Linking Watershed Scales through Altered Waterways

Nested hierarchy theory advances the idea that rivers have a fractal dimension where processes at the catchment scale (>1 km) control processes at the reach or mesoscale (100 m) and microscale (1 10 m). Largely absent from this work is a mesoscale link to the larger and smaller scales. We used stream alteration classifications to provide this link. We used orthophotographs, land cover, and LiDAR derived terrain models to classify stream alterations within four watersheds. We compared phosphorus point data with watershed, sub-watershed, and 100-meter buffers around the point data. In the predominately urban watershed, the 100 m buffer scale correlated better with phosphorus levels. In the predominately agricultural watershed, the sub-watershed scale correlated with phosphorus levels better. We found adding the classification of the stream alteration type clarified anomalously low phosphorus levels.

depth, and slope to affect changes in runoff and sediment. The RCC study focused on the linkage between organic matter inputs and watershed landuse. The RCC theory differs slightly from NHT in that RCC views the river as a smooth continuous ecosystem without discontinuities and sees heterogeneity within the system as "noise" [3]; whereas, NHT associates these discontinuities with the influence of different scales within the watershed.
NHT assumes large scale processes within a river influence smaller scale processes.
Of interest is landuse's role at the watershed and reach scale, particularly agriculture, urban, forest, and wetland. In general water quality samples usually have lower sediment and/or pollutants in regions with lower percentages of agriculture and urban.
Prior studies have determined that human alteration to the landscape yields a response across many scales [4]- [8]. For example, agricultural landuse is associated with higher volumes of sediment and nutrients than most other landuse types. High levels of sediment and nutrient levels measured at a predominantly agricultural watershed's outlet are expected. At the reach scale, the increased sediment load may increase scour and alterations to the channel's slope, depth, and width. Sedimentation at the microhabitat scale can change substrate and drastically alter the benthic zone and habitat availability for benthic invertebrates. The resulting loss of habitat and food source affects larger invertebrates at multiple scales. These changes at the smaller scale provide responses that influences peak flow, sediment storage, and erosion at downstream reaches at the watershed scale.
The net result of 30 years of projects investigating watershed scales and their relationship to landuse has not resulted in a consensus [9]. Two central issues concerning NHT and landuse remain unclear. The first issue is which land use scale (watershed or reach) provides the best predictor of water quality or ecological health. For example, how relevant is the landuse of the watershed at the reach scale. The second issue is how to use information determined at one scale to link it to others. For example, how can information be upscaled from point or reach scales to the watershed scale or downscaled from the watershed scale to point or reach scales.
Studies investigating the optimal scale of landuse to predict water quality have yielded opposing results. Research in a wide variety of landuses has found that watershed scale landuse factors may not be strongly correlated with water quality indicators at the reach or microhabitat scale. Researchers have found that riparian vegetation, bank condition, and landuse directly adjacent to the site are better indicators of water quality [3] [10]. In a comprehensive study of 79 watersheds within Minnesota, Wisconsin, and Michigan, Wang et al. (2003) found that the range of biotic and abiotic indicators collected at the watershed and riparian scales had only a baseline impact on the prediction of fish assemblages while data collected at the point scale yielded a prediction [7].
Other studies suggest that watershed scale factors play an important role in understanding the smaller scales [11]. In a study including 72 sampling sites in Belize, Esselman and Allan (2010) found data at the watershed scale explained the variation in fish assemblages better than data collected at the reach scale (mesoscale) [12].
Some papers suggest that the confounding factor reducing the efficacy of a landuse evaluation of water quality and habitat is the location of environmental factors [6]- [9].
Controls, such as impoundments, culverts, and bridges, impact upstream reaches differently than downstream reaches. Both segments suffer from a lack of flow variability, but upstream segments are more prone to flooding and sedimentation, while downstream regions are sediment-starved with low flows.
In a comprehensive review of the landscapes and their interactions with streams, Allen, 2004 identifies and summarizes six principle environmental factors that affect rivers: 1) sedimentation; 2) nutrient enrichment; 3) contaminant pollution; 4) hydrologic alteration; 5) riparian clearing and opening; and 6) loss of large woody debris [13].
These environmental factors are often correlated with types of landuse, particularly urban and agricultural landuse, which have a high degree of human impact. Of the six, hydrologic alteration can be identified and classified using Geographical Information Systems (GIS) techniques. Hydrologic alterations are also highly correlated with reaches that exhibit the other environmental factors. Indications where stream channels have been altered or modified due to anthropogenic changes can assist in providing the mesoscale or reach scale link to the larger and smaller scales and may explain the inconsistencies in studies that focus on the watershed and micro scales.
Management of watersheds, especially those impacted by anthropogenic landuse (e.g. agricultural and urban) is driven by a need to maintain water quality at set levels and to manage sampling costs. Predictive water quality models using landuse characteristics at the watershed scale often fail to target specific regions within the watershed requiring remediation [9].
In this paper, we identify and classify hydraulic alterations using spatial data. Our hypothesis is twofold. First, we expect a higher percentage of altered waterways will be found in watersheds with higher amounts of anthropogenic landuse; and second, that water sampling points will correlate best with landuse buffers from the immediate location. We also expect that anomalously low or high readings will be better accounted for by their hydraulic classification. We define our research scales as follows: large scale (watersheds: approximately 130 km 2 ), intermediate scales or mesoscales (sub-watershed (10 -0.05) km 2 and stream segments 10 km), and microscale (100-meter buffers around water quality sampling points, 0.1 km 2 ).

Site Description
The selected hydrologic unit code 12 (HUC12) watersheds reside in four regionally dif- Creek is predominantly forest and woody wetland in Iron and Ashland counties. West

Altered Stream Data Acquisition
Data used for classification of altered hydrology came from state and federal sources.
An existing geospatial stream vector file from the WDNR [16] [17]. Lake data was also obtained from the WDNR (WDNR, 2007).
One-meter Light Detection and Ranging (LiDAR) data was downloaded from the Wisconsin View website and extracted (Wisconsin View, 2011). Hill shade was derived from one-meter LiDAR data. Ten-meter digital elevation data was used for one of the watersheds devoid of LiDAR data. The United States Department of Agriculture's National Agricultural Imaging Program (NAIP) 2013 aerial imagery was also obtained from the Wisconsin View website for Crawford and Rock Counties [18]. Imagery from 1938 was downloaded from the Wisconsin Historical Aerial Image Finder [19]. The

Altered Stream Data Pre-Classification
Prior to determining altered hydrology, fields were populated in the stream vector feature class table. Altered water course type (AWC-type) with domains: (1) Altered, (2) Natural, (3) Impounded, and (4) No Definable Channel. Sinuosity was also added and calculated using the ArcPy Sinuosity tool (ArcGIS ver 10.2). This takes the distance along all vertices on the stream channel from start point to endpoint and divides by the straight-line distance between the two points. A sinuosity value approaching zero represents more sinuous while a value approaching one is less sinuous or straighter.
Each stream segment was assessed separately. If a stream segment lies in two different classifications, the segment would be split and classified appropriately. For example, a long stream segment that crossed into an agriculturally dominated area and became less sinuous would be classified as altered. If the other portion of the same segment were more sinuous and resided in a forest-dominated area with no apparent anthropogenic changes, this portion would be classified as natural. Classification of altered hydrology followed the Minnesota Geospatial Information Office (MnGeo) for altered watercourse framework [24]. MnGeo prepared a statewide-altered hydrology geospatial layer for the Minnesota Pollution Control Agency (MPCA) using the USGS National Hydrography Dataset (NHD) flow lines. The MPCA used this dataset for water quality monitoring and assessment programs used to provide information about stream habitat. Although classification closely followed the altered watercourse determinations by MnGeo, some modifications were necessary for this study. MnGeo's classification only included streams that exceeded 150 meters in length, while this study analyzes streams of any length. Instead of using the USGS NHD flow line vector and its corresponding Hydro Event Manager Toolbar for editing, the Wisconsin hydrology dataset was used with basic ArcMap editing.

Altered Stream Data Classification: Altered
In order to classify a stream reach as altered, adjacency to agricultural operations, sinuosity, and general anthropogenic alterations were assessed. Each stream flagged as potentially altered was compared to certain acquired spatial layers.  prior aerial imagery are unaltered even though increased agriculture is present. We determined if channel sinuosity is retained and there are no structures or evident changes, the stream segment is classified as natural. Stream segments that closely border a road or other obstructions such as water and sediment control basins are classified as altered.
Stream segments with forested regions around them but are adjacent to roads are also considered altered because roads change the channel's natural flow path ( Figure 4).

Altered Stream Data Classification: Natural
Sinuosity in headwater streams was less important in steeper gradients. Headwater streams in high relief regions do not have the hydraulic capability of creating meandering channels. Stream segments adjacent to an expanse of forest and distant from agricultural impacts and any manmade structures were indefinitely classified as natural ( Figure 5). However, a stream reach surrounded by agriculture can still be considered natural. If the stream resided in an actively farmed region, sinuosity would be analyzed from previous aerial imagery ( Figure 6). Higher order streams, such as Milwaukee River, exhibit a decreased sinuous channel. A decrease in sinuosity can be attributed to higher discharge resulting in a less defined channel [25]. Even though heavy urban landuse surrounds the stream, structures are built far enough away to not alter flow.
Wetlands information was accessed by NWI wetlands inventory and was draped over the stream layer. Noticeable wetlands that surround a stream and are considered functional were classified as natural. However, indications of a "dried-up" wetland are considered altered [24]. Wetlands placed on a ditched watercourse to remove excess nutrients or improve habitat that essentially dry-up wetland are not considered a natural phenomenon.

Altered Stream Data Classification: Impounded
Water body features that interrupt river systems are often anthropogenic. Using the Wisconsin lakes data and prior aerial imagery, impounded watercourses were identified. Water body features that were not observed in 1938 aerial imagery but were found

Altered Stream Data Classification: No Definable Channel
Streams delineated from historic maps and aerial imagery can change course or disappear altogether over time. It is often hard to define any path that these streams take which in turn makes them harder to classify. Stream vectors that cross tillage land with no distinct flow path are considered to have no definable channel. Even if a stream looks distinct and well definable from an aerial imageries view, analyzing LiDAR data can assist in determining if a watercourse exists. The non-existent stream that is not definable through LiDAR draped over hill shade is classified as no definable channel.
Another indicator of a lack of definable channel is a stream's path in relation to tillage.
If a stream's path is perpendicular to tillage lines, the effective drainage direction of that reach is destroyed. A path is no longer definable from either aerial imagery or LiDAR.

Water Quality Data
Water quality sampling data were obtained from the WDNR sampling and volunteer monitoring program and Environmental Protection Agency (EPA) Storage and Retrieval and Water Quality Exchange (STORET) and combined into a statewide database [26]. Data points were collected from 1961 to 2015, with the majority of data collected in the last three years. Within the data set is a wide array of water quality indicators.
We chose total phosphorus (mg/l) as a water quality indicator because it is often paired with landuse and the abundance of the data for this indicator [4].

Watershed Scale
We divided Milwaukee and Bass-Stevens watersheds into sub-watersheds using the sampling points as outlets. The watersheds were delineated using Arc Map 10.2's hydrology toolbox. For Milwaukee River, the sub-watersheds ranged from 8 to 104 km 2 .
For Bass-Stevens Creek watershed, the sub-watersheds ranged from 0.02 to 5.6 km 2 . We designate these as "watershed scale". We also used the Arc Map 10.2's buffer tool to create a 100 m buffer around each point. We designate these as "reach" scale. For each sub-watershed and 100 m buffer, we extracted landuse [21] percentages and compared them to Storage and Retrieval and Water Quality Exchange's (STORET) average total phosphorus (mg/L).

Statistics
We compared total phosphorus from the point data to major landuse within the subwatersheds and the 100 m buffers. The Mann Kendall trend test, a nonparametric statistical hypothesis, was used to evaluate the significance of upward or downward

Altered Waterways
USGS 2011 and USDA Cropland 2014 data percentages were extracted for each watershed ( Table 1). The percentage of altered waterways were determined for each watershed and summarized in Figure 11. For the agriculturally dominated watershed, Bass-Stevens Creek, the results showed a greater percentage of altered streams (95 per-   Fork Knapp Creek watershed has greater deciduous forest cover (63 percent) than the other watersheds based on USDA Cropland 2014 [22] and USGS 2011 [21] and also the highest percentage of natural streams (65 percent). Swamp Creek, which is mainly deciduous forest and woody wetland land cover had the highest percent of natural streams (96 percent). A small percentage of Swamp Creek's hydrology (4 percent) were classified as impounded. The Milwaukee River has 80 percent of its landuse as urban developed. The watershed contained the most diverse waterways with 34 percent of classified as altered, 32 percent classified as natural, 20 percent classified as impounded, and 14 percent classified as no definable channel.

Phosphorus Landuse and Altered Waterways
As summarized in the methods section, the Milwaukee River (n = 25) and Bass-Stevens Creek (n = 8) watershed has several sampling points with total phosphorus (mg/L) data collected mostly during the last five years. The average of the data points for the Milwaukee River total phosphorus is 0.2 with a range of 0.09 to 0.47 mg/L. The average for the Bass-Stevens watershed was 0.094 with a range of 0.04 to 0.16 mg/L.

Sub-Watersheds and Landuse
The twenty-five sub-watersheds of the Milwaukee River watershed's landuse (ranging could not be used due to insufficient number of points (ten or more points are required). In Figure 12, the relationship between total phosphorus versus sub-watershed cultivated crop landuse percent has a correlation coefficient of 0.57, which indicates that approximately 60 percent of the linear model presented is explained.

Reach Scale (100-m Buffer) and Landuse
For the Milwaukee River watershed, the 100-m buffers yielded a different story, perhaps in part due to a much higher heterogeneity, especially in the developed landuse percent than the sub-watershed. The landuse ranges percentages for the 100-m buffers are as Around approximately 80 percent developed landuse and higher, the water quality sampling points show anomalously lower levels of total phosphorus Figure 13(a).
These points were found on natural waterways. We removed the points with greater than eighty percent developed landuse. In Figure 13

Landuse and Altered Waterways
Overall, the more anthropogenic landuse found within a watershed, the higher the amount of altered waterway classification ( Figure 11). It is a little surprising to find the forested watershed (West Fork Knapp Creek) with over thirty percent of the waterways as altered, but it is unclear how much of that watershed supported logging activities that were prevalent in the 1800 s in Wisconsin. The agricultural (Bass-Stevens) watershed's results conform to assumptions associated with agricultural practices such as the straightening of streams [5]. The wetland (Swamp Creek) watershed waterway analysis results (less than 5 percent altered) conforms with studies of low anthropogenic impact [13]. The urban watershed (Milwaukee River) results, with the most heterogeneous classification of the waterways, may be indicative of heavily urban areas.

Altered Waterways Link between Sub-Watershed and Reach Scale
The Milwaukee River's is a predominately urban watershed. The 100-m buffer developed landuse percent ranged from 0 -100 percent. The eleven points in the range from 80 -100 percent developed do not follow the increasing total phosphorus trend found in the other data points (between 0 -80 percent developed). Six of those eleven points are on first order streams near the headwaters of the watershed (see Figure 9). The other five points are close to the main watershed's outlet and have over 95 percent developed landuse; therefore, we would expect them to have high total phosphorus levels.
What is unique about these points is that they fall on or near the natural portion of the streams (see Figure 9). Both the position of the points within the watershed and the type of waterway that the point is on help to explain all of the points' significantly lower total phosphorus levels.
In the case of Bass-Stevens Creek, landuse within the 100-m buffer point does not have a statistically significant relationship with total phosphorus. Rather, the sub-watershed scale correlates better than the 100-meter buffer scale for cultivated crops versus total phosphorus. This difference from the Milwaukee River may be for several reasons. The sub-watersheds for Bass-Stevens Creek are smaller, where the largest sub-watershed (5.6 km 2 ) was smaller than all of the Milwaukee River sub-watersheds. Unlike the Milwaukee River, Bass Steven's landuse is more homogenous. In Bass-Stevens Creek 90 percent of the landuse is associated with agriculture (where pasture and hay are considered agricultural landuse) while the Milwaukee River watershed is 80 percent developed (including low intensity development). Another difference between the two watersheds is the type of hydrologic classification. Bass-Stevens Creek is more homogenous based on classification, with approximately 70 percent altered and only 5 percent natural. The Milwaukee River has over 30 percent of its streams classified as natural. Finally, all of the Bass-Stevens Creek STORET sampling points (n = 8) were located onhydrologically altered waterways and several (n = 5) of Milwaukee River were on natural streams.

Conclusions
Altered waterways are highly correlated with watersheds that have more anthropogenic landuse. While this result confirms what is well documented in the literature, it also gives us additional information as to why larger scales (sub-watershed) landuse may not correlate well with water quality data such as phosphorus levels. In this paper, we found the Milwaukee River reach scale (100-m buffer) had a statically significant relationship between the dominant landuse (urban) and STORET phosphorus point data. In Stevens-Bass Creek, the sub-watershed scale showed a statistically significant relationship between the dominant landuse (agriculture) and phosphorus. Other studies have shown either a mix of scales to better correlate with water quality indicators [30] depending upon the indicator, while other studies found specific scales to be the best indicators [12] [31].
The Bass-Stevens Creek watershed finding of catchment scale landuse having a higher correlation to water quality indicators is similar to Silva and Williams (2013) who compared two scales (100-m buffer zones and whole catchment scale (sub-watershed) in a 332 km watershed of diverse landuse (forest, agricultural, and urban) [31]. They found the catchment scale more useful to predict water quality indicators [31]. Of note is the 100-m buffers landuse percentages were homogenous and the dominant landuse was less than 5 percent for most watersheds. By contrast, the catchment scale landuse was more heterogeneous, and had a wider range of the dominant landuse found in the watershed. Esselman and Allan (2010) also found at the catchment (sub-watershed) scale, abiotic factors were more statistically significant than at the reach scale (100-m buffer) scale for fish assemblages [12]. Esselman and Allan (2010) also note that the reach scale is still important and explained about 30 -50 percent of the variance within their models [12]. Silvia and Williams (2013) [31] study found a similar finding to Esselman and Allan (2010). In the Silvia and Williams (2013) study, the correlation coefficient for reach (100-m buffer) and catchment (sub-watershed) scale factors differed by 10 percent [31]. The reach scale (100-m buffer) was not insignificant, just not as significant as catchment.
In this study we examined two watersheds with significant anthropogenic impacts to both landuse and the stream channels themselves, but we found differences in which scale explained the total phosphorus levels better. Bass-Stevens Creek and the Milwaukee River both have over 80 percent landuse that is associated with human alteration, but the Milwaukee River has 30 percent of its channels which are considered natural, by comparison to Bass-Stevens Creek, which has nearly all channels altered. In contrast to the Milwaukee River, all sampling points for Bass-Stevens were on the same type of stream alteration. The homogeneity of the Milwaukee River data combined with the observation that lower phosphorus levels were found along the natural portions of the river may suggest that alteration type is a factor that should be considered.
This study uses remote sensing and filtering techniques to identify stream alterations.
Without using a GIS method developed by MNGeo (2011) [25] or one similar, the identification of stream alteration may have been difficult or impossible. Additionally, this method provides a method to link the larger sub-watershed scale to the reach scale and may address a way to remove autocorrelation associated with stream alteration and anthropogenic landuse.