National framework for ranking lakes by potential for anthropogenic hydro-alteration

Abstract

Lakes face multiple anthropogenic pressures that can substantially alter their hydrology. Dams and land use in the watershed (e.g., irrigated agriculture) can modify lake water regimes beyond natural ranges, and changing climate may exacerbate anthropogenic stresses on lake hydrology. However, we lack cost-effective indicators to quantify anthropogenic hydrologic alteration potential in lakes at regional and national extents. We developed a framework to rank lakes by the potential for dams and land use to alter lake hydrology (HydrAP) that can be applied at a national scale. The HydrAP framework principles are that 1) dams are primary drivers of lake hydro-alteration, 2) land use activities are secondary drivers that alter watershed hydrology, and 3) topographic relief limits where land use and dams are located on the landscape. We ranked lakes in the United States Environmental Protection Agency National Lakes Assessment (NLA) on a HydrAP scale from zero to seven, where a zero indicates lakes with no potential for anthropogenic hydro-alteration, and a seven indicates large dams and/or intensive land use with high potential to alter lake hydrology. We inferred HydrAP population distributions in the conterminous US (CONUS) using the NLA probabilistic weights. Half of CONUS lakes had moderate to high hydro-alteration potential (HydrAP ranks 3–7), the other half had minimal to no hydro-alteration potential (HydrAP ranks 0–2). HydrAP ranks generally corresponded with natural and man-made lake classes, but >15% of natural lakes had moderate to high HydrAP ranks and ~10% of man-made lakes had low HydrAP ranks. The Great Plains, Appalachians, and Coastal Plains had the largest percentages (>50%) of high HydrAP lakes, and the West and Midwest had the lowest percentages (~30%). Water residence time (τ) and water-level change were associated with HydrAP ranks, demonstrating the framework’s intended ability to differentiate anthropogenic stressors that can alter lake hydrology. Consistently across ecoregions high HydrAP lakes had shorter τ. But HydrAP relationships with water-level change varied by ecoregion. In the West and Appalachians, high HydrAP lakes experienced excessive water-level declines compared to low-ranked lakes. In contrast, high HydrAP lakes in the Great Plains and Midwest showed stable water levels compared to low-ranked lakes. These differences imply that water management in western and eastern mountainous regions may result in large water-level fluctuations, but water management in central CONUS may promote water-level stabilization. The HydrAP framework using accessible, national datasets can support large-scale lake assessments and be adapted to specific locations where data are available.

1. Introduction

Lakes, whether natural or man-made, provide multiple water-related ecosystem services that support human populations and activities, including drinking water, water for irrigation and industrial cooling, flood control, recreation, and hydropower (Carpenter et al. 2011). Dams, impoundments, and modifications to the landscape facilitate and enhance these services but in turn can substantially alter lake hydrology (Haddeland et al. 2006b, 2014) and impair other components of lake integrity such as nearshore physical habitat (Gaeta et al. 2014; Carmignani and Roy 2017), water quality (Leira and Cantonati 2008; Carpenter et al. 2011; Zohary and Ostrovsky 2011), and biotic integrity (Jeppesen et al. 2015; Evtimova and Donohue 2016; Kann and Walker 2020). With changing climate conditions, demands for water-related ecosystem services are likely to increase and put more pressure on lake and watershed hydrology (Vörösmarty et al. 2010; Carpenter et al. 2011; Haddeland et al. 2014; Nazemi and Wheater 2015). A system for classifying the potential for anthropogenic alteration of lake hydrology at broad spatial extents that goes beyond “natural” and “manmade” classes could aid in understanding these interactions and better support watershed management activities under changing environmental conditions.

Initiatives to protect freshwater ecosystems acknowledge the need to assess the relative influence of anthropogenic and climatic stressors on lake hydrology. The United States Environmental Protection Agency, other federal agencies, states, and tribes implement the National Lakes Assessment (NLA) surveys to provide a population wide estimate of the extent of the nation’s lakes and reservoirs that support water quality and ecological integrity goals of the Clean Water Act. The NLA evaluates the extent and severity of several key stressors, including hydrologic disturbances to lake and reservoir condition in the conterminous US (CONUS). Human activities and structures can alter lake hydrologic regimes and lead to frequent, extreme water-level fluctuations and/or water-level stabilization beyond natural ranges (Hill et al. 1998) and altered water residence times. These hydrologic alterations may be widespread in the US. Fergus et al. (2020) showed that in 2007 (a dry year) more than half the population of natural and man-made lakes in CONUS had lower than expected water levels. In a wetter year (2012) most natural lakes had returned to normal water levels, but drawn down water levels persisted in about 20% of lakes, most of which were man-made lakes in arid regions. These patterns suggest that continued water-level decline between survey years may be attributed to human causes, but anthropogenic metrics are lacking to tease apart human and climatic drivers of lake hydrology.

Lake water levels naturally fluctuate as a function of water inflows, outflows, and water loss from evaporation (Hofmann et al. 2008), but anthropogenic infrastructure including dams, canals and ditches, and land use activities also have the potential to substantially alter lake water balance. Dams store water and modify the magnitude and timing of water release downstream to meet human water needs (Magilligan and Nislow 2005; Poff et al. 2007). Land use such as irrigated agriculture and urban development can alter watershed hydrology by converting natural land cover to anthropogenic structures and activities that consume (e.g., irrigated agriculture) or redistribute water across the landscape (Poff et al. 2006a; Nazemi and Wheater 2015). Dams and land use are important drivers of altered lake water levels but have not been commonly incorporated in lake hydrology assessments.

Regional understanding of anthropogenic stressors on lake levels is constrained by the lack of suitable, cost-effective measures of the human potential to alter lake levels that are applicable to populations of lakes across diverse hydrologic settings (Poff et al. 2006b). This deficiency is not only a problem for lake monitoring and assessment but also for addressing macroscale empirical research questions. For example, one outstanding challenge to constructing more accurate global and continental hydrologic cycles is the need to integrate anthropogenic hydrologic stressors into large scale models (Wood et al. 2011). Understanding the relative influence of anthropogenic versus natural climate controls is a pressing challenge to lake hydrologic condition assessments (Wine and Laronne 2020; Kraemer et al. 2020) and these relationships have implications for water management decisions to protect lake ecological integrity (Cobbaert et al. 2015; Kann and Walker 2020).

Classifying lakes by lake origin as either natural or man-made, have long served as a framework for differentiating lakes according to their level of anthropogenic pressures. The NLA defines natural lakes as water bodies existing prior to European settlement even if artificial control structures are present, and man-made lakes as water bodies intentionally created by constructing and filling lake basins and/or damming river outlets where no lake basin existed prior to European settlement (USEPA 2009). These binary lake classes differ in important characteristics including lake morphology, watershed land use, and water quality (Hayes et al. 2017; Doubek and Carey 2017). However, lake origin classes were not developed to explicitly characterize the potential influence of humans on lake hydrology, and consequently do not accurately portray the spectrum of anthropogenic hydro-alteration potential. Human modifications to lake ecosystems are continuous and multidimensional, and a more informative framework is needed that characterizes these modifications along a gradient of potential hydro-alteration rather than as a dichotomous classification.

We developed a framework to rank the potential for anthropogenic hydrologic alteration (HydrAP) in lakes. The HydrAP framework sets lake and landscape criteria using data on dam capacity, land use activities, and topographic relief to rank lakes at a national scale on a gradient of potential anthropogenic hydro-alteration. The HydrAP framework is intended to be adapted as new information and data become available and/or to fit specific locations. Our HydrAP framework was developed to rank lakes across CONUS and does not include region-specific anthropogenic stressors if they are not quantified in national geospatial datasets. The premise of the HydrAP framework is that dam and land use activities are indicators of the potential for humans to alter lake water balance as measured by lake water levels and water residence time. The theoretical principles behind the framework are: 1) dams are the primary, direct drivers of lake hydrologic alteration, 2) land use activities are secondary drivers that alter watershed hydrologic processes and indirectly affect lake water balance, and 3) topographic relief in the watershed affects where human land use activities and dam construction can occur and imposes limits on the magnitude of these effects. Actual measures of anthropogenic hydrologic alteration, such as dam operation records, are not available at regional and national extents for the vast majority of lakes (Poff and Hart 2002; Nazemi and Wheater 2015). However, dam physical attributes and land use metrics can provide indirect measures of the potential for human modification of lake water balance.

We applied the HydrAP framework to lakes in the 2007 and 2012 NLA surveys (USEPA 2009, 2016). We describe the HydrAP framework structured as a decision tree and present population distributions of HydrAP ranks at national and ecoregional extents in CONUS using the NLA probabilistic survey design. Although HydrAP ranks potential alteration, we relate HydrAP ranks to lake and watershed morphometry, water-level fluctuations, and water residence time to examine compositional differences among ranks and how they may relate to various dimensions of actual lake hydrology. Finally, we discuss limitations of the HydrAP rankings and ways the framework could be improved.

2. Material and methods

2.1. HydrAP decision tree

The HydrAP framework is organized as a decision tree and ranks lakes on a scale from zero to seven along a gradient of lake hydrologic alteration potential (Figure 1) using best professional judgement guided by literature. We recognized the need for best professional judgement because the combined impacts of human infrastructure and land use on lake hydrologic alteration are poorly understood at a national scale. We created multiple decision tree versions that had similar HydrAP distribution outputs. We selected this version to present because it followed simple logical steps and cutoff values. In the HydrAP framework, a zero HydrAP rank signifies lakes with no apparent anthropogenic structures or activities that could alter lake hydrology (i.e., no dams or human land use in the watershed) and a seven signifies a lake with anthropogenic activities and/or structures with great potential to alter lake hydrology.

Figure 1.

Lake HydrAP decision tree. Sampled 2007 and 2012 NLA lakes were ranked from 0 to 7 based on dam characteristics, land use and topographic relief. Branches in the decision tree (A – D) were guided by specific criteria. Branch A: Lakes without dams. Land use gradients in the watershed (Ws) were the primary driver of HydrAP rank for lakes without water control structures. Land use disturbances in the watershed were represented by summing total agriculture and urban development proportions. Within high land use watersheds, specific land use types (irrigated agriculture, artificial agricultural drainage, and urban development) in the direct drainage area (Adj) were expected to directly impact watershed hydrology. Branch B: Lakes with dams located in low topographic relief watersheds. Dams were the primary drivers and land use gradients were the secondary drivers of HydrAP ranks. Dams were represented by a ratio between dam height and maximum lake depth at full pool (zmax). Watersheds with <50% proportional cover of agriculture and urban land use had more relative dam height bins to differentiate lakes with different degrees of hydro-alteration potential. Watersheds with >50% proportional cover of agriculture and urban land use referenced proportions of irrigated agriculture, artificial agricultural drainage, and urban development in the direct drainage area. Branch C: Lakes with dams located in moderate topographic relief watersheds. Dams were the primary drivers and land use gradients were secondary drivers. Specific land use types were less prevalent in this topographic range and were not referenced. Branch D: Lakes with dams located in high topographic relief. Dams were the primary and only drivers of HydrAP ranks.

The HydrAP framework is based on the following principles ordered by relative influence on lake hydrology: 1) dam infrastructure directly facilitates water storage and release in lakes, 2) land use activities in the watershed (e.g., irrigation, urban impervious surfaces) can alter surface and subsurface hydrology and subsequently affect lake water balance, and 3) watershed topographic relief is associated with slope, lithology, and soils that can influence distributions of land use activities and dams and their potential effects on hydrology. These factors are grounded in theory based on published literature and provide a robust framework to rank lake hydro-alteration potential across different lake types and landscape contexts in CONUS. We describe below the three factors influencing the potential for altered lake hydrology, the theory behind them, and the variables that represent them in the decision tree (Table 1).

Table 1.

Dam and landscape variables in the HydrAP decision tree.

Variable	Description	Source
Dam/Zmax	Ratio of dam height and maximum lake depth at approximate full pool	Dam: National Anthropogenic Barrier Database (NABD) (Cooper et al. 2017) Lake depth: NLA and online lake records
Ag+Urb Ws	Aggregated land use classes summing total agriculture and urban development class proportions in the lake watershed (Ws)	USGS National Land Cover Database (NLCD) 2006, 2011
Irrig Ag Adj	Estimated extent of irrigated agriculture based on satellite-derived estimates of vegetative vigor, NLCD land cover, and US Dept. of Agriculture irrigation statistics. Proportion is estimated in the lake adjacent direct drainage area (Adj)	Brown and Pervez 2014
Ag Drain Adj	Estimated artificial drainage on agricultural lands based on soil drainage (SSURGO), crop productivity surveys (US Dept. of Agriculture), and NLCD land cover. Proportion is estimated in the lake adjacent direct drainage area (Adj)	Christensen et al. 2013; Vanderhoof et al. 2017
Urb Adj	Proportions of total urban development classes in the lake direct drainage area (Adj)	USGS National Land Cover Database (NLCD) 2006, 2011
Topographic relief	Difference in maximum elevation in the watershed and lake elevation	USGS National Elevation Dataset

2.1.1. Dams and outlet controls

Dams are human-constructed structures designed to store water and modify the magnitude and timing of water release downstream (Poff and Hart 2002). These point sources of hydrologic alteration are expected to have the greatest potential to alter lake hydrology compared to human land use activities, which generally result in more diffuse effects on lake hydrology via modifications to watershed hydrology. Dam operation and management decisions dictate the volume and timing of water release from lakes and ultimately determine dam effects on lake hydrology. But dam operation records are not often available for most lakes with control structures, and many dams in the US are not actively managed. Rather we used dam height as a robust and more widely available indicator of a dam’s potential to alter lake hydrology because it is directly associated with lake water storage volume (Poff and Hart 2002). Additionally, dam height is associated with altering other important aspects of lake integrity such as thermal regime dynamics, biotic migration and sediment transport, and biogeochemical processing (Poff and Hart 2002; Crook et al. 2015). We represented the influence of dams by calculating the ratio between dam height and maximum lake depth at full pool (DamHt/Z_max)), where Z_max is the sum of field-measured maximum depth plus the height from present lake water level to full-pool water level estimated by field crews. This ratio scales the height of the dam to maximum lake depth and estimates the relative influence of the dam on lake water levels. Larger values indicate dams with greater potential to alter lake water levels.

2.1.2. Land use

Human land use alters fundamental watershed processes that regulate the magnitude and rate of water delivery to receiving lakes and streams (Poff et al. 2006a). While dam presence is expected to be the main driver of HydrAP, land use activities (i.e., agriculture and urban development) can substantially alter lake and watershed hydrology. For lakes that do not have dams, land use is the primary driver of hydrologic alteration in the HydrAP framework. Agriculture alters watershed hydrology through land conversion, surface drainage, surface and groundwater withdrawals for irrigation, impoundments, and channelization and straightening of water courses (Blann et al. 2009; Carlisle et al. 2019). Urban development, characterized by impervious surfaces such as paved roads and buildings, prevents precipitation from infiltrating into the ground and results in increased surface runoff and flashy stream hydrography (Shuster et al. 2005; Eng et al. 2013; Nazemi and Wheater 2015). Clearing native vegetation for human land use alters evapotranspiration and soil infiltration patterns and can result in increased runoff and flashy flow conditions (Shuster et al. 2005; Poff et al. 2006a). We summed agriculture and urban land use classes to represent a general measure of the potential for hydrologic alteration in the lake watershed. We expect that with larger proportions of human land use in the watershed, there is greater potential for altered lake and watershed hydrology.

Certain land use types are expected to have greater effects on lake and watershed hydrology even in small proportions on the landscape, specifically irrigated agriculture, agricultural drainage, and urban land use. Irrigated agriculture is supported by water withdrawals from streams, reservoirs, lakes, and groundwater, and has been associated with decreased stream discharge and increased evapotranspiration in the watershed (Haddeland et al. 2006a). This consumptive water use has been associated with lake water-level declines and desiccation (Tang et al. 2009; Wine and Laronne 2020). Where water is in excess, artificial agricultural drainage via surface (ditches) and subsurface (tiles) infrastructure are constructed to reduce water pooling/flooding on the landscape and support crop and agricultural practices. These drainage networks can substantially alter water and nutrient transport capacity in watersheds (Fraterrigo and Downing 2008) and impact water budgets (Blann et al. 2009) by depressing shallow groundwater levels, thereby reducing stream baseflows, and by redistributing water to other locations in the watershed. Irrigated agriculture, agricultural drainage, and urban development may have disproportional effects on watershed hydrology such that even small areas of these land uses may substantially alter hydrology (e.g., Booth and Jackson 1997; Eng et al. 2013). We expect that greater levels of irrigated agriculture, agricultural drainage, and urban development result in greater potential for hydro-alteration.

2.1.3. Geomorphology

Geomorphic characteristics such as topography, geology, and soils shape the physical structure and dynamics of aquatic ecosystems but also can constrain human activities that alter hydrology. Elevation and topography are associated with valley confinement, slope, lithology, and soils that not only drive aquatic ecosystem patterns (Dodds et al. 2019), but also influence the location of land use activities (e.g., agriculture), and constrain the location of dams and their capacity to store water (Bemmelen et al. 2016).

Topographic relief in the watershed was used in the HydrAP decision tree as an indicator of underlying terrain and topographic features that influence dam capacity, direction of water flow, and land use activities. Topographic relief was calculated as the difference between the maximum elevation in the watershed and the lake elevation. Lakes were separated into three topographic relief bins to distribute the range of observed values at the national scale: low topographic relief (≤ 200 m), moderate topographic relief (200–700 m), and high topographic relief (>700 m). Topographic relief bins are associated with regional distribution patterns, where most low relief lakes are located in the central and southeastern coastal plains of the CONUS and the majority of high relief lakes are located in the western mountains and Appalachians. The relative influences of dams and land use are expected to vary within each of the topographic bins. In watersheds with high topographic relief, dams are expected to have a greater effect than land use on altering lake hydrology; where topographic relief is low, surface waters may be affected by both land use activities and dams.

2.2. Datasets

2.2.1. National Lakes Assessment surveys

The NLA surveys apply probability-based designs to assess the ecological condition of lakes across CONUS. The geographic sample frame used to identify lake sample sites is based on lake polygon features in the National Hydrography Dataset (NHD Plus v.1; 1:100,000) that includes natural and man-made lakes and ponds and excludes the Laurentian Great Lakes, the Great Salt Lake, commercial treatment ponds, and coastal and ephemeral lakes. Target lakes in the NLA 2007 survey included all NHD permanent waterbodies with surface area ≥ 0.04 km², estimated maximum depth ≥1 m, and ≥ 0.001 km² of open water. The target lake definition was the same in NLA 2012 but expanded to include smaller sized lakes 0.01 – 0.04 km². In total, NLA 2007 sampled 1,028 lakes and NLA 2012 sampled 1,038 lakes, with about 30% of the 2007 lakes being resampled in 2012 (n = 350). With the probabilistic survey design, each NLA sampled site is associated with a sampling probability weight that allows inference from the sampled lakes to the target lake population. The NLA population expansion weights account for unequal probability of selection among the lake size classes, such that small lakes that are abundant in the landscape are assigned greater weights than larger lakes.

Lakes in the NLA are classified as natural or man-made based on the lake origin classification developed for the surveys (USEPA 2017). NLA analysts determined lake origin using multiple lines of evidence that included field observations, maps with background imagery or topographic maps (e.g., Google Earth, Google Maps, GIS software), expert opinion, and other available records (e.g., Army Corps of Engineers reservoir database, etc.). Natural lakes were considered to be lakes that existed prior to European settlement; and man-made lakes were defined as water bodies intentionally created by humans where no lake existed prior to European settlement (USEPA 2009). Close to half of the sample lakes in NLA 2007 and 2012 are classified as natural (42.6% and 44.2%, respectively) and the other half are classified as man-made (57.4% and 55.8%, respectively) (Table S1). Population ratios between natural and man-made lake classes are roughly equal at the national-scale but vary across the five aggregated ecoregions (Table S2). For example, natural lakes are over three times more abundant than man-made lakes in the Midwest, whereas nearly all lakes in the Southern Appalachians are man-made.

Lake physical characteristics recorded in the NLA that were used in the HydrAP framework include lake surface area, maximum depth (Z_max), and elevation. Lake Z_max is measured by NLA field crew at the approximate deepest spot of the lake. We modified Z_max to represent the maximum lake depth at full pool by summing the field recorded Z_max and vertical water-level drawdown height at the time of sampling. This adjustment provided a more accurate estimate of maximum lake depth for lakes that had drawn down water levels at the time of the NLA field visit. For a small subset of very large, deep lakes (NLA Z_max > 40 m), Z_max measured in the field was not a reliable approximation of maximum lake depth and we substituted Z_max reported in online lake records where available. These reported Zmax values were reliable measures of maximum lake depth at full pool, and we did not add vertical water-level drawdown height to those values.

NLA lakes are distributed across a wide range of hydrogeomorphic, climatic, and land use gradients that are expected to reflect regional variation in lake hydrologic characteristics and human water use activities. To account for this regional variation, CONUS was organized into five ecoregions: West, Great Plains, Midwest, Appalachians, and Coastal Plains. These ecoregions were defined by aggregating Omernik-Level III ecoregions that delineate areas on the landscape with similar natural geographic and climate features (Omernik 1987; Herlihy et al. 2008). These aggregated ecoregions capture broad-scale variation in hydrology and human water use activities that will aid in interpretation of the HydrAP patterns across CONUS.

2.2.2. Lake-Catchment (LakeCat) and other geospatial datasets

Lake HydrAP criteria are based on national-scale geospatial data (Table 1). Most geospatial information came from the LakeCat Dataset, a database of watershed features for over 370,000 lakes in CONUS (Hill et al. 2018). Land use/land cover variables in LakeCat are from the National Land Cover Database (NLCD; United States Geological Survey) for 2006 and 2011. Irrigated agriculture and artificial agricultural drainage percentages were summarized following the same techniques used in LakeCat. Landscape variables were quantified at two spatial scales: 1) the direct drainage area (referred to as “catchment” in LakeCat), and 2) the lake watershed area (Figure S1). We contend that irrigated agriculture, artificial drainage agriculture, and urban development in the direct drainage area of a lake would have a more pronounced effect on lake hydrology than these classes summarized at the watershed scale. The direct drainage area is the sum of the National Hydrography Dataset (NHD) Plus version 2 catchments with stream segments that flow into the focal lake. For lakes without stream connections, the direct drainage area was derived using rasters depicting the flow direction across the landscape. The watershed is the sum of all connected NHD catchments that flow into the focal lake. For lakes that are not on a stream network and do not have upslope lakes, the direct drainage area and lake watershed are equivalent.

Land use for NLA lakes were summarized for the NLCD layer at the time closest to the respective NLA survey period. Agriculture and urban developed land subclasses in the NLCD were aggregated to create total agriculture and urban estimates. Irrigated agriculture data came from Moderate Resolution Imaging Spectroradiometer Irrigated Agriculture Dataset (MIrAD-US) (Brown and Pervez 2014). Irrigated agriculture was approximated from satellite-derived estimates of vegetation vigor, NLCD land cover, and U.S. Department of Agriculture irrigation statistics. Artificial agricultural drainage layer was estimated using soil drainage classes (SSURGO), crop productivity surveys performed by the U.S. Department of Agriculture, and NLCD land cover data (Christensen et al. 2013; Vanderhoof et al. 2017). Summed land use classes were quantified at the watershed scale, and specific land use classes (i.e., irrigated agriculture, agricultural drainage and urban development) were summarized in the direct drainage area to indicate land use activities in closer proximity to lakes (Figure S1).

Dam attributes came primarily from the National Anthropogenic Barrier Dataset – NABD (Ostroff et al. 2013). The NABD is a database of dams in the US that has been quality checked with geospatial information (Cooper et al. 2017). The dam data were originally derived from the US Army Corp of Engineers National Inventory of Dams (NID) that include dams that are considered to have high hazard potential or dams that exceed 7.62 m in height and 1.85 ha-m of storage volume or exceed 1.83 m in height and 6.17 ha-m of storage volume. NABD dams located on NLA lakes were selected through geospatial queries. We noted if multiple NABD dams were associated with a lake and selected the largest dam for the HydrAP decision tree. NLA field crews recorded the presence of lake outlet control structures but did not report dam size or other characteristics. We performed additional searches for dam attribute information in West NLA lakes that were reported to have dams by NLA field crews but were missing NABD dam records. This exercise was done to examine what types of lakes and dam attributes may be underrepresented in the NABD for a subset of the NLA lakes. Additional dam observations in the West came from the US Bureau of Reclamation website (https://www.usbr.gov/).

2.2.3. Lake hydrological test variables

After ranking NLA lakes using the HydrAP decision tree, we examined how the HydrAP ranks were related to select lake hydrologic variables in the NLA. We selected lake water-level change and water residence time as examples of lake water balance characteristics that are influenced by diverse hydrologic processes and capture different aspects of lake hydrology. Vertical and horizontal water-level decline are measured as the mean height or distance from the lake water level to the apparent high-water mark at 10 equidistant stations around the lake during the summer sample visit (Kaufmann et al. 2014a). Vertical and horizontal water-level decline estimate short- to medium-term declines in water levels that may be caused by drought and/or water management activities over monthly to decadal time scales. Horizontal drawdown affects littoral and riparian habitat and is of concern to lake physical condition. Vertical and horizontal water-level decline were scaled to account for the influence of lake morphology on absolute measures of water-level drawdown. Vertical water-level decline was divided by Z_max, and horizontal decline was divided by the square root of the lake surface area. Scaled vertical and horizontal water-level declines were log10 transformed. To estimate the deviation in water-level declines from expected water-level fluctuations under least-disturbed conditions, we calculated a z-score index from Log (observed water-level decline) estimates and Log (expected decline in regional least-disturbed lake sites) for vertical and horizontal decline separately. Regional least disturbed sites in the NLA were identified based on water chemistry, nearshore and surrounding human influences, and evidence of human water extraction/diversion. More details of least disturbed site identification are described in the NLA 2012 Technical Report (USEPA 2017) and Herlihy et al. (2013). We calculated a z-score index of water-level change by dividing the Log of the ratio of observed water-level decline (O) to mean expected water-level decline in reference sites (Exp) by the standard deviation (SD) of the Log of water-level decline in least-disturbed reference lakes. Below are example equations for estimating deviation in scaled vertical water-level decline.

$$\mathit{\text{Log}}{\left(O/\mathit{\text{Exp}}\right)}_{\mathit{\text{scaled}}\mathit{\text{vert}}}=\mathit{\text{Log}}{(O)}_{\mathit{\text{scaled}}\mathit{\text{vert}}}-\mathit{\text{mean}}{\left(\mathit{\text{Log}}\mathit{\text{Exp}}\right)}_{\mathit{\text{scaled}}\mathit{\text{vert}}}$$

$$\Delta \mathit{\text{scaled}}\mathit{\text{vert}}\mathit{\text{water}}\mathit{\text{level}}=\frac{\mathit{\text{Log}}{\left(O/\mathit{\text{Exp}}\right)}_{\mathit{\text{scaled}}\mathit{\text{vert}}}}{\mathit{\text{SD}}\mathit{\text{Log}}{\left(\mathit{\text{ref}}.\mathit{\text{lake}}\mathit{\text{decline}}\right)}_{\mathit{\text{scaled}}\mathit{\text{vert}}}}$$

Water residence time (τ) in the NLA was derived from water stable isotope values (δ²H, δ¹⁸O), estimated lake volume (Hollister and Milstead 2010), and modeled potential evapotranspiration from the lake surface (Brooks et al. 2014; Gibson et al. 2016; Fergus et al. 2020). Stable isotope-based τ estimated the time that sampled water has resided within a lake and is largely influenced by precipitation, runoff, and temporally stable characteristics like lake basin and watershed morphology. Lake hydrologic characteristics were intentionally not included in the HydrAP decision tree to maintain independence when associating them with the HydrAP ranks.

2.3. Analysis

Data processing and analysis were performed using R software (v. 3.6.3). HydrAP ranks were assigned to NLA sampled lakes following the decision tree criteria (Figure 1). We used the R package spsurvey: Spatial Survey Design and Analysis (Kincaid et al. 2019) to estimate the national and ecoregional population distributions of lake HydrAP ranks for the 2007 and 2012 surveys separately. We examined the compositional differences among HydrAP ranks in sampled lakes through boxplots on lake size (surface area and maximum depth), watershed-to-lake area ratio (WA:LA), and lake hydrologic characteristics (change in vertical and horizontal water-levels and τ) by the five aggregated-ecoregions. Kruskal-Wallis tests were performed to determine whether lake morphometry and hydrologic characteristics differed among HydrAP ranks and post-hoc Dunn tests were performed for pair-wise comparisons.

3. Results

3.1. Summary

We ranked 907 lakes in the 2007 NLA survey and 887 lakes in the 2012 NLA survey by their potential for anthropogenic alteration of lake hydrology (Figure 2, Table 2). Lakes ranged from having no apparent dams or land use activities (HydrAP rank 0) to lakes with large dams and/or land use activities with the potential for substantial hydrologic alteration (HydrAP rank 7). Lakes with low HydrAP ranks 0 – 2 had small percentages of land use in the watershed or direct drainage area and/or had small dams relative to lake Z_max with little influence on lake water levels. Lakes with HydrAP ranks 3 – 5 had intermediate percentages of land use activities and/or medium sized dams with moderate potential to alter lake water levels. Lakes with high HydrAP ranks 6 – 7 had large percentages of land use and/or tall dams relative to lake Z_max with great potential to alter lake water levels.

Figure 2.

Map of NLA (2007, 2012) lakes by HydrAP ranks across five ecoregions in CONUS. Lakes not assigned HydrAP ranks are represented as dark gray dots (NA).

Table 2.

Number of lakes within HydrAP ranks by NLA survey year and by ecoregion. Sampled lakes in the NLA 2007 and 2012 surveys were ranked by potential anthropogenic hydrologic alteration on a score from zero to seven across five aggregated ecoregions in CONUS (Figure 2).

	2007						2012
HydrAP	West	Great Plains	Midwest	Apps	Coastal Plains	Total	West	Great Plains	Midwest	Apps	Coastal Plains	Total
0	28	10	25	8	3	74	44	10	11	17	1	83
1	38	11	28	12	1	90	32	7	38	15	7	99
2	18	23	67	15	16	139	18	29	87	18	23	175
3	4	1	20	8	5	38	13	2	15	4	2	36
4	8	17	44	23	9	101	13	14	49	16	17	109
5	37	21	23	42	21	144	31	7	16	24	15	93
6	55	35	26	39	10	165	48	23	11	33	9	124
7	33	44	21	39	19	156	39	43	32	31	23	168
Ranked	221	162	254	186	84	907	238	135	259	158	97	887
Not ranked	17	31	29	27	17	121	25	31	36	30	29	151
TOTAL	238	193	283	213	101	1028	263	166	295	188	126	1038

HydrAP ranks were not assigned to 121 and 151 lakes in the 2007 and 2012 NLA surveys, respectively, because dam attribute information was missing. NLA field crews reported observing dams on these lakes but the NABD did not have dam attribute data for them. We found that NLA lakes missing dam data were smaller (median _{2007; 2012} = 0.41 km²; 0.17 km²) and shallower (median _{2007; 2012} = 4.50 m; 3.70 m) than lakes with dam data (median area _{2007; 2012} = 0.94; 0.53 km² and median Z_{max 2007; 2012} = 6.60 m; 5.10 m) based on non-parametric Mann-Whitney tests; (W_size = 19428; W_Zmax = 24042, p < 0.05). These patterns indicate that the dam database used in the HydrAP framework may not adequately record dams on small lakes and subsequently result in small lakes not being assigned HydrAP ranks.

Of the 297 lakes that were resampled in 2012 and assigned HydrAP ranks, 70% did not change HydrAP rank class between survey year (Table 3). Changes in HydrAP ranks resulted from differences in land use percentages between years. Of the resampled lakes that changed HydrAP ranks, most changed by one HydrAP rank (18% increased by one class; 6% decreased by one class). Only 6% of resampled lakes changed by two HydrAP rank classes, with the majority of these lakes increasing in HydrAP rank. Individual ecoregions followed similar patterns.

Table 3.

Confusion matrix of HydrAP ranks (0 – 7) for resampled NLA lakes in 2007 and 2012 (n = 350 total, 53 lakes were not ranked because of missing dam attributes).

					2012
2007	0	1	2	3	4	5	6	7
0	13	9	1	–	–	–	–	–
1	–	26	5	–	–	–	–	–
2	–	2	43	1	3	–	–	–
3	–	–	–	9	–	–	–	–
4	–	–	–	–	25	6	2	–
5	–	–	–	–	6	22	11	8
6	–	–	–	–	2	7	27	22
7	–	–	–	–	–	1	4	42

Across CONUS, lakes followed a north-south pattern of low to high HydrAP ranks (Figure 2). Lakes with low HydrAP ranks were commonly located in northern, mountainous or glaciated areas, and lakes with high HydrAP ranks were in mid- to southern ecoregions. Within ecoregions, lakes exhibited a wide range in HydrAP ranks that generally followed this north-south gradient of low to high ranked HydrAP lakes. However, in the Coastal Plains, lakes with low HydrAP ranks tended to cluster in central Florida.

3.2. Lake population estimates

We applied the NLA population weights to estimate the percentages of HydrAP ranks in the target lake populations for each survey. NLA lakes assigned HydrAP ranks represented approximately 37,600 lakes in 2007 and 53,600 lakes in 2012 in CONUS (Table 4). The 2012 lake population was larger because it included lakes 1 – 4 ha in size whereas the 2007 lake population was restricted to lakes 4 ha and larger. Percentages of HydrAP ranks in the lake populations were similar between survey years. Nationally, about 50% of CONUS lakes were estimated to have moderate to high HydrAP ranks in both survey years, with about 14% of CONUS lakes ranked as HydrAP 6 and 7. The remaining portions of the populations were estimated to have HydrAP ranks between 2 and 0. The percentage of CONUS lakes with no apparent hydrologic alteration potential (HydrAP rank 0) changed from 23% in 2007 to 13% in 2012. After adjusting the 2012 target population to match the 2007 target population lake size, we estimated an even greater decrease in low hydrologic alteration potential lakes in 2012 with only 10% of CONUS lakes ranked HydrAP 0.

Table 4.

National population estimates by HydrAP ranks in CONUS by NLA survey year

	2007		2012
HydrAP	Population (std error)	% Population (std error)	Population (std error)	% Population (std error)
0	8689.2 (1209.8)	23.1 (2.7)	6886.0 (1015.4)	12.8 (1.8)
1	4331.0 (627.6)	11.5 (1.7)	9803.9 (1270.0)	18.3 (2.2)
2	6958.6 (933.9)	18.5 (2.3)	13457.2 (1277.6)	25.1 (2.3)
3	1888.0 (508.8)	5.0 (1.3)	2937.2 (635.8)	5.5 (1.2)
4	6257.6 (923.7)	16.6 (2.2)	8542.4 (1713.9)	15.9 (2.8)
5	4229.1 (586.7)	11.2 (1.5)	4154.0 (799.3)	7.7 (1.4)
6	2671.5 (394.2)	7.1 (1.1)	3210.2 (626.5)	6.0 (1.2)
7	2652.7 (436.9)	7.0 (1.1)	4625.5 (705.5)	8.6 (1.3)
TOTAL	37677.8 (1914.9)	100	53616.4 (2653.7)	100

HydrAP ranks varied within lake origin classes, showing that both natural and man-made lakes exhibit a broad range in anthropogenic hydrologic alteration potential. About 75% of natural lakes had low HydrAP ranks, while over 75% of man-made lakes had moderate to high HydrAP ranks both survey years (Figure 3). These patterns in hydro-alteration potential followed expectations between lake origin classes, but there were exceptions. About 15% of natural lakes in CONUS were estimated to have moderate to high HydrAP ranks that were characterized by having large dams and/or land use activities with the potential to substantially modify lake and watershed hydrology. Similarly, about a tenth of man-made lakes in CONUS had low HydrAP ranks, and these lakes had either no dams (e.g., quarry lakes, farm ponds) or small dams with little capacity to alter lake water levels. Although low HydrAP ranked man-made lakes were constructed by humans at one point in time, they lacked dam infrastructure or land use activities to alter lake water inflows and outflows.

Figure 3.

Population distributions (%) of lake HydrAP ranks in CONUS by lake origin type and NLA survey year. HydrAP ranks not visible in the bar chart had percentages less than 0.5%.

HydrAP population distributions varied across the five ecoregions (Figure 4). Close to 70% of lakes in the Great Plains, 60% in the Appalachians, and 60% in the Coastal Plains were estimated to have moderate to high HydrAP ranks in 2007. In contrast, only about 30% and 28% of lakes in the West and Midwest, respectively, had moderate to high HydrAP ranks in 2007. These ecoregional HydrAP patterns were similar between survey years.

Figure 4.

Population distributions (%) of lake HydrAP ranks in CONUS by ecoregion and NLA survey year.

The percentage of HydrAP rank 0 lakes varied by ecoregion and by survey year. Across ecoregions, the percentage of HydrAP 0 lakes was greatest in the West and lowest in the Coastal Plains in both 2007 and 2012 (Figure 4). Similar to the national pattern, the percentage of HydrAP rank 0 lakes decreased within all ecoregions between survey years except the Appalachians, and this ecoregional pattern remained after adjusting the NLA 2012 target population size to match the NLA 2007 population (Figure S2).

3.3. Lake morphometry and hydrologic characteristics

HydrAP ranks were associated with lake and watershed attributes from sampled NLA lakes in both survey years (NLA 2007 Figure 5, NLA 2012 Figure S3). Kruskal-Wallis group mean tests indicated significant differences in mean lake size, Z_max, and WA:LA among HydrAP ranks by ecoregion. In the West, Appalachians, and Coastal Plains, lakes with high HydrAP ranks were larger and deeper compared to lakes with low HydrAP ranks (Figure 5). Across all ecoregions, lakes with high HydrAP ranks had significantly larger WA:LA compared to low ranked lakes. Bimodal distributions between lake morphometry and HydrAP ranks in the West and Appalachians were related to lake origin class. In these regions, the HydrAP ranks separated natural (low HydrAP ranks) from man-made (high HydrAP ranks) lakes. Natural lakes in the West and Appalachians exhibited a large range in lake area and Z_max values. Associations between HydrAP ranks and morphologic characteristics suggest that some lake and watershed types may be more suitable than others for human water-use infrastructure and activities.

Figure 5.

Distributions of lake and watershed characteristics by lake HydrAP ranks and ecoregion for sampled lakes in NLA 2007 survey. Kruskal-Wallis tests (p-value) were performed to test for differences in mean lake and watershed attributes among HydrAP ranks within each ecoregion.

We found HydrAP ranks, representing potential anthropogenic hydro-alteration, were associated with actual lake hydrologic measures in sampled NLA lakes in the 2007 survey. In the West and Appalachians, vertical and horizontal water-level change were greater than expected in lakes with moderate and high HydrAP ranks (Figure 6). In the Great Plains, Midwest, and Coastal Plains, lakes with high HydrAP ranks had less vertical and horizontal change compared to lower ranked lakes, suggesting that dam infrastructure in these regions may be used to stabilize water levels. Lakes in the Great Plains with low HydrAP ranks had greater than expected water-level change. These lakes tend to be small and shallow and may experience large water-level fluctuations from weather and climate conditions.

Figure 6.

Distributions of deviations in lake water-level from least-disturbed conditions by lake HydrAP ranks and ecoregion for sampled lakes in NLA 2007 survey. Excessive and suppressive lake water-level fluctuation were estimated by calculating differences in log-transformed observed and mean expected lake water-level decline scaled by lake morpometry. Expected lake water-level decline was based on regional least-disturbed conditions in natural lakes. Vertical water-level decline was scaled by maximum lake depth, and horizontal water-level decline was scaled by the squareroot of lake surface area. Values were log10 transformed and standardized (z-score) by subtracing from the mean and dividing by the standard deviation for regional reference lakes. Red horizontal line represents expected water-level change under least-disturbed conditions. Kruskal-Wallis tests (p-value) were performed to test for differences in mean lake water-level change among HydrAP ranks within each ecoregion.

Association of lake τ with HydrAP ranks within and among ecoregions were consistent with our expectations (Figure 7). Mean τ among HydrAP ranks were significantly different and were shorter with higher HydrAP rank. However, within a subset of West and Appalachians lakes with low HydrAP ranks, water residence time increased with HydrAP rank because lake size and depth also increased with HydrAP rank in these mostly natural lakes. Large lakes in these ecoregions commonly had some land use activities in their watersheds and/or direct drainage areas that gave them a HydrAP ranking of 1 or 2. Relationships between HydrAP ranks and τ may be attributed to a combination of factors such as correlations between HydrAP ranks and lake and watershed morphometry, as well as anthropogenic structures and activities that affect water inflows and outflows.

Figure 7.

Distributions of lake water residence time by lake HydrAP ranks and ecoregion for sampled lakes in NLA 2007 survey. Kruskal-Wallis tests (p-value) were performed to test for differences in mean water residence times among HydrAP ranks within each ecoregion.

4. Discussion

4.1. Summary

We developed a framework to rank lakes on a gradient of potential for human-induced hydrologic alteration based on theory, published literature (e.g., Poff and Hart 2002; Nazemi and Wheater 2015), and best professional judgement. Using widely available geospatial information on dam infrastructure and land use activities, the HydrAP framework can be applied and adapted to diverse lake types spanning regional to national extents. We applied this framework to rank lakes in the NLA surveys across CONUS according to the prevalence and spatial distribution of human activities and structures that have the potential to alter lake hydrology. We found that anthropogenic activities and structures with the potential to alter hydrology were prominent in both natural and man-made lakes, and that the percentage of lakes with no apparent hydro-alteration potential (HydrAP rank 0) decreased between surveys in 2007 and 2012 across all ecoregions in CONUS. These patterns suggest that anthropogenic modifications to lake and watershed hydrology are prevalent in CONUS and may be increasing, evidenced by a decrease in the percentage of lakes with no human land use in their watersheds between survey years. The HydrAP framework provides a tool to integrate anthropogenic impacts in large-scale lake hydrology assessments and can be adapted as new information and data become available or to fit specific regions. With this framework future studies can examine the effects of anthropogenic hydrologic stressors on lake condition across different regional and climatic contexts.

4.2. HydrAP rank spatial patterns

The spatial patterns and associations between HydrAP ranks and lake and watershed attributes aligned with expectations. Both natural and man-made lakes in CONUS exhibited a range of hydrologic alteration potential from pristine to highly modified systems (Figure 3). This variation highlights the complexity and diversity of lake types and the limitations of using a dichotomous classification scheme based on lake presence from pre-European times that does not account for contemporary human hydrologic modifications. For example, the number of dams in CONUS increased from the 1950s to the late 1970s and augmented reservoir storage capacity in both natural and man-made lakes (Graf 1999).

In a similar study to ours, lakes from the Environmental Monitoring and Assessment Program (a probability-based lake survey in the northeastern US) were characterized based on lake origin, hydrogeomorphology, and human modifications (Whittier et al. 2002). A total of 235 lakes were grouped into the following lake classes: natural seepage and drainage lakes, highly augmented natural lakes, quarry lakes, and impoundments. Lakes were classified using best professional judgement based on detailed and time-consuming evaluations of bathymetric and topographic maps, field observations, historical records, and other available information. They found ~ 20% of natural drainage lakes in the northeastern US had a dam-like water control structure or barriers that could alter lake water levels, implying that a large proportion of natural lakes are hydrologically modified. For comparison, we found about 29% of natural lakes in the Northern Appalachians had moderate to high levels of human activities and structures that can alter hydrology (HydrAP 3 – 7), based on the 2007 NLA survey. These patterns suggest that the HydrAP ranks may capture similar attributes to the lake classes in Whittier et al. (2002), but our ranking approach is easy to replicate and apply to thousands of lakes distributed across large and diverse landscape settings using readily available geospatial datasets.

The spatial patterns in lake HydrAP ranks followed lake abundance distributions in CONUS supporting observations that humans tend to hydrologically modify natural lakes or create lakes by impoundments in drier, lake-poor areas (Whittier et al. 2002; Smith et al. 2002; Lehner and Döll 2004). In general, northern lake-rich ecoregions had higher percentages of lakes with no to minimal hydro-alteration potential; and southern, lake-poor ecoregions had higher percentages of lakes with high HydrAP ranks. For example, only about 9% of lakes in the NLA 2007 population were in the Great Plains, but over 67% of these lakes had moderate to high HydrAP ranks. As human water demands are expected to increase in arid locations, lakes with high HydrAP ranks may be vulnerable to over-exploitation.

The percent of lakes with no apparent potential for anthropogenic hydro-alteration (HydrAP = 0) decreased between 2007 and 2012 across all CONUS ecoregions, except the Appalachians. After adjusting the target population sizes to be restricted to lakes 4 ha in size to be comparable between NLA surveys, these differences persisted and were even larger in the West and Midwest ecoregions (Figure S2) indicating that changes in the percent of hydrologically undisturbed lakes were not artifacts of the change in survey design between years but may be attributed to land cover changes. Lakes that changed from HydrAP rank 0 in the resampled subset increased by one or two rank classes such that these lakes went from having no apparent human land use in their watersheds to having up to 30% agriculture + urban cover in their watersheds. These changes are supported by NLCD land cover change studies (Homer et al. 2015, 2020; Wickham et al. 2017). Between the 2006 and 2011 NLCD time periods, overall forest cover declined in CONUS particularly in the southeast, northwest, and northern parts of the country largely due to forest harvest (Homer et al. 2015). Developed area increased by 0.14% (7,631 km²) across CONUS from 2006 to 2011 (Homer et al. 2015). Landscape conversions from natural cover to human land use may be reducing the portion of lakes with no human hydro-alteration potential in CONUS.

4.3. HydrAP rank and lake and watershed characteristics

Lake HydrAP ranks were associated with lake and watershed attributes that have been used in other studies to characterize differences between natural lakes and reservoirs. These associations may be important to understand how anthropogenic stressors and changing climate conditions may impact lake integrity (Hayes et al. 2017; Doubek and Carey 2017). We found that lakes with high HydrAP ranks had large watershed-to-lake area ratios, similar to characteristics observed in reservoirs. Watershed area is a large potential source of materials and nutrients from the landscape to lakes (Hayes et al. 2017); and anthropogenic hydrologic stressors, like dams and agricultural drainage, can affect the delivery and processing of these nutrients by altering hydrologic regimes (Jones et al. 2004; Kröger et al. 2008). Examining WA:LA and HydrAP characteristics together can provide insight into nutrient and sediment loading to lakes and downstream water (Poff and Hart 2002), and potentially water quality and trophic condition. We also found HydrAP ranks were associated with lake size and depth. Lake depth and basin morphometry coupled with anthropogenic hydrologic alteration can affect internal biogeochemical processes (Coops et al. 2003; Zohary and Ostrovsky 2011) and nearshore habitat condition (Leira and Cantonati 2008; Kaufmann et al. 2014b). Recognizing these associations can help identify lakes that may be sensitive to altered hydrologic conditions and offer insight into best water management practices to minimize negative effects on lake condition (Cobbaert et al. 2015; Kann and Walker 2020).

4.4. HydrAP rank and lake hydrologic characteristics

HydrAP ranks were associated with lake water-level change and water residence time. These associations demonstrate the potential of the HydrAP ranks in capturing lake, watershed, and anthropogenic characteristics that can influence different aspects of lake hydrology. Regional differences in lake water-level patterns with HydrAP ranks are likely related to management strategies and dam purpose. Although water management records are not available for the nation, dam purpose reported in the NID may provide some insight into how lakes are used to meet regional human water needs. We characterized each of the NID dam purpose classes as either hydrologically dynamic (NID classes: “irrigation”, “hydroelectric”, “flood control”, “water supply”) or hydrologically stabilizing (NID classes: “recreation”, “navigation”, “fish and wildlife pond”). For lakes with multiple NID dam purposes recorded, we selected the first purpose to represent the dominant function of the dams. Over 80% and 40% of lakes with dams in the West and Appalachian ecoregions have dam purposes related to water release (Table S3) and may be why water-level drawdown was greater than expected in lakes with high HydrAP ranks. Irrigation was the most common dam purpose in the West and flood control was the most common in the Appalachians. In contrast, most NID dams in the Midwest (54%) and Coastal Plains (86%) were built for purposes to retain water (e.g., recreation, fish and wildlife habitat) and may explain why water-level drawdown was less than expected in lakes with high HydrAP ranks. These patterns should be interpreted with caution because the current dam purpose and management may differ from that recorded in the NID when the dams were first constructed. And water-level drawdown activities may occur during other times of the year and not be observed by the summer NLA field sampling (e.g., Carmignani and Roy 2017). But these associations provide a first look at regional variation in CONUS lake water management practices.

Relationships between lake HydrAP ranks and water residence times are similar to observations between reservoirs and natural lakes in other studies (Hayes et al. 2017; Doubek and Carey 2017). Shorter water residence times in lakes with greater anthropogenic hydrologic alterations potential may be related to water management activities with frequent water release and to correlations with lake and watershed morphology. Lakes with high anthropogenic hydro-alteration potential tended to be stream connected systems with large WA:LA. These lake and watershed characteristics are associated with shorter water turnover times compared to deep lakes without prominent stream connections.

4.5. HydrAP ranks and underlying hypothesized principles

The lake HydrAP rank assignments were driven primarily by dam height:Z_max ratios as was intended in the decision tree. Dams were hypothesized to directly control lake outflows and have the greatest potential to substantially alter lake water levels. In the CONUS lake populations, dams are prevalent with about 26% of lakes in the 2007 population and 20% of lakes in the 2012 population having dams meeting the NID criteria. Dams were ubiquitous across the five ecoregions but not evenly distributed. The percentages of lakes with dams were lowest in the Midwest (9% in 2007) and highest in the Great Plains (40% in 2007). Percentage of dams in the lake populations were similar between survey years. In CONUS, dam height: Z_max ratios ranged from 0.02 to 30 with the majority of lakes with dams having dam height: Z_max ratios greater than 1.2. These characteristics indicate that most dams on CONUS lakes can have potentially large effects on lake water level fluctuations, which have implications for water, sediment, and nutrient transport (Poff and Hart 2002; Carlisle et al. 2019), thermal stratification (Zohary and Ostrovsky 2011), and internal nutrient processing (Jeppesen et al. 2015). Conversely, these water level controls have the potential for best management practices to mimic natural water level regimes to protect lake ecological integrity and sensitive species under changing environmental conditions (Cobbaert et al. 2015; Kann and Walker 2020) but depends on how lake water levels are managed.

Land use proportions were also used to differentiate potential anthropogenic hydrologic alteration. Irrigated agriculture and drainage agriculture were expected to have large effects on lake and watershed hydrology. However, at the national extent, the proportions of these land use classes were small and not equally distributed across ecoregions. Irrigated agriculture was most prominent in the West, Great Plains and Coastal Plains, and artificial agricultural drainage was most prevalent in the Midwest. These land use classes likely did not have large effects on assigning HydrAP ranks for most lakes. But in specific ecoregions, these agricultural practices can greatly modify watershed hydrology processes and we tried to account for that in the HydrAP decision tree. Tile drainage in the Midwest can reduce water storage in soils and increase conveyance across the landscape (Fraterrigo and Downing 2008; Blann et al. 2009). These landscape modifications combined with precipitation drivers can affect nutrient transport (Kröger et al. 2008), which make them relevant to lake hydrology and water quality assessments. Similarly, irrigated agriculture can substantially alter hydrology and has been attributed to lake shrinking and desiccation patterns (Tang et al. 2009; Wine and Laronne 2020). These agricultural land use practices have reciprocal interactions with regional climate and thus can be viewed as dynamic landscape drivers of altered hydrology. Urban development, on the other hand, is a more permanent landscape modification that was ubiquitous across the five ecoregions. Lakes assigned HydrAP rank 3 tended to have large percentages of urban development in their direct drainage area, and for some lakes, urban development surrounded the lake perimeter. Impervious surfaces from roads, buildings, and pavement can alter surface runoff and infiltration patterns (Shuster et al. 2005) and ultimately affect lake water balance. Belowground pipes and drain infrastructure in urban areas may divert water across the landscape, but this information is not readily available in national geospatial datasets. Across CONUS, other region-specific land use activities could modify lake and watershed hydrology, but geospatial information characterizing these practices may be lacking. Advances in remote sensing technology and improvements in land use resolution and accuracy offer promise to characterize specific landscape modifications at broad extents using standardized methods that then could be incorporated into a framework such as HydrAP.

4.6. Limitations and improvements to HydrAP framework

The HydrAP framework to rank the potential for anthropogenic alteration on lake hydrology is limited by its reliance on national-scale geospatial data availability and quality (e.g., resolution and accuracy), and the working assumptions we assert on relationships between human infrastructure and water use. Dam attributes from the NID are important components in the HydrAP framework, but the NID dam database does not provide a comprehensive inventory of all dams in the US. The NID catalogs dams with heights and/or storage capacity that pose potential hazards under dam failure; it likely underrepresents a substantial portion of small dams across the US. In New England (US), the number of dams recorded in the NID (4,000) was substantially lower than the total number of dams from compiled state inventories (14,000) (Magilligan et al. 2016). We found that about 10% of NLA lake sites were not assigned HydrAP ranks because NID data were not available. These lakes missing NID data were smaller and shallower compared to lakes in the NID database. These patterns reflect the under-representation of small dams and reservoirs in the NID database at a national scale (Poff and Hart 2002) that in turn limits the characterization of hydro alteration potential in our HydrAP framework. This limitation is related to deficiencies in dam data at a national scale as opposed to limitations of the HydrAP conceptual framework.

Dam height information can be found from other sources such as from the US Bureau of Reclamation (USBR), but the data are limited and available mostly for larger lakes. For example, in the West, 75 of 501 NLA lakes with dams were missing dam information in the NID records. We found dam height records for 37 of these lakes in the USBR data, but these were only large lakes (median = 7.2 km²) with tall dams (median = 50 m). The remaining 38 NLA lakes missing dam data were smaller (median = 0.52 km²) and shallower (median = 4.1 m). Although the USBR database is a useful resource for dams, it is restricted to the western half of the CONUS and tends to record information on large reservoirs. Small, shallow lakes with dams are underrepresented in the NID and USBR dam databases and gathering this information may require more labor-intensive searches.

Our HydrAP rank assignments could be improved, particularly for small lakes and reservoirs, if dam inventories were developed by combining state data into a comprehensive national database. Although small reservoirs and impoundments are certainly abundant across the landscape (Smith et al. 2002; Downing et al. 2006; Fergus et al. 2020), the effects of human activities on their hydrology are poorly understood (Wisser et al. 2010). The hydrologic regimes of these small water bodies may be particularly sensitive to changing climate conditions and human water use, and lake hydrologic assessments should more explicitly include them. A comprehensive dam inventory would enable a more complete national assessment of the potential for anthropogenic modification of hydrology for lakes of all sizes.

A caution in interpreting the HydrAP framework is that it characterizes potential anthropogenic hydrologic alteration rather than actual anthropogenic hydrologic alteration. The actual effects on hydrology are determined by dynamic water management decisions that are shaped by multiple factors such as hydrology, human water demands, management trade-offs, and weather and climate conditions (Coerver et al. 2018; Song et al. 2019). In this study, we lacked information on water management directives (e.g., water release) and detailed temporal patterns in lake levels to thoroughly evaluate the effects of dams and land use activities on lake hydrology. Water management records are rarely available and would be difficult to incorporate in large-scale, multi-lake assessments. However, availability of remote sensing resources on water surface area dynamics offer promise to evaluate lake hydrologic changes over time (e.g., Pekel et al. 2016) and the accuracy of these measures will only increase with improvements in spatial and temporal resolutions.

Another consideration of how the HydrAP framework could be improved or adapted to other areas is evaluating the appropriate spatial scale to characterize land use activities and their influence on lake hydrology. In some regions, human water demands may come from outside of the lake watershed. Agricultural activities or urban populations downstream from dammed lakes may influence water release decisions and subsequently affect lake water-level fluctuations. In addition, water may be extracted from lakes directly to geographically separated watersheds through pipes and groundwater resources may be extracted and alter lake water balance (Rosenberry et al. 2015). However, information on surface and groundwater abstraction are not available with reliable spatial resolution at broad spatial extents (Döll et al. 2014). It would be challenging to incorporate these human influences that do not clearly follow conventional hydro-geophysical boundaries (e.g., upstream surface watershed) in national-scale analyses without detailed information on the water demand and transport mechanisms. This information may be available for individual lakes and could be incorporated in the HydrAP decision tree to estimate the degree of human hydrologic alteration potential for sub-region assessments.

4.7. Conclusions

Despite its potential limitations, the HydrAP framework, similar to other conceptual lake classification frameworks (Whittier et al. 2002; Poff and Hart 2002; Hayes et al. 2017), recognizes the need for more informative human hydrologic disturbance metrics to meet the challenges of assessing and understanding lake ecological condition under changing climate and human water use conditions. The HydrAP framework provides a transferrable, national-scale metric that distributes lakes along a continuum of anthropogenic hydrologic alteration potential that goes beyond a simple classification by natural and man-made lake origin. Alterations in lake hydrology continue to be a concern in lake management. HydrAP is an accessible tool to explicitly incorporate anthropogenic impacts in large-scale lake hydrologic assessments, facilitating evaluation of the relative influence of human versus natural drivers of ecologically and economically important changes in lake water levels.