1 1 3 5 2 9 5 3 3 7 3.6 9.4 0.1326 6 6 9 **This used as a surrogate for wetlands 6 2 4 3 9 5 2 5 Total: 5.2 72.4 7.8 70.6 0.1105 1.00 Slope **Proximity to Successional Lands Proximity to Protected Openspace Private Forests Table 3: Rank Scale Weighting Scheme 3/30/2006 (MA_Methodology.doc) Pg. 11 of 15 2. Quantitative analysis The theoretical efficacy of using weights can be verified to a large extent through Quantitative analysis of the results of Overlay analysis using each Weighting Scheme. Overlay models were thus run using the Relative Weights derived from both the Weighting schemes as well as a non-weighted Analysis, where every datalayer was assigned equal weight. The dataset or ‘Sample Population’ used for Statistical testing comprised a thousand random locations over the State. Data was extracted at each of these locations from each of the overlay results; the nonweighted and the two weighted overlays. The process of extracting this data was as follows: • A Random Point Generator Extension in ArcView was used to generate Random locations with a minimum of 50 meters between points within the State boundaries. A point feature vector file was created as a result that held three thousand points. (The Extension is available at http://arcscripts.esri.com/details.asp?dbid=11605) • A ESRI Avenue script was then run to populate each Random Point in the file with the value of the grid cell coinciding with it within each of the three overlay Grids. In other words, this was done for the non-weighted grid as well as the grid weighted according to both the weighting schemes. (The script is available at http://arcscripts.esri.com/details.asp?dbid=10200) • The Random Points shapefile was queried to discard all features with Null values (where the points were in Masked areas which had Null cells) and the first thousand Random Points exported to a database. • This data was then subjected to Paired Two Sample for Means t-tests to test for variance. All Statistical analysis was conducted in Microsoft Excel using the Add-In ‘Analysis ToolPak VBA’. Initially the basic overlay outputs were considered for quantitative analysis; three continuous grids with large ranges of values. Close examination of the data however suggested that any test of correlation or variation between the different datalayers was likely to yield misleading results. The large range of possible values in each grid leads to a very high likelihood of there being a significant difference between corresponding cells from different grids. Statistical Testing on these grids supported this premise: there was a significant difference at a 95% Confidence Level between the Un-weighted grid and each of the weighted grids as well as between the 2 weighted grids. In other words, the test results suggest that the spatial distribution of areas of varying suitability for Forest Stewardship is very different in each of the overlay results. While this scenario is not impossible it may mask the real trend of the data. As discussed before, the results of the overlay Aggregated Aggregated Grid Grid 1 1 are most useful when represented as broad classes of 1 1 6 L L M Stewardship Potential such as High, Medium and 1 7 7 L H H Low. The use of Natural Breaks to create these three Reclassified as: 3 7 7 L H H Continuous 0-3 L classes reveals the true trend of the data by preserving Aggregated grid 2 Grid 2 4-6 M the distribution of the respective grids and makes 7 - 10 H 1 2 9 L L H 1 8 8 L H H statistical testing more relevant. Diagram 4 illustrates 4 9 9 M H H such a hypothetical situation where the similarity in 7 of 9 cells correlate Only 2 of 9 cells correlate No Significant Difference Significant Difference the spatial patterns of two grids is hidden due to their large ranges of data. The output grids from each of the Diagram 4: Results of aggregation three overlays were thus reclassified into groups of High, Medium and Low using their respective Natural Breaks. Statistical testing on these aggregated grids yielded the following results: there was no significant difference at a 95% Confidence Level between the Un-weighted grid and the Ordinal Weight Scheme grid, but a significant difference between the Un-weighted grid and the Rank Weight Scheme as well as between the 2 Weighting Scheme grids. Table 4 below illustrates these results and illustrates that the Ordinal Weighting Scheme is actually comparable to using no weights at all (as there is no significant difference in distribution between these two grids). 3/30/2006 (MA_Methodology.doc) Pg. 12 of 15 Statistical Parameters Mean Variance Observations Pearson Correlation Hypothesized Mean Difference df t Stat P(T<=t) one-tail t Critical one-tail P(T<=t) two-tail Un-weighted Ordinal Un-weighted Rank Ordinal Rank Scheme Scheme Scheme Scheme Scheme Scheme 1.789 1.773 1.789 1.64 1.773 1.64 0.452931932 0.568039039 0.452931932 0.462862863 0.568039039 0.462862863 1000 1000 1000 1000 1000 1000 0.834978042 0.861454182 0.873169848 0 0 0 999 999 999 1.213243069 13.2254695 11.42793689 0.112661843 3.24935E-37 8.19918E-29 1.646380952 1.646380952 1.646380952 0.225323686 6.49871E-37 1.63984E-28 t Critical two-tail 1.962343958 Statistical Significance: No significant Difference 1.962343958 Significant Difference 1.962343958 Significant Difference Table 4: Comparison of Test Results Table 5 on the right shows similar Statistical Testing done on the Continuous, un-aggregated grid in addition to the above tests on the aggregated t-test results at 95% Confidence Level data. There is significant difference between Continuous Output Aggregated Output Un-weighted Overlay all the continuous grids because of the large No significant difference vs. Ordinal Weighting Significant difference range of values in each grid. Distribution Un-weighted Overlay vs. Rank Weighting Significant difference Significant difference trends in neither of the three Ordinal vs. Rank Weighting Significant difference Significant difference grids corresponding to each Weighting scheme are apparent in the continuous data Table 5: Tests on Continuous vs. Aggregated Results and the process of a weighted overlay rendered almost moot. 3. Determination of final Weighting scheme It would be fair to conclude that Statistical Testing indicates with a fairly high degree of certainty that incorporating weights into the overlay analysis has a significant affect on the spatial distribution of Forest Stewardship Potential. This supports the theoretical supposition made earlier about input factors bearing different degrees of influence on the final output. The decision to incorporate a weighting scheme in the overlay analysis is thus strengthened by this quantitative analysis. Test results also show that the Ordinal Weighting Scheme has no significant influence on the outcome of the overlay analysis compared to a straight, un-weighted approach. The Rank Weighting Scheme becomes the obvious choice of weighting schemes. Further justification for using the Rank over Ordinal weighting scheme is an inconsistency in the Ordinal weighting scheme results. Detailed examination of the Foresters’ datalayer evaluation using weighting results from the Rank Weighting Scheme showed that the Relative weights were not conforming well to the Foresters’ assigned weights for a few critical datalayers. For instance, ‘Risk of Development’ was a datalayer rated as highly significant and influential on both weighting schemes by most Foresters. The relative weight on the Rank weighting scheme however skewed downwards leading this Datalayer to have the lowest Relative weight. The Interval Scale did not have any such distortion and seemed to more accurately reflect the consensus. Generating statistics from the overlay analysis A series of maps were created to demonstrate trends in Forest Stewardship Potential over the entire state as well as in specific areas, such as within Private Forests and within Non-forest – Nondeveloped lands. In addition, un-weighted overlays were also run on datalayers falling into the 3/30/2006 (MA_Methodology.doc) Pg. 13 of 15 ‘Resource Threat’ and ‘Resource Potential’ categories and the results aggregated according to a scheme similar to the main Weighted Overlay (that uses all the datalayers). Since these smaller overlays have no assigned weights they serve as demarcations of concentrations of Threats and Resources, respectively. They may be helpful in understanding the datalayers and in getting better acquainted with the Natural Resources profile of the state. Patterns in conservation and Stewardship potential do in fact seem to emerge in these grids but have limited value and restricted applicability to forest management and planning given that the trends represented are one-sided and incomplete. For instance, a certain area may fall into the ‘Low Threat’ category in the Resource Threat grid, suggesting perhaps that there is less Stewardship Potential there than say, a nearby area that falls into the ‘High Threat’ class. However, its value in the Resource Potential grid may be comparatively higher, giving it greater total Stewardship Potential. Statistics were generated to calculate areas of varying Stewardship Potential over the State from the following three grids: 1. Statewide Stewardship Potential – results of the Weighted Overlay Analysis 2. Stewardship Potential on Private Forests – the Private Forest Mask grid 3. Stewardship Potential on Non-forested – Non-developed (NFND) land – from the NFND Mask grid In addition, a vector datalayer was created of areas currently under Forest Stewardship, called the ‘Stewardship Plans’ layer. This was the result of digital conversion of known Stewardship Plans with the input of the State Service Forestry Department who hold both documentary and intrinsic, working knowledge and of their locations and extents. Area calculations of these plans and their degree of coincidence with areas of Stewardship Potential as indicated by the Overlay Analysis, may be used to gauge the efficacy of past and current Stewardship planning programs. In addition, the Stewardship Plans may provide valuable guidance to future Stewardship planning and in identifying priority areas. Most calculations were done by either summing cell counts in attribute tables of grids or with the ArcView Spatial Analyst ‘Tabulate Areas’ tool. The format of the tables and methods of calculation are annotated on the following page. Maps created had the following titles and contents: “Potential for Forest Stewardship Program Benefits and Existing Stewardship Plans” “Forest Stewardship Potential on Private Forest Lands and Existing Stewardship Plans” “Resource Richness” : the un-weighted overlay of ‘Resource Potential’ datalayers “Resource Threats”: the un-weighted overlay of ‘Resource Threats’ datalayers “Forest Stewardship Potential on Non-forested – Non-developed Lands and Existing Stewardship Plans” “Regional Analysis”: each State participating in this Pilot Study represented the Stewardship Potential grid in a focus area of choice. The Massachusetts focus area was the Connecticut River Watershed running North-south through the State of Massachusetts and continuing through Connecticut on the South to its terminal in the Atlantic. The continuous Forest Stewardship Potential grid (direct from the Overlay Analysis) was clipped to the EPA 6-digit HUC watershed for the river and Natural Breaks re-calculated for the range of data occurring within that extent. It was then aggregated (reclassified) to High, Medium and Low classes corresponding to the Natural Breaks. The same was done with the watershed in Connecticut and both Stewardship Potential grids depicted on the Regional map. The considerable similarity in patterns of High, Medium and Low Stewardship Potential at the boundary between the two States illustrates the scalability of this Analysis; that it works on different areal extents. “Potential for Forest Stewardship Program Benefits”: without the Stewardship Plans 3/30/2006 (MA_Methodology.doc) Pg. 14 of 15 Map: “Potential for Forest Stewardship Program Benefits and Existing Stewardship Plans” Stewardship Potential Forest Acres 1 % of Tot. Forested Stew Capable Lands Non-forest - Non-developed 2 Acres % of Tot. non-For. Acres Total all HML % of Total Acres 3 % of total Low 1,101,352 42% 447,251 74% 1,548,603 48% 1,548,603 48% Med 1,178,504 45% 135,293 22% 1,313,797 41% 1,313,796 41% High 354,744 13% 22,147 4% 376,891 12% 376,890 12% Total: 2,634,600 604,691 3,239,291 3,239,289 Map: “Potential for Forest Stewardship Program Benefits and Existing Stewardship Plans” Calculation Key Stewardship Potential Low Total: Medium High 4 Areas capable of Stewardship : 1,548,603 1,313,797 376,891 3,239,291 5 Stewardship Plan (acres) : Stewardship Plan vs. Areas Capable (%): 59,572 4% 70,266 5% 23,230 6% 153,068 5% Acres Acres of Stewardship Stewardship 5 Plan (acres) Areas capable Private Forest Lands Stewardship Potential Private Forest Lands Low 6 Areas capable of Stewardship : 7 Stewardship Plan (acres) : Stewardship Plan acres as a % of acres Private Forest: Medium Total: High 1,178,504 354,744 2,634,600 49,324 67,173 22,747 139,245 6% of Stewardship 6% 5% 7 Plan (acres) Areas capable of Stewardship Stewardship Plan (acres) Map: “Forest Stewardship Potential on Non-forested – Non-developed Lands and Existing Stewardship Plans” Non-forest - Non-developed Lands Stewardship Potential Non-forest - Non-developed Low 8 Areas capable of Stewardship : 9 Stewardship Plan (acres) : Stewardship Plan acres as a % of acres Non-for - Non-devpd: 3/30/2006 (MA_Methodology.doc) Medium Total: High 447,251 135,293 22,147 604,691 10,248 3,093 483 13,824 2% 2% 2% 4 2% Pg. 15 of 15 9 From ‘Total’ column of first table Spatial Analyst Æ Tabulate Area Æ Row theme = Stew. Plans, Column Theme = Stewardship Potential Grid 6 Stewardship 1,101,352 4% 2 3 Acres Areas capable Map: “Forest Stewardship Potential on Private Forest Lands and Existing Stewardship Plans” Private Forest Mask Grid, info tableÆ ‘count’ field for each value Æ convert to acres NFND Mask Grid, info table Æ ‘count’ field for each value Æ convert to acres Stewardship Potential Grid, info tableÆ ‘count’ field for each value Æ convert to acres. Compare with the ‘Total’ column to verify that all land-use categories were accounted for. 1 From ‘Forest’ column of first table (same as Private Forest Mask Grid, Info table) Spatial Analyst Æ Tabulate Area Æ Row theme = Stew. Plans, Column Theme = Private Forest Mask Grid 8 From ‘Non-forest – Non-developed’ column of first table (same as NFND Mask Grid, Info table) Spatial Analyst Æ Tabulate Area Æ Row theme = Stew. Plans, Column Theme = Non-forest – Non-developed Mask Grid