REVISED MS - MULTISCALE GEOSPATIAL ANALYSES - EMA.DOC

Abstract River science and management can be conducted at a range of spatiotemporal scales from reach to basin levels as long as the project goals and questions are matched correctly wit

Critical Role for Hierarchical Geospatial Analyses in the Design of Fluvial Research, Assessment, and Management

James H Thorpa, Joseph E Flotemerschb, Bradley S Williamsa, and Laura A Gabanskic

a Kansas Biological Survey and Department of Ecology and Evolutionary Biology, University of

Kansas, Lawrence, KS, USA

b U.S Environmental Protection Agency, Office of Research and Development, Cincinnati, OH,

USA

c U.S EPA, Office of Water, Washington, D.C., USA

Corresponding Author: Prof James H Thorp

Kansas Biological Survey and Department of Ecology and Evolutionary Biology, University of Kansas, Higuchi Hall, 2101 Constant Ave., Lawrence, KS 66047-3759 USA

Email: thorp@ku.edu; Tel: 1-785-864-1532; Fax: 1-785-864-1534

Keywords: aquatic ecoregions • functional process zones • GAP analysis • hydrogeomorphic

patches • riverine ecosystem synthesis • stream classification assessments

Abstract River science and management can be conducted at a range of spatiotemporal scales from reach to basin levels as long as the project goals and questions are matched correctly with the study design’s spatiotemporal scales and dependent variables These project goals should also incorporate information on the hydrogeomorphically patchy nature of rivers which isonly partially predictable from a river’s headwaters to its terminus This patchiness significantly affects a river’s habitat template, and thus community structure, ecosystem function, and

responses to perturbations

Our manuscript is designed for use by entry-level river scientists through senior administrators at government agencies It analyzes common challenges in project design and recommends solutions based partially on hierarchical analyses that combine geographic

information systems (GIS) and multivariate statistical analysis to enable self-emergence of a stream’s patchy structure These approaches are useful at all spatial levels and can vary from primary reliance on geospatial techniques at the valley level to a greater dependence on field-based measurements and expert opinion at the reach level Comparative uses of functional process zones (FPZs = valley-scale hydrogeomorphic patches), ecoregions, hydrologic unit codes(HUCs), and reaches in project designs are discussed along with other comparative approaches for stream classification and analysis of species distributions (e.g., GAP analysis) Use of

hierarchical classification of patch structure for sample stratification, reference site selection, ecosystem services, rehabilitation, and mitigation are briefly explored

Policies for managing watersheds, establishing or enforcing environmental policies, and

rehabilitating riverine landscapes have inherent assumptions that the grain and extent of the spatiotemporal sampling scale employed and the data variables selected are appropriate to their study goals (cf Kondolf 1998) They may also assume implicitly while selecting monitoring sitesthat the physical characteristics of the river channel and its valley change continuously and predictably from headwaters to the river terminus Too often, however, these vital assumptions are never critically examined, with the result that many studies are conducted at questionable spatiotemporal scales, using less compatible dependent variables (cf Williams et al 1997; Dollar

et al 2007), and ignoring river patchiness while introducing systematic sampling errors (Thorp et

al 2006, 2008) Our manuscript’s goals are to help improve the protection, management, and rehabilitation of “rivers”, which are defined here as all fluvial systems from headwater streams togreat rivers

Common spatiotemporal scales and designs in river assessment and management

Historical focus in bioassessment and management

Since passage of U.S federal legislation mandating improvements in water quality (e.g., the Federal Water Pollution Control Amendments of 1972), bioassessment approaches have usually concentrated on smaller streams (e.g., Barbour et al 1999) The primary independent variables were often stratified by stream order and derived from field-based habitat analyses of reaches,

such as the still popular Rosgen Method in the USA (Rosgen 1994, 1996) and the River Styles approach in Australia (e.g., Brierley et al 2002) On-site hydrogeomorphic analyses were often combined with land use and coverage data obtained from aerial or satellite imagery The primary dependent variables reflected the smaller extent and grain of the spatial scale and focused

initially on physicochemical measurements of water quality and later on biotic metrics,

especially related to easily collected macroinvertebrates Later variables included the abundance

and diversity of fish (longer-lived organisms with greater home ranges) and Chl-a biomass

estimates of algae (rapid turnover)

With the transition to a new century, scientists and government agencies (e.g Newson 1992; Gardiner et al 1994; Council of the European Union 1999; Hooper 2005) increasingly recognized the need to expand the diversity of monitoring and assessment from a primary focus

at the reach level to larger (grain and extent) spatial scales needed for watershed management (Tarlock 2008; Thorp et al 2008; Heathcote 2009; Southerland et al 2009) As a partial result, government agencies are increasingly developing sampling protocols for large rivers (Humphries

et al 1998; Parsons et al 2004; Flotemersch et al 2011), but a consensus on ideal protocols for rivers lags behind those for shallow streams Protocols for sampling non-wadeable rivers were not even published by the U.S Environmental Protection Agency (EPA) until the current century(Lazorchak et al 2000)

Typical bioassessment project designs

The motivation for quantitative bioassessments at basin-wide scales probably stems from both legal mandates requiring watershed assessments and a growing recognition of threats from multiple and co-varying stressors (e.g., pollutants, habitat destruction, altered flow regimes, channel modifications, invasive species, and climate change) Bioassessment allows evaluation

of a waterbody’s biological integrity, as defined by the ability to support and maintain a

balanced, integrated, and adaptive community having a biological diversity, composition, and functional organization comparable to those of natural aquatic ecosystems in the region (Frey 1977; Karr and Dudley 1981; Karr et al 1986)

Bioassessments occur at spatial scales ranging from targeted site-specific sampling for particular program needs (e.g., compliance or stressor-specific studies of the EPA's National Pollutant Discharge Elimination System) to sampling at national scales, such as EPA’s National Aquatic Resource Surveys (USEPA 2010) When assessment objectives emphasize a spatial extent beyond individual sites, sample surveys using stratified random site selection help accountfor spatial variability and control bias, and they contribute to more objective statements of area-wide conditions (Larsen 1997; Urquhart et al 1998; Larsen et al 2007) Common sample area categories used in such survey designs include those based on ecoregions, physiographic

provinces, vegetative classes, stream size or order, and other natural biogeographic factors directly influencing ecosystem structure, with the spatial grain and extent of sample units ideally approximating information needs, as defined by assessment objectives (Barbour et al 1999; Flotemersch et al 2006) Sampling has also been stratified by political boundaries (e.g., U.S EPA’s Missouri River EMAP)

Although biomonitoring has mostly occurred at the reach level in wadeable streams, reach definitions and boundaries are often arbitrary (Frissel et al 1986; Flotemersch et al 2011) Nineteenth-century river boat captains defined reaches as straight stretches of variable length

between river bends (see Life on the Mississippi by the former riverboat pilot Samuel L Clemens

[Mark Twain]) To 21st century lotic scientists, a reach could be, for example: (a) a repeatable stream unit (e.g., pool-riffle sequences); (b) a specified number of wetted channel widths (e.g., 40); (c) the distance between two minimum-sized tributaries; (d) a consistent, pre-selected stream length around a mid-point determined from a geospatial grid; or (e) an operationally defined, arbitrary sample segment length for a specific study Reaches can also be defined more precisely as a stream length between substantial geomorphic breaks in channel slope, local side-slopes, valley floor width, riparian vegetation, and bank material (e.g., Frissell et al 1986), although this is a more time-intensive and expensive approach While this variability in reach definitions poses challenges for extrapolating among projects and sites (especially among

hydrogeomorphically distinct areas), the reach stream unit is useful for describing medium- and long-term effects of human activities (Frissell et al 1986), at least for local disturbances or rehabilitation projects

Matching the appropriate concepts, scales, and variables in study designs

River concepts influence study design

River science progresses by building and improving on past conceptual models and empirical studies, with their associated sampling strategies A decade into this new millennium, most

published environmental research and assessment studies seem to have assumed that physical conditions change gradually and mostly continuously from headwaters to the mouth, as proposed

by the river continuum concept, or RCC (Vannote et al 1980) Consequently, past study designs incorporated random sampling or data collection stratified by linear downstream segments, as has been employed in many studies including the Missouri River EMAP project (Schweiger et al.2005) A reasonable assumption that followed from the RCC’s important but scale-free

conceptual model was that adjacent areas were more similar to each other than they were to moredistant patches (but see Poole 2002) Many monitoring studies also focused on the main channel and intentionally ignored lateral, and often species rich areas of the riverine landscape to

simplify sampling strategies, data interpretation, and costs

Conceptual ecological studies in the last two decades and access to high resolution data atthe basin scale (Fonstad and Marcus 2010) are altering our perspective of rivers from linear, main channel systems to complex, 4-dimensional riverine landscapes (Ward 1989) composed of

a hierarchically nested, hydrogeomorphic patch structure (e.g., Montgomery 1999; Poole 2002, Carbonneau et al 2011) Rivers can also be considered as “directionally nested networks” (Melles et al 2011) to account for differences in hierarchical interactions within upstream vs downstream components of the river network These hierarchical patches may be repeatable and only partially predictable in position according to the riverine ecosystem synthesis (Thorp et al

2006, 2008) This patch perspective engendered the hypothesis that ecosystem structure,

function, and services will be more comparable in two distantly separated patches of the same hydrogeomorphic type than in adjacent patches with different physical structure (Poole 2002; Thorp et al 2006, 2008, 2010) Modern landscape perspectives also emphasize the importance of

considering the two major components of the riverine landscape: the riverscape (sub-flood components, including main channel, side channels, and the mostly isolated backwaters; cf Leopold and Marchand 1968; Wiens 2002) and the floodscape (floodplain lakes, isolated

oxbows, wetlands, and the normally terrestrial floodplains; Thorp et al 2008)

Effective river management demands an appreciation of the system’s hierarchical organization in order to select proper variables and sampling designs For example, a river scientist needs to understand how effects of biotic or physicochemical processes on system functioning vary with the study’s focal scale (Parsons et al 2004; Dollar et al 2007) Past studieswere often shaped by widespread views of rivers as continuous gradients of physical structure and by limited access to geospatial techniques and models that would enable the project designer

to understand better the river’s patch structure In some cases, investigators can reinterpret old sampling data using new hydrogeomorphic analyses as long as the sampled variables conformed

to the correct spatiotemporal scales

Examples of mismatched spatiotemporal scales

A mismatch of scales, ecological processes, and dependent variables may occur if dependent variables are not simultaneously expanded in scale with increases in the physical area sampled and the temporal scale of the research or management question This problem most often occurs when reach scale analyses are employed to answer valley or basin scale questions The followingfour examples illustrate some of the many areas where mismatches may occur

Home range: Reach-level monitoring studies may be appropriate only when the

dependent variables are equally limited in scale For example, if the spatial extent of

environmental impacts is local, it is appropriate to measure the density and diversity is of benthicinvertebrates because these organisms generally have very limited home ranges In contrast, mostfish species have larger home ranges than invertebrates, and some migrate long distances,

especially diadromous species Therefore, measurement of fish species at the reach level may be problematic in terms of cause-and-effect determinations (cf Fausch et al 2002; Boys and Thoms2006; Springe et al 2006) unless the species in question has a reach-scale home range

Regulation by density independent disturbances: A temporal mismatch can easily result if

researchers focus on short-term processes affecting species diversity and density because the community may be subject to periodic reset from natural, density independent processes These include, in particular, hydrologic flow and flood pulses as well as droughts that reduce or

eliminate discharge, especially when surface water channels dry to isolated pools, as in many arid-zone rivers of Australia (e.g., Bunn et al 2006; Feld and Hering 2007)

Measurement and interpretation of ecosystem processes: If a study focuses on responses

of ecosystem function (e.g., system metabolism and nutrient spiraling) to disturbances, then system-level sampling that is restricted to, or interpreted for the reach level is misleading at best, especially if the spatiotemporal scales do not encompass major flow fluctuations and lateral connections with floodscape and riverscape slackwater components Reach-level perturbations can alter local measurements, but information on upstream and lateral processes are needed to adequately interpret disturbance to ecosystem function (e.g., Feld and Hering 2007)

Watershed impacts: River scientists and managers have long recognized the importance

of terrestrial components of watersheds to community structure and ecosystem function While the local riparian zone can alter reach-level structure and function, the larger floodscape and the local terrestrial basin (i.e., floodscape plus valley side-slopes) exert their influences much farther up- and downstream from local sample areas and over longer periods through the flow of water, sediment, and organic matter

Recommended improvements to study design

Hierarchical designs for a hierarchical world

Designing and conducting hierarchical studies require a multi-stage process which proceeds fromgoal establishment, through determining appropriate spatiotemporal scales and data variables, to final conclusions – with opportunities for reassessment of scales and questions as more

knowledge is gained An example of this step-by-step process, Figure 1 shows a decision tree which uses ecosystem management of riverine Least Terns (a federally endangered shore bird) Some factors influencing study design are described below

Every activity within a riverine landscape has a primary governing domain (e.g., scientific, policy/management, or conservation) with various organizational categories and levels For instance, the river management domain could include functional levels related to policy development, implementation, monitoring/assessment, and enforcement (e.g., see the first two steps in Figure 1) These functional levels are often associated with specific hierarchical spatial levels, most of which can be defined hydrogeomorphically, as shown in Table 1 for a

portion of the Mississippi River Such levels can also be linked in the sample design to

freshwater (Table 1) and terrestrial ecoregions Different geomorphic levels (e.g., macrohabitat tobasin) have corresponding hydraulic and hydrologic processes (e.g., fluid mechanics to floods) operating over different time scales (e.g., seconds to 100 yr or more) (Fig 2) The appropriate dependent variables for ecological structure and functions (Table 2) are also hierarchically nestedwithin landscapes, ecosystems, communities, population demes, and individual organisms

The management focus and core research question prescribe a project’s appropriate spatial level for analysis (Fig 2) Data should be collected to characterize the next

hydrogeomophic level below the focal level of the question (O’Neill et al 1989; Dollar et al 2007) For example, if basin management is the goal, then assessment should initially occur at the basin’s first constituent stratum: the valley-to-reach level or “functional process zone” (= FPZ; Thorp et al 2006, 2008) An individual FPZ is then characterized by averaging variables from representative reaches If the goal were instead to manage a small, homogenous watershed comprised of a single FPZ type, then reaches are the units of measure and dependent variables constituting the overall reach value should be averaged from samples within functional units (e.g., pools or riffles)

Once study questions and spatiotemporal levels are identified, the investigator can select the independent and dependent variables matching the study’s hierarchical level Unfortunately, acommon error is to collect and merge fine grain and small extent data and then apply the

summarized results at coarser grain and larger extent scales without considering whether the smaller scale metrics are still appropriate to the question Some examples of these mismatches ofscale were discussed above Table 2 includes recommendations for types of data to collect at

different hierarchical levels and components of a riverine landscape, but the project goals and specific questions being addressed will influence the choice of variables.

Hierarchical, geospatial analysis at valley and reach levels

Significant systematic errors can hide within data sets when investigators fail to account for the system’s patch structure at valley or reach levels For example, a stratified random monitoring program based on expectations of continuous physical change from upstream to downstream in awatershed will encounter systematic errors if the investigator ignores that some patches are in braided patches and others in meandering patches and the patches are not distributed in an entirely predictable way (Thorp et al 2006, 2008) This results because the community similarityreflects the nature of the patch at least as much as the upstream-downstream position (cf Poole 2002) Fortunately, geospatial and statistical techniques now allow the river’s physical structure

to self-emerge statistically at different spatial levels with reduced chances of human bias or need

to match sections of riverine landscapes with a general manual’s pre-determined categories (see discussion in Williams et al., in review) By asking questions at a valid spatiotemporal scale and using independent variables appropriate for the selected spatial levels, one can detect differences among river sections and incorporate this information into study designs to reduce systematic error

GIS-based programs for classifying hydrogeomorphic patch structure are currently available at the FPZ level, and one approach1 is summarized below; similar techniques are being developed for reaches For FPZ-level analyses of U.S rivers, the technique relies entirely on freely available data Rivers in other countries can be analyzed using similar techniques (e.g., Rayburg et al 2007, 2008 for Australian rivers), with the possible limits being data precision (especially surface elevation) and availability (e.g., coverage of geological and precipitation data) For example, one can use DEMs (digital elevation model data) with 10-90 m resolution (we generally use 10 m DEMs for medium to large rivers) or the more precise LIDAR (light detection and ranging) data with resolutions of 0.5-3 m In much of the developing and under-developed world, 90 m DEMs derived from overflights of the Shuttle Radar Topography Mission(SRTM) may be the best option To resolve channel features in smaller streams, LIDAR data should be used if available

Reach-level analyses require a different suite of finer grain variables (e.g., bedform composition), some of which must be collected in the field rather than remotely sensed or

accessed from existing databases However, higher resolution remote sensing techniques are being developed which show promise for analyzing reach patch structure (e.g., Fonstad and Marcus 2005) in some stream types (Carbonneau et al 2011) Gross measures of substrate size and fine grain determinations of channel form might be obtained from a series of reaches with helicopters and drones to supplement on-ground field surveys The focus on intensive field data collections makes reach-scale analyses more expensive per unit area than at the FPZ level, but

1Free copies of our hydrogeomorphic structural analyses program “RESonate” for ArcGIS® 10 and instruction manuals are available from James H Thorp or Bradley S Williams

only a small portion of the total basin is typically analyzed We recommend a greater focus on statistical self-emergence to delineate groups of similar reaches This would reduce both the reliance on occasionally problematic “expert opinion” (cf., Davies et al 2010) and the

categorization of rivers into broad types based on extrapolation of data collected at one field site and applied to many others

This is not to imply that stream classification approaches relying primarily on satellite and aerial imagery and measurements are without problems (e.g., radial distortion of imagery) However, the limits are generally known and correctable, and the approach is more efficient and less expensive when working at larger spatial scales Moreover, the imagery is characterized by greater geometric accuracy and spatial resolution and than many related spatial measurements made on the ground

Summary of patch analysis at the functional process zone level

GIS-based analyses of a river’s hierarchical patch structure are cost-effective techniques at the large spatial scales required for basin management, and they can be integrated with, and

supplement current approaches in environmental analysis (Rayburg et al 2007, 2008; Thorp et

al 2008) Our quantitative, hierarchical classification program (RESonate) currently incorporates13-15 relevant hydrogeomorphic variables at three spatial scales which bracket the scale of basinmanagement interest (e.g., the valley or FPZ level) The appropriate variables used to distinguishFPZs may vary slightly among major river types and biomes Automated routines in ArcGIS®enable rapid measurement of these variables in equally spaced sample segments along the entire

course of a river network We typically extract data at 10-km segments for basins of at least 50,000 km2 but may employ more appropriate 5-km segments for smaller basins File size and data processing time increase with a reduction in segment length, which may be a limiting factor for the investigator Additional automated GIS routines organize the data into a sample segment

by hydrogeomorphic data matrix that can be imported into various multivariate statistical

analysis packages Multivariate cluster analysis is then used to identify distinct groups of sample segments with similar hydrogeomorphic character (i.e., different types of FPZs) This

classification procedure differs from most in that it allows relevant groups to self-emerge from

data measured at the most appropriate scales as opposed to placing sections of a river into a priori groups based on surveys of other river systems or expert opinion of varying quality The

statistical significance and the hydrogeomorphic properties of identified groups can be further assessed with ordinations and additional discriminate analyses

Once groups have been identified and statistically validated, individual sample segments can be coded by group and mapped over the river network in GIS A subset of representatives from each FPZ group can then be "ground truthed" using on-the-ground surveys and/or virtual reconnaissance with remote sensing imagery Statistical procedures only generate numerical groups and, thus, later application of names to the groups based on their hydrogeomorphic character may facilitate communication among scientists, managers, and policy makers

A small but increasing number of rivers have been typed on multiple continents using these top-down, newly developed ArcGIS programs by research groups in the USA (a

collaboration between J H Thorp’s team at the University of Kansas and J.E Flotemersch’s group at EPA in Cincinnati) and Australia (M C Thoms’ team at the University of New

England) Results from these studies show that the number and nature of the FPZs vary

considerably among river basins in response to the structural nature of the encompassing riverinelandscape Figure 3 illustrates this for the highly contorted and constrained channels in the Kanawha River Basin located primarily in the Appalachian Mountains of the eastern USA

A related software approach for characterizing patches, known as the Fluvial Information System, has been developed by P Carbonneau and S J Dugdale in collaboration with APEM Ltd and operates within the MATLAB® environment It is reviewed in Carbonneau et al (2011)

Ecoregions, HUCs, FPZs, and reaches

A common question in discussions on integrating a river’s hydrogeomorphic structure into research and monitoring studies is how FPZs relate to ecoregions, HUCs (hydrologic unit code), and stream reaches, all of which are commonly employed in project design and analysis While these four terms are based on different environmental features, they can be employed to some extent in a single analysis to provide additional information to the investigator

Ecoregions are relatively large areas of land or water containing identifiable and partially distinct, natural communities (Ricketts et al 1999) The purpose of ecoregional classifications is

to minimize within-group variability and maximize among-group variability for aquatic or terrestrial biota as well as for patterns of soil type, precipitation, temperature, topography, and other habitat variables (Omernik 1995) Terrestrial ecoregions, which are smaller than biomes and major North American bioregions, are commonly named for major botanical communities and reflect the effects of different climates and soil characteristics (Bailey 1994) In contrast,

aquatic ecoregions (e.g., Abell et al 2000) are delineated mostly by major basins and are usually named for the geographic area or river basin section Little concordance exists between

boundaries of terrestrial and aquatic ecoregions, with overlap most often occurring in areas of marked topographic change (Ricketts et al 1999, p 27) In a similar way, the hydrogeomorphic patch structure can shift in areas of substantial topographic change and thus affect the local FPZ type in a broader watershed analysis Ecoregions are widely employed in environmental

management by government agencies and non-government organizations Their use is based on the premise that aquatic ecosystem structure and function are related to the quality and integrity

of the surrounding terrestrial ecosystem Although effects of land cover conversions on stream and river health are well known (e.g., Richards et al 1996; de la Crétaz and Barten 2007), the relationship merits more study, particularly where quantitative source-stressor-response linkages are sought for individual stressors at single locations Use of aquatic ecoregions to design aquaticsampling studies may be more appropriate than for terrestrial ecoregions because aquatic

communities are more similar within than between watersheds Unfortunately, aquatic

ecoregions can be too large for a watershed-scale monitoring study to stratify sampling and thus may contribute most as a correlative factor in statistical analyses

Hydrologic Unit Codes (HUCs) are hydrological analysis reporting units delineated by the U.S Geological Survey (USGS) based on surface hydrologic features from very large (e.g., Great Lakes) to very small areas Hydrologic unit area is inversely related to length of their unit code For example, 2-digit codes describe very large areas (up to 459,878 km2), while 12-digit codes encompass areas as small as 41-60 km2 HUCs delineate hydrologic areas at various scalesbut do not imply changes in the structural character of the riverine landscape HUCs are not

related to terrestrial ecoregions, but at some code level may correspond to aquatic ecoregions based on similar watershed designations They delineate the area but not the hydrogeomorphic characteristics of a watershed, and thus they only marginally overlap with FPZs Investigators may employ HUCs to stratify the watershed into sample areas as long as they remember that these units do not remove the underlying systematic error in the sample design resulting from differences among HUCs in the hydrogeomorphic character of the encompassed streams.

Functional process zones are based on the patch characteristics of the riverine landscape, and their use in sampling design can control much of this systematic error among river sites A given ecoregion may contain many types of FPZs, and similar types of FPZs often occur in multiple ecoregions Plant communities (terrestrial) or basin connections (aquatic) are

emphasized by ecoregions, while FPZs characterize the physical, macrohabitat template of the riverine landscape upon which the abiotic and biotic factors of the ecoregion interact The

location of FPZs are repeatable and partially predictable in location within a basin, but the degree

of predictability decreases when comparing ecoregions with different topographic structures While only a handful of U.S basins been analyzed to date, once a given watershed is delineated, the results should be applicable for future studies unless major changes occur in the channel (e.g., construction of reservoirs or major levees)

As described earlier, the definition and boundaries of reaches vary considerably among

investigators even though they are the most common aquatic sample unit area In general, the reach is defined in the context of how one chooses to model the system, and this varies among countries, government agencies, and individual investigators For example, in a stratified-randommonitoring design for stream condition (e.g., Level 2 Ecoregion or 1st - 4th Strahler order