ࡱ > 1 ] _ R S T U V W X Y Z [ \ } A C E G I
K x z
| ~ B D F H J L ` @ p# bjbjFF -
, , { $ P 0 l Ô n L " 0 ? n | ) @ B D D D D D D $ 1 R h V ? V V h 0 y } a a a V 0 B a V B a a z } 0 0E] Y l { ց l 0 Ô { \ # O^ # 0 } } # ~ i 8 3
a [A
WL
i i i h h < D ` < Connecticut River Floodplain Analysis
Final Report, April 18, 2007
M.G. Anderson, C. Ferree, A. Olivero, and F. Zhao
Eastern Conservation Science
The Nature Conservancy: Eastern Regional Office
11 Avenue De Lafayette, Boston 01130
Background: Floodplain forests and related riparian communities provided critical habitats for a variety of plants, mammals, birds, reptiles and invertebrates. During spring floods, silt laden waters replenish floodplain soils and serve as feeding and nursery grounds for fish (overview in Zimmerman 2006). Unfortunately the once magnificent and extensive floodplain forests of the northeast have been reduced to isolated fragments by agricultural clearing, road building and hydrologic alteration. Even so, in the Connecticut River watershed, ecologists have located, mapped and evaluated over 80 remaining stands more than any other watershed in the Northeastern US. What was lacking until now, was a unifying floodplain assessment of the for the entire stream network to give context to the remnants, identify suitable restoration areas, highlight broad-scale patterns and suggest strategies for conserving this remarkable resource.
The Nature Conservancys Eastern Conservation Science team (ECS) has been compiling and analyzing information about the Connecticut River watershed for over a decade. The extensive information base underpinned several broad-scale assessments designed to determine critical areas for ecosystem and species protection (Anderson et. al.2006 a, b) and a basin-specific analysis of ecological communities and Neotropical migrant birds (Anderson et al. 1998). Reports and data are posted on a collaboratively maintained web site dedicated to the watershed (http://nh.water.usgs.gov/projects/ct-atlas).
The focus of this project was on an integrated floodplain assessment and creating an analysis tool to support conservation decision-making in the watershed. Having previously compiled data on the entire stream network, wetlands, dams, toxic release points and road stream crossings, we were well positioned to develop an accurate riparian and floodplain map for the basin. Additionally, techniques for mapping floodplains using cost-surface models (models that estimate the resistance of the topography to flood waters spilling outward from a stream - Strager et al. 2000) have proven to be surprisingly accurate. We hoped to test the cost-surface methods and to locally improve the resulting models using existing data layers on bedrock and surficial geology, land cover, landforms, topography, soil moisture indices, National Wetland Inventory (NWI) wetlands and Natural Heritage element occurrences. The results were used to address questions on the quality and restorability of floodplain examples. Final products will be available for all TNC and partner organizations working within this region.
Objectives: Our overarching goal was to identify high quality existing examples or the best restorable examples of floodplain forest systems in the Connecticut River watershed. To this end we employed a five-step process.
1) Construct an Accurate Model and Map of all Floodplain Areas: develop a spatially explicit, geographically comprehensive, and topographically-based model for places along the river network where geomorphic characteristics favored the development of riparian communities.
2) Verify Current Flooding: determine the regions along the main stem and tributaries that were still experiencing some level of spring flooding.
3) Determine which Floodplain Occurrences were likely to contain Floodplain Forest: develop a predictive model to separate floodplain forest occurrences from other riparian systems such as alluvial marshes, conifer swamps and other basin wetlands.
4) Evaluate the Characteristics of each Potential Floodplain Forest Occurrence: characterize each unit (floodplain area) with respect to its size, condition, and landscape context. Compile corroborative information on the occurrence from existing ground inventory or other studies.
5) Prioritize the Occurrences with respect to Suitability for Conservation Action: determine suitable areas for conserving existing floodplain forests or the best candidates for restoration at appropriate and functional scales.
Methods and Results
Section 1: A Topographically-based Model of Riparian Areas
The base model used a cost-surface to estimate the relative resistance of the landscape to flood water moving outward from any given point along the stream bank (Stager et. al. 2000). Constructing the model requires two necessary inputs; Lines defining the river/stream network and shapes for the ponds and lakes (GIS-arcs and polygons) and a digital elevation model for determining slope. We used stream and river arcs from the River Reach Files, version 3 (RF3, US EPA, nominal scale 1:100,000), water bodies in the National Hydrography Dataset (NHD, US Geological Survey, same scale) and the National Elevation Dataset (NED) 30 meter digital elevation model (DEM -USGS, nominal scale 1:24,000). Software used was the suite of GIS analysis tools in workstation ArcInfo 9.1 (ESRI, 2005).
In the cost-surface model, high slope translated into high cost. To generate the surface it was necessary to rasterize the river and lake features to 30m cells and to derive a floating point slope grid from the DEM. We then attached to each output cell the accumulating cost of moving upslope on a least cost path away from a stream or lake cell, using the slope grid as the cost-surface (Stager et al. 2000). This was done with the Arc Info Grid costdistance function.
The upper cost threshold must be set by the user to match the area where the cost of floodwaters moving upslope is presumed to be too high and the riparian zone ends. To set an upper cost threshold and calibrate the cost accumulation grid, we overlaid polygons of known floodplain forest occurrences in the basin, NWI data on wetland occurrences and Flood Insurance Rate Maps of the Federal Emergency Management Agency. Grid cells with values below the threshold were assumed to be in low, flat areas adjacent to streamways or ponds/lakes and prone to seasonal flooding. Cells with values above the threshold were assumed to be above the riparian zone and out of range of overbank flows (Figure 1).
Figure 1: Floodplain/riparian model from slope cost surfaces in the Northampton, MA area. Model in grey, known floodplain occurrences in red, Connecticut River in horizontal hatch, smaller rivers in blue, major roads in black. Area shown is about 15 km (9 miles) across.
The size class of each stream reach and the shapes of water bodies were built into the model following an existing classification (Table 1, Olivero et al). To do this the RF3 stream arcs were coded by the size of the area they drained before running the costdistance function, and generating separate surfaces for gridded arcs in each of 4 watershed size classes. The lake and ponds grids were run separately, as were a grid of tidally influenced water features (from the NWI and NHD datasets). The results of all six runs were merged into a common data set. Because this process occasionally over-models low wet areas adjacent to lakes, lakeside model occurrences were cut off at a distance of 250 meters from the water body before combining with the other costdistance output grids.
Table 1. Size classes for RF3 rivers and streams; values are in drainage area.
Link AllocationFeatureDefinition1Size 1 stream 0- 30 square mile watershed2Size 2 stream30-2000 square mile watershed3Size 3 stream 200-1000 square mile watershed4Size 4 stream1000 + square mile watershed6Waterbody 6Non-tidal lake or pond7Waterbody 7Tidal lake or pond
The river-centricity of the resulting model were appealing and intuitive (Figure 1); however corroborative tests using known riparian wetlands indicated that the model underrepresented many known locations. The model was so slope-sensitive that in many cases light elevation changes adjacent to a streamway pushed the accumulated costs above the threshold in a single cell-step. To correct for this we incorporated existing topographic data on wet flat landforms from The Nature Conservancys Ecological Land Units (ELU) dataset into the model (Anderson et al. 1998). In the ELU data set flat landforms (areas of less than 2% slope) were split into wet and dry sub-units using a simple moisture index based on flow accumulation and slope, both derived from the 30m DEM (Ferree 2005, Fels and Matson 1997).
Combining the topographic wet flat data with the cost-distance data created a complementary analysis that benefited the model. While the costdistance algorithm analyzes each cell, one by one, as it moves away from a river, the landforms and the moisture index aggregate elevation information over a broader area and are less sensitive to minute changes in slope. Adding wet flats that came into contact with the slope cost surface-defined model occurrences often filled out modeled occurrences realistically and made them roughly equivalent to the size and shape of known occurrences. Only the portions of ELU-derived wet flats within a river/lake/estuarine slope cost zone of 200 were added to the modelthis threshold was set, again, with reference to the boundaries of known floodplain and riparian ecosystems. The final model corresponded roughly to the FEMA 100-year flood zones (Figure 2).
Figure 2. Riparian-floodplain models (light grey shapes) in south-central Connecticut, with wet flats added in (colored shapes: wet flats associated with size 1 streams in green, with size 2 in pink, with mainstem in orange, with lakes/ponds in pale blue; open water in darker blue). FEMA flood zone data is shown in dark red hatch. Image covers an area about 20 km (12 miles) across.
Creating individual floodplain occurrences
The initial modeled riparian areas included their adjacent waterbody. To separate them, all open water river, stream, lake, estuary was removed from the merged model grid. Removing the water from the model had the side effect of leaving small holes in the riparian occurrences and sometimes splitting an occurrence on either side of any stream or river. Holes were filled using the Arc Info Grid expand and shrink functions. We expanded the model by one cell, shrank it by two cells, and then expanded it again by one. This expansion-contraction-expansion reconnected the model grid across a one or two cell gap and gave them more natural and realistic shapes. This was important in the next step, in which we used the Grid regiongroup function to break the merged model grid into spatially distinct riparian occurrences, each of which has a unique identifier and is tagged with its source a value that indicates whether the occurrence was associated with a size 1 stream, for example, or a larger river, or a lake. Only if two occurrences on either side of a river were separated by more than two 30 meter cells in the model grid from which water had been removed are they recognized as separate riparian sites in the regiongrouped grid.
Finally, we removed urban and residential developed landcover classes from the models (but not agricultural classes). To focus our attention on potentially important riparian occurrences we eliminated from further analysis several thousand occurrences less than two acres in size.
Results of the Topographic Floodplain Model construction
For the entire watershed, the model identified 17,313 riparian occurrences that were two acres or larger. Acreages for the model occurrences (MOs) totaled to 331,073 acres distributed across all stream sizes and states in the watershed (Table 2).
Table 2. Floodplain/riparian model area summaries by state and by the water
feature type the model occurrences are associated with; areas are in acres
EMBED Excel.Sheet.8
Section 2: Verifying Current Flooding
Many of the potential floodplain occurrences were located on sections of the Connecticut River and its tributaries where the hydrology has been altered in varying ways, sometimes radically. To identify the most ecologically intact examples, and the examples most readily restorable, we wanted to determine which occurrences still experience some level of seasonal flooding.
Methodologies for evaluating the extent of current flooding across large areas have been developed using processed Landsat Thematic Mapper imagery (Hudson & Colditz 2003; Wang, 2004). Landsat imagery owes its utility for this purpose to its wide coverage, a 16 day return interval, and favorable spectral and spatial resolution (six bands in the visible and near and middle infrared regions of the electromagnetic spectrum at a spectral resolution of 30 meter resolution, and a panchromatic band of 15 meters). To identify those areas flooded in the spring yet dry in the fall (e.g. seasonal flooding) requires imagery from both a spring flooding event and an autumn dry period. To this end, we acquired Landsat Thematic Mapper (ETM+) imagery for the entire river basin for April 14, 2001, and September 30, 2001. This set of matched imagery answered our need for high, but not atypical, spring flows and a typically low September flow with which to compare it (Figure 3) All image analysis took place in the ERDAS Imagine software environment (ERDAS, 2005).
Figure 3. Monthly mean discharge at a gauging station on the Connecticut River
in Hampden County, MA, from 1993-2002.
We tested three indices for their ability to define inundated areas in the spring imagery, intending to use the best in a change detection analysis that would highlight areas under water (or at least very wet) in April, and dry in September. These change detection procedures have proven to be less error-prone than slicing a single water-oriented index band in each image and comparing the results, or performing a post-classification image differencing analysis (Cohen and Fiorella, 1998; Hodgson, 2003).
Tests were performed on a pilot area in central Massachusetts (Figure 4), using
1) The wetness band of the Tasseled Cap Transformation (Crist and Cicone, 1984), a linear combination of all of the Thematic Mappers multispectral bands;
2) The Normalized Difference Water Index (NDWI) (McFeeters, 1996), an index derived especially to enhance the water signature by maximizing water information while minimizing vegetation and soil information; and
3) An additive index using bands 4 (near infrared) and 7 (mid-infrared) developed by Wang (2002) to enhance flood-inundated areas.
Figure 4. Landsat 7 ETM+ images, acquired April 14, 2001 (left) and September 30 2001 (right) Displayed with RGB = bands 4:3:2.
The combined bands 4 and 7 (Wang 2002) resulted in the sharpest definition of water features (Figure 5) and this index was chosen for further analysis over the entire river basin.
Figure 5. Comparison of three water enhancement indices. a) Imagery in April; b)Imagery in September; c) Addition of band 4 and band 7 of ETM imagery; d) Wetness index derived from Tasseled Cap transformation; e) NDWI ( Normalized Difference Water Index)
A standard technique in change detection analysis is to generate a multi-date composite
image combining three bands that will capture and reflect the changes that have occurred over the time period. For this study, bands 1 and 2 were derived from the addition of the ETM+ bands 4 and 7 for April and September, respectively. For the third band in the composite image we used a slope map derived from the USGS 30 meter DEM. This was advisable because topographic shadows in areas of moderate to high relief can produce spectral reflectance signatures similar to that of water. Incorporating the slope band in the composite image alleviated this problem by allowing the separation of flooded flats and shaded slopes. A clip from each of the three derived bands and from the basin-wide composite image appears in Figure 6.
Figure 6. a) Additive image of ETM+ bands 4 and 7 for September. b) Same for April. c) Slope layer. d) Composite image (R:G:B = September:April:slope). Areas of flooding are in pink.
In order to extract pixels representing flooded areas into their own dataset, we performed an unsupervised classification on the composite image. Use of the latter was justified because the flooded areas were well enhanced, and because an unsupervised classification does not require a priori information and thus offers a more efficient and reliable solution than a supervised classification. The resultant data set contained 15 statistically separable classes. With the help of an aerial photo taken in April 2001, the 15 classes were collapsed into a simple binary image of flooded/non-flooded pixels. As a final step we removed any discrete areas that were not within 100 meters of a water feature in our base dataset, or a riparian model occurrence, or a known floodplain occurrence, or on any landform other than flats or gentle slopes. Results for an area upriver and downriver from Northampton, MA, are shown in Figure 7.
Figure 7. Extraction of spring flooded areas, Northampton, MA. a) Original imagery of Sept., 2001. b) Original imagery of April, 2001. c) Composite image. d) Binary grid, with blue representing areas of overbank flows.
A classification accuracy assessment was performed using digital orthophoto tiles for the central Massachusetts pilot area of 1 meter resolution for April 26, 2001, and randomly extracted 150 reference flooded points and the same number of non-flooded points. Daily peak discharge rates for April, 2001, for the gauging station in Hampden County, MA, show that peak discharges for the 14th and 26th of the month were quite close, indicating that there were likely similar flooding conditions for the two dates. Overall accuracy for the flooded and non-flooded pixel classes was over 96 percent. In this area the methodology proved efficient and accurate. From the strength of this test we extrapolated the data to the whole watershed. Further testing and/or field sampling will be performed by TNCs Connecticut River program in 2007-2008
As our intent was to predict floodplain forests, the inability of the ETM+ optical sensors to penetrate the forest tree canopy raises the possibility of under-prediction of area of inundation with this methodology. Comparing the orthophotography and the satellite imagery and image analysis products, however, suggested that those concerns may be largely unfounded (Figure 8). A moderately dense tree and shrub canopy along a stretch of Bachelor Brook in South Hadley near its junction with the mainstem did not prevent the ETM+-based analysis from correctly identifying it as a riparian wetland occurrence that experienced an extensive spring flooding event. (This is one of three A-ranked small river silver maple-green ash floodplain forests in Massachusetts.) While there are undoubtedly some mixed water-forest pixel effects that cut into classification accuracy, it did not seem to be a serious problem.
Figure 8. Response of forested area to the flooding as shown in aerial photo (upper), ETM+ data acquired in April (middle) and composite image(lower).
Results
In sum, the imagery analysis identified 48,956 acres of spring flooding in the watershed of which 45% overlapped directly with the modeled riparian areas (Table 3).
Table 3. Acreage and distribution of spring flooding based on the imagery analysis.
Inundation Analysis: Total Acres of Spring Flooding: 48,956 acres
FeatureSize1 StreamsSize2 StreamsSize3 StreamsSize4 StreamsLake/pondEstuarineTotal overlap with model%14.710.44.59.15.40.644.5Acres7,1975,0922,2034,4552,64429421,786
Section 3: Developing a Predictive Model for Floodplain Forests
The focus of this study was on floodplain forest ecosystems that are among the most diminished and degraded of all of New Englands natural communities (Thompson and Sorenson 2000, Sperduto and Nichols 2004). The topographically-derived model identified 17,313 riparian occurrences but did not differentiate between types of riparian communities such as alluvial marshes, conifer swamps, black willow thickets or tidal woodlands. The objective of this section was to develop a predictive model to separate the potential floodplain forests from other riparian systems.
To accomplish this we compiled a corroborating data set of 85 element occurrences (EOs) of floodplain forests located and inventoried by the Natural Heritage Programs of Connecticut, Massachusetts, New Hampshire and Vermont. In addition we compiled inventory points for 197 known non-floodplain forest riparian communities with locations that coincided with the riparian model occurrences. We attached the ground survey information to the modeled occurrences and used a classification and regression tree analysis (CART, Steinberg and Colla 1997) to determine whether the 82 floodplain forest occurrences could be consistently separated from the other communities using a set of ecological attributes developed for each model polygon. These attributes included information on its size, shape, elevation, landcover, the substrate it occurred on, the local landforms, the size of the river it is adjacent to, and whether it experienced any spring flooding. (Table 4).
The CART algorithm looks for patterns in information about known data elements to construct a decision tree that enables the analyst to classify or make predictions about a larger set of unknown elements. CART builds its decision tree by going through the learning set (in this case 282 polygons) and recursively partitioning it into increasingly homogenous subsets of the two groups (floodplain forest or non-floodplain forest). It bases its selection of which variable it uses to divide the data, and the dividing value, on a quantitative measure of how cleanly the measures distinguish between the floodplain forest and non-forest riparian occurrences. Each parent node splits into two child nodes and the process is repeated. Improvement of classification accuracy is evaluated at each split. The analyst can choose the optimal tree size (measured in number of terminal nodes based on accuracy, efficiency, and economy.
Table 4: Attributes attached to 17,313 riparian/floodplain model occurrences (MOs) for the CART analysis.
EMBED Excel.Sheet.8
We performed several CART runs; restricting the variable set the program had access to in various ways. Important variables were consistent from run to run but minor variables differed from one run to the next with small effects on the classification accuracy of the resulting tree. Because CART is mechanistic in the criteria it applies to choose splitting variables, it cannot adjust early variable choices based on what it finds later in the tree. Withholding certain key variables allowed us to identify alternative splitters at parent nodes. New trees were created that in some cases offset some known weaknesses in our data set (such as the problem created by remnant spring ice in the far north) and for this reason our set of predicted occurrences is a combination of the best decision trees.
The ability to estimate the prediction accuracy of a given decision tree model before applying it to unknown occurrences is built into the CART program. This is important because very high classification success rates for the learning data, (the 282 known occurrences) may give poor results when applied to new data. In addition, using the learning data to measure classification success enforces a bias to large, complex and idiosyncratic decision trees. CART uses a cross validation procedure to sidestep this problem. For the 10-fold cross validation we called for in our study, the program built 10 independent trees every time we ran the program, withholding 10% of the data records every time against which to test tree classification accuracy. Sampling is done without replacement, so every observation was used exactly once in a test set. Error rates are calculated for every stage of decision tree construction, with the addition of every new terminal node, based on the mean error rates for all ten trees. Research has shown that cross validation error rates are typically very close to actual rates, even tending to slightly overestimate them (Steinberg and Phillip 1997).
Results of the Classification and Regression Tree (CART) analysis
Among various runs, a repeating set of important variables emerged. Variables consistently selected included: the amount of overlap with the spring flood model; elevation mean and minimum; stream size or water feature type the occurrence was adjacent to; the percent of water features (i.e. ponds and sloughs and rivers) within a 500 meter buffer of the modeled occurrence, the percent of the more dramatic landforms (e.g. summits and steep slopes) in that buffer; the percent of mixed/coniferous forest and some of the wetland cover types within the occurrence.
The simple tree shown in figure 9, with seven terminal nodes, was an efficient and fairly accurate classifier. The spring flooding variables, consistently used as a primary splitter in exploratory runs, were excluded from this run to force the model to search for alternative variables. The root node represented by the box at the tope of the tree contains the entire set of 85 floodplain forest (class 1) and the 197 non-floodplain forest majority class, the number of modeled occurrences at the node and the breakdown in numbers and percent of actual group membership and the histogram bar gives a simple graphic display of node heterogeneity. Just above each child node is the criterion used to make the split from the parent node above.
Stream size (the LNK_ALLOC variable) was the primary variable used in this tree to split a high proportion of the known floodplain forest model occurrence (67 of the 85) from the learning set of 282 and put them on the right hand branch of the tree (these were occurrences associated with size 2,3, and 4 rivers). Eighteen of the true floodplain forests, which occur on size 1 streams and lakes (LNK_ALLOC = 1, 6, 7), remained on the left side of the tree, and are subsequently separated from the non-floodplain riparian communities due to their higher amounts of mixed conifer-deciduous landcover and larger amounts of nearby lakes and ponds (LF_POLYWAT), leaving nearly pure terminal nodes 4 and 1. On the right side of the tree none of the variables provided were as good at purifying nodes, with the most useful separator being dramatic landforms (LF_DRAMA). The later was a synthetic number calculated as the total percentage of cliffs + steep slopes + slope crests + summits + cove slopes + cove bottoms). The misclassification rates for this decision tree are given in the node boxes.
Figure 9. The classification tree resulting from trimming the tree and manipulating parameters to manage error. See text for an explanation of the tree and its splitting variables.
The decision tree is equivalent to a dichotomous key, and all 17,000 unknown model occurrences were scored and classified into one of the two groups, by working them through the decision tree to the terminal nodes. Based on the composition of the terminal node a model occurrence ended up in, it was assigned a probability of class membership. For example, if 20 of the original 282 occurrences in the learning set occupied a given terminal node, and 15 of them were non-floodplain, then any of the other 17,000 new occurrences that had the same characteristic as the 20 in the original node were given a probability of 0.75 of being a non-floodplain occurrence
Applying the final classification keys to the entire set of modeled occurrences identified 3272 potential floodplain forests (including the 82 confirmed examples) in the watershed. These were distributed across all states and stream sizes, and totaled to 93,000 acres (Table 5)
Table 5. Predicted floodplain forests: area summaries by state and by the water feature type the model occurrences are associated with; areas are in acres
Section 4: Evaluate and Describe the Characteristics of each Modeled Occurrence:
The classification and regression tree analysis identified floodplain areas that were likely to contain, or be restorable, to floodplain forest. Each occurrence was attributed with the metrics data described below and we evaluated and ranked it based on five factors. The first three factors were used in ranking and ordering the occurrences. We expect the latter two factors to be used by conservationists in deciding where and how to work on the ground.
Size
Condition (3 variables)
Landscape/Watershed Context Index
Feasibility score
Correspondence to TNC portfolio and NH occurrences
Occurrence Size: The area of each floodplain forest occurrence was automatically generated for each occurrence in the model development
Occurrence Condition
The condition of each occurrence was estimated and ranked based on three variables:
1) Percent Natural Cover: the percent of the floodplain area (polygon) covered by forest, wetland, shrubland or any other natural cover class (based on NLCD 2001). The remaining area by definition was covered by agriculture or low level development. Range 0 -100%
2) Percent Verified Flooding: the percent of the floodplain area (polygon) inundated by flooding in spring but dry in fall, based on the satellite image analysis of flooding described above. Range 0-100%
3) Reach-level Hydrologic Intactness; This indicator measures the ratio of total upstream dam storage to the amount of annual runoff for each stream reach associated with the floodplain occurrence. High ratios indicate more hydrologic alteration. To calculate this metric the total volume of water stored behind upstream dams (NID_STOR attribute) was summed for each stream reach. The mean annual runoff for each reach was estimated from the total upstream drainage area using a standard coefficient (Zimmerman and Lester 2006, Olivero, 2003; Fitzhugh, 1999 details in Appendix A). Values ranged from 0 to 4,900, average 41. Due to its skewed distribution, this measure was transformed to a rank/percentage value for inclusion in the analysis.
The preceding three variables were independent and uncorrelated. Each contained critical information about the occurrence and collapsing the variables into a single index yielded poor results. Never-the-less we did calculate a Condition index and used it primarily for graphing purposes. To create the index, the three condition values were summed after first being arcsin transformed (1 and 2) or rank transformed (3) and normalized to a standard between 0 and 100 with 100 always being the best and zero the worst. The unweighted condition index followed the formula:
CI = (norm%NC + normPVF+ normHI)
This index had a maximum value of 300 and gave a preference to hydrologic variables.
Landscape / Watershed Context
These variables were derived from the landscape or watershed surrounding the floodplain occurrence:
1) Landscape Context Index: this variable was calculated for each polygon and an 1140 meter buffer area (1000 acres if the polygon were a single point) immediately surrounding the floodplain occurrence. The LCI was calculated based on the amount of agriculture, development and quarries found in the land cover maps as well as the amount and densities of roads. Values range from 0, indicating pristine natural cover to 400 which indicates the occurrence and its buffer are intensely developed.
2) Flow Alteration within the Watershed: This indicator measured the ratio of total dam storage capacity to the amount of annual runoff for each watershed containing a floodplain occurrence. High ratios indicate more flow alteration. To calculate this metric the total volume of water stored in the dams (NID_STOR attribute) was summed for each watershed. The mean annual runoff for each watershed was estimated from the total drainage area using a standard coefficient (Zimmerman and Lester 2006, Olivero, 2003; Fitzhugh, 1999 details in Appendix A). Values ranged from 0 to 4,900. Due to its skewed distribution, this measure was transformed to a rank/percentage value for inclusion in the landscape / watershed context index.
3) Miles of Connected Network: This metric estimated the length of stream network between dams that the floodplain occurrence was situated within. The larger the network, the larger the area for which the floodplain could theoretically provide spawning habitat, nutrients, sediment and other related processes that are disrupted by barriers. The network length was calculated using all dams tracked by the Army Corps of Engineers National Inventory of Dams (NID) augmented by hundreds of small dams tracked and mapped by the states of Connecticut, Massachusetts, New Hampshire, and Vermont (Lester and Olivero, 2006 more details in Appendix A). Values ranged from 0-761 miles.
Landscape/ Watershed index. Landscape values were summarized in an index by summing the three values. To create the index values were first log (1 and 3) or rank (2) transformed then normalized to a standard between 0 and 100 with 100 always being the best and zero the worst. The unweighted landscape index followed the formula:
LI = (normLCI + normFAW+ normMCN)
This index had a maximum value of 300 and gave a preference to hydrologic variables. The weighted index used in the final results followed the formula
WLI =((2* normLCI) + normFAW+ normMCN))
and gave equal weight to terrestrial and hydrologic values, with a maximum score of 400
Feasibility
Each polygon was scored for feasibility based on the following three values
1) Percent of GAP 1 or 2 land: This refers to lands with ownerships or easement that guarantee the intention of the land is the management of biodiversity and natural processes (for example TNC reserves, Federal Research Natural Areas) Range 0-100
2) Percent of GAP 3 land: This measures the percent of the polygon that falls on land secured from conversion but managed for multiple uses. The uses may include biodiversity values but typically also includes resource extraction such as logging. Range 0-100
3) Total upstream Hydrologic, Water Supply and Flood Control Dams: This metric measured number of hydropower, flood control, and water supply dams upstream of a given modeled floodplain occurrence. We assumed that for these types of dams the economic and human safety concerns inherent in their operations will make them difficult to remove or to renegotiate their flow release policies to meet more natural regimes - at least in relationship to recreational, wildlife and farm pond dams. Thus the metric is used as a rough indicator of how much work it would take to restore a natural flood cycle. Values ranged from 0 288
Feasibility Score: Calculated by the following formula
FS = (%GAP 1,2)+ (%GAP 3) + (normDAM3MAJ)
The two GAP indices were mutually exclusive with respect to a single point, collectively totaling to a maximum of 100% of a floodplain area. The maximum value for the feasibility index was 200
Correspondence with The Nature Conservancy Portfolio or Natural Heritage Element Occurrence
These metrics indicated whether the floodplain occurrences co-occurred with any other critical occurrences identified in the ecoregional assessments. Specifically they note if the floodplain area:
Was along a portfolio stream
Was within a forest matrix site
Contained a rare species or community target
Matched an existing floodplain or wetland basin target
Was corroborated by a natural heritage floodplain forest element occurrence
Results: Characteristics of the Floodplain Forest Modeled Occurrences
The characteristics were calculated for every occurrence (Table 6), however a subset of occurrences over 50 acres (Table 7) were used in ranking the floodplain forest occurrences and identifying high quality examples (section 5).
Table 6. Descriptive statistics of floodplain modeled occurrences
in the Connecticut River watershed
Floodplain OccurrencesAcresCondition IndexLand/ Watershed IndexFeasibility ScoreMean2813116685Standard Deviation77495241Median8131165100Mode2100276100Range1,286299338200Minimum20110Maximum1,288299349200Sum93,053427,109544,117278,094Count3,2723,2723,2723,272
Table 7. Descriptive statistics for occurrences 50 acres or larger.
ACRESCOND
INDXL/W Context
INDXFEAS.
INDX%_Natural
Cover%_ Flooding
Hydrologic
IntactnessMean161.0144.1162.885.267.014.240.3Median99.0147.1158.6100.075.16.810.1Mode54.0189.5221.9100.0100.00.00.0Standard Deviation175.648.844.736.328.218.2213.6Minimum50.023.220.40.01.30.00.0Maximum1288.0256.7328.9195.0100.094.92466.2Range1238.0233.5308.5195.098.794.92466.2Sum60,68854,32661,36732,12625,2455,36214,940Samples N=377377.0377.0377.0377.0377.0377.0371.0
Section 5: Identifying High Quality Floodplain Occurrences
In this section we identified screening criteria and thresholds relative to the size, condition and landscape/watershed context of each occurrence. Occurrences and their sites are ranked as to quality and potential for conservation activity
Size:
The natural size of a potential floodplain forest occurrence in the watershed ranged from less than one acre to over 1,288 acres. We posed the question: how large does an individual occurrence need to be in order to sustain critical processes and to maintain sufficient habitat for characteristic floodplain breeding species?
Floodplain forests in this watershed forests are generally comprised of silver maple, cottonwood, black willow, American elm and green ash. Less common trees, or trees restricted to certain geographies, include sycamore, box elder, river birch, swamp white oak, pin oak, and sandbar willow. Commonly these forests have a tangled understory of vines such as Virginia creeper, poison ivy, river bank grape and disturbance tolerant shrubs such as honeysuckle, blackberries and currents. Herbaceous cover can be very diverse and usually includes ferns, sedges and grasses such as ostrich fern, riverbank rye, long-beaked sedge and cattail sedge, along with herbs like green dragon, stinging nettle, wood nettle, false nettle, jewelweed, wild cucumber, and crooked stem aster (Thompson and Sorenson 2000, Sperduto and Nichols 2004).
Typical breeding fauna include red-shouldered hawk, wood duck, hooded merganser, yellow-billed cuckoo, veery, warbling vireo, yellow-throated vireo, blue-gray gnatcatcher, and mammals such as mink and raccoon. Butterfly and dragon fly fauna are diverse and overlap with those for the river. We collected information on typical breeding habitat territories of floodplain species from a number of current sources (Poole and Gill 2000, DeGraaf and Yamasaki 2000) and calculated the average sized territory of a breeding female to get a rough idea of how much area would be needed to contain twenty five breeding populations (Figure 10).
Figure 10. The area needed to contain 25 average-sized female breeding territories for characteristic floodplain forest fauna.
Although we were focused on floodplain forests we wanted the high quality examples to also encompass a range of floodplain features and communities such as back water sloughs, braided channels, vernal ponds, alluvial marshes, levees and scour barrens of cobble, sand and silt. Such habitat complexity is correlated with higher levels of diversity (Harper et al. 1997) and complexity tends to be correlated with size and extent.
Using these results as a guide, the variability of forests and their associates suggested that we aim for a range of forest sizes distributed across a variety of river classes. Recent assessments in the North Atlantic Coast (Anderson et al 2007b) and Northern Appalachian regions (Anderson et al. 2007a) both used a minimum size of 50 acres based partially on the co-occurrence of known rare species or communities. Based on these sources we set a minimum of 50 acres for the detailed assessment of floodplain occurrences. Although small examples less than 50 acres were not specifically addressed in this report the compiled data sets were include all floodplain occurrences down to 2 acres in size.
Condition and Landscape/Watershed Context
Our objective was to identify those occurrences that had the highest probability of either containing an intact floodplain forest system or to be restorable to an intact floodplain system. To meet this objective we focused our analysis on the 377 large examples, each over 50 acres in extent, that had the highest ranks for condition and landscape context.
Specifically we wanted to identify the occurrences with high ratings for each of four variables:
1) % Natural Cover
2) % Flooding
3) Hydrologic Intactness
4) Landscape/Watershed context index
There was evidence in the 73 ground inventoried floodplain occurrences that the first three variables, along with the occurrence size, were positively correlated with the condition of the inventoried forest sites. Based on the inventoried element occurrences, A-ranked examples were larger in size, had more active flooding, and had greater hydrologic intactness (Table 8). A-ranked and B-ranked examples ranked higher than C examples for natural cover.
Table 8: Average values for heritage ground inventoried floodplain forest examples.
Floodplain Forest Occurrences by Rank
A, ABB,BCCD TotalSize (acres)241142818138%Flood42%33%16%11%29%Nat73%79%58%53%70Hydrologic Intactness (rank%)72%57%56%0%0.59L/W Index162179171260174Count143325173
Top Ranking Groups
Using this information as a guide we grouped the high quality floodplain occurrences into four sets: In each set the members of the group were above the mean for all variables or for 3 out the 4 variables listed above.
Group 4: Examples 50 acres or larger with scores above the mean for all four variables.
Group 3H: Examples 50 acres or larger with scores above the mean for all variables except hydrologic intactness
Group 3N: Examples 50 acres or larger with scores above the mean for all variables except % natural cover.
Group 3L: Examples 50 acres or larger with scores above the mean for all variables except the landscape/watershed index
Group 3F: Examples 50 acres or larger with scores above the mean for all variables except % flooding.
In aggregate these four groups defined the set of occurrences that ranked highest in both condition and landscape quality (Figure 11, quadrant HH). (Detail on all 3272 floodplain polygons may be found in the accompanying GIS data). The four sets of high quality occurrences each contained 14 to 55 examples in the watershed. Figures 12 -16 and Tables 9-16 are included and the end of this paper due to their paper size and orientation
Figure 11. Conceptual model for identifying floodplain occurrences that ranked above the mean value in both condition and landscape context setting. The graph is divided into four quadrants such that occurrences in upper right quadrant (HH) meet this criterion.
Occurrences in the this quadrant were selected for detailed examination and are presented in the accompanying tables
Group 4: Thirty-two examples ranked above the mean for all four variables. The examples were associated with 19 small rivers (size 1 or size 2) and a few lakesides (Figure 12 and Table 9a) Correspondence with TNC portfolio areas was very high with 13 examples being found on critical streams, 6 examples within matrix forest sites, 6 examples corroborated by NHP ground inventory element occurrences and 10 examples being evaluated as candidate areas for the portfolio floodplains and wetlands (Table 9b). See tables 9a and 9b for individual grid codes corresponding to each polygon.
Ammonoosuc River size 3
Ashuelot River size 6 & 2: (2) confirmed with sycamore floodplain forest
Cold River size 2
Elmer Brook size 1: confirmed with small river floodplain forest
Hockanum River size 2
Johns River size 2 (2)
Mascoma River size 2: (2) confirmed with red maple floodplain forest
Mill River size 6 & 2: (2) confirmed with 2 examples of small river floodplain forest
North Branch Sugar River size 2
Otter River size 3
Piper Brook size 1 (2)
Podunk River size 1 (2)
Priest Brook size 2
Roaring Brook Number 2 size 1
Scantic River size 2: (4)
South Branch Ashuelot River size 2
Stockwell and Priest Brook size 2
Sunco Brook size 1
Third Branch White River size 2: (2)
Group 3H: Top ranked examples with 3 variables above the mean but hydrologic intactness below the mean (Figure 13 and Table 10a). This set of 14 examples from 7 rivers captured many larger river systems that, due to the sheer size of the rivers had storage to run ratios similar to that of the whole watershed. (The storage to run variable, our measure of hydrologic intactness, was somewhat negatively correlated with stream size). Twelve of these were found on TNC portfolio streams, 3 in Tier1 matrix forest blocks, 8 had candidate occurrences of wetland/floodplain systems and 4 were confirmed by NHP ground inventory occurrences (Table 10b). See tables 10a and 10b for individual grid codes corresponding to each polygon.
Ashuelot River Size 3 (3) confirmed with floodplain forest element occurrence
Coginchaug River Size 2 (1)
Connecticut River Size 4 (5) confirmed with high quality alluvial swamps and marshes
Manhan River Size 2
Mascoma River Size 2
Millers River Size 2
West River Size 3 (2) confirmed with silver maple ostrich fern element occurrence
Group 3F: Top ranked examples with 3 variables above the mean but with spring flooding, as measured by % inundation, below the mean (Figure 14 and Table 11a) This set of 55 examples from 35 rivers captured many river systems that also had examples from group 4 in addition to these. Individual scrutiny of the raw data is necessary on this set as some examples still showed significant flooding but others did not. Thirty four of these examples were on TNC critical stream, 21 were in matrix forest blocks, 15 corresponded to candidate wetland sites but only 1 (Upper Ammonoosuc) was corroborated by a ground inventory element occurrence (Table 11b). See tables 11a and 11b for individual grid codes corresponding to each polygon.
Ammonoosuc River size 2 (4)
Beaver Brook size 1
Boweyns Brook size1
Connecticut River Size 2
Deckers Brook size1
Dickinson Creek size1
East Branch Nulhegan River size 2
East Branch Passumpsic River size 2
East Swamp Brook size 1
Eightmile River size 1 and 2 (2)
Falls River size 1
Fort River size 2 (2)
Ham Branch size 2
Hockanum River size 2
Indian River size 2
Israel River size 2
Jeremy River size 2
Lower Connecticut Mainstem Tributaries size 7
Mascoma River size 2 (2)
Mill River size 2
Moose River size 2 (2)
Ompompanoosuc River size 2
Onion Brook size 1
Ottauquechee River size 6
Phillips Brook size 2
Podunk River size 1
Priest Brook size 2 (2)
Salmon Brook size 1
Scantic River size 2 (2)
Simms Stream size 2
South Branch Ashuelot River size 2
Stockwell and Priest Brook size 2
Stony Brook size 1
Upper Ammonoosuc River 2 & 3 (6) one confirmed with silver maple-ostrich fern forest.
West Branch Westfield River size 2 (3)
Unknown name size 6 and 1 (5)
Group 3N: Top ranked examples with 3 variables above the mean but % natural cover below the mean (Figure 15 and Table 12a). This set of 20 examples from 11 rivers captured many river systems with significant proportions of agriculture on the occurrences. Some of these may be restorable to floodplain forest or may have small examples of existing forest in parts of the occurrence. Correspondence with TNC portfolio sites was somewhat weaker for this group with 7 being found on critical streams, none being in matrix forest blocks, 2 being identified as candidate portfolio wetlands and 2 being corroborated by NHP ground inventory occurrences of floodplain forest or alluvial marsh (Table 12b). See tables 12a and 12b for individual grid codes corresponding to each polygon.
Cow Bridge Brook size 1
Dickerson Brook size 6
Goff Brook size 1
Halls Brook size 1
Hubbard Brook size 1 (2)
Mill River size 2 (2) confirmed with small river floodplain forest element
Moose River size 2
Roaring Brook size 1
Saxtons River size 2 (2)
Scantic River size 2 (2)
Unknown size 1 (5) size 6 (1) one confirmed with alluvial marsh
Group 3L: Top ranked examples with 3 variables above the mean but the Landscape / Watershed context index was below the mean (Figure 16 and Table 13a). This set of 8 examples from 7 rivers were mostly from the southern end of the basin were development is heaviest. While the current condition may be high these examples may be hard to restore over the long term due to their setting. Only the examples on the Quaboag River corresponded to a critical stream. One occurrence (sunny valley) was in a matrix forest block and no occurrences were corroborated by NHP ground inventoried occurrences (Table 13b). See tables 13a and 13b for individual grid codes corresponding to each polygon.
Coginchaug River size 1
Farmington River size 2
Mill Brook size 1
Ottauquechee River size 3
Quaboag River size 2 (2)
Still River size 6
Sunny Valley size 2
Evaluation by River System: The high quality examples of floodplain occurrences were concentrated in a set of 62 rivers. Some of these, such as the Scantic (8), Ammonoosook (6), Connecticut (6), and Asheolot (5) had several high ranking examples scattered along the river floodplain. Other had only one or two examples (Table 14). Table 14. Summary of top examples by river system. Examples are shown with their quality ranks: AA4, 3F, 3H, 3L, 3N, 2, 1, 0 and the number and element occurrence rank of any confirming inventory sites. Table is ordered by the number of AA4 and AA3 ranked examples.
The twenty top ranking streams collectively contained 75 floodplain areas totaling to 12,800 acres excluding the occurrences smaller than 50 acres (Table 15 and see expanded Table 16 in the back pages)
Table 15: The top twenty rivers for existing and potential floodplain conservation.
TOP RANKED RIVERSNumber of High Quality FOs by Stream SizeSize (acres)CONDLWCFEASNAME123467SUMAVE SUMAVEAVEAVEScantic River882241,789267204124Connecticut River6281751,39723320156Upper Ammonoosuc River5271961,375242248112Ammonoosuc River4268752025021291Mascoma River44196784272185119Mill River3143011,205252177101Fort River4418473720719289Priest Brook33140421263233123Podunk River33304913241185100Ashuelot River11133991,196241183139Moose River33103309236223100Saxtons River3363190223232100Hubbard Brook3366198223231106West Branch Westfield River3355165222225114West River12320962620218892Stockwell and Priest Brook22106211310211105Piper Brook2284168263172100Eightmile River112100200262208129Hockanum River227615224616668Johns River2214128224519610094976407512,838
Correspondence with the Nature Conservancys portfolio
About 1846 polygons corresponded directly with the TNC portfolio either adjacent to a critical stream, within a matrix block, matching a critical ecosystem example or containing a rare species occurrence. Another 200 were included if you counted some alternative portfolio sites such as tier 2 blocks or streams, or candidate occurrences that were not verified by ground information. Lastly 119 were adjacent to streams chosen for their connectivity values only (Tables 9b, 10b, 11b, 12b).
Appendix 1: Details on the calculations of the Ranking Attributes
Measure of Risk of Hydrologic Regime Alteration: The ratio of total dam storage to mean annual runoff is a useful indicator of the potential for hydrologic alteration in rivers ADDIN REFMGR.CITE Graf1999793Dam nation: a geographic census of American dams and their large-scale hydrologic impactsJournal793Dam nation: a geographic census of American dams and their large-scale hydrologic impactsGraf,W.L.1999damsgeographic variationhydrologic alterationIMPACTIMPACTSNew EnglandIn File13051311Water Resources Research354Water Resources Research1(Graf 1999, Dynesius and Nilsson 1994), particularly in absence of site-specific flow data. In the Connecticut River Watershed, an examination of all stream gages with at least 20 years of flow records found estimates of dam storage to mean annual runoff correlated to documented changes in the timing, magnitude, frequency, and duration of the natural flooding regime (Zimmerman and Lester 2006). Tributaries with maximum dam storage <10% of mean annual runoff were classified as low flow alteration risk while tributaries with dam storage >50% of runoff were classified as severely impacted. Tributaries with storage between 10 and 30% of runoff were classified as moderate flow alteration risk and tributaries storage between 31 and 50% of runoff were classified as high flow alteration risk (Zimmerman and Lester 2006).
A. Reach Risk of Hydrologic Regime Alteration: This metric gave us a measure of potential alteration to the hydrologic regime on the stream reach adjacent to each modeled floodplain occurrence. The hydrologic regime in this reach directly impacts the floodplain occurrence in terms of the floodplain receiving overbank flows from this reach. Ideally, we would choose to protect/restore a floodplain occurrence adjacent to a stream reach with a more functional hydrologic regime to maximize the potential that the floodplain would receive overbank flows with the most natural timing, magnitude, frequency, and duration. Because we do not have adequate stream gage data to examine site-specific hydrologic impacts for every stream reach in the Connecticut basin, we relied on reach estimation of dam storage/mean annual runoff as a means to estimate the potential of hydrologic alteration due to dams. For each reach, we calculated the total volume of water stored behind dams upstream using the National Inventory of Dams NID_STOR attribute and an EPA Rf3 stream routing accumulation script (Fitzhugh, 2001) which allowed us to sum all the storage of all dams upstream for each reach in the basin. We calculated the mean annual runoff of each reach by generating the total upstream drainage area (Olivero, 2003; Fitzhugh, 1999) of each reach and applying a coefficient of .19 which was developed within the Connecticut River for mean annual flow estimation (Zimmerman and Lester 2006). Calculations using the following formulas allowed generation of the reach flow alteration index in comparable units (acre-feet).
Total Dam Storage: NID_STOR acre-feet
Mean Annual Runoff: area in sq.mi * 2589988 (to get area in sq.meters) * 0.19 (apply coefficient to get run off in cubic feet) * 0.000022957 (convert cubic feet to acre-feet)
B. Watershed Risk of Hydrologic Regime Alteration: This metric gave us a measure of potential alteration to the hydrologic regime within the larger watershed that each modeled floodplain occurrence was located within. Ideally, we would choose to protect/restore a floodplain occurrence that was located within a more functional larger watershed hydrologic system to maximize the potential for the occurrence to interact with larger watershed hydrologic processes related to floodplain development and maintenance (e.g. seed dispersal to maintain network of floodplains downstream). Because we do not have adequate stream gage data in each of the major watersheds of the Connecticut River to examine site-specific hydrologic impacts, we relied on the estimation of dam storage/mean annual runoff done for the 44 major tributary watersheds by Zimmerman and Lester 2006. For the mainstem Connecticut River and the small tributary watersheds (<30 sq.mi.) entering directly into the mainstem Connecticut river without being within one of the 44 major tributaries, we relied on the cumulative Reach Hydrologic Regime Alteration Estimates described above.
Length of Connected Stream Network: This metric gave us a measure of length of stream network between dams that the floodplain occurrence was situated within. Ideally, we would choose to protect/restore a floodplain occurrence that was located within a larger network of accessible streams to maximize the potential for the floodplain to be accessible and serve as source of spawning habitat, nutrients, sediment and other related processes for the instream biota and ecosystems connected to it via the related accessible stream channel network. The network length was calculated using dams as barriers to connectivity. The dam database for the Connecticut River Watershed was updated in 2006 to include not only the larger dams that are tracked by the Army Corps of Engineers National Inventory of Dams but also the hundreds of small state dams mapped by Connecticut, Massachusetts, New Hampshire, and Vermont (Lester and Olivero, 2006). The hydrography network used for length calculations was the National Hydrography Dataset (NHD) 1:100,000k flowline dataset from the USGS (2006).
# of Hydropower, Floodcontrol, and Water Supply Dams: This metric gives us a measure of the number of hydropower, floodcontrol, and watersupply dams upstream of a given modeled floodplain occurrence. These types of dams were chosen to report in a metric due to their purpose which includes strong economic and human safety concerns for their existence and management policy. It is hypothesized that these types of dams will be more difficult to remove and it may be more difficult to renegotiate flow release policies with their owners to meet more natural regimes. It is hypothesized that other types of dams such as recreational, wildlife and farm ponds etc. will be easier to remove, reoperate to meet natural flow regimes over time, and likely already have less hydrologic regime impacts as many are run-of-the-river dams. This metric thus gives a measure of the number of more difficult dams whose location and management would need to be addressed to restore a more natural hydrologic regime.
Literature Cited
Anderson, M. G., Lombard, K., Lundgren, J., Allen, B., Antenen, S., Bechtel, D., Bowden, A., Carabetta, M., Ferree, C., Jordan, M., Khanna, S., Morse, D., Olivero, A., Sferra, N., Upmeyer, A. 2006b. The North Atlantic Coast: Ecoregional Assessment, Conservation Status Report and Resource CD. The Nature Conservancy, Eastern Conservation Science, Boston, MA.
Anderson, M.G., Vickery, B., Gorman, M., Gratton, L., Morrison, M., Mailet, J., Olivero, A., Ferree, C., Morse, D., Kehm, G., Rosalska, K., Khanna, S., and S. Bernstein. 2006a. The Northern Appalachian / Acadian Ecoregion: Ecoregional Assessment, Conservation Status and Resource CD. The Nature Conservancy. Boston, MA.
Anderson, M.G. 1999a. Viability and Spatial Assessment of Ecological Communities in the Northern Appalachian Ecoregion. PhD dissertation. Univ. of New Hampshire. 224 pp.
Anderson, M.G., Comer, P., Grossman, D., Groves, C., Poiani, K., Reid, M., Schneider, R., Vickery, B., and A. Weakley. 1999b. Guidelines for representing ecological communities in ecoregional conservation plans. The Nature Conservancy. 74 pp.
Anderson, M.G., M.D. Merrill, and F. D. Biasi. 1998. Connecticut River Watershed Analysis: Ecological communities and Neo-tropical migratory birds. Final Report to USGS Biological Resources Division. The Nature Conservancy, Boston MA.. 236 pp.
Cohen, W.B., and M. Fiorella. 1998. Comparison of methods for detecting conifer forest change with Thematic Mapper imagery. In R. S. Lunetta & C.D. Elvidge (Eds.), Remote sensing change detection: environmental monitoring methods and applications (pp. 89-102). Chelsea, Michigan: Ann Arbor Press
Crist, E.P., and R.C. Cicone. 1984. A physically-based transformation of Thematic Mapper data: The TM tasseled cap. IEEE Transactions on Geoscience and Remote Sensing. Vol. GE-22, pp. 256-263
Dynesius M. and Nilsson, C. 1994. Fragmentation and Flow Regulation of River Systems in the Northern Third of the World. Science 266:753-761.
DeGraaf, R. M. and M. Yamasaki. 2000. New England Wildlife: Habitat, Natural history and Distribution. New England Press
ESRI 2005: GIS Arc Info
ERDAS 2005: Imagine software
Fels, J, and K.C. Matson. 1997. A cognitively-based approach for hydrogeomorphic land classification using digital terrain models. HYPERLINK "http://www.ncgia.ucsb.edu/conf/SANTA_FE_CD-ROM/sf_papers/fels_john/fels_and_matson.html" www.ncgia.ucsb.edu/conf/SANTA_FE_CD-ROM/sf_papers/fels_john/fels_and_matson.html
Ferree, C.E. 2005. Supplementary metadata for the ELU30 raster dataset. The Nature Conservancy, Boston, MA, unpublished paper
Fitzhugh, T. 1999. TNC GIS Freshwater Tools. The Nature Conservancy.
FitzHugh, T. 2000. Watershed Characteristics and Aquatic Ecological Integrity: A Literature Review. The Nature Conservancy. (unpublished report)
Graf, W.L. 1999. Dam nation: a geographic census of American dams and their large-scale hydrologic impacts. Water Resources Research 35:1305-1311.
Groves, C.R., D.B. Jensen, L.L. Valutis, K.H. Redford, M.L. Shaffer, J.M. Scott, J.V. Baumgartner, J.V. Higgins, M.W. Beck, and M.G. Anderson. 2002. Planning for biodiversity conservation: Putting conservation science into practice. BioScience. 52(6): 499-512.
Harper, D., Mekotova, J., Hulme, S., White, J. and J. Hall. 1997. Habitat heterogeneity and Aquatic Invertebrate Diversity in Floodplain Forests. Global Ecol.and Biogeogr. Letts. 6: 275-285. Blackwell Science, London.
Hodgson, M.E., J.R. Jensen, J.A. Tullis, K.D. Riordan, and C.M. Archer. 2003. Synergistic use of lidar and color aerial photography for mapping urban parcel imperviousness. Photogrammetric Engineering and Remote Sensing 69: 973-980
Hudson, P.F., and R.R. Colditz. 2003. Flood delineation in a large and complex alluvial valley, lower Panuco basin, Mexico. J. Hydrology 280: 229-245
Jain, S.K., R.D.Singh, M.K.Jain, and A.K.Lohani. 2005. Delineation of Flood-Prone Areas Using Remote Sensing Techniques. Water Resources Management 19: 333-347
Jospe, A. and Olivero, A.P. 2007. Fragmentation Analysis in Connecticut River Watershed. The Nature Conservancy, Connecticut River Program and Eastern Conservation Science Department. (unpublished report)
Olivero, A.P. 2003. Development of a GIS based freshwater classification and conservation prioritization in Lower New England. Masters Thesis. Department of Biology. University of Massachsuetts-Boston. 203p.
Poiani, K.A., B.D. Richter, M. G. Anderson, and H. E. Richter. 2000. Biodiversity conservation at multiple scales. BioScience 50(2): 133-146.
Poole, A, P. Stettenheim and F. Gill. 2000. Birds of North America No 1-600. Philadelphia: THe Academy of Natural Sciences; Washingon, D.c: THe American Ornithologists Union
Sperduto, D.D., and W.F. Nichols. 2004. Natural Communities of NH. UNH Cooperative Extension, Durham, NH
Strager, J.M., C.B. Yuill, P.B.Wood. 2000. Landscape-based riparian habitat modeling for amphibians and reptiles using Arc/Info Grid and ArcView GIS, West Virginia GAP Project, unpublished paper
Steinberg, D. and P. Colla. 1997. CARTClassification and Regression Trees. San Diego, CA; Salford Systems.
Thompson, E.H., and E.R. Sorenson. 2000. Wetland, Woodland, Wildland. University Press of New England, Hanover, NH
Wang, L., J. Lyons, and R. Gatti. 1997. Influences of watershed land use on habitat quality and biotic integrity in Wisconsin streams. Fisheries 22 (6) 6-12.
Wang, Y. 2004. Using Landsat 7 TM data acquired days after a flood event to delineate the maximum flood extent in a coastal floodplain. Int. J. Remote Sensing 25: 959-974
Wang, Y., J.D. Colby, and K.A. Mulcahy. 2002. An efficient method for mapping flood extent in a coastal floodplain using Landsat TM and DEM data. Int. J. Remote Sensing 23, 3681-3696
Zimmerman J. and A. Lester. 2006. Spatial distribution of hydrologic alteration and fragmentation among tributaries of the Connecticut River. The Nature Conservancy, Connecticut River Program, Northampton, MA (unpublished report)
Zimmerman, J. 2006. Response of physical processes and ecological targets to altered hydrology in the Connecticut River basin. The Nature Conservancy, Connecticut River Program, Northampton, MA (unpublished report)
Figures and Tables
These pages contain landscape formatted figures and tables to large to fit within text pages
Figures 12 16.
Figure 12. Group 4: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for four variables 1) % natural cover, 2) % flooding, 3) hydrologic intactness and 4) landscape/watershed context.
Figure 13. Group 3H: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % natural cover, 2) % flooding, 3) landscape/watershed context but below the mean for hydrologic intactness.
Figure 14. Group 3F: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables 1) % natural cover, 2) hydrologic intactness, 3) landscape/watershed context but was below the mean for % flooding. .
Figure 15. Group 3N: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % flooding, 2) hydrologic intactness, 3) landscape/watershed context but below the mean for % natural cover.
Figure 16. Group 3L: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % flooding, 2) % flooding, 3) hydrologic intactness but below the mean for landscape/watershed context.
Tables 9 - 16
Table 9a. Group 4: Attributes of Size, Condition and Landscape Context for Floodplain model occurrences over 50 acres in size. The table is ordered by column 4, the sum of the Condition and Landscape indices
Table 9b. Group 4: Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table 10a. Group 3H: Attributes of Size, Condition and Landscape Context for Floodplain model Occurrences. The table is ordered alphabetically by the label
Table 10b. Group 3H: Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table R11a. Group 3F: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table R11b. Group 3F Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table 12a. Group 3N: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table 12b. Group 3N Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table 13a. Group 3L: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table 13b. Group 3L Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table 16. Top ranked river systems based on 378 over 50 acres average size 163 (average Condition 196, average Landscape / Watershed 163). Each column shows the distribution of the variable by stream size and average or sum across sizes.
Figure 12. Group 4: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for four variables 1) % natural cover, 2) % flooding, 3) hydrologic intactness and 4) landscape/watershed context.
Table 9a. Group 4: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences over 50 acres in size. The table is ordered by column 4, the sum of the Condition and Landscape indices
Table 9b. Group 4: Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Figure 13. Group 3H: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % natural cover, 2) % flooding, 3) landscape/watershed context but below the mean for hydrologic intactness.
Table 10a. Group 3H: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences. The table is ordered alphabetically by the label
Table 10b. Group 3H Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Figure 14. Group 3F: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables 1) % natural cover, 2) hydrologic intactness, 3) landscape/watershed context but was below the mean for % flooding. .
Table R11a. Group 3F: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table R11b. Group 3F Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Figure 15. Group 3N: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % flooding, 2) hydrologic intactness, 3) landscape/watershed context but below the mean for % natural cover.
Table 12a. Group 3N: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table 12b. Group 3N Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Figure 16. Group 3L: High ranked floodplain model occurrences over 50 acres in the Connecticut River Watershed. Insert Box provides a close-up view of those occurrences that ranked higher than the mean for three variables: 1) % flooding, 2) % flooding, 3) hydrologic intactness but below the mean for landscape/watershed context.
Table 13a. Group 3L: Attributes of Size, Condition and Landscape Context for Floodplain Model Occurrences.
Table 13b. Group 3L Floodplain occurrences: Attributes of Feasibility and Correspondence with the TNC portfolio
Table 16. Top ranked river systems based on 378 over 50 acres average size 163 (average Condition 196, average Landscape / Watershed 163). Each column shows the distribution of the variable by stream size and average or sum across sizes.
PAGE
PAGE 32
a)
c)
b)
d)
EMBED Excel.Sheet.8
c.)
b.)
d.)
a.)
e.)
d.)
c.)
b.)
a.)
% & ; < C I Q S V ^ a j q s u
= >
;
Z { ʿwwww hI B*ph h9 B*ph h R B*ph h6 B*ph h - B*ph hI B*ph h7* B*ph hj# B*ph h|% B*ph h6 h6 B*ph h6 h6 5B*\ph hM. h2; hW h]) hu % 5 h]) 5 hf 5hE hf 5 . & C D v 9 + & ' gd
& F d d [$\$gd6
& F gdE
& F gd` d d [$\$gd6 " " o# U V ^ m & ' 9 P Q f s y z 9 ? ߹Ĺıġ̡̡ߡ玙~ ht B*ph hy` B*ph h6 h R B*ph hJz B*ph hM. B*ph h{ B*ph hhk B*ph h6 hJz B*ph hN B*ph hF B*ph h6 hI B*ph hI B*ph hI B*ph h R B*ph h/5 B*ph 1? F G | } ~ 9 : D E F J Z b w * xpxxxc h` h` >*B*ph hE B*ph h R B*ph hm B*ph h|% B*ph hdG B*ph hy` B*ph h6 B*ph h6 h6 B*ph h6 h6 5B*ph h6 5B*ph hF 5B*ph h{ B*ph hF B*ph h6 h R B*ph ht B*ph hM. B*ph $* + 3 4 9 A D I S Y Z [ c ) * + . 5 > E G H Q ] ·ʨҔ~zrmrhmrcrmrh hw" >* hy` >* hE >*hE hE >*hE h` h` B*ph h` h R B*ph h R B*ph h6 B*ph h` h` h` hdG B*ph h6 h6 B*ph hy` B*ph h` B*ph h` h R >*B*ph hM. >*B*ph h` h` >*B*ph h` >*B*ph & 1 0 A D I J T U l % & ( ) E P l o ǽުǽ}uh CJ aJ hjO 5h6 hf 5h p+ B*ph hdG B*ph hE B*ph h|% B*ph h6 h6 B*ph hy` B*ph hy` >*B*ph h` h R >*B*ph h` >*B*ph h R B*ph hE hE B*ph hE hy` h` h` . ' ( # X ] ` % ' , ; g h i ⿻ַֻûַַֻ h p+ ht h~. hfP ht ha h p+ hm hm h s< h"
[ h9 hM. hw" h p+ h h p+ h p+ 5 hy` 5 hjO 5 E' ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! gd gd F \ e i ! ! ! ! ! ! ! ! ! ! ! " " " " " # # ]# b# e# j# մը՜Վqjcjjj h p+ h9 h p+ h. j h p+ hZ UmH nH u hy h 6CJ aJ hy h 6CJ aJ y( hy hZ 6CJ aJ hy hMT 6CJ aJ hy hf\ 6CJ aJ ha 6CJ aJ hy hN8 6CJ aJ hy h 6CJ aJ hN8 h h s< h~. h. ht h p+ h &! ! ! ! ! % % T% U% e% m% x% $If gd gd. gd j# t# ~# # # # $ $ !$ % % )% .% B% E% J% M% R% S% T% U% % & & 3& 5& 6& [& \& }& ~& & & & & & & & & & E' N' P' Q' T' Z' ]' ^' `' p' v' ' ' ' ' ( !( >( Q( R( X( p( ( ( ( ƿ h4 hp`l hf\ hZ h h#> h p+ h h p+ h]) ha huk h]) hy hA 6CJ aJ hy h 6CJ aJ hy h. h p+ h. @x% y% {% % % L C C C $If gd kd $$If l F \
z N V
t 0 6 4 4
l a p % % % % % q h h h $If gd kd $$If l F \
z N V
t 0 6 4 4
l a % % % % & q h h h $If gd kd5 $$If l F \
z N V
t 0 6 4 4
l a & & & & 4& q h h h $If gd kd $$If l F \
z N V
t 0 6 4 4
l a 4& 5& 7& C& Z& q h h h $If gd kd $$If l F \
z N V
t 0 6 4 4
l a Z& [& ]& i& |& q h h h $If gd kd $$If l F \
z N V
t 0 6 4 4
l a |& }& ~& & ) ) 1- 2- . . . . q l g g g g g b g g g gd4 gd gd kd $$If l F \
z N V
t 0 6 4 4
l a ( ( ( ( ( ( ( ( ( ( ( ( ( ) $) *) ) ) ) ) ) ) ) .* S* T* + + , , , , , , )- ,- 1- 2- 3- . . . . . . . .
/ / // gdjO gd 0 0 0 K0 L0 M0 O0 S0 T0 a0 b0 k0 s0 0 0 0 0 0 0 1 &1 (1 U1 1 1 1 2 2 C4 H4 4 4 4 4 4 4 4 4 4 5 5 %5 @5 [5 d5 p5 z5 5 5 5 5 5 6 6 6 6 6 6 6 6 6 6 ̼꼮 hy hjO 6CJ aJ h p+ hjO h !' h! 5 hjO 5hjO h: hD1- hl hbq h p+ h hMT h p+ h9 hA h hZ h p+ h4 hq h p+ h =6 6 6 6 6 6 6 6 6 6 7 7 7 07 ;7 <7 H7 7 7 7 7 8 8 +8 .8 58 H8 S8 8 8 8 8 8 8 8 8 8 8 9 9 9 9 =9 @9 P9 : : : : : /; ; ; I< K< Ͼ hL hL h p+ hL h y> hL h.0[ hj h*K h p+ h h p+ h 5 hA 5hjO h p+ hjO 5 hjO 5j~ h p+ h3 Uj`aI
h p+ hjO UV h p+ hjO j h p+ hjO U 6K< L< R< S< U< < < < < < < < < < %= &= 7= 9= = > ? 2? 6? 8? z? ? ? ? ? ? ? &@ '@ (@ +@ /@ @ @ @ A A A A A A A B B ڱ hy h 6CJ aJ hy h hy hx hy h h h hx h j h p+ h UmH nH u hy hMT 6CJ aJ hy hZ 6CJ aJ hy h y> 6CJ aJ h h y> h h p+ h h y> hZ /%= '= (= )= *= += ,= -= .= /= 0= 1= 2= 3= 4= 5= 6= 7= 8= 9= 7? 8? ? ? '@ (@ @ @ A A gd A B B B B B B C C C C C C C C C C C C C C C C C C C C C C gd gd B B B
B B B 'B 3B jB mB pB qB rB sB B B B B C C C C C C 8D 9D G G G G G %G &G G ݼ݃yleae]eY hF h6y hx h p+ hF h p+ hx 5B*
ph3 hx 5B*
ph3 h 5B*
ph3 ,j h p+ h 5B*
UmH nH ph3 u hy h6y 6CJ aJ hy hF 6CJ aJ hx CJ aJ hF h9 h9 hF hF hF h h p+ h h h p+ h j h p+ h UmH nH u!C C C C C 9D G G G G G G G G G G G G G G G G G G G G G G G gdF gd G G G G G G
H H H H H H I 'I 8I 0J 5J LJ SJ VJ DK ZK _K K K K K K K K K K ~L L L L L L L L L L L L L M M +N -N 4N 6N N O O
O 7O O O O O O O O 躧 h h
\ h p+ h H* h p+ hMT hOr h Q h p+ hF hF hF h6y hx hCf hMT hF hy h p+ h j h p+ h UmH nH u>G G K K ~L L L L L L L L L L L L L L L L L L L L L L L L L gdF gd L L L L L L L L L L L L L O O S S T !T )T *T +T ,T -T .T /T 0T 1T 2T 3T gd O O $P 0Q 2Q dQ iQ _S bS S S S T T T CT KT LT T T T T T KU U U U U U V V !V "V (V QV RV SV ^V V ykykykyfa h7 5 hC\ 5hC\ CJ OJ QJ ^J aJ hC\ CJ aJ hC\ CJ OJ QJ ^J aJ hC\ 5CJ OJ QJ \^J aJ huk hC\ 6CJ aJ huk h 7b 6CJ aJ h 7b hC\ h 7b 5 hjO 5 j h p+ h UmH nH u h6y h6y h h6y hx h p+ h hOr h p+ h
\ &3T 4T 5T 6T 7T 8T 9T :T ;T T ?T @T AT BT CT LT T T KU U U $A &P#$/ If gdC\ gd U U U U U U U U U U V V V V V FfF $$A &P#$/ If a$gdC\ $A &P#$/ If gdC\ V kd, $$If l F
& 'v&
6PA
v&6 4 4
l a p
V V V V !V "V (V .V 4V :V @V FV JV QV RV SV V V X X \ \ =` ` gd gd FfT $A &P#$/ If gdC\ Ff $$A &P#$/ If a$gdC\ V V V V V V V V +W -W :W BW GW IW bW cW eW iW W W W W W W FX cX X X X X Y
Y Y Y Y Y %Y ;Y ` ` ` ` ` ` ` ` ` a ?a qa ra a a Kb Tb b b b b c c Fc ec kc c c c c c d d d d d Se be e e -f 8f kh mh h h h h h h h h h h h i i k k k m m m Ƿǯ hP hjO h% 5hjO hjO 5hP h/ h% hb h wS hJ h& h9 h p+ h9 h h p+ h j h p+ h UjP h p+ h U >` ` c c h h h k k m m 5r 6r r r r r r r r r r r r r r r r r r gd m n n n 2n 4n n n *o p p q q q q r 4r 5r 6r 7r =r >r ?r Ur Xr r r r r r s s s s s ?s Is os ps vs ws |s s s t t )t +t .t /t :t $u Gu `u cu u ̴̴🛗 hq h& h h p+ h j h1
UmH nH u hH1 hH1 6CJ aJ hH1 h
{ 6CJ aJ hH1 h1
6CJ aJ hH1 hC\ 6CJ aJ hb h1
h7h hH1 h9 hQd h`J 7r r s s s s s s s s s s
s s u u v v [w `w aw bw cw dw ew fw gw hw gdf gd& gdA gd u u u u u 3v 4v Yv v v v v v v v v v w w [w \w aw bw fw hw qw sw tw w w w x x x x x x x y !y -y Wy ey ny qy y y ҺҬޝ{{{ haF hN hf hfR hk hk 5B*ph hk 5hk h& 5 h p+ h& j h& UmH nH u hM@ h& 6CJ aJ hM@ hC\ 6CJ aJ hM@ h 6CJ aJ h|B h p+ hA hC\ hq h& h h p+ h .hw w w y y y y y
z z z z z { { | | | | ' ( ~ `gdf/ gddG gdk
& F gdf gdf y y y y y y y y y z
z z z z z z z z z { { { :{ @{ G{ H{ { { {
| | (| H| N| U| V| w| {| | | | | | | | } } } %} &} +} ,} 4} 5} 8} 9} C} G} Q} R} X} Y} ^} ʾƺƺƶƲƾƺƺƺƺƲƭơơơơơơơhUF h hR h >* haF >*hzy h ? hfo hR hb >*hb h 6 h)U h 6 5h~< hk hk hk 5 hfR 5 h> 5hV hT hf h/ >^} c} g} } } } } } } } } } } ~ ~ !~ Y~ o~ ~ ~ ~ ~ ~ ! + , / 6 > @ L M
@ E S ^ ` a À π 5 < H U Y Z f j m hMT hR h2U >*hN8 haF h2; hjM h2U hfo h t hUF hzy hb hUF h L " $ ' ( R d } ~ A R T m v σ փ ڃ n v z { # ! & [ q ̆ ч և ) S »»»»»»»»ֳ¯¯ h hR h| >* h t hCZF hCZF hMT hR hR hf/ >*h| hT 5h)U hf/ 5hw, hfo h 6hf/ hfo 6hf/ E ҇ Ӈ ~ ˋ ͋ l m e f j k ޑ `gd| gd| gdf S W k ĉ ɉ Չ ߉ # = Z \ ^ } $ % 1 ͋ ڋ k l p Y d i ~ i j n o t } ټٴٰٴٰٰ hX hR >*hX h)U >*h)U hR h| >*h)U h| 5hf/ h| 6 h| 6hMT h h| hN8 hT >*hN8 h| >*hf/ h( hCZF hCZF h =} ~ 5 Y Đ ǐ Ӑ ِ * N r Ñ ˑ ϑ ё ב ߑ C ֒ 2 ; b c y z ͓ ( * F G ̿Ŀ hW h h
4C hX hV 5hW h)U 5 h
4C 5 h|B 5h hq\N hMT hw* hpwG h h)U hX h)U >* hX >* Eޑ ߑ B C c ) G h Δ b c K N ˖ і $$If a$gd3 gdmI gdk gdf G Δ # a b ƕ ܕ ݕ ޕ ߕ M N T U
= > S T g h ї җ 7 9 ǘ Ș ǻǯǞǻLj hmI CJ OJ QJ ^J aJ hmI CJ aJ hmI 6CJ OJ QJ ]^J aJ hM@ hMT 6CJ aJ hM@ hC\ 6CJ aJ hM@ hmI 6CJ aJ hmI hmI hMT hC\ hmI hmI 5 hk 5hW] hV h hW 5
Y P D D D D $$If a$gd3 $If gd3 kd/ $$If l r R 0 6 4 4
l a] p2 1 4 7 : = Y P D D D D $$If a$gd3 $If gd3 kdD0 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 = > E G K O S Y P D D D D $$If a$gd3 $If gd3 kd1 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 S T Y [ _ c g Y P D D D D $$If a$gd3 $If gd3 kd2 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 g h n t x | Y P D D D D $$If a$gd3 $If gd3 kd4 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 Y P D D D D $$If a$gd3 $If gd3 kd`5 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 Y P D D D D $$If a$gd3 $If gd3 kd6 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 ɗ ї Y P D D D D $$If a$gd3 $If gd3 kd7 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 ї җ ؗ ޗ Y P D D D D $$If a$gd3 $If gd3 kd59 $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 7 9 ? Y T O T F : $$If a$gd3 $If gd3 gdk gdmI kd|: $$If l r ' 'R ' '0 ' 6 4 4
l a] p2 ? E J V [ a g q x $$If a$gd3 kd; $$If l ִ
9 p$ % % % % % %c %W % X6 4 4
l a pP ǘ $$If a$gd3 $If gd3 ǘ Ș kdA= $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP Ș Ϙ Ԙ ژ $$If a$gd3 $If gd3 Ș X Y S T V ` a ޚ ) ՜ ݟ ġ ˡ ϡ С ӡ ƻƻͳِٜͬ٘٘ٔ٘ hz1 hCm ht^ h6 hk 6hAz hAz 6 hAz hAz hAz hmI 6 hAz hmI hjM hAz hk hAz h6 hk 5hk hk hk 5 hk 5hmI CJ aJ hmI CJ OJ QJ ^J aJ 8 kd> $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP $$If a$gd3 $If gd3 kd=@ $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP 3 9 > C H M R X $$If a$gd3 $If gd3 X Y kdA $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP Y a f k p t x | $$If a$gd3 $If gd3 kd9C $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP $$If a$gd3 $If gd3 kdD $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP ș Ι ԙ ٙ ޙ $$If a$gd3 $If gd3 kd5F $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP
$$If a$gd3 $If gd3 kdG $$If l ִ
9 p$ % % % % % %c %W %X6 4 4
l a pP ) / 5 ; A G M S $$If a$gd3 $If gd3 S T kdMI $$If l ִ
9 p$ % % % % % %c %W % X6 4 4
l a pP T U V ל ! " # Ф Ѥ gdk ƥ ǥ z è Ψ " # : J K V l \ g Ȫ ɪ ժ ) / : P X Y Z ǿ hC\ hAz h h13 hQj hR hz hK hW h hV hvY h} hAz hAz 5hk hz1 h6 hk j h|B UmH nH u hM@ hC\ 6CJ aJ hM@ hk 6CJ aJ hk 6Ѥ # * , $If gdxj `gdQj gdf gdk Z ^ e k s , - E F d e ˬ ̬ ͬ ά Ϭ լ ඪ~v~v~v~h~h~v~Z~Z~v~ h CJ OJ QJ ^J aJ hz1 CJ OJ QJ ^J aJ h13 CJ aJ h13 CJ OJ QJ ^J aJ h h13 5CJ aJ #h h13 5CJ OJ QJ ^J aJ h h 5CJ aJ #h h 5CJ OJ QJ ^J aJ hM@ hW 6CJ aJ hM@ hC\ 6CJ aJ hM@ h13 6CJ aJ hAz hV hxj hW h13 #, - / 5 : < > E $If O kdJ $$If l ' '
6 4 4
l a] p
E F S $If kd^K $$If l ֈ @ x4 'O '8 ' '8 ' '8 < 6 4 4
l a] p< S W [ ^ ` d e 8 kdM $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< $$If a$ e l p t x | $$If a$ $If A ; 2 2 2 $$If a$ $If kdzN $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< 8 2 $If kdO $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< $$If a$ ì Ƭ ˬ ̬ 8 kdVQ $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< $$If a$ ̬ ֬ ڬ ެ $$If a$ $If A ; 2 2 2 $$If a$ $If kdR $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< 8 3 3 gdxj kdNT $$If l ֈ @ x4 'O '8 ' '8 ' '8 6 4 4
l a] p< $$If a$ * 9 E R w x í ѭ ֭ G H I Q [ n ̮ ߮ % - 7 J ï H K ñ شؙ hM@ h 6CJ aJ hM@ hM@ 5 hAz hAz hM@ hz1 hw h h h7s hAz hW 5hMT hAz hW hz h he hAz hAz >*h13 h13 CJ aJ : I % ò 9 : ; > ? @ V m )
& F gdD gd
& F gdW gdxj ñ ű ò IJ ϲ ײ ܲ 9 : < = ? E F H 4 5 9 µ Ź粮|n|n| h1 CJ OJ QJ ^J aJ hD CJ OJ QJ ^J aJ hz1 h%# h. h hG he hiPG hz h7s h h h hM@ hMT 6CJ aJ +j hM@ h 6CJ UaJ mH nH uhM@ h%# 6CJ aJ hM@ h 6CJ aJ hM@ hw 6CJ aJ )µ õ N U a h z ɶ ! ( 4 ; G N _ g w ~ ҷ ٷ ' 0 6 < > x y ¸ ߸ 0 o r ļĸİİ h>_ h h%# hE h( h. h h hiPG hz h/iR CJ OJ QJ ^J aJ h1 CJ OJ QJ ^J aJ hD CJ OJ QJ ^J aJ h{K CJ OJ QJ ^J aJ >) < S k ڷ 5 6 ` } ӻ e f g D E gd( ] ^]
& F gd,ho gdz1 gdD
& F gdD + _ { | һ
& d f g o p ۼ ܼ # % M N e l ~ D E U ` m t ſ ʿ ܚܚܚܚܚ hiPG CJ OJ QJ ^J aJ h[ h>_ h. h h hiPG hD CJ OJ QJ ^J aJ h CJ OJ QJ ^J aJ h( CJ OJ QJ ^J aJ h,ho CJ OJ QJ ^J aJ h( hz1 _ hO( h h. h[ h{K 0 1 I [ u ! " # gdxj gd{K gdO(
& F
gdO( m y | < @ Z r ] xgVg h2; 5CJ OJ QJ \^J aJ hR 5CJ OJ QJ \^J aJ hM@ h5@- 6CJ aJ hM@ hR 6CJ aJ h5@- hM@ h9 h2; hR hG h hG h{K h{K CJ OJ QJ ^J aJ j h2z UmH nH u hM@ h2z 6CJ aJ hM@ hyb 6CJ aJ hM@ h2z 6CJ aJ hM@ hE 6CJ aJ # $ % & ' ( ) * + , - . / 0 1 2 3 4 5 6 7 8 9 : ; < ]