Reconstructing plant invasions using historical aerial imagery and pollen core analysis: Typha in the Laurentian Great Lakes

Abstract

Aim

Determining the spatial-temporal spread of an invasive plant is vital for understanding long-term impacts. However, invasions have rarely been directly documented given the resources required and the need for substantial foresight. One method widely used is historical photography interpretation, but this can be hard to verify. We attempt to improve this method by linking historical aerial photos to a paleobotanical analysis of pollen cores.

Location

Laurentian Great Lakes coastal wetlands, United States of America.

Methods

We chose invasive cattail ( Typha) as our model species because it is identifiable from aerial imagery and has persistent, identifiable pollen, and its ecological impacts appear to be time-dependent. We used Geographic Information Systems, aerial photo-interpretation and field verification to post-dict the invasion history of Typha in several wetland ecosystems. Using ²¹⁰ Pb and ¹³⁷ Cs sediment dating and pollen classification, we correlated the temporal dominance of Typha to our estimates of per cent coverage at one site. The pollen record was then used to estimate the Typha invasion dynamics for dates earlier than those for which aerial photos were available.

Results

Typha spread through time in all study wetlands. Typha pollen dominance increased through time corresponding with increased spatial dominance. Hybrid cattail, T. × glauca increased in pollen abundance relative to T. angustifolia pollen through time.

Main conclusions

This study illustrates the value of generating historical invasion maps with publically available aerial imagery and linking these maps with paleobotanical data to study recent (< 100 years) invasions. We determined rates of Typha expansion in two coastal wetland types, validated our mapping methods and modelled the relationship between pollen abundance and wetland coverage, enhancing the temporal precision and breadth of analyses. Our methodology should be replicable with similar invasive plant species. The combination of pollen records and historical photography promises to be a valuable additional tool for determining invasion dynamics.

Introduction

The ecological effects of invasive species tend to vary temporally (Witkowski & Wilson, 2001; Lovett et al., 2006; Marchante et al., 2008; Carlsson et al., 2010), as the invading species, the invaded biological community, abiotic ecosystem components and invader-ecosystem interactions undergo dynamic nonlinear changes through time (Strayer et al., 2006). Thus, the importance of considering temporal context in invasive species research is increasingly recognized (Blossey, 1999; Strayer et al., 2006). Understanding the long-term effects of invasive species would ideally be accomplished by studying a specific ecosystem pre- and post-invasion. However, this approach is seldom possible as pre-invasion data collection is rare, obtaining funding for long-term ecological research is difficult (Callahan, 1984; Hobbie et al., 2003), and the time frame necessary to study long-term invasion processes is often too great (but, see Meiners et al., 2001, 2002; Carlsson et al., 2010). Different approaches are thus necessary to study invasive species in a temporal context. A chronosequence approach, which has been successfully used to study plant succession (e.g. Foster & Tilman, 2000), is a pragmatic alternative to study the ecological effects of plant invasions through time (Witkowski & Wilson, 2001; Zavaleta & Kettley, 2006; Springsteen et al., 2010).

Establishing an invasion chronosequence requires an accurate spatial-historical reconstruction of the spread of the invasive species, which can be accomplished using remotely sensed data analysed with Geographic Information Systems (GIS) technology. Hyperspectral aerial imagery has been utilized effectively to detect, map and model invasive plant species in the landscape (Underwood et al., 2003; Hestir et al., 2008; He et al., 2011), yet its high cost and the lack of historical imagery makes chronosequence development unfeasible (Jensen et al., 1984, 1986; Underwood et al., 2003). In contrast, historical aerial imagery is widely available, free or inexpensive, and in the United States, temporal coverage extends from the late 1930s through the present (USGS, 2010). However, the resolution of aerial photography is limited and may not allow for accurate species-specific spectral signature identification (Huang & Asner, 2009). Despite limitations, when used in concert with ancillary geospatial data, field demarcation, and ground-truthing for accuracy and precision improvements, historical aerial imagery may be utilized to accurately determine the spatial extent of target plant species at various points in time (Robbins, 1997; Zavaleta & Kettley, 2006; Boers & Zedler, 2008; Wilcox et al., 2008). However, because traditional ground-truthing can only be directly applied to the most recent image, errors are likely to occur in historical interpretation, and alternative methods are necessary for accuracy validation.

Paleobotanical data could be used to evaluate historical image interpretation accuracy and enhance analyses of invasion dynamics. Fossil-pollen data reflect relative vegetation abundance at the time of deposition and can illustrate plant community shifts over time (Moore, 1991) and could provide the historical data necessary to evaluate photo-interpretation accuracy. Paleoecological studies typically examine changes in vegetation at the centennial or millennial timescales (Bunting et al., 1997; Finkelstein & Davis, 2006). However, ²¹⁰Pb and ¹³⁷Cs sediment dating and pollen core analysis together can be used to identify more recent vegetation shifts (50–100 yr bp) (Goldberg, 1963; Jackson, 1997). Although numerous studies have combined remote sensing analyses and paleoecological analyses to model vegetation at a landscape scale (e.g. Broström et al., 1998; Williams & Jackson, 2003; McLauchlan et al., 2007), no known studies have applied the novel approach of linking site-specific aerial photography interpretation and palynology to examine invasion dynamics.

Invasive Typha spp. in Great Lakes wetlands

Invasive Typha spp. are appropriate model species for chronosequence development and paleo-dating because of their invasion dynamics, ecological impacts, physical structure, and persistent and readily identifiable wind-distributed pollen. In the Laurentian Great Lakes region, the invasive narrow-leaved cattail (Typha angustifolia L.) and hybrid cattail (Typha × glauca Godr.), a hybrid between native T. latifolia L. and T. angustifolia (Smith, 1987), are dominant and ecologically disruptive species (Mills et al., 1993; Galatowitsch et al., 1999). Recent genetic analyses has revealed that in the upper Great Lakes region, in sites where both parent species are present, F₁ hybrids tend to dominate, but backcrossing and advanced generation hybrids also occur (Snow et al., 2010; Travis et al., 2010). Additionally, T. angustifolia and T. × glauca are structurally similar, making differentiation from aerial imagery difficult. Because of physical similarities, habitat overlap, and similarities in invasion dynamics, invasive Typha spp. are commonly undifferentiated in the ecological literature (e.g. Frieswyk & Zedler, 2007; Trebitz & Taylor, 2007; Tulbure et al., 2007; Chun & Choi, 2009; Vaccaro et al., 2009; Mitchell et al., 2011). Therefore, in this study, we considered both T. angustifolia and T. × glauca as ‘invasive Typha’.

Invasive Typha has become abundant in Great Lakes regional wetlands (Mills et al., 1993; Trebitz & Taylor, 2007) because of increased propagule pressure (Zedler & Kercher, 2004; Lockwood et al., 2005), alterations in hydrology (McDonald, 1955; Wilcox et al., 1985; Shay & Shay, 1986; Wilcox & Nichols, 2008; Wilcox et al., 2008; Farrell et al., 2010) and anthropogenic nutrient enrichment (Crosbie & Chow-Fraser, 1999; Trebitz et al., 2007; Trebitz & Taylor, 2007; Morrice et al., 2008). Typha tolerates a wide range of water levels (Harris & Marshall, 1963; Waters & Shay, 1990), and recent historically low water levels have been linked to invasions into Lake Michigan and Lake Huron coastal wetlands (Frieswyk & Zedler, 2007; Tulbure et al., 2007; Lishawa et al., 2010). Climate change is predicted to further reduce water levels over the next 50–100 years (Mortsch & Quinn, 1996; Lofgren et al., 2002; Angel & Kunkel, 2009), likely increasing Typha dominance (Lishawa et al., 2010). Following establishment, Typha can spread rapidly (Tulbure et al., 2007; Boers & Zedler, 2008) and is typically much larger than the native species it replaces (Woo & Zedler, 2002). Because of high rates of primary productivity and slow decomposition (Davis & Van der Valk, 1978; Freyman, 2008), litter accumulates in Typha beds (Vaccaro et al., 2009) eventually excluding other macrophytes (Larkin et al., 2012). In Great Lakes coastal wetlands, T. × glauca dominance reduces plant community diversity (Tuchman et al., 2009; Lishawa et al., 2010) and sediments in T. × glauca stands tend to have unique physical composition, microbial communities (Angeloni et al., 2006) and elevated nutrient concentrations (Tuchman et al., 2009; Farrer & Goldberg, 2009; Lishawa et al., 2010). Typha's great biomass and persistent litter give Typha stands novel structure, allowing for remote sensing demarcation. Historical aerial and satellite imagery have been used successfully to assess Typha spread rates (Boers & Zedler, 2008) and increased dominance through time (Wilcox et al., 2008; Farrell et al., 2010). Invasive Typha are also appropriate species for paleoecological analyses because they have distinct and persistent pollen. Typha latifolia produces tetrad pollen grains, whereas T. angustifolia produces monads and hybrid T. × glauca produces monads, dyads, triads and tetrads. Thus, presence of dyads and triads is indicative of T. × glauca (Finkelstein, 2003), and the ratio of pollen types indicates relative dominance of Typha species. Furthermore, Typha pollen is widely wind-dispersed and persists as a significant extralocal component of the pollen record (Janssen, 1984; Clark & Patterson, 1985; Finkelstein & Davis, 2005), allowing for wetland-scale interpretation of Typha dominance from a single pollen core.

Our goals were to develop replicable methods for accurately reconstructing the spatial-temporal spread of dominant invasive plant species and link aerial photo-interpretation with paleobotanical analyses. First, we mapped invasive Typha's distribution through time in two Great Lakes coastal wetlands to create an invasion chronosequence. Second, we validated our interpretation with paleobotanical sediment core data and explored invasion dynamics with spatial and paleo-data. The results of this research will be used to examine the effects of invasive species residence time on wetland ecosystem structure and function (e.g. Mitchell et al., 2011) and the effectiveness of Typha management treatments.

Methods

Study area

We selected a northern and southern Great Lakes coastal wetland with old, extensive stands of Typha to conduct our research. The northern study site, Cheboygan Marsh, is a Lake Huron lacustrine, open-embayment wetland (Albert et al., 2005) in northern lower Michigan (Fig. 1). The upland portion of the wetland is dominated by a large stand of T. × glauca, comprising more than 99% of the relative biomass in the invaded areas (Angeloni et al., 2006; Tuchman et al., 2009). On the lake fringe of the wetland, an emergent-marsh zone approximately 50-m wide is dominated by Juncus, Eleocharis, Carex, Schoenoplectus species and T. angustifolia. The southern study site, Illinois Beach wetlands, is a barrier ridge and swale complex (Albert et al., 2005) located along the south-western Lake Michigan shoreline at the Illinois-Wisconsin border (Fig. 1). This site contains three contiguous park and preserve land parcels: Chiwaukee Prairie Preserve, the largest intact coastal wetland complex in southern Wisconsin (Epstein et al., 2002); Spring Bluff Nature Preserve; and Illinois Beach State Park; together, the only remaining undeveloped Great Lakes ridge and swale complex in Illinois. A patchwork of Typha stands, and native wet meadow and emergent-marsh plant communities characterize the swales.

Figure 1. Map of the Illinois Beach and Cheboygan Marsh study wetlands.

Aerial photo-interpretation

We implemented five steps to develop the invasion chronosequence that are detailed below.

Imagery acquisition

We conducted a survey of historical aerial imagery from national, state, county, university and private data sources with the goal of creating a collection of at least one usable image per site for each decade from 1950 through 2009. The United States Department of Agriculture (USDA), Farm Service Administration (FSA), Aerial Photography Field Office distributes 1–2 m resolution, leaf-on condition images as county-wide mosaics. FSA aerial imagery from 2007 to 2009 was available for all study sites from the USDA, Natural Resources Conservation Service's Geospatial Data Gateway (USDA, 2009). We collected historical aerial imagery from the State of Wisconsin, State of Illinois, and State of Michigan websites and through data sharing agreements with Cheboygan County, Michigan and Lake County, Illinois. We surveyed publicly available imagery from Google (Google Inc., Mountain View, CA, USA) and Microsoft (Microsoft Inc., Redmond, CA, USA); both companies provide web-based and desktop applications that include dated high-resolution imagery. Bing Maps^™ (Microsoft Inc., Redmond, CA, USA) houses high resolution (under 1 m at nadir), low oblique aerial imagery, which were available for the Illinois Beach site. We also obtained historical imagery (1939–2005) from the United States National Archives aerial photography collection (USNARA, 2009), Michigan State University's Aerial Imagery Archive (Michigan State University, 2008) and the USGS Earth Explorer (USGS, 2009). Available imagery occurred in varied formats, including scanned black-and-white photographs, scanned colour-infrared photographs, scanned colour photographs and digital orthoimagery. After evaluating the spatial, spectral and temporal resolution of the available imagery, we downloaded or purchased at least one photo-interpretable image per decade. Additionally, we collected a wide range of publicly available ancillary geospatial data to support the image classification including digital elevation models, elevation contours, land use and land cover classifications, soils and surface geology classifications, Great Lakes water levels (USACE, 2010), vegetation survey data and ecological management data.

All imagery and ancillary spatial data were imported to a GIS, arcgis 9.3 (Environmental Systems Research Institute, Redlands, CA, USA) and standardized prior to interpretation. The georeferencing tool in arcmap (Environmental Systems Research Institute, Redlands, CA, USA) was used to georeference images to the base map, as is commonly necessary with scanned photographs, by correlating three or more objects from each image with the same object in a spatially referenced image. Additionally, images were manipulated in arcmap to enhance contrast or balance colours. Histogram stretches were applied to many of the images to increase contrast. Choices of spectral bands were limited, as most historical imagery was scanned from black-and-white film, resulting in a single panchromatic spectral band. Original band assignments were used for natural colour or false-colour-infrared images that contained multiple spectral bands.

Field mapping and digitization of Typha stands

We determined the current extent of Typha in a portion of each study wetland by collecting field vegetation data. Hand-held mapping GPS units (Garmin GPSMAP 60Cx Global Positioning System; Garmin International Inc., Olathe, KS, USA) were used to collect a series of points around the perimeter of contiguous stands of Typha. Stand boundaries were demarcated by determining where Typha litter and living biomass had more than 50% of aerial coverage. Field teams repeatedly calibrated their assessment to ensure compatibility of data sets and replicability of results. GPS data were loaded into arcgis and overlaid on the most recent imagery (2007–2009) base map. A new, empty polygon shapefile was created, and features were added through onscreen digitizing, tracing the sequence of points. Because some stands had convoluted shapes, points were labelled with sequential numbers during data collection to ensure the correct succession when digitizing.

Photo-interpretation

Traditional manual photo-interpretation techniques (Lillesand et al., 2008) were used to identify discrete Typha polygons. The most recent imagery was overlaid with the Typha polygons, and the characteristics of Typha stands were examined carefully at a range of scales for their shape, size, hue and texture. This detailed assessment revealed that Typha stand texture is homogeneously flat with some striations, whereas the surrounding vegetation (dominated by Carex spp., Schoenoplectus spp., and Juncus spp.) is more heterogeneous in texture (see Fig. S1 in Supporting Information). Following Typha characterization, we manually digitized new Typha polygons outside of the field surveyed area where the majority of the ground cover shared Typha visual characteristics. Ancillary data such as soil type and topography were used in conjunction with the above layers to corroborate Typha presence and to refine Typha polygons. When available, vegetation surveys and management practice history data were utilized to further refine image interpretation. In locations where data were available, current high resolution, low oblique aerial imagery (Bing Maps^™) was referenced to further improve interpretation.

Accuracy assessment and interpretation refinement

An accuracy assessment was conducted to find errors and improve interpretation of Typha delineation. Interpretation confidence levels for each Typha polygon were classified: a value of 1 was given to areas where the interpreter was highly confident that the vegetation was Typha; a value of 2 represented moderate confidence; and a value of 3 represented low confidence and was assigned to polygons with some variability in texture or colour. Randomly located points (RLPs) were generated within all polygons using the ‘Generate Random Points’ tool in ET Geo Wizards (Tchoukanski, 2008) within arcmap. A stratified random sample of RLPs from each confidence category was selected for field ground-truthing. Twenty-five per cent of the RLPs from each confidence category were loaded onto a hand-held GPS unit and visited in the field. A one-metre square vegetation plot was established at each point, and Typha presence or absence, per cent cover, and stem density were recorded.

The accuracy of interpretation was assessed for each confidence category and for each site. For instance, in Spring Bluff, overall interpretation accuracy was 84%; interpretation was 100% accurate for the highest-confidence category, 85% accurate for the moderate category and 67% accurate for the low-confidence category. Using information gained from the accuracy assessment, particularly incorrectly assigned Typha stands, polygon delineations were refined and manual interpretation of digital images was improved through examination of subtle differences in colour and texture between Typha and native vegetation types.

Historical imagery Typha delineation

Typha is actively expanding its range in our study wetlands (Debbie Maurer, personal communication; Tuchman et al., 2009). Therefore, historical stand extents were typically smaller in each subsequently older image and, with few exceptions, older Typha stands lay entirely within the more modern Typha polygons. Working counter-chronologically from current imagery, the next most recent images were analysed first. Current Typha polygons were overlaid to help facilitate the determination of historical Typha presence. The interpreter manually digitized the reduced perimeter of contiguous stands of Typha within each historical image in succession, creating year-specific Typha extent polygons. Again, differences in texture and colour, as compared to the surrounding ground cover, allowed for delineation. These characteristics differed somewhat between image dates because of variability in image seasonality, sun angles, atmospheric conditions and other image distortion. As successively older images were examined, interpretation was recalibrated in the light of variability in image quality.

Paleobotanical analyses

Sediment dating

Sediment cores were collected from Cheboygan Marsh in August 2006 (core A, 32 cm) and in August 2007 (core B, 40 cm) for paleobotanical analyses. Paleobotanical analyses were limited to a single site because of limited available funds. Both cores were collected from within large stands of Typha, proximal to the oldest stands in Cheboygan Marsh and well within the pollen source area where wind-dispersed Typha pollen would be deposited (Janssen, 1984), and which likely experienced similar invasion history because they shared similar sediment profile structure and were equidistant from the Typha invasion edge (approximately 75 m). Core A was used for ²¹⁰Pb and ¹³⁷Cs dating. To reduce analytical costs, four-centimetre subsections of the core were homogenized, dried at 105 °C for 48 h and sent to MicroAnalytica LLC for radiometric dating analysis of ²¹⁰Pb and ¹³⁷Cs. Assuming a constant flux of unsupported ²¹⁰Pb from the atmosphere (Appleby & Oldfield, 1978), ²¹⁰Pb values from each subsection of the core were used to determine a constant flux: constant sedimentation rate (CF:CS; Robbins, 1978). The CF:CS was used to calculate a dry-mass sedimentation rate. Additionally, the depth of the core subsection which contained a peak of ¹³⁷Cs, from nuclear testing in 1963 (Ritchie & McHenry, 1990), was determined for ²¹⁰Pb sediment accumulation rate validation.

Sediment characterization and pollen analyses

Sediment core B was used to analyse organic-matter content and for pollen analyses; subsamples were taken at 1-cm intervals throughout the length of the core, dried at 105 °C for 24 h, weighed, then placed in a muffle furnace at 550 °C for 2 h (APHA, 2005). The non-volatile ash remaining was subtracted from the initial dry mass, and organic matter was calculated as per cent of dry mass (APHA, 2005).

Preparation of pollen samples followed standard methods (Fægri & Iversen, 1989), with successive treatments of 10% KOH to remove humates, 10% HCl to remove carbonates, 49% HF to remove silicates and acetolysis solution to remove cellulose. Samples were sieved with a 7-μm Nitex screen to remove clay and other fine particles (Cwynar et al., 1979). After dehydration with tert-Butanol, samples were mounted in silicone oil. A spike of 0.5 ml of a suspension of polystyrene microspheres (15 μm) of known concentration was added to each volumetric pollen sample, and these were counted along with pollen to calculate pollen concentration. A minimum of 300 pollen grains per sample were counted under 400 × magnification (Moore, 1991). In each sample, the four types of Typha pollen grains (monads, dyads, triads and tetrads), native sedge pollen (Cyperaceae) and pine (Pinus) pollen, were counted; pollen grains from other species were identified as other. Pure T. × glauca pollen has an average of 20% dyad/triad pollen, whereas pure T. angustifolia pollen has approximately 1% dyad and 0% triad, and T. latifolia has 0% dyad/triad pollen types (Finkelstein, 2003). Therefore, for each pollen sediment section, we calculated the estimated per cent of Typha pollen, that is, T. × glauca pollen by determining the proportion of dyad/triad pollen types divided by 20, ±5% (representing up to 5% dyad contribution from T. angustifolia): T. × glauca (±5%) = [(dyad + triad)/20].

Statistical analyses

We performed ordinary least-squared linear regression analysis of sediment depth and ²¹⁰Pb dating derived ages to determine sediment accretion rates, and we extrapolated continuous sediment profile age estimates between sampled mid-points using the resulting regression function. We evaluated correlations between Cheboygan Marsh Typha pollen abundance and sediment organic matter using a generalized linear model (GLM) with binomial error distribution (Typha pollen abundance% ~ sediment organic matter%). Additionally, for those years when aerial photo-interpretation Typha coverage data were generated, we evaluated the relationship between Cheboygan marsh dominance by Typha and Typha pollen abundance using a GLM (Cheboygan Marsh Typha cover (ha) ~ Typha pollen abundance%); the resulting model was used to post-dict Typha marsh dominance values at the temporal resolution of pollen core samples. To relate Typha pollen abundance data and aerial imagery, which did not always occur from the same years, we assumed a linear change in Typha pollen data between data points and used average values for those years when dates did not align. All statistical analyses were conducted using r 2.12.1 (R Development Core Team, 2010).

Results

Aerial photography interpretation

Photo-interpretation and GIS-based delimiting allowed for the spatial reconstruction of the spread of Typha through time and the creation of mapped mosaics representing the age of Typha stands (Figs 2 and 3). Additionally, photo-interpretation revealed detailed quantifiable patterns of Typha establishment and invasion in our two study wetlands. In Cheboygan Marsh, Typha was well established by 1963, the first year where high-quality aerial photography was available. In 1963, Typha dominated 8.2 ha of the marsh (35% of area; Tables 1 and 2); between 1963 and 2009, marsh coverage expanded linearly at an annual rate of 0.14 ha year⁻¹, dominating 14.6 ha (62% of marsh area) by 2009 (Tables 1 and 2; Fig. 2).

Figure 2. Map of the extent of Typha stands in Cheboygan Marsh, Michigan between 1963 and 2009. The locations where core samples were collected for sediment dating and pollen analyses are indicated.

Figure 3. Map of a subset of aged Typha stands within Illinois Beach State Park, Illinois.

Table 1.  Typha spp. dominance in each of four wetlands at the date of the earliest and most recent aerial image analysed and the average annual increase in Typha area over the study period

	Cheboygan Marsha		Chiwaukee Prairieb		Illinois Beach State Parkc		Spring Bluff Natural Aread
^a Photo dates analysed: 1963, 1970, 1980, 1987, 1998, 2004, 2009. ^b Photo dates analysed: 1976, 1981, 1986, 1995, 2000, 2005, 2009. ^c Photo dates analysed: 1955, 1972, 1974, 1976, 1981, 1986, 1994, 1998, 2002, 2007. ^d Photo dates analysed: 1974, 1976, 1981, 1986, 1988, 2000, 2002, 2004, 2005, 2006, 2007. ^e Determined from National Land Cover Dataset 2001 (NLCD 2001), aerial photo-interpretation, and field assessment. ^f Represents the extent of aerial photography interpretation, not the entire property.
Wetland area (ha)	23.2		53.9e		72.8e ^, f		59.5e
Year	1963	2009	1976	2009	1955	2007	1974	2007
Typha area (ha)	8.21	14.60	0.21	2.71	3.29	28.85	10.78	38.57
Per cent of wetland	35.4	62.4	0.4	5.0	4.5	39.6	18.1	64.8
Average annual increase (ha)	0.14		0.08		0.49		0.84

Table 2. Results of aerial photography interpretation of Typha coverage and Typha pollen dominance by year in Cheboygan Marsh

Year	Typha aerial coverage			Typha pollen
	Area (ha)	Per cent of marsha	Rate of increase (ha year^−1)b	(Per cent of total)	(grains ml^−1 sediment)
^a Marsh extent varied with fluctuating Lake Huron water levels. Per cent of marsh is a measure of the 2009 marsh area. ^b Average spread rate between aerial photo-interpreted years: (year 2 area–year 1 area)/(year 2 date–year 1 date). ^c When pollen strata dates did not correspond directly with aerial photo dates, pollen data were averaged from the two years straddling the year of photo-interpretation (±3 years).
2009	14.60	63.0	0.62	–	–
2004	11.48	49.5	0.12	–	–
1998	10.74	46.3	0.10	95.6c	537c
1987	9.63	41.5	−0.02	87.9c	254c
1980	9.79	42.2	0.05	90.3	239
1970	9.28	40.0	0.15	85.0	116
1963	8.22	35.4	–	74.1b	48c

Within the Illinois Beach wetland complex, Typha was first identifiable in 1955 in the Illinois Beach State Park (IBSP; Fig. 3), aerial coverage increased through time and came to dominate 28.9 of 73 ha of habitat (40% of wetland) by 2009 (Tables 1 and 3). In the Chiwaukee Prairie (CP) portion of the complex, a small patch of Typha (0.21 ha) was identified in 1963, also spread through time eventually dominating 2.7 of 53 ha (5%) of habitat by 2009. In Spring Bluff natural area (SB), large stands of Typha (10.8 ha) were first identified on aerial imagery in 1974, spread through time eventually dominating 38.6 ha (65% of wetland habitat; Tables 1 and 3). By 1986, Typha had replaced all of the native vegetation in some Illinois Beach swales and by 2007 only pockets of native plant communities remained un-invaded throughout much of the complex (Fig. 3).

Table 3. Results of aerial photography interpretation of Typha coverage by year in Illinois Beach State Park, Chiwaukee Prairie and Spring Bluff wetlands

Year	IBSP			CP			SB
	ha	%a	ha year^−1 b	ha	%a	ha year^−1 b	ha	%a	ha year^−1 b
^a Per cent of the wetland area evaluated. ^b Spread rate between aerial photo-interpreted years: (area 2 − area 1)/(year 2 − year 1).
2009	–	–	–	2.71	5.0	0.28	–	–	–
2007	28.8	39.6	0.52	–	–	–	38.6	64.8	0.11
2006	–	–	–	–	–	–	38.5	64.6	−1.21
2005	–	–	–	1.57	2.9	0.02	39.7	66.7	0.63
2004	–	–	–	–	–	–	39.0	65.6	0.86
2002	26.3	36.1	1.95	–	–	–	37.3	62.7	−0.34
2000	–	–	–	1.43	2.7	0.06	38.0	63.9	1.33
1998	18.5	25.3	0.42	–	–	–	–	–	–
1995	–	–	–	1.16	2.1	0.04	–	–	–
1994	16.8	23.0	0.14	–	–	–	–	–	–
1988	–	–	–	–	–	–	22.0	37.0	1.26
1986	15.7	21.4	1.57	0.76	1.4	0.09	19.5	32.7	2.11
1981	7.8	10.7	1.15	0.28	0.5	0.01	8.9	15.0	−1.00
1980	–	–	–	–	–	–	–	–	–
1976	–	–	–	0.21	0.4	–	13.9	23.3	1.55
1974	11.4	15.7	1.47	–	–	–	10.8	18.1	–
1972	8.5	11.7	0.31	–	–	–	–	–	–
1955	3.3	4.5	–	–	–	–	–	–	–

Pollen analyses

A significant linear relationship between unsupported ²¹⁰Pb and sediment depth was observed (unsupported ²¹⁰Pb ~ sediment depth (cm); R ² = 0.890; F = 24.32; P = 0.016; Table 4). Assuming a constant flux, constant sedimentation model (Robbins, 1978), we determined the rate of sediment deposition in Cheboygan Marsh to be 0.4 cm year⁻¹ through 24 cm of core length. The deposition rate was corroborated by ¹³⁷Cs data which peaked in the sediment profile at approximately 17 cm, indicating an age of AD 1963 ± 2 years (Ritchie & McHenry, 1990).

Table 4. Sediment core depth, total ²¹⁰Pb, unsupported ²¹⁰Pb, ¹³⁷ Cs and modelled age/year

Sediment depth (cm)	Core mid-point depth (cm)	Total ^210Pb (dpm g^−1) (SD)	Unsupported ^210Pb (dpm g^−1)	^137Cs (dpm g^−1) (SD)	Apparent age	Estimated yeara
^a Assuming a linear sediment deposition rate within each subsection, homogenized cores ^210Pb and ^137Cs values correspond with core depth mid-point.
0–8	4	21.32 (0.68)	20.69	2.17 (0.15)	9.90	1996
8–12	10	12.06 (0.60)	11.39	3.59 (0.16)	24.75	1981
12–16	14	11.28 (0.37)	10.34	7.16 (0.15)	34.65	1971
16–20	18	10.96 (0.59)	9.95	6.95 (0.23)	44.55	1961
20–24	22	3.94 (0.43)	3.32	4.09 (0.13)	54.46	1952
24–30	26	–	–	–	–	–

Typha pollen increased rapidly in relative abundance (% of total pollen grains) between 1945 and 1955, when Typha becoming the majority pollen type in Cheboygan Marsh (Fig. 4). Typha dyads and triads, indicative of T. × glauca (Finkelstein, 2003; Table 5), first appeared in 1948 indicating an approximate date of T. × glauca establishment. Relative abundance of Typha pollen increased linearly between 1955 and 2000 (Fig. 4; see Table S1 in Supporting Information). Relative abundance of Cyperaceae pollen and concentration (grains per ml sediment) peaked in 1950, declined in the late 1950s, and disappeared from the record in the mid-1990s (Fig. 4; see Table S1). The initial increase in Typha abundance, concurrent with the brief (5 year) increase in Cyperaceae pollen, corresponded with a period of declining Lake Huron water levels; after peaking in 1952, water levels steadily declined through 1965, when water levels reached an 85--year low (USACE, 2010).

Figure 4. Percentage pollen and organic carbon derived from Cheboygan Marsh sediment cores. Dates were assigned based on ²¹⁰ Pb and ¹³⁷ Cs sediment analyses; analyses were limited to the top 24 cm (1945–2000).

Table 5. Per cent abundance of Typha monads, dyads, triads and tetrads averaged from the full pollen core, compared to documented values for pure Typha × glauca (Godr.) and T . angustifolia (L.) pollen, from Finkelstein (2003)

	% abundance: minimum–maximum (mean)
	This study	Finkelstein (2003)
	Per cent of Typha a	T. × glauca	T. angustifolia
^a Per cent abundance of Typha spp. pollen.
Monads	88–97 (89.8)	47–92 (75)	96.5–100 (99)
Dyads	1–7 (5.3)	7–30 (17)	0–3 (1)
Triads	0–1 (0.8)	0–10 (3)	0 (0)
Tetrads	1–5 (4.2)	0–14 (5)	0–0.1 (0.01)

The relative per cent abundances of Typha pollen types for the full core are between the diagnostic ranges for T. × glauca and T. angustifolia (from Finkelstein, 2003), indicating the presence of both species within the wetland over the last 65 years (Table 5). Between 1945 and 2000, the estimated per cent abundance of T. × glauca pollen increased from 3.3% (±5) to a maximum of 40.5% (±5) in 1995, indicating a substantial increase in wetland dominance (see Table S1). Prior to 1945, Typha pollen averaged 98.2% monad and tetrad types, indicating nearly pure T. angustifolia and T. latifolia dominance (Finkelstein, 2003). Finally, we found a significant positive correlation between relative abundance of Typha pollen and sediment organic-matter content (Typha pollen abundance% ~ sediment organic matter%; F = 13.321; P < 0.01).

Pollen core and aerial photography correlations

A significant positive correlation between Typha marsh coverage and Typha pollen relative abundance was observed from Cheboygan marsh between 1963, the first year of photo-interpretation, and 1998 (Cheboygan Marsh Typha cover (ha) ~ Typha pollen relative abundance; F = 47.82; P < 0.01; Fig. 5a). Using the GLM function, we modelled the marsh area dominated by Typha in the years between 1945 and 1963, the year with the first usable aerial image, and modelled marsh coverage values for the years intervening those with aerial photo-interpretation values (Fig. 5b).

Figure 5. (a) Correlation between logit-transformed relative dominance of Typha pollen and Cheboygan Marsh Typha cover. (b) Typha cover (ha) between 1945 and 2009 illustrating aerial photo-mapped Typha cover (1963–2009) overlaid on generalized linear model modelled Typha cover (Cheboygan Marsh Typha cover (ha) ~  Typha pollen dominance (%); 1945–2000).

Discussion

We recreated the spread of a dominant invasive macrophyte, Typha, through time in two Great Lakes coastal wetlands using historical aerial photography, field demarcation and ancillary geospatial data. We generated invasion chronosequences of aged Typha stands in a GIS database. Stand-age maps will be used to assess the relationship between experimental Typha restoration outcomes and stand-age and have been used to evaluate effects of Typha residence time on a range of ecological conditions (Mitchell et al., 2011). The methods employed in this study allow for the spatial recreation of an invasion and should be replicable for other problematic invasive graminoid and forb species that grow in highly dominant stands and accumulate litter, such as Phragmites australis (Cav.) and Phalaris arundinacea (L.), which have both been successfully identified using remote sensing and aerial imagery interpretation (Rice et al., 2000; Jakubowski et al., 2010).

We employed a novel approach to analyse a species invasion by linking paleobotanical pollen analyses with historical mapping. A strong correlation between Typha dominance in the paleo-data and mapped cover data validate the accuracy of our historical mapping interpretations. Furthermore, we determined the date of historical invasion by T. × glauca. Additionally, the significant regression between Typha pollen dominance and wetland coverage allowed us to model the extent of Typha cover in Cheboygan Marsh from 1945 to 2009 at the temporal resolution of the pollen data and for 18 years prior to the first analysed aerial image (1963). Thus, linking the two data sets increased both the temporal scope and resolution of our historical understanding of Typha dominance in the study wetland. This study demonstrated the usefulness of paleoecological methods for investigating recent (< 100 year) vegetation changes in wetland ecosystems, revealing aspects of the history and dynamics of a Typha invasion. For example, pollen analysis allowed us to distinguish the relative abundance of different Typha species, which is not possible solely using historical imagery.

Pollen data provide insights into the historical invasion dynamics of two co-occurring invasive Typha spp. in a Great Lakes marsh. Typha angustifolia established early and rapidly, followed by a slower, linear increase in T. × glauca over the last 50 years. Prior to 1945, Typha pollen was dominated by T. angustifolia and T. latifolia. Following 1948, a mixture of T. × glauca and T. angustifolia pollen was present with T. × glauca pollen dominance increasing through time, revealing a successional trend between the two invasive Typha species (see Table S1). Additionally, a significant correlation between Typha pollen dominance and sediment organic-matter accumulation was observed, providing further support for the hypothesis that Typha may be a driver of ecological change in the wetlands it invades by fundamentally altering sediment conditions (Angeloni et al., 2006; Farrer & Goldberg, 2009; Tuchman et al., 2009; Lishawa et al., 2010). However, only a single sediment core was analysed in this study, limiting the scope of interpretation.

In Cheboygan Marsh, Typha pollen became the majority pollen type in the 1950s, corresponding with a long-term decline in Lake Huron water levels (USACE, 2010). Varying water levels play an important role in the establishment and dominance of emergent macrophytes in the Great Lakes including Typha seedlings, which establish successfully on exposed mudflats (Keddy & Reznicek, 1986; ter Heerdt & Drost, 1994) and are often successful following high water or flooding periods (Miller, 1973; Farney & Bookhout, 1982; Wilcox et al., 1985). Further research into the relationship between water levels and invasive species establishment is warranted in the light of predicted climate change driven Great Lakes water level declines (Mortsch & Quinn, 1996; Lofgren et al., 2002; Angel & Kunkel, 2009).

Conclusions

Determining whether invasive species are driving ecological degradation or are responding to degraded abiotic conditions remains an important question in the field of invasion ecology (Didham et al., 2005; MacDougall & Turkington, 2005). Stand-age mosaic maps, like those created in this study, can allow researchers to evaluate biodiversity, abiotic conditions and assessment of ecological function along stand-age gradients, thereby shedding light on this question. For instance, Mitchell et al. (2011) found significant Typha stand-age-dependent relationships in the biotic and abiotic ecosystem components: Typha density, litter mass, plant diversity and sediment organic matter. Mapping the spread of invasive plant species through time by interpretation of publically available historical aerial imagery in a GIS is a low-cost method which could greatly enhance invasion ecology research. The traditional aerial photography interpretation methods utilized in this study proved to be highly accurate for determining the extent of a dominant invasive macrophyte over the last 60 years.

Typha spp. are particularly suitable for wetland-scale paleo-analyses because of long-range wind-dispersal of pollen (Janssen, 1984; Clark & Patterson, 1985; Finkelstein & Davis, 2005). However, pollen distribution distance is of critical importance when extrapolating vegetation cover relationships and should be considered when applying our techniques to other invasive species; additional pollen cores may be required to accurately determine wetland-scale dominance. Future studies examining invasive species pollen dominance at a range of distances from the source vegetation would yield important clarifying data.

Coupling historical photo-interpretation with an invaded site's pollen record is a novel approach to invasive species research. As we demonstrated, by correlating these two distinct historical data sets, researchers can generate invasive species cover models with longer historical perspective and higher temporal resolution than is possible using aerial imagery alone. This approach can provide a wealth of high-resolution data about relatively recent invasions by wetland macrophytes, improving understanding of invasion dynamics by enhancing the spatial and temporal scope of invasive species research.

Acknowledgements

This research was supported by a grant from the Illinois-Indiana Sea Grant College Program to N. Tuchman, D. Larkin, P. Geddes and D. Treering. We thank D. Larkin, P. Geddes, D. Maurer and B. Lawrence for their advice on various aspects of the research; M. Freyman, K. Jankowski, D. Miceli and M. Mitchell for assistance collecting data; and J. Wilson and three anonymous reviewers, whose comments greatly improved the manuscript.

Biosketches

Shane Lishawa is a Research Associate at the Center for Urban Environmental Research and Policy (CUERP), Loyola University Chicago. His research focuses on the interplay between invasive macrophytes, community dynamics, nutrient cycling and water level change in Laurentian Great Lakes coastal wetland ecosystems and explores the outcomes of ecological restoration treatments.

This work was a collaborative effort by the Illinois State Museum, where Eric Grimm is Curator and Chair of Botany, Nancy Tuchman's Aquatic Ecology Lab, and CUERP, where David Treering is a GIS Specialist, Lane Vail is a Research Associate and Owen McKenna was an Undergraduate Researcher.

Author contributions: N.T., L.V., and S.L. conceived the idea; E.G. provided palynological expertise; L.V., S.L., D.T. and O.M. collected data; and SL conducted the analyses and led the writing.

Editor: John Wilson