SUSTAINING ENVIRONMENTAL FLOWS IN SOUTHERN NEW ENGLAND RIVERS : EFFECTS OF WATERSHED FACTORS AND LAND USE BY

Rivers and river systems serve as conduits for nutrients and organisms, function as corridors for fish and wildlife passage, and provide resources for humans. Streamflow has been called the master variable in a river because it affects habitat diversity and availability through its impact on physical factors that influence habitat quality. However, land use changes such as urbanization and irrigation, can have major effects on stream hydrology. Modifications of the land surface due to urbanization alters natural stream hydrographs by increasing flood peaks, decreasing time to peak flows, and causing higher runoff velocities. Irrigation may produce the opposite effects. In order to preserve a spectrum of stream functionality, rivers must maintain seasonally adequate flows. For example, low flows can affect stream connectivity, restrict movement of aquatic organisms, concentrate prey into limited areas, purge invasive species from riparian corridors, and enable recruitment and evolution of floodplain plants. State agencies throughout the Northeast U.S. are considering policies linked to low-flow thresholds that sustain these ecosystem services. Methods that set minimum flow standards often result in conflicting values, due to differing environmental goals and levels of protection they aim to achieve. Two such methods, the USFWS Aquatic Base Flow (ABF) method and the Wetted Perimeter method have been widely used. The USFWS ABF method recommends using the median of August flows and has been refined for Rhode Island (RIABF). The wetted perimeter method uses stream cross-sections at riffle locations to determine critical flow values to maintain flow based on the wetted perimeter of the channel. In addition to setting flow standards, methods to minimize the adverse effects of urbanization have also been proposed. Low impact development (LID) has emerged as a strategy to reduce the hydrologic impacts of urbanization on aquatic ecosystems by combining site planning and design processes with runoff reduction and treatment practices. Within a given climatic region, water resource managers seeking to optimize stream ecosystem services need a clear understanding of the importance of land use, physical/climatic characteristics, and hydrography on different components of stream hydrographs. Within 33 Southern New England watersheds (average area 80 km), we assessed relationships between watershed variables and a set of low flow parameters: 1-, 7and 30day minimum flows. We used an information theoretical approach to develop regression models to identify relationships between landscape attributes and parameters that describe different components of the flow regime. The key variables identified by the AIC weighting factors as generating positive relationships with median annual minimum flow events included percent stratified drift (greater infiltration and storage), mean elevation (likely related to higher snowfall), drainage area and mean August precipitation. The extent of wetlands in the watershed was negatively related to low flow magnitudes likely due to the capacity of those ecosystems to remove water from the basins via evapotranspiration during drought conditions. Of the various land use variables, the percent developed land was found to have the highest importance, but it was less important than wetlands and physical/climatic features. The extent of impervious cover in the study watersheds was primarily less than 10% and the study watersheds were generally larger than watersheds used in other studies relating impervious cover to stream health. Our results suggest that even with watersheds located within close spatial proximity, strategies focused on balancing water extraction to sustain low flows in fluvial systems can benefit from attention to select watershed features. We draw attention to the finding that streams located in watersheds with high proportions of wetlands may require more stringent approaches to withdrawals to sustain these ecosystems during drought periods. We then determined the minimum flow requirements at three locations (riffle zones) along the Beaver River, located in southern Rhode Island, using both the wetted perimeter method and the RIABF method. In order to determine stream flow at ungaged locations, runoff was modeled using the HEC-HMS rainfall/runoff model. To assess biological conditions, we reviewed macroinvertebrate, fish and temperature data obtained within the watershed. Biological conditions of the Beaver River indicated that the Beaver River is a well-functioning stream habitat. Minimum stream flow requirements using ABF and WP methods were investigated. Stream flows were found to be below the ABF value between six to 21% of the time and below the WP flow between 37 to 72% of the time. Physical and biological sampling done in the watershed indicate the river is a well-functioning, river, comparable to pristine sites; however minimum flow criteria set by the wetted perimeter method suggest that the river is flowing below critical flow values over 50% of the time. Our results suggest that minimum flow values obtained from the wetted perimeter method for southern New England rivers should be approached with caution and should be compared to results obtained from other methods to determine the accuracy and applicability of the critical flows prior to using these values for any type of instream flow regulations. We also assessed the effect of increased impervious cover for both conventional and LID-based urbanization on low flow metrics and flow depths in riffle habitats in a small, relatively undeveloped watershed located in southern Rhode Island. We employed a hydrologic model to simulate stream flow, base flow and storm flows under different land cover scenarios and then compared these results to the effects of direct stream withdrawals from agricultural irrigation. We found baseflow to be negatively correlated to impervious area. On pervious surfaces, direct runoff is likely to be infrequent during the summer months, when most of the precipitation that falls is utilized for the soil moisture deficit. In contrast, connected impervious area (IA) will generate immediate runoff to streams from rainstorms that would have otherwise infiltrated the soil. During periods of excess precipitation, the falling limbs of those hydrographs generated prolonged periods of comparatively elevated flows. Combining baseflow and storm flow shows that increased values of IA can generate higher flow values during the summer months during periods with excess precipitation. As IA increases through the different land use scenarios, storm related runoff increases immediately following precipitation events, causing higher stream flows. The small decreases in base flow input to the stream due to increased IA are negated by the impacts of the higher storm flows, causing summer stream flows to be higher under the developed land use scenarios than existing conditions. Changes to the channel depth of the riffles were also relatively minor. During a year with median precipitation, the model predicted a lower frequency of low flows with both conventional development and with LID compared to the predictions for the limited development present in current conditions. Both conventional development and LID also display fewer low flow periods during a dry year, but the pattern reverses, with LID predicted to have slightly lower frequencies of low flows than the conventional development. Over the summer, storm runoff and the associated falling limb of the runoff hydrograph that results from connected impervious cover occurs with enough frequency to influence the low flow thresholds we use for metrics. During the dry year, rainfall occurrences were very infrequent and the higher baseflow associated with LID accounts for the slight increase in flows compared to the conventional development. Irrigation scenarios decreased both flows and depths. Changes in land use generally increase river flows while water withdrawals decrease river flows. The occurrence of low flows within the Beaver River was found to be relatively resilient to the extent of development and water withdrawals simulated by this study. These analyses will help inform future water management decisions in watersheds with the diversity of land uses that occur in southern New England.

major effects on stream hydrology. Modifications of the land surface due to urbanization alters natural stream hydrographs by increasing flood peaks, decreasing time to peak flows, and causing higher runoff velocities. Irrigation may produce the opposite effects.
In order to preserve a spectrum of stream functionality, rivers must maintain seasonally adequate flows. For example, low flows can affect stream connectivity, restrict movement of aquatic organisms, concentrate prey into limited areas, purge invasive species from riparian corridors, and enable recruitment and evolution of floodplain plants. State agencies throughout the Northeast U.S. are considering policies linked to low-flow thresholds that sustain these ecosystem services.
Methods that set minimum flow standards often result in conflicting values, due to differing environmental goals and levels of protection they aim to achieve. Two such methods, the USFWS Aquatic Base Flow (ABF) method and the Wetted Perimeter method have been widely used. The USFWS ABF method recommends using the median of August flows and has been refined for Rhode Island (RIABF). The wetted perimeter method uses stream cross-sections at riffle locations to determine critical flow values to maintain flow based on the wetted perimeter of the channel. In addition to setting flow standards, methods to minimize the adverse effects of urbanization have also been proposed. Low impact development (LID) has emerged as a strategy to reduce the hydrologic impacts of urbanization on aquatic ecosystems by combining site planning and design processes with runoff reduction and treatment practices.
Within a given climatic region, water resource managers seeking to optimize stream ecosystem services need a clear understanding of the importance of land use, physical/climatic characteristics, and hydrography on different components of stream hydrographs. Within 33 Southern New England watersheds (average area 80 km 2 ), we assessed relationships between watershed variables and a set of low flow parameters: 1-, 7-and 30-day minimum flows. We used an information theoretical approach to develop regression models to identify relationships between landscape attributes and parameters that describe different components of the flow regime.
The key variables identified by the AIC weighting factors as generating positive relationships with median annual minimum flow events included percent stratified drift (greater infiltration and storage), mean elevation (likely related to higher snowfall), drainage area and mean August precipitation. The extent of wetlands in the watershed was negatively related to low flow magnitudes likely due to the capacity of those ecosystems to remove water from the basins via evapotranspiration during drought conditions. Of the various land use variables, the percent developed land was found to have the highest importance, but it was less important than wetlands and physical/climatic features. The extent of impervious cover in the study watersheds was primarily less than 10% and the study watersheds were generally larger than watersheds used in other studies relating impervious cover to stream health. Our results suggest that even with watersheds located within close spatial proximity, strategies focused on balancing water extraction to sustain low flows in fluvial systems can benefit from attention to select watershed features. We draw attention to the finding that streams located in watersheds with high proportions of wetlands may require more stringent approaches to withdrawals to sustain these ecosystems during drought periods.
We then determined the minimum flow requirements at three locations (riffle zones) along the Beaver River, located in southern Rhode Island, using both the wetted perimeter method and the RIABF method. In order to determine stream flow at ungaged locations, runoff was modeled using the HEC-HMS rainfall/runoff model.
To assess biological conditions, we reviewed macroinvertebrate, fish and temperature data obtained within the watershed. Biological conditions of the Beaver River indicated that the Beaver River is a well-functioning stream habitat.
Minimum stream flow requirements using ABF and WP methods were investigated. Stream flows were found to be below the ABF value between six to 21% of the time and below the WP flow between 37 to 72% of the time.
Physical and biological sampling done in the watershed indicate the river is a well-functioning, river, comparable to pristine sites; however minimum flow criteria set by the wetted perimeter method suggest that the river is flowing below critical flow values over 50% of the time. Our results suggest that minimum flow values obtained from the wetted perimeter method for southern New England rivers should be approached with caution and should be compared to results obtained from other methods to determine the accuracy and applicability of the critical flows prior to using these values for any type of instream flow regulations.
We also assessed the effect of increased impervious cover for both conventional and LID-based urbanization on low flow metrics and flow depths in riffle habitats in a small, relatively undeveloped watershed located in southern Rhode Island. We employed a hydrologic model to simulate stream flow, base flow and storm flows under different land cover scenarios and then compared these results to the effects of direct stream withdrawals from agricultural irrigation.
We found baseflow to be negatively correlated to impervious area. On pervious surfaces, direct runoff is likely to be infrequent during the summer months, when most of the precipitation that falls is utilized for the soil moisture deficit. In contrast, connected impervious area (IA) will generate immediate runoff to streams from rainstorms that would have otherwise infiltrated the soil. During periods of excess precipitation, the falling limbs of those hydrographs generated prolonged periods of comparatively elevated flows.
Combining baseflow and storm flow shows that increased values of IA can generate higher flow values during the summer months during periods with excess precipitation. As IA increases through the different land use scenarios, storm related runoff increases immediately following precipitation events, causing higher stream flows. The small decreases in base flow input to the stream due to increased IA are negated by the impacts of the higher storm flows, causing summer stream flows to be higher under the developed land use scenarios than existing conditions. Changes to the channel depth of the riffles were also relatively minor.
During a year with median precipitation, the model predicted a lower frequency of low flows with both conventional development and with LID compared to the predictions for the limited development present in current conditions. Both conventional development and LID also display fewer low flow periods during a dry year, but the pattern reverses, with LID predicted to have slightly lower frequencies of low flows than the conventional development. Over the summer, storm runoff and the associated falling limb of the runoff hydrograph that results from connected impervious cover occurs with enough frequency to influence the low flow thresholds we use for metrics. During the dry year, rainfall occurrences were very infrequent and the higher baseflow associated with LID accounts for the slight increase in flows compared to the conventional development. Irrigation scenarios decreased both flows and depths. Changes in land use generally increase river flows while water withdrawals decrease river flows. The occurrence of low flows within the Beaver River was found to be relatively resilient to the extent of development and water withdrawals simulated by this study.
These analyses will help inform future water management decisions in watersheds with the diversity of land uses that occur in southern New England. xiv

Abstract
Rivers and river systems serve as conduits for nutrients and organisms, function as corridors for fish and wildlife passage, and provide resources for humans.
In order to preserve a spectrum of stream functionality, rivers must maintain seasonally adequate flows. For example, low flows can affect stream connectivity, restrict movement of aquatic organisms, concentrate prey into limited areas, purge invasive species from riparian corridors, and enable recruitment and evolution of floodplain plants. State agencies throughout the Northeast U.S. are considering policies linked to low-flow thresholds that sustain these ecosystem services. Within a given climatic region, water resource managers seeking to optimize stream ecosystem services need a clear understanding of the importance of land use, physical/climatic characteristics, and hydrography on different components of stream hydrographs. Within 33 Southern New England watersheds (average area 80 km 2 ) we assessed relationships between watershed variables and a set of low flow parameters: 1-, 7-and 30-day minimum flows. We used an information-theoretical approach to develop regression models to identify relationships between landscape attributes and parameters that describe different components of the flow regime.
The key variables identified by the Akaike Information Criteria (AIC) weighting factors as generating positive relationships with median annual minimum flow events included percent stratified drift (greater infiltration and storage), mean elevation (likely related to higher snowfall), drainage area and mean August precipitation. The

Introduction
Rivers and river systems serve as conduits for nutrients and organisms, corridors for fish and wildlife passage, and provide resources for humans; such as fresh water, food, and opportunities for recreation (Puth and Wilson, 2001). In order to preserve stream functionality, rivers must maintain seasonally adequate flows. Richter et al. (1997) defined a natural flow paradigm where "the full range of natural intra-and inter-annual variation of hydrological regimes and associated characteristics of timing, duration, frequency and rate of change are critical in sustaining the full native biodiversity and integrity of aquatic ecosystems." These characteristics affect the integrity of a stream through their effects on water quality, energy sources, physical habitat, and biotic interactions (Bunn and Arthington, 2002).
Although annual flow in a stream is largely controlled by the amount and timing of precipitation and evapotranspiration within its watershed, the amount of water at any given time (generally measured by stream flow) may also be influenced by watershed characteristics such as mean elevation, basin slope, hydrography, geology, soils, and land use. Landscape attributes can be classified as land use/land cover variables; physical/climatic variables, including features such as drainage area, geology, precipitation across the watershed, and slope; and hydrography which includes the extent of open water and wetlands within a watershed. Land use changes occur on a rapid temporal scale, are largely driven by human activity and management actions, and can have major effects on stream hydrology. For example, intensive urbanization can increase runoff, cause larger storm peaks, and reduce low flows (Leopold, 1968;Meyer et al., 2009). Agriculture can exert a variety of effects on flow regimes based on cropping systems and management (Poff et al. 1997;Meybeck, 2003). In contrast, physical/climatic variables may have large impacts on stream flow but are relatively resistant to change due to human activities. Allan (2004) found that when anthropogenic and physical features covary and are used for evaluation, the influence ascribed to land use can be overestimated. Arthington et al. (2006) argue that management of flow regimes can benefit from analyses that focus on classifying river into management units that have comparable climatic and physiography attributes. Within a given unit, water resource managers can then engage in analyses that further the understanding of the effects of land use, physical/climatic characteristics, and hydrography on different components of the flow regime and related ecological conditions. Stream flow is often statistically analyzed to characterize the magnitude and probability of various components of the flow regime such as low flows, high flows, and average or median discharges (Richter et al., 1996;Allan, 1995;. In this study we focused on the low flow portion of the annual flow regime. Within a set of watersheds with similar climate and physiography, low flows can affect stream connectivity, restrict movement of aquatic organisms, concentrate prey into limited areas, purge invasive species from riparian corridors, and enable recruitment and evolution of floodplain plants . In particular, extreme events, such as those caused by drought, can be particularly important to the vitality of biotic communities (Naiman et al., 2008). While a number of approaches have been recommended to establish sustainable flow regimes (Arthington et al., 2006;James et al. 2012), state agencies throughout the Northeast U.S. are considering policies linked to low-flow thresholds that sustain these ecosystem services (CTDEEP, 2011;Richardson, 2005).
We used model selection procedures to assess relationships between watershed basin characteristics and the following parameters of the flow regime: We used an information theoretical approach (Burnham and Anderson, 2002) to develop regression models to identify relationships between landscape attributes and flow components, with the flow components as the dependent variable and the watershed characteristics as the explanatory or independent variables. Exploring the strength of relationships between specific watershed variables and flow regime can contribute to the development of land use and water extraction policies that sustain fluvial ecosystems.

Site Selection
Study sites were selected using NHDPlus StreamGageEvent data (www.horizon-systems.com) and from the United States Geological Survey (USGS) National Water Information System (NWIS) Real Time Water Data website (http://waterdata.usgs.gov/nwis/rt). Site selection focused on watersheds that were similar in size (area), physiography, dates of continuous flow data, and within a single ecoregion. Ecoregions combine biotic and abiotic phenomena that are expected to influence ecological integrity and environmental quality (Olson, et al., 2001). There are 766 stream gages located in New England in the NHDPlus StreamGageEvent data set. We eliminated the 279 gages located to the west of the Connecticut River to restrict the physiography to settings dominated by glacial deposits. The surficial geologic materials in eastern New England are primarily glacial. These unconsolidated glacial deposits vary in thickness across the region. The most common glacial deposit in New England is glacial till, a well-graded material with grain sizes ranging from clay to large boulders. Glacial stratified deposits occur in the coastal or valley areas and consist of layers of sand and gravel. Till deposits have a much lower permeability than the coarse-grained stratified deposits and often have restrictive layers that generate seasonal wetness. We further eliminated 92 gages that had drainage areas under 5 mi 2 (12.95 km 2 ) and 189 gages that had drainage areas over 75 mi 2 (194.25 km 2 ). To establish a common set of long-term flow 8 patterns the number of potential gages was reduced by 165 as we sought watersheds that had the same 30 year period of daily stream continuous flow gaging data, from January 1, 1980to December 31, 2009, by the USGS. Lastly, sites were selected from watersheds within the Northeast Coastal Ecoregion (Omernik, 1995). Using these criteria, 33 sites (Table 1, Figure 1) were selected for analysis. Watersheds for the selected stream gages were delineated using NHDPlus Basin Delineator software and were visually checked for accuracy using USGS 7.5 minute quadrangles (1:24,000) and associated NHDPlus catchment areas (1:100,000).

Watershed Variables
The United States Geologic Survey has tailored sets of state-specific regression equations that can be used to determine components of the flow regime  and Friesz, 2000). For this study, we developed regression models using the collection of all variables that were used in the final regression equations for those states within the study area -Massachusetts, Rhode Island, New Hampshire (Olson, 2009), and Connecticut ( Table 2). The variables were elevation difference (used to represent the variables of mean basin slope and mean channel slope), percent stratified drift, drainage area, mean April precipitation, mean August precipitation, mean basin elevation, and drainage density (the ratio of the total stream length to the watershed area, referred to as stream density in the RI models).
Channel slope (used in New Hampshire) and mean basin slope (used in Massachusetts) were not included in the regression models due to their high correlation (p<0.001) with elevation difference. The regional factor used in Massachusetts was also not included as it could not be used across our study region.
In addition to these physical and climatic variables, we also included land use variables and one additional hydrographic variable (extent of open water) in the models. Additional land use categories were aggregated into five groups -developed, impervious, forest, cultivated, and pasture/hay -and the percent cover for each category was tabulated and then normalized by area for each watershed.
The extent of open water was also calculated and normalized by basin area for each and percent forested area within a watershed was found to be highly significant (p<0.001) and negatively correlated (r=-0.92), so percent forested area was not included in the final regression models (Table 3). Additionally, the correlation between percent impervious cover in the watersheds and percent total developed area was found to be highly significant (p<0.001) and positively correlated (r=0.94).

Variables used in Regression Equations
New Hampshire (Olson, 2009)

Flow Components
Flow components (

Statistical Methods
Multiple linear regression models between the flow components and landscape variables were developed using PROC REG in SAS 9.1 (SAS Institute 2002. Multiple regression techniques provide mathematical equations to describe the empirical relationship between a dependent or response variable, such as one day minimum flow, and two or more independent, or explanatory, variables, such as Instead of using a null hypothesis method, we used a model selection approach where many competing hypotheses are tested to identify a set of possible models. Those models are then compared by evaluating relative support for each model as well as each included variable. Model selection approaches are becoming more widespread among ecologists and are ideal for making inferences from observation data (Johnson and Omland, 2004). Model selection criteria, such as AIC, account for both model complexity and fit. These approaches do not simply compare models by calculating a measure of fit, such as R 2 , and then maximizing that value.
They also recognize that parsimony is a bias-variance tradeoff. Using too few parameters in a model can underfit the model and may fail to identify all important variables, while conversely, using too many variables or overfitting a model may lead to spurious correlations. Because R 2 will always increase with the addition of more variables, simply maximizing R 2 will always favor fuller, more variable-rich models.
This approach, however, ignores problems with overfitting and parsimony. AIC is a measure of goodness of fit where lower values indicate that less information is lost.
Regression models were developed from all possible combinations of available independent variables. Once the full suite of regression analyses were run, rather than selecting a single "best" predictive model, our objective was to use the

Results
In comparing the components of the low flow regime across the top five models (Table 5), we found all models were significant at the p<0.001 significance level. Selected physical/climatic variables and the percentage wetlands were the most important variables (Table 6)

Minimum Flows
Prolonged low flow periods reduce plant cover and diminish plant diversity (Taylor, 1982). Low flows help eliminate invasive species from the floodplain by reducing soil moisture and nutrients, restrict movement of aquatic organisms, and concentrate prey into smaller available habitats (Mathews and Richter, 2007). Low flows are not expected to represent storm-related runoff but rather reflect summer base flow, a product of inputs from groundwater or wetland storage and potential evapotranspiration losses from phreatophytes associated with wetlands connected to the stream network.
The landscape variables assessed in this study can be grouped into two general categories -anthropogenic or natural. The anthropogenic variables consist of watershed characteristics that may be altered by human activities. These attributes include watershed land cover, such as developed area and cultivated area, as well as drainage density, which can be altered by increased development. Elevation difference, precipitation values, stratified drift, drainage area, and mean elevation are natural variables that are not easily altered by human activities. Open water and wetlands were historically subject to change due to dams or artificial drainage, but the rate of change in these variables has diminished markedly due to regulations governing wetland loss, flooding associated with new reservoirs and the complexities associated with dam construction or removal.
For low flow magnitudes, natural features, such as percent stratified drift, drainage area, mean August precipitation, mean elevation, and percent wetlands, were found to be most important predictors for the more extreme low flow events (the 1, 7 and 30 day minimum flows) with only one anthropogenic variable, the percent of developed lands, found to be important but of lesser consequence.
Wetlands and developed lands were not found to be important for explaining the variability in the low pulse threshold, the flow that represents the lowest 25 th percentile flow or the lowest flows over roughly 90 days. Mean April precipitation, although found to be important in the New Hampshire regression equations, was not found to be an important explanatory variable for any descriptors of low flows examined in this study, likely due to the fact that the minimum flow for the study area streams was usually found to occur during September, thus August precipitation values would be expected to exert more influence on the minimum flows.
Stratified drift, which comprises most of the deep productive aquifers within the study region, was positively correlated to minimum flow magnitudes. There are many substantial groundwater withdrawals for cropland irrigation as well as municipal supplies occurring within stratified drift deposits on some watersheds within the study area. However, even with these withdrawals, the proportion of stratified drift was positively correlated with higher summer flows. Stratified drift has much higher infiltration and permeability rates than glacial till and is often located in valleys and coastal regions. Stratified drift deposits can store considerable volumes of recharge during wet periods through water table rise. In the summer months, groundwater flow from the stratified drift contributes to stream base flow, resulting in higher low flows than streams located in areas of till (Wandle and Randall, 1994).
Mean watershed elevation was found to be positively correlated to minimum flow magnitudes, indicating higher elevation watersheds are likely to have higher values of minimum stream flows. Mean monthly precipitation, as well as snow depth increases with elevation in the study region. Dingman (1981) has shown that both floods and low flows increase with mean basin elevation and suggested that the effects of elevation are so strong in portions of New England that it can be used as the only dependent variable in estimating many stream flow statistics.
The percent of wetlands within the watershed was negatively correlated to the more extreme minimum flows that are associated with drought conditions. As the amount of wetlands increased in a watershed, the minimum flows for the 1-, 7and 30-minimums decreased. Evidence from many studies (Bullock et al. 2003) suggests that wetlands generate substantially more summer evapotranspiration than other land uses, such as upland forests or pasture during dry periods. During dry periods, the elevated water tables and soil wetness of wetlands promote conditions that permit these areas to meet evaporative demand, while upland ecosystems undergo a number of changes that constrict evapotranspiration (e.g., stomatal closure, declines in soil upwelling due to low levels of unsaturated hydraulic conductivity). Kellogg et al. (2008) found that riparian wetland forests in southern New England intercepted virtually all groundwater flow during drought conditions, essentially starving the rivers of baseflow during those periods. For the low pulse threshold, however, the extended time frame may encompass time periods that are not entirely within the period of highest evaporation stress, thus reducing the relative importance of wetlands on seasonal low flows under these conditions.
In many locales, watershed management is focused on reducing the extent of impervious cover to protect or restore stream health (Finkenbine et al., 2007;Snyder et al., 2005). Our finding that the extent of development in the watershed (which was strongly related to % impervious cover) was not as important as other factors for explaining the variability of flow regimes parameters is in line with the features, scale, and focus of our study. A meta-analysis conducted by Schueler et al. (2007) found few studies which researched the effects of impervious area on hydrologic factors, and those studies were either contradictory or ambiguous. Specifically, they found that an inverse relationship between impervious cover and base flow to streams was not always present. Most papers that confirmed hydrologic effects related to impervious cover mainly studied small watersheds (between 5 to 50 km 2 ).
Contradictory studies sampled watersheds that were generally larger (between 75 and 100 km 2 ). The average watershed size of our study was 87 km 2 generally larger than watersheds where other studies found strong relationships between impervious cover and hydrologic effects. In addition, 29 out of the 33 watersheds in our study had % impervious cover less than 10%, the frequently used threshold for degradation in stream health. The extent of impervious cover is not unexpected for watersheds of this size in southern New England, where outside of the Boston and New York City metro regions, the land use patterns typically encompass extensive amounts of open space and lower density suburbia.

Management Implications
Our results were obtained from watersheds with a number of similar characteristics (e.g., similar size, ecoregion, and geomorphology). Yet even with these similarities, important differences emerged suggesting that management of water extraction to sustain low flows can benefit from attention to select watershed features. Based on our analyses of regression models, minimum flow components of the flow regime are more likely to be governed by natural features of the watershed.
For example watersheds with high proportions of stratified drift were less likely to have extremely low minimum flow levels, while in watersheds where wetlands comprise a relatively high proportion of the watershed minimum flows tend to decrease. Management insights that follow from these findings may imply that watersheds with high proportion of stratified drift may be less susceptible to summer withdrawals for irrigation or municipal uses. In contrast, more stringent requirements on water extraction during drought conditions may be warranted in watersheds with high proportions of wetlands to avoid pushing those fluvial systems into extreme drought stress. Although the importance of development on low flows during drought were weaker than natural watershed features, we expect that the large area and low proportion of development within the study watersheds may mask the effects that can occur in smaller, more developed watersheds. Our analyses did not include the proximity of wetlands to fluvial systems; thus, we were not able to distinguish the role of isolated wetlands (Leibowitz, 2003) versus riparian wetlands and floodplains on the low flow components of the flow regime. Effects of wetlands proximity to fluvial systems is a critical question for future research, particularly given recent interests in the courts related to regulatory questions on the extent of connections between wetlands and "navigable" waters (Nadeau and Rains, 2007).
Our study suggests that we need to recognize that a variety of watershed factors can that influence low flows. The increasing availability of geospatial data can assist in future, management decisions regarding environmental flow recommendations that will, ultimately, support healthy river ecosystems and communities.  We determined the minimum flow requirements at three locations (riffle zones) along the Beaver River, located in southern Rhode Island, using both the wetted perimeter method and the RI-ABF method. Field surveys of stream characteristics at each of the three riffle zones were conducted for evaluating the wetting perimeter method. In order to determine stream flow at ungaged locations, runoff was modeled using the HEC-HMS rainfall/runoff model.
To assess biological conditions, we reviewed macroinvertebrate, fish, and temperature data at locations within the watershed. Biological condition of the Beaver River indicated that the condition of the Beaver River, in terms of macroinvertebrate taxa, is equal to or better than the reference site. Fish sampling locations showed an abundance of fluvial specialists and fluvial dependent species indicating for fish species, the Beaver River is a well-functioning stream habitat.
Temperature readings in the Beaver River show stream temperatures never exceeded 22 0 C, indicating suitable temperature for the maintenance of a brook trout population.
Minimum stream flow requirements using RI-ABF and WP methods were investigated during a wet year and a dry year for the modeling time interval. During the wet year, stream flows were below the ABF value between 6% to 12% of the time and below the WP flow 37% to 46% of the time. During the dry year, stream flows were below the RI-ABF value 18% to 21% of the time but below the WP flow 46% to 72% of the time.
Based on physical and biological sampling done in the watershed, the river is, comparable to pristine sites; however minimum flow criteria set by the WP method suggest that the river is flowing below critical flow values over 50% of the time.
Critical flow values as determined by the RI-ABF method suggest that the river flows below critical flow values under 10% of the time. Our results suggest that minimum flow values obtained from the wetted perimeter method for southern New England rivers should be approached with caution.

Introduction
Streamflow has been called the master variable in a river because it affects habitat diversity and availability through its impact on stream geomorphology, channel substrate, water depth, velocity, and other factors that, in turn, affect habitat quality, such as water temperature and water quality (Poff et al., 1997;Wilding and Poff, 2008;Poff and Zimmerman, 2009). Flow also influences habitat variables, such as the shape and size of channels, as well as distribution of riffle and pool habitats. Physical conditions within a habitat mediate levels of food resources available (Rabeni and Minshall, 1977) and may constrain the roles of predator competition (Peckarsky and Dodson, 1980). Minimum flow requirements for rivers aim to provide protection to habitat. Currently, there exists a wide variety of methods that result in conflicting minimum flow standards, largely due to differing environmental goals and levels of protection they aim to achieve. There are currently over 200 methods for determining low flow instream flow requirements (Annear and Conder, 1984), ranging from simple methods that use historical flow data to field-based reconnaissance for critical aspects of stream morphometry.
More complex simulation models can link flow, velocity, and stream depth to habitat requirements.
Standard setting methods set limits to define a threshold flow regime or minimum flows, below which water cannot be diverted. Examples of standard setting methods include the United States Fish and Wildlife Service Aquatic Base Flow method (USFWS ABF). To define minimum flow requirements, the USFWS used historical flow records for New England to describe stream flow conditions that will sustain and perpetuate indigenous aquatic fauna. The USFWS ABF method assumed that the most critical flows occur in August when the metabolic stress to aquatic organisms is at its highest due to high water temperatures, diminished living space, low dissolved oxygen, and low or diminished food supply. Where adequate records exist, the USFWS recommends using the median of the monthly means of August flows as a minimum threshold to sustain benthic organisms.
Field-based hydraulic methods use the hydraulic geometry of stream channels to estimate low-flow discharge thresholds. The hydraulic geometry is based on surveyed cross-sections, from which parameters such as width, depth, velocity, and wetted perimeter are determined. Hydraulic models can predict water depth and velocity within a specific habitat (i.e. riffles) (Gippel and Stewardson, 1998) or throughout a reach (i.e PHABSIM) (Milhous et al., 1984). These are then compared with habitat suitability criteria to determine the area of suitable habitat for the target aquatic species. When this is done for a range of flows, it is possible to see how the area of suitable habitat changes with flow.
The WP method (Gippel and Stewardson, 1998), one type of hydraulic method, uses stream cross-sections -typically at riffle habitat -to determine critical flow values expected to maintain flow based on the wetted perimeter of the channel.
The wetted perimeter is the distance along the bottom and sides of a channel crosssection in contact with the water. The wetted perimeter method assumes that the fish carrying capacity of a stream is related to food production and that food production is related to the amount of wetted perimeter in riffle sections. As discharges decrease, riffle habitats are often the first locations to be exposed or go dry. For a specific cross-section, the flow rates that cover a reasonable proportion of bed area of riffles with flowing water are determined in order to provide adequate minimum flows for benthic macroinvertebrates communities and allow for fish passage. The method uses plots of wetted perimeter vs. discharge to identify a break or inflection point. Critical discharge corresponds to this inflection point (Gippel and Stewardson, 1998). At this inflection point, food production is assumed to approach optimum levels (Gippel and Stewardson, 1998). Below the inflection point, aquatic habitat will decline and thus support lower populations of benthic species. Above the upper inflection point, the flow regime is expected sustain thriving benthic aquatic populations which then contribute to a robust food web.
We compared the two approaches for estimating the minimum flow requirements for a third order stream network in Rhode Island. Rosenblatt et al. (2001) found that first and second order streams compose 70% of total stream length in RI based on digitized hydrographic data from the 1:24,000 scale USGS 7.5 minute topographic quadrangle maps. In addition, headwater streams have smaller average flows , resulting in low flow stress more frequently occurring in headwater streams (Richardson and Danehy, 2007). Third order watersheds, therefore, may have many ramifications for low flow management.
We used the wetted perimeter method and the Rhode Island Aquatic Base

Study Watershed
The study area is the 32 km 2 Beaver River watershed, a third-order stream located in southern Rhode Island (Figure 1).  The USGS Gage 01117468 is located at the intersection of Kingstown Road with the Beaver River.

Field Based Measurements
Riffles along the Beaver River were field located during the summer of 2009.
USGS standard methodology (Rantz, 1982) was used to measure discharge at the riffle locations at each subbasin outlet (Buchanan and Somers, 1969). At each riffle location, the stream channel cross section was divided into one-foot (0.3048 m) subsections. In each subsection, the depth at the center of the subsection was measured with a surveyor's rod (Figure 2), and the area was estimated by multiplying the depth times the width (0.3048 m). Water velocity was determined using a Global Water Flow Probe FP101 current meter. Stream bed elevations were determined using a CST/berger automatic level (Figure 3). Stream discharge was then calculated using the mid-section method:

Q=∑
where the X i are the distances to successive measurement points along the transect, where stream velocity (U i ) and water depth (Y i ) are measured, starting with X 1 being the initial point on one bank and X n being the final measuring point on the opposite bank.
A slope-area method was used to determine Mannings roughness coefficient (n) for each riffle location by rearranging Mannings Equation to solve for n: Where A is the cross sectional area of the stream measured from survey data, n is Mannings n which is an index of the roughness of the stream bed, R is the hydraulic 50 radius which is the ratio of the cross section area of the stream to its wetted perimeter (which is the cross-sectional distance along the stream bed and banks that is in contact with the water), and S is the change in elevation of the stream over a

Wetted Perimeter Breakpoint Analysis
The channel in the Beaver River is roughly rectangular throughout its length. Gippel and Stewardson (1998) found that for a hypothetical rectangular crosssection, the relationship between wetted perimeter and discharge was logarithmic and had the general form: Where WP is wetted perimeter, Q is the discharge and a is a constant. Gippel and Stewardson (1998)

Aquatic Base Flow
The United States Fish and Wildlife Service (USFWS) used historical flow gage records for New England from 48 unregulated rivers to prescribe stream flow conditions that will sustain and perpetuate indigenous aquatic fauna (Lang, 1999).
This policy has been widely used in New England. Richardson (2005)

Assessment of Benthic Macroinvertebrate Condition
To assess biological conditions, we used benthic macroinvertebrate data that were collected each summer within the Beaver River over a 11 year period (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)) (Gould, 1999;Pomeroy, 2000;Da Silva, 2002). A single evaluation score of biological condition was derived by scoring a set of metrics according to the type, abundance and diversity of taxa found at each site and ranking each sample in comparison to the metrics of a reference sample station taken during the same year from the Wood River in Richmond, RI ( Figure 1). The Wood River is mostly surrounded by the Arcadia Management Area, and thus receives minimal human impacts. Scoring criteria for each metric were derived from Plafkin et al. (1989) and modified according to specifications for the region (Jessup, 2000). A maximum score of 50 represents excellent biological condition.

HEC-HMS Model
In order to determine stream flow at ungaged riffle locations, runoff was  (Feldman, 2000).

Model calibration and validation
Initial model parameters were estimated using the guidelines given in the HEC-HMS Technical Reference Manual (Feldman, 2000). The automatic parameter optimization tools, available in the HEC-HMS model, were used to find the optimum set of parameters (groundwater storage coefficients for the linear reservoirs) for each sub-basin.
Nash-Sutcliffe efficiency (NSE) ) was used to determine "goodness of fit" of the model. NSE ranges from -∞ to 1.0 with values between 0 and 1 being acceptable levels of performance .
Root mean square error (RMSE) is also a commonly used error index statistics (Singh et al., 2005). Singh et al. (2005)  combines both an error index and the additional information recommended by . RSR varies from the optimal value of 0, which indicates zero RMSE or residual variation -perfect model simulation -to large positive values.

Flow Analysis
Stream discharge data is a continuous variable that is often summarized by frequency distributions. The values for the streamflow were first ranked from smallest to largest, and then plotted using a Weibull distribution (Weibull, 1951) where: Where F(x) is the non-exceedance probability, i is the rank of the flow observation  Year

Yearly Precip
Average Summer Temperature

Model Calibration and Validation
Statistical indices of NSE and RSR, for both the calibration and the verification periods, were calculated using the results of daily time steps (Table 2) (Moriasi, 2007).

Breakpoint Analysis
For each riffle section, normalized discharge vs. normalized wetted perimeter, calculated from the field cross-section data, was plotted, and the breakpoint of the resulting curve was determined ( Figure 6, Table 3

Macroinvertebrates
Indicator species are taxa that are highly sensitive to pollution or anthropogenic disturbance and are the first to disappear with disturbance or pollution. Biological condition based on macroinvertebrate taxa from riffle sites was compared between the Beaver River and a pristine river site on the Wood River that is used as a reference station for all Rhode Island benthic macroinvertebrate assessments. Over the 10 years of sampling, the Beaver River had a median score of 36, with an interquartile range of 27 to 38, while the Wood River had a median score of 24, with an inter-quartile range between 20 and 36 ( Figure 7) (Da Silva, 2003).
These values indicate that the biological condition of the Beaver River is comparable to a pristine river in terms of macroinvertebrate taxa and is likely to be very capable of sustaining high levels of biological integrity. generalists, such as eel, redfin pickerel, and brown trout, which are found in both lentic and lotic systems (Galat et al., 2005) and were found in lower numbers at both sites.
To provide additional context on the status of the Beaver River fish communities, the results were compared to the Ipswich River ( Figure 9) (Armstrong et al., 2003). The Ipswich target fish community is used to show a healthy fish community in a small coastal river (Armstrong et al., 2003). Fish in the Ipswich River have a population of 49% fluvial specialists, 19% fluvial dependents and 32% macrohabitat generalists, suggesting that the Beaver River, with a substantially higher proportion of fluvial specialists, sustains a healthy fish community relative to its location and physiography.

Temperature Data
Water temperature is a key factor affecting fish (Fry 1971). Temperature regimes influence such life cycle stages as migration, egg maturation, spawning, incubation success and growth as well as resistance to disease, parasites and pollutants (Armour, 1991). In warmer streams, trout populations have been found to be almost nonexistent (Barton et al., 2002). Brook trout, the native trout for New England and Vermont's official cold-water fish, are found primarily in streams with maximum weekly average water temperatures less than 22 0 C ( Barton et al.., 2002).
Sustained water temperatures over 25.3 0 C are considered to be lethal for brook trout (Mullen, 1958).
The Beaver River maintained stream temperatures conducive to brook trout, with daily maxima temperatures at or below 22 0 C and sustained average temperatures of 17 0 C for the two site monitored during 2004 ( Figure 10). These temperatures are substantially lower than the average summer air temperature of 25.8 0 C.
For lower order streams, phreatic groundwater inputs and riparian shade have the highest influence on sustaining cool summer water temperature (Poole and Berman, 2001). Virtually the entire stream network within the Beaver River is shaded by forested riparian zones, and the extensive forest cover and minimal extent of impervious cover enhances the potential for excess rainfall to enter the stream as baseflow, rather than as storm-generated overland flows.

Minimum Flows
Minimum stream flow requirements using ABF and WP methods were investigated during a wet year (1983) and a dry year (1993) for the modeling time interval (Figure 11). We focused our attention on the summer months, when flow is at its lowest, since our goal was to examine the efficacy of these methods for establishing minimum flow requirements. Marked differences between the two methods were observed in the summer non-exceedance flow probabilities predicted for the riffles targeted in our study. Summer flows were usually below the minimum thresholds determined by Wetted Perimeter method for both wet and dry years. At the riffles located in the upper and middle subbasins, the flows met or exceeded the minimum WP thresholds less than 20% during the summer for both wet and dry year (  Given the many indications that the Beaver River sustains a healthy coldwater fishery, we suggest that the current flow regime is not generating impairment.
The thresholds set by the WP method resulted in high frequencies of summer flows that failed to meet those thresholds, thus it appears that this method may be far too conservative. In contrast, summer flows in the Beaver River routinely meet the minimum thresholds determined from the simpler RI ABF, which may be more suited as a low flow threshold method that can reflect the required minimum flow conditions of the river.
In addition to minimum flows predicted by the two methods, we also examined the minimum flow depths at riffle habitats that would be expected for each method. These findings were obtain by plotting ABF flow values and WP flow values, along with the cross-section data to determine maximum depth of flow under both criteria ( Figure 12, Table 5). USDA (1975) suggests that a stream depth of at least 0.12 m is required for trout passage. Except for the upper subbasin, the threshold flows predicted by the ABF method would provide adequate depth for trout passage at the riffle sites (e.g., the riffles located on the outlet and middle watersheds). The WP method will provide at least twice the minimum depth required for trout passage in the lower two watersheds, suggesting that the WP method produces extremely conservative estimates of minimum flow depths for this river system. which defines cold water fisheries as waters in which naturally occurring water quality and/or habitat allow the maintenance of an indigenous coldwater fish populations (RIDEM, 2006).
Macroinvertebrate data obtained from a riffle habitat on the Beaver River also indicate that the Beaver River maintains high biotic integrity -comparable to the results obtained at the pristine reference site in the Wood River, which is surrounded by the Arcadia Management area and has almost no anthropogenic alterations in its watershed.
Despite these indicators that the river is currently in excellent condition, We found baseflow to be negatively correlated to impervious area (IA). On pervious surfaces, direct runoff is likely to be infrequent during the summer months, when most of the precipitation that falls is utilized for the soil moisture deficit. In contrast, connected IA will generate immediate runoff to streams from rainstorms that would have otherwise infiltrated the soil. During periods of excess precipitation, the falling limbs of those hydrographs generated prolonged periods of comparatively elevated flows.
Combining baseflow and storm flow showed increased values of IA can generate higher flow values during the summer months during periods with excess precipitation. The small decreases in base flow input to the stream due to increased IA are negated by the impacts of the higher storm flows, causing summer stream flows to be higher under the developed land use scenarios than existing conditions.
Changes to the channel depth of the riffles were relatively minor.
During a year with median precipitation, the model predicted a lower frequency of low flows with both conventional development and with LID compared to the predictions for the limited development present in current conditions. Over the summer, storm runoff and the associated falling limb of the runoff hydrograph that results from connected impervious cover occurs with enough frequency to influence the low flow thresholds we use for metrics. During the dry year, rainfall occurrences were very infrequent and the higher baseflow associated with LID accounts for the slight increase in flows compared to the conventional development.
Irrigation scenarios decreased both flows and depths.
The occurrence of low flows within the Beaver River was found to be relatively resilient to the extent of development and water withdrawals simulated by this study. The analyses will help inform future water management decisions in watersheds with the diversity of land uses that occur in southern New England.

Introduction
Riverine systems serve as conduits for nutrients and organisms, corridors for fish and wildlife passage, and provide resources for humans; such as fresh water, food, and opportunities for recreation (Puth and Wilson, 2001). In order to preserve stream functionality, rivers must maintain seasonally adequate flows (Richter et al. 1998). Characteristics, such as the duration and frequency of flow, affect the integrity of a stream through their effects on water quality, energy sources, physical habitat and biotic interactions (Bunn and Arthington, 2002). Although flow in a stream is controlled by the amount and timing of precipitation and evapotranspiration, the amount of streamflow at any given time is also influenced by watershed characteristics, such as elevation, hydrography, drainage area, water abstractions for irrigation and domestic uses. (Richter et al., 1998;Allan, 1995;Piao et al., 2007). Land use changes such as urbanization can have major effects on stream hydrology, generating changes in both low flow and flood conditions (Brabec, 2002;Walsh et al., 2009). Modifications of the land surface due to urbanization alters natural stream hydrographs by increasing flood peaks, decreasing time to peak flows, and causing higher runoff velocities (Paul and Meyer, 2008). Urbanization can also generate higher frequencies and durations of low flow conditions (Leopold, 1968;Meyer, 2005). Direct water withdrawals for agricultural use from streams and rivers has become a common occurrence in Rhode Island since the 1980's when center pivot and linear move irrigation systems were introduced by turf growers (Gold et al., 1988).
Increasingly, states in the Northeast are developing management strategies to protect riverine ecosystems against stresses imposed by low flow conditions (CT DEEP, 2011;RI WRB, 2011). Low flows can affect stream connectivity, restrict movement of aquatic organisms, concentrate prey into limited areas, purge invasive species from riparian corridors, and enable recruitment and evolution of floodplain plants (Cushman 1985;Gehrke et al., 2006;Scheidegger and Bain, 1995;Mathews and Richter, 2007;Humphries and Baldwin. 2003). Low flows are expected to reflect summer baseflow and can be reduced by evapotranspiration of riparian wetlands (i.e., phreatophytes) and withdrawals for irrigation and other uses (Winter, 2007).
Low impact development (LID) has emerged as a strategy to reduce the hydrologic impacts of urbanization on aquatic ecosystems. It combines site planning and design processes with runoff reduction and treatment practices (Dietz, 2007;Coffman and France, 2002;Davis et al., 2009). LID is intended to mimic the natural devices (Booth and Jackson 1997). Effective IA, or connected IA, is the proportion of IA that is directly connected to the stream network. There have been many studies that show connected IA affects changes in runoff much more than total IA (Brabec et al. 2002).
Stream flow is often statistically analyzed to characterize the magnitude and probability of various components of the flow regime, such as low flows, high flows, and average or median discharges (Richter et al., 1998;Allan, 1995;. In this study, we assessed the effect of increased impervious cover for both conventional and LID-based urbanization on statistical metrics related to low flow in the Beaver River, a small, relatively undeveloped watershed located in southern Rhode Island. We also evaluated the flow depths associated with these low flow metrics at specific riffle habitats where abnormally low flows is expected to degrade aquatic habitat. In order to assess the flow depths, we performed field surveys of riffle sections at three locations along the Beaver River to determine cross-section morphometry. We employed a hydrologic model using HEC-HMS (USACE, Hydrologic Engineering Center, Davis, CA) to simulate stream flow, base flow and storm flows under different land cover scenarios over a 26-year period. We then compared these results to the effects of direct stream withdrawals from agricultural irrigation. The analyses were undertaken to inform future water management decisions in watersheds with the diversity of land uses that occur in southern New England.

Study Watershed
The study area is the 23 km 2 Beaver River watershed, located in southern Rhode Island (Figure 1). This watershed is a sub-watershed of the Pawcatuck River, which drains into the ocean at Little Narragansett Bay. Current land use in this rural watershed is approximately 82% forested, 9% agricultural, 6% residential housing,

Stage-Discharge Relationships at Riffle Cross-Sections
At each riffle location that corresponded to the outlet of each subbasin a stage-discharge relationships was developed to relate simulated discharge from the various scenarios to minimum flows and depths at riffle habitats. To generate the stage-discharge relationships, the stream channel cross section was divided into onefoot (0.3048 m) subsections. In each subsection, the depth at the center of the subsection was measured with a surveyor's rod, and the area was estimated by multiplying the depth times the width (0.3048 m). Water velocity was determined using a Global Water Flow Probe FP101 current meter. Stream bed elevations were determined using a CST/berger automatic level. Stream discharge was then calculated using the mid-section method:

Q=∑
where the X i are the distances to successive measurement points along the transect, where stream velocity (U i ) and water depth (Y i ) are measured, starting with X 1 being the initial point on one bank and X n being the final measuring point on the opposite bank.
A slope-area method was used to determine Manning's roughness coefficient (n) for each riffle location by rearranging Manning's Equation to solve for n: Where A is the cross sectional area of the stream measured from survey data, n is Manning's n, which is an index of the roughness of the stream bed, R is the hydraulic radius, which is the ratio of the cross section area of the stream to its wetted perimeter (i.e., the cross-sectional distance along the stream bed and banks that is in contact with the water), and S is the change in elevation of the stream over a specified distance. Stream bed elevations and water surface elevations for points about 7m upstream and downstream of the riffle section were measured.

Model Selection
Criteria for selecting a basin scale model for this study was ease of use, compatibility of model parameters with available site-specific data, ease of calibration, model availability, and lastly, whether the model is commonly used for hydrologic studies. An integrated, physically based, distributed model (MIKE SHE; DHI, www.dhisoftware.com) was given extensive attention, since it is intended to simulate most major hydrological processes of water movement, including canopy and land surface interception after precipitation, snowmelt, evapotranspiration, overland flow, channel flow, unsaturated subsurface flow, and saturated ground water flow, including exchanges between surface water and ground water. However, the MIKE SHE modeling system required extensive parameterization with highresolution data and a large number of parameters, such as detailed soil and vegetation attributes. For the Beaver River watershed, a number of key model parameters are either not available or not available at the required scales, negating the value of many of the process-oriented, distributed aspects of the model (Beven, 1989;Jakeman and Hornberger, 1993). In addition substantial complications with model calibration was encountered when default values for required parameters were estimated or used.

HEC-HMS Model
The Hydrologic Engineering Centers Hydrologic Modeling System (HEC-HMS) model, developed by the Army Corps of Engineers was then selected for use. HEC-HMS is a lumped parameter model that incorporates the spatial pattern of development by subdividing the watershed into areas that are approximately homogeneous in land use, soil type, and slope. The HEC HMS model has been used for a variety of different hydrological studies, such as studying the effects of urbanization on runoff (Hejazi and Markus, 2009;Du et al., 2012) and flood modeling (Harris, 2007;Amengual et al., 2007) Runoff was modeled using the HEC-HMS rainfall/runoff model to simulate

Runoff volume
Daily runoff volume was computed by the deficit and constant rate loss model. The model simulates connected impervious areas by assuming that all rainfall onto "connected" impervious surfaces results in direct surface runoff to the stream.
Connected impervious surfaces, also known as effective impervious cover (Brabec et al., 2002) include only those areas that drain directly into a storm conveyance system that discharges to surface water (Brabec et al., 2002).
For pervious areas, the deficit constant loss method employs a quasicontinuous model of precipitation loss that uses a single soil layer to account for daily changes in moisture content. This method has been widely validated in many studies, it is easy to use and is parsimonious, requiring only a few input parameters .
The deficit constant loss method for pervious surfaces employs a daily soil water balance to assess the depth of water storage capacity, known as the deficit field.
Infiltration represents the input to the daily soil water balance. Evapotranspiration and soil percolation to the groundwater are the outputs. Rainfall onto pervious surfaces first fills the initial soil deficit until the maximum storage depth is reached at which point runoff can occur.
The initial daily soil deficit at the beginning of the modeling simulation indicates the amount of water that is required to saturate the soil to the maximum storage and reflects the topography, land use, hydrologic soil group, type, infiltration capacity and antecedent moisture condition. This combination of interception, the precipitation required to fill the soil water deficit, and depression storage are considered watershed losses and is also termed the initial loss (I a ). The potential evapotranspiration computed by the meteorological model of HEC-HMS is used to dry out the soil layer between precipitation events. Evapotranspiration was based on monthly average values for Rhode Island (Farnsworth and Thompson, 1982). The maximum potential rate of precipitation loss due to infiltration, referred to as the constant loss rate (f c ) was assumed to be constant throughout an event. The loss rate is the long-term infiltration capacity of the soil. Skaggs and Khaleel (1982) published estimates for f c based on hydrologic soil types. Both the fc and I a values in the validated model were determined by calibration (Feldman, 2000). Precipitation excess (pe t ) is obtained by subtracting all soil and watershed losses (Ia and infiltration) from precipitation. The precipitation excess (pe t ) during the time interval t to t+Δt was then calculated as follows: The direct runoff was then generated from pe t by using Clark's unit hydrograph model.

Clark's Unit Hydrograph
The Clark's Unit Hydrograph (UH) model was used to perform runoff simulations. This model derived the subbasins' UHs by representing two critical processes in the transformation of pe t to runoff: (1) the movement of pe t from its origin through the drainage area to the outlet and (2) attenuation, the storage effect of the stream channel (Feldman 2000). Short-term storage of water in the watershed was represented using a linear reservoir approach, represented by the equation: Where dS/dt = time rate of change of water storage at time t; I t =average inflow to storage at time t ; and O t = outflow from storage at time t .
Along with the linear reservoir model for groundwater flow, the storage at time t is related to outflow as:

St=ROt
Where R is a constant linear reservoir parameter (storage coefficient). These equations are combined and solved using a simple finite difference approximation, yielding: Ot = CAIt + CBOt-1 Where C A and C B are routing coefficients and were calculated as follows: and CB= 1 -CA The average watershed storage outflow for each time interval was:

̅̅̅
Conceptually, the reservoir for the watershed was located at the outlet of each subbasin and represents the aggregate impacts of all watershed storage (Feldman, 2000). Clark's UH model also accounted for the time required for water to move to the watershed outlet by using a linear channel model that routed the water from remote locations to the linear reservoir at the outlet without attenuation. The time delay was represented implicitly with a time-area histogram, included within HEC-HMS. If the area is multiplied by unit depth and divided by t, the result is the inflow to the linear reservoir. Since the unit depth for the simulation was pe t , solving for the reservoir outflow ordinates generated the Unit Hydrograph (Feldman, 2000).
The other parameter required for by HEC-HMS for the Clark's UH simulation was the storage coefficient, or R. R is an index of the temporary storage of pe t in the watershed as it drains to the outlet, estimated for this study using the autocalibration feature of the model (Feldman, 2000).

Linear reservoir for subsurface flow
Base flow was modeled using a linear reservoir approach, which simulated the storage and movement of subsurface flow as water moving between two linear reservoirs and is used along with Clark's UH. The initial baseflow was specified for the beginning of the simulations. The groundwater storage coefficient was a time constant, measured in hours, giving a sense of the response time of the subbasin.
Groundwater flow was the sum of volumes of groundwater from each layer and is computed by: Where GwFlow t and GwFlow t+1 were the groundwater flow rates at the beginning of the time interval t and t+1, ActSoilPerc was the actual soil percolation from the soil profile to the groundwater layer, computed from the constant infiltration rate input in the deficit and constant loss method and obtained from model calibration.
CurGw i Store was the calculated groundwater storage for the groundwater layer, RoutGw i Store was the groundwater flow routing coefficient from groundwater storage, TimeStep was the simulation time step.

Kinematic wave routing for channel flow
Channel flow was modeled using a kinematic wave routing model, based on a finite difference approximation of the continuity equation and a simplification of the momentum equation. Values for Manning's n (roughness coefficient of the channel) were estimated from visual inspection, field measurements and comparison to other channels (Barnes, 1967). The cross-sectional area of the channels were approximated by rectangles.

Model calibration and validation
Initial model parameters were estimated using the guidelines given in the HEC-HMS Technical Reference Manual (Feldman, 2000). The automatic parameter optimization tools, available in the HEC-HMS model, were used to find the optimum set of parameters (groundwater storage coefficients for the linear reservoirs) for each sub-basin.
Model calibration was based on 10 years of continuous flow data at the USGS Beaver River gauging station and validation was performed on a separate 15 years of daily runoff records at the same location. Validation was also examined for just the summer months to assess the low flow performance of the calibrated model. Both Nash-Sutcliffe efficiency (NSE) and root mean square ratio (RSR) were used to determine "goodness of fit" of the model . The NSE is a normalized statistic that determines the relative magnitude of the residual variance compared to the measured data variance.
NSE indicates how well the plot of observed versus simulated data fits a 1:1 line and was computed as: Where Y i obs is the i th observation for the stream flow, Y i sim is the i th simulated value and Y mean is the mean of the observed data and n is the total number of observations. NSE ranges from -∞ to 1.0 with values between 0 and 1 being acceptable levels of performance  Root mean square error (RMSE) is also a commonly used error index statistics (Singh et al., 2005). Singh et al. (2005) stated that RMSE values less than half the standard deviation of the observed data may be considered appropriate for model evaluation. Based on the recommendation by Singh et al. (2005), a model evaluation statistic, RMSE-observations standard deviation ratio (RSR), was developed. RSR standardizes RMSE using the standard deviation of the observed values, and it combines both an error index and the additional information recommended by . RSR was calculated as the ratio of the RMSE to the standard deviation of observed data: RSR varies from the optimal value of 0, which indicates zero RMSE or residual variation, perfect model simulation, to large positive values. In general, models can be considered "very good" if 0.75 < NSE < 1.00 and 0.00 < RSR < 0.50, (Moriasi et al.,< 0.60.

Climate Data
Daily precipitation data were obtained from the National Weather Service Cooperative Observer Station 37-4266-01, Kingston, Rhode Island, located approximately 11.4 km to the east of the watershed. Monthly average evaporation data was obtained from the National Weather Service (Farnsworth and Thompson, 1982). Both rainfall and evaporation rates were assumed to be constant over the entire watershed. For the modeling time interval (1982 to 2007), the median annual precipitation was 1300 mm, with 1983 the wettest year with an annual precipitation of 1783 mm. The driest year was 1993 with 1110 mm of precipitation.

Low Impact Development Land Use Scenarios
To Applying these practices to conventional zoning for subdivisions with ½ acre lots, IA can be reduced from approximately 25% IA per lot to between 11 and 18% per lot (CWP, 1998).
The upper watershed using Scenario B was used to evaluate the potential impacts of LID on stream flow in the Beaver River. The developed area was reduced by half, but the housing density was doubled to 1/4 acre lots and the connected IA per lot was increased from 25% to 38%. The RI Stormwater design manual (RIDEM, 2010) was used to guide assumptions in the LID scenario. It requires that IA be disconnected and that a portion of the IA runoff (based on the NRCS Soil Hydrologic Group at the site) be directed to recharge structures. Given the soil attributes of the developed areas, a recharge factor of 35%, was used. This was reflected in the simulation by reducing the percent connected IA under LID scenarios by 35%. The combination of less area in development and partial recharge of runoff from disconnected IA resulted in a substantial change in connected impervious area from 14.3% to 7.9%.

Irrigation Scenario
The effects of direct river withdrawal for irrigation on the probability of low flows in the upper subbasin of the Beaver River was explored. The irrigation scenario represents the daily water withdrawals from a 50 ha turf field in summer (mid-June through August) when potential evapotranspiration is most elevated (average month ET of 0.126-0.144 m/month). Withdrawals from a linear move system that traverses the field in 22 hours and operates seven days a week were modeled. Irrigation was assumed to occur for a total of 40 days between mid-June and August 31, (representing dry periods punctuated by occasional rains). This level of irrigation does not represent a worst case drought situation. For the 66 days between June 1 and August 7, 1999, the Kingston RI weather station recorded a total of 41.2 mm of rainfall, warranting much more extensive periods of irrigation. An application rate of 0.035 m/day was selected to meet the daily ET demand fully. Irrigation was scheduled for 5 consecutive days followed by 5 days without rainfall to mimic intermittent rainfall. Withdrawals could be substantially higher in some watersheds where the area of irrigated agriculture is higher. In addition, irrigation systems are usually not operated continuously, since time is needed for maintenance and repair, so irrigation demand is satisfied through somewhat higher rates of pumping and withdrawal.

Flow Analysis
Stream flow data is a continuous variable often summarized by frequency distributions. The values for the streamflow were ranked from smallest to largest and plotted using a Weibull distribution (Weibull, 1951) where: such as the Q90, the flow that corresponds to discharge equaled or exceeded 90% of the time (Smakhtin, 2001) or the Q95 or Q96 (Pyrce, 2004;Shokoohi and Hong, 2011), while the Q99 is often used to quantify more extreme drought conditions. (Price et al, 2011). In this study two exceedance levels of low flows, Q90 and Q95 were assessed. These exceedance levels were used as metrics to compare the flow regime of the various land development scenarios to the flow regime that is expected under current watershed conditions.

Model Calibration and Validation
The observed and model predicted stream flow hydrographs for the calibration period of January 1982 to December 1992 are shown (Figure 3a). The calibrated model was then applied to predict the stream flow for the validation period of January 1993 to December 2007 ( Figure 3b).
Statistical indices of NSE and RSR, for both the calibration and the verification periods were calculated using the results of daily time steps (Table 2). In our simulation, model calibration statistics for the calibration period, the validation period and overall time period were classified as "very good" or "satisfactory" and indicated that this generated model was acceptable. The results would likely improve if longer time steps were used, i.e., monthly (Engel et al., 2007); however the focus was on daily flow metric for management applications. For example, in a study conducted by Fernandez et al. (2005) (Moriasi, 2007).

Changing Land Use
In order to quantify the changes in flow conditions due to changing impervious area, the model results were examined in two ways. First, the flows associated with the Q90 and Q95 for each of the scenarios were obtained. This permits comparison of changes in the actual flow rates between different scenarios (Table 3). Second, the flow associated with each exceedance (and companion nonexceedance) metric (e.g., Q95, Q90) from the current watershed development condition was used as the basis for comparison. This is refered to as the "basis" flow rate. Then the exceedance (and companion non-exceedance) probabilities for each land use scenario were determined for the "basis" flow in each land use scenario (   Table 4 -Non-exceedance probabilities for subbasins using daily flow rate computed for existing conditions as baseline. The model indicates slightly higher levels of flow for the low flow metrics with increasing impervious cover. Changes to the channel depth of the riffles were also relatively minor ( Numerous studies suggest that base flow will be negatively correlated to connected impervious area (Klein, 1979;Finkenbine et al., 2007). Our modeled scenarios agreed with these findings (Figure 4; Table 6). In the HEC-HMS model baseflow originates as percolation from the soil profile to the groundwater.
Connected impervious areas within HEC-HMS do not contribute to the baseflow. On pervious surfaces, percolation from the soil profile reflects both the extent of the soil moisture deficit and the magnitude of daily rainfall. During the summer in the study region, monthly evapotranspiration usually exceeds precipitation and soil moisture can be depleted substantially. In 1995, soil moisture depletion dropped to 53% of its full storage capacity ( Figure 5), while in 1993, soil moisture depletion dropped to less than 30% of its full storage capacity. Baseflow from LID-based development is higher than from conventional development and this difference is most pronounced during a median year, with less differences noted for a dry year when the soil moisture deficit is expected to be higher.
On pervious surfaces in the study region, direct runoff (storm runoff) is likely to be infrequent during the summer months, when most of the precipitation that falls is utilized for the soil moisture deficit. As seen in Figure 6, direct runoff for the study watershed, with its current condition of 2% IA, was negligible during much of the summer of both a median and a dry summer. Summer rainfall must fill the soil voids of the pervious areas before runoff begins. In contrast, connected IA will generate immediate runoff to streams from rainstorms that would have otherwise infiltrated the soil when those areas were in pervious surfaces (Lull and Sopper, 1969). It is noteworthy that at least one period of excess precipitation occurred in the summers of both the median and dry years and the falling limbs of those hydrographs generated prolonged periods of comparatively elevated flows ( Figure 6).
Total flow to the stream is the total of baseflow and direct runoff. The     Table 7 includes the low flow metrics for the 1995 (median year) and 1993

Comparative Effects of Impervious Cover, Irrigation and LID
(dry year) of the upper watershed for current conditions, Scenario B (14% impervious), LID and irrigation. Changes in flow conditions due to either implementing LID or accounting for potential irrigation losses were examined as before, comparing the flows associated with the Q90 and Q95 for each of the scenarios as well as comparing the exceedance (and companion non-exceedance) probabilities for each land use scenario based on existing probabilities.
During 1995, a year reflecting median rainfall conditions, the flow predicted to occur with 10% non-exceedance probability under the existing conditions decreased to 5.7% under Scenario B (convention development with 14% IA) and to 6.7% under an LID scenario (Table 7). In other words, during a year with median precipitation, the model predicts a lower frequency of low flows with both conventional development and with LID development compared to the predictions for the limited development present in current conditions. Both conventional development and LID also display fewer low flow periods during a dry year, but the pattern reverses, with LID predicted to have lower frequencies of low flows than the conventional development (Table 7). As noted above, connected impervious cover generates more storm-generated flow, but lower baseflow. Over the summer, storm runoff and the associated falling limb of the runoff hydrograph that results from connected impervious cover occurs with enough frequency to influence the low flow thresholds used for metrics (i.e., the flow rate that coincides with the lowest 18th or 133 37 th day of a year). During the dry year, rainfall occurrences were very infrequent and the higher baseflow associated with LID accounts for the slight increase in flows compared to the conventional development.
Irrigation within the upper watershed was the only scenario that resulted in a decrease in flows compared to current conditions. Irrigation scenarios decreased both flows and depths. For example, while daily flows of < 0.032 m 3 s -1 occurred 10 % of the time (Q90) in the upper sub-basin under current conditions in a dry year, during the irrigation scenario, this level of daily flow occurred more than 15 % of the time (Table 7). Based on hydraulic measurements taken at the riffle cross section at the outlet of the upper basin, this change in flow will lower the water depth from 7.2 cm with present conditions to 6.9 cm with irrigation (Table 8).
Changes in land use generally increase river flows while water withdrawals decrease river flows (Gerten et al, 2008). Eheart and Tornil (1999) found both surface water and groundwater withdrawals have the potential to deplete streams to dangerous levels. Caldwell et al. (2012) found that water withdrawals decreased river flows by an average of 1.4% nationwide. Stream height (above thalweg) associated with Q95 and Q90 for existing conditions and development scenarios. 1993 b) Stream height above thalweg for development scenarios Table 8 -Flow metrics for upper watershed for Scenario B, Scenario B with LID and existing conditions with irrigation. a) daily flow rates associated with Q95 and Q90; b) stream height above thalweg for development scenarios

Conclusions and Limitations
The occurrence of low flows within the Beaver River was found to be relatively resilient to the extent of development and water withdrawals simulated by this study. Generally, any changes observed in the Q90 and Q95 flow values due to different land use scenarios were not dramatic. A meta-analysis conducted by Schueler et al. (2007) found few studies which researched the effects of impervious area on hydrologic factors, and those studies were either contradictory or ambiguous. Specifically, they found that an inverse relationship between impervious cover and base flow to streams was not always present. Winter (2007) found that base flow is more sustained in watersheds with extensive aquifers, like the Beaver River aquifer (Dickerman and Ozbilgin, 1985), but transpiration from riparian vegetation can causes notable loss of stream flow. Morrison et al., in a statistical study of the importance of watershed attributes to low flow metrics in 33 watersheds in southern New England (Chapter 1), found that the proportion of developed areas (which was highly correlated with IA) was not as important to the magnitude of low flows as natural attributes within a watershed i.e., the proportion of wetlands (negatively correlated to low flow magnitudes) and the extent of stratified drift which was positively correlated to low flow magnitudes. These natural attributes were unchanged for all scenarios investigated in this chapter.
The Beaver River study watershed has approximately 14% wetlands soils and 60% of the length of the river abuts riparian wetlands. In riparian areas, groundwater is closer to the land surface and riparian vegetation will derive most of its water from the groundwater. During spring and summer months when evapotranspiration is high, riparian vegetation will draw water from the stream and reduce streamflow (Winter, 2007). In a meta-analysis of wetland functions, Bullock and Acreman (2003) found that floodplain wetlands reduce the flow of water in streams during dry periods. Evaporation was also found to be higher in wetlands than in non-wetland portions of a watershed. In a study of riparian wetlands in southern Rhode Island on soils similar to those found in our study site, Kellogg et al. (2010) found that transpiration from riparian wetlands intercepted virtually all base flow to the river during the summer months. Rowe (1963) found that streamflows are greatly increased when woody riparian vegetation is removed, which would suggest that the vegetation was drawing water from the streams.
A lumped parameter model, such as HEC-HMS does not differentiate location of soil types within subbasins, but rather calculates an overall value for soil properties such as infiltration rate. Also, the HEC model does not account for losses due to increased water demand from riparian vegetation, perhaps overestimating stream flow during summer months when the evapotranspiration demands are highest. However, the simulations relied on a calibration step which may have partially accounted for the role of riparian zone on the flow regime of the stream.
In addition to the lack of explicit representation and modeling of riparian wetlands, there are other limitations to the HEC-HMS model as well as factors that were not included in the changing land use scenarios which may affect the results of the low flow analysis. That is, the simulated scenarios did not consider the increased well water usage that typically coincides with increased development. Depending on the location of the wells, distance from the river, withdrawal rates and hydrogeologic setting, installation of wells will have differing impacts on the river. Long term studies of stream discharges have found groundwater withdrawals have decreased stream flow significantly as well as to become disconnected from downstream reaches or dry up altogether (Wahl and Wahl, 1988;Sophocleous, 2000).
The effects of increased effluent from septic systems were also not investigated. Burns et al. (2005) found that base flow during dry periods was higher in high density residential areas, perhaps due to discharge from septic systems. They suggest that while development and increased IA will increase peak magnitude and accelerate the conveyance of storm runoff to streams, the combined effects of natural landscape features such as wetlands and human alterations can change the expected effects of human development on both storm runoff and groundwater recharge. In addition, Hirsch et al. (1990) suggest that the effects of septic system effluent may mitigate the effects of increased impervious area on baseflow recharge.
The effects of groundwater withdrawals for human consumption coupled with septic system groundwater recharge and reinfiltration from lawn watering may be insignificant as overall, they may negate each other (Foster, 1990).