Citizen science data validates aerial imagery to track the ‘rise and fall’ of woody vegetation through extremes of climate

Keywords: citizen science, freshwater wetland, Friends of Thirlmere Lakes (FOTL), Melaleuca linariifolia, Open Data Kit (ODK), peatland, public participation, Thirlmere Lakes National Park (TLNP), woody encroachment.

One of the challenges for species discrimination in mapping vegetation is the requirement for fieldwork and ground truthing of remotely sensed imagery, which is labour intensive, costly, and time consuming (Husson et al. 2016). Harnessing the value of crowdsourcing (Brabham 2013), especially with its revival in the past 20 years for collecting environmental data (Fienen and Lowry 2012) may be an under-utilised resource that scientific research, especially vegetation mapping could progress. Even though public participation in the collection of scientific data (that is, citizen science, also referred to with the more inclusive term ‘community science’) has been active for hundreds of years (Dickinson et al. 2010; Miller-Rushing et al. 2012; McInnes et al. 2020), it has more recently gained widespread support and recognition (Bonney et al. 2009, 2014, 2020; Silvertown 2009; Miller-Rushing et al. 2012; Pecl et al. 2015; Irwin 2018) with peer-reviewed publications increasing exponentially worldwide, especially in the past decade (Follett and Strezov 2015; Huddart et al. 2016; Fritz et al. 2017; McKinley et al. 2017).

While citizen science data in Australia has benefited from high-profile environmental research (e.g. Nancarrow 1996; Pecl et al. 2019; Steven et al. 2019; Pirotta et al. 2020; Roger et al. 2020; Kirchhoff et al. 2021), its use for vegetation monitoring and mapping is limited to studies in saltmarsh (Buldrini et al. 2015; Dykman and Prahalad 2018), reference image collection to assist in the qualitative assessment of vegetation condition (Roger et al. 2019), trampling (Simpson et al. 2020), or as part of more general observations for mapping the effects of fire on biodiversity (Kirchhoff et al. 2021). Overseas, citizen scientists (CS) observations, while still limited to a few studies, highlight their value for identifying and categorising plant phenology for trees and invasive weeds have been shown to complement other data applications for monitoring dryland ecosystems, mapping invasive species, habitat modelling, vegetation classification (Elmore et al. 2016; Wallace et al. 2016) and mapping the location and intensity invasive weed species (Hawthorne et al. 2015). Integrating CS observation data is also valuable in the calibration and validation of remotely sensed (RS) imagery for earth observations (Fritz et al. 2017).

One of the key benefits of CS is their ability to collect more frequent, longer-term data (Lottig et al. 2014; Blake and Rhanor 2020). It can also increase data coverage in environmental research, with the potential for volunteers to collect vastly more data than government agencies and professional researchers (Storey et al. 2016). Also, involving the community in monitoring increases not only their ability to contribute data, but also their capability to discuss their knowledge with experts (Carolan 2006). Participating community members also show increased scientific literacy, greater awareness of local ecosystems and wider environmental issues, stronger social networks including relationships with local government, and a greater interest in planning (Storey et al. 2016 and references within) that results in further community engagement with government in decision making (Roger et al. 2019). Uncertainty about the reliability of data collected by the community (Hyder et al. 2015; Roger et al. 2019) has repeatedly been shown to be unwarranted when CS programmes are appropriately supported (training, robust protocols) and augmented by professionally collected data (Kosmala et al. 2016; Storey et al. 2016; Dumakude and Graham 2017; Blake and Rhanor 2020). More recently, a key enabler for CS involvement is the ubiquitous ownership of mobile devices and smartphone technology. This technology has enabled the community to use well-designed and user-friendly apps to collect observations of the environment, which are uploaded into a cloud storage repository for the researcher to download and analyse (Herrick et al. 2017).

Continuous ground truthing of vegetation in a dynamic wetland system is rare, yet knowledge of their distribution and extent is critical as these habitats are threatened by climate change. As such, we suggest that by harnessing the willingness of the public to participate in appropriately designed scientific research, we can increase the frequency of observations and lower the costs for projects to ground truth and validate high-resolution imagery for vegetation monitoring and surveys. This is especially true in the constant financial challenges of monitoring programmes, with a lack of resources (time and money) accounting for 85% of river restoration projects lacking some form of monitoring in Australia (Brooks and Lake 2007) and globally (Huddart et al. 2016).

In this study, we examined high-resolution aerial imagery in the rush-dominated peat wetlands of Thirlmere Lakes National Park, NSW, Australia, from 1955 to 2022 and identified the encroachment of woody vegetation during a dry successional state into the herbaceous zone. Initially, as a pilot programme, the aim of the CS programme was general surveillance monitoring of wetland vegetation to validate the high-resolution aerial imagery. This further evolved to target the presence of woody vegetation during the dry event from mid−2018. However, rainfall from March 2020 to June 2021 followed by higher water levels in the lakes, changed the focus again from June 2021 to monitor the dieback of some woody species. General surveillance of multiple species has an important role to play as the first line of attack (Dickinson et al. 2010), and as evident in this evolution of this project, can illuminate unexpected trends for intensive, targeted monitoring.

This project aimed to show that CS programmes and the data they collected can augment researchers’ data to ground truth the high-resolution aerial imagery to monitor the response of woody vegetation encroachment into the wetlands of Thirlmere Lakes National Park.

Thirlmere Lakes are a group of five peat lagoons (Lakes Gandangarra, Werri Berri, Couridjah, Baraba, and Nerrigorang) located within Thirlmere Lakes National Park (Fig. 1), located 100 km south-west of Sydney CBD (34°13′S, 150°32′E). They are part of the Greater Blue Mountains World Heritage Area and protected under the National Parks Act 1974 and the Sydney Water Catchment Management Act 1998 (New South Wales Department of Planning, Industry and Environment (NSW DPIE) 2019).

Fig. 1.  (a) Location map of Thirlmere Lakes National Park in south-east Australia. (b) Lake Werri Berri (WB) is the largest of the five lakes (G, Gandangara; C, Couridjah; B, Baraba; N, Nerrigorang) within the Thirlmere lakes system. Aerial imagery from 13 November 2019. Inset highlights boundary for aerial images of the northern end of Lake Werri Berri: (c) 1955 when water levels were high enough to flow from Lakes Gandangara to Werri Berri; (d) 2014 when woody vegetation is easily detectable from the aerial imagery (red arrows); (e) 2020 when another cohort of woody vegetation can be detected closer to the centre of the lake (yellow arrow); (f) 2022 when water levels were again high enough to flow between Lakes Gandangara and Werri Berri. 1955 aerial imagery sourced from NSW Department of Lands Inquiry Report July 1955, page 1811 (Riley et al. 2012). Other aerial imagery sourced from Nearmap Australia (http://maps.au.nearmap.com/, accessed May 2022).

Although their water levels have fluctuated over time, declines since 2010 have caused community concern that anthropogenic activities such as nearby longwall mining and groundwater extraction might be contributing to the water level declines (Pells and Pells 2016). While significant investment by government and scientific researchers has been undertaken to better understand the geomorphology and hydrology of the system (e.g. Chen et al. 2020; Cowley et al. 2020; Forbes et al. 2021), there is limited documentation of the ecological condition and state (Black et al. 2006; Rose and Martin 2007; Ling and Jacobs 2011; Kobayashi et al. 2020; Forbes et al. 2021) of this unique wetland system. The lakes are dominated by large sedges (e.g. Lepironia articulata, Lepidosperma longitudinale, and Machaerina articulata) and support 22 threatened flora and fauna species including the wetland plants Commersonia prostrata (dwarf kerrawang) and Persicaria elatior (tall knotweed) (Daly et al. 2015; NSW Government 2016).

To highlight the establishment of woody vegetation towards the centre of the lakes during the dry phase at Thirlmere Lakes (see Fig. 1), discrete polygons were digitised by outlining the boundary of the trees and shrubs in each wetland from aerial imagery. Historical aerial imagery were derived from the NSW Department of Planning and Environment’s (DPE). Data from Spatial Imagery Services (1962–2004), Google Earth Pro (2009–2014) and Nearmap (Nearmap Australia http://maps.au.nearmap.com/, accessed 3 May 2022) (2010–2022) were collated either JP2, TIF or JFIF formats and georeferenced in ArcGIS 10.4. The 1955 aerial imagery was sourced from NSW Department of Lands Inquiry Report July 1955 (Riley et al. 2012). Older imagery were georectified using ArcGIS tools. The area (m²) inside each polygon was calculated using ArcGIS automation tools and then converted to hectares (ha) for reporting.

We used the Open Data Kit (ODK) tool to develop an application to use with android mobile devices. ODK is a free suite of tools developed at the University of Washington’s Computer Science and Engineering Department (https://www.google.com/earth/outreach/learn/odk-collect-and-odk-aggregate-to-store-and-manage-your-data/, accessed January 2021). The ODK allows data collection without internet service at the time of collection, and later data submission to an online server. The bespoke application for this project was designed for use by non-scientists. That is, it is user-friendly with photographs and descriptions of the common wetland plant species. The application records the photographs taken by CS to enable validation and for quality control. Once installed, users record the date, time, GPS location, plant names, and photographs.

CS volunteers were sourced from the Friends of Thirlmere Lakes (FOTL) who meet monthly at the lakes to discuss issues and attend other activities such as bushwalking and bush regeneration. Training of the application and plant species identification occurred at least three times with key volunteers, who then trained and aided fellow volunteers. While the intention was for researchers to attend multiple field trips with volunteers, travel restrictions due to the COVID-19 pandemic, floods and road closures prevented researchers from accessing the area. Regular updates of survey observations were mapped for the FOTL monthly newsletter, as was regular feedback between the project researcher and the field survey organiser.

The use of high-resolution imagery highlighted the growth of woody vegetation, such as Melaleuca linariifolia (flax-leaved paperbark) established towards the centre of the lakes (Fig. 2). To highlight this establishment of woody vegetation towards the centre of the lake, the area inside the visible woody vegetation was calculated for the five lakes in Thirlmere Lakes National Park for 33 aerial imagery from 1955 to 2020. The 1955 aerial imagery clearly shows that Lakes Gandangara (LG), Werri Berri (LWB) and Couridjah (LC) are a single large waterbody (Fig. 3) and therefore used as the baseline for the maximum high-water level. Using the 1955 baseline boundary, there was a distinct decline in the digitised area within the woody vegetation in 2009 of 8–15%. By May 2020, the area inside the tree-lined polygons was nearly halved (up to 45%) (Fig. 4).

Fig. 2.  Digitised area of the lake boundary (defined within the limit of the woody vegetation) for July 1955 (baseline), April 2020 and May 2022. This shows the contraction in area over time in 2020, and subsequent expansion again by 2022. LG, Lake Gandangara; LWB, Lake Werri Berri; LC, Lake Couridjah; LB, Lake Baraba; and LN, Lake Nerrigorang.

Fig. 3.  Digitised area (ha) within the woody vegetation boundary from 1955 to 2022. LG, Lake Gandangara; LWB, Lake Werri Berri; LC, Lake Couridjah; LB, Lake Baraba; and LN, Lake Nerrigorang.

Fig. 4.  A total of 90 survey points were mapped as part of a trial for ground truthing wetland vegetation monitoring in Thirlmere Lakes. Survey points collected from March 2020 to February 2021 by researchers (24 records) (yellow) and CS (19 records) (green) show the established woody species. Survey points recorded from May 2021 to May 2022 show the dieback of some woody species (red) (47 records). LG, Lake Gandangara; LWB, Lake Werri Berri; LC, Lake Couridjah; LB, Lake Baraba; and LN, Lake Nerrigorang. Aerial imagery from Nearmap Australia (http://maps.au.nearmap.com/, accessed 3 May 2022).

The July 2021 high-resolution imagery, and personal observation (M. Krogh, pers. comm.), showed that water levels have risen enough for water to flow between LG and LWB, although the area calculations did not change since trees were still visible.

The monitoring and ground truthing of wetland vegetation by CS validated the location of established woody vegetation within the lakes. Along with other wetland species monitoring, 90 survey points were identified (and subsequently verified) as woody species (Fig. 4) recorded from March 2020 to May 2022. Established plants closer to the centre of the lake were mainly M. linariifolia with some of the plants being over 2 m in height. Other species also included Leptospermum trinervium (flaky-barked tea-tree), Kunzea ambigua (tick bush), Casuarina cunninghamiana (river she-oak), Banksia ericifolia, Acacia spp., and Eucalyptus spp. (Fig. 4).

This study documented the phenology of wetland vegetation in Thirlmere Lakes through cycles of climate and weather changes, such as El Niño dry phases and La Niña wet phases. It recorded the dynamic shift in the extent and location of woody vegetation during a dry successional phase, and then its dieback through a wet successional phase using data collected by CS. CS were able to record the location of woody vegetation growing within the rush-dominated zones of the wetland with high precision and accuracy using mobile devices. Initially, only the quality of the photographs was a limiting factor for validation, but this was mitigated with ongoing feedback and training sessions.

Analysis of historical aerial imagery showed the initial changes from a rush-dominated peat wetland to an alternative shrub and tree-dominated state with the establishment of woody species during a 20-year dry phase. The dry successional phase, during the Millennium Drought, provided ideal conditions for woody species to germinate and establish within the stands of Lepironia. By 2009, they were large enough to be visible on the high-resolution aerial imagery. CS data verified that these well-established woody species recorded amongst the L. articulata grey rush were mainly M. linariifolia, as well as more terrestrial woody species such as L. trinervium, Eucalyptus spp., Acacia spp., and C. cunninghamiana.

The dry weather conditions during the El Niño from 2014 to 2016, and again in 2018 (with September 2018 recorded as the driest September in 119 years of records in Australia), are reflected in the low water levels in Thirlmere Lakes (0 m; September 2018–January 2020) (WaterNSW Real Time Data https://realtimedata.waternsw.com.au/). These dry conditions continued into 2019, with rainfall the fifth-lowest on record in NSW and the driest since the severe El Niño drought of 2002 (Australian Government Bureau of Meteorology (AGBOM) 2018). As part of the Thirlmere Lakes Research Program (Department of Planning and Industry 2022), water balance predictions suggested that the lakes are climate sensitive and primarily driven by rainfall and evaporation (Chen et al. 2020). These reports also demonstrated that the low water levels in 2018–2019 were not unprecedented (Chen et al. 2020; Forbes et al. 2021), and these studies supported anecdotal reports and photographs that indicated the presence of trees in the centre of the lake in 1958–1959 that likely established during the long El Niño drought event of the 1940–1950 (World War II Drought), with water returning in 1954 (AGBOM http://www.bom.gov.au/climate/influences/timeline/, accessed 31 May 2020). This research of historical changes though sedimentology and climate highlight the dynamic (i.e. not static) nature of this system over different temporal scale, and it is repeatable, yet the variable patterns are dependant on the window of time that is examined.

Other studies have also shown that woody seedling recruitment was also highest in the dry phases (Tipping et al. 2012; Saintilan et al. 2021). The woody encroachment into graminoid freshwater wetlands has been associated with changes in water levels (Warren et al. 2007), shortened hydroperiods and drainage (Johnston et al. 2003; Duever 2005; Knickerbocker et al. 2009), altered flood regimes (Colloff et al. 2014), and cumulative changes in climate (Berg et al. 2009). In Australia, Johnston et al. (2003) found that 50% of a coastal floodplain wetland had been encroached by the native Melaleuca quinquenervia (broad-leaved paperbark) from 1942 to 1998 due to changes in water levels.

While this study could only detect woody species when they reached a size visible on the aerial imagery, the on-ground survey coupled with the validation of photographs by the CS has allowed the identification of changes to the phenology of the wetland vegetation, especially during the dry phase. Before the La Niña event of March 2021, the CS record the presence of both well-established woody vegetation, and seedlings amongst the rushes. As water levels increased (3.6 m; 4 February 2021) (WaterNSW Real Time Data https://realtimedata.waternsw.com.au/) with the La Niña in 2020–2021, the on-ground photographs document the survival of M. linariifolia established near the centre of the lakes, and demise of other terrestrial species (eucalypts, acacias and Leptospermum species). The survival of M. linariifolia is not unexpected since it can grow in moist and swampy ground as well as dry sclerophyll forests or (https://plantnet.rbgsyd.nsw.gov.au, accessed 30 November 2020). It is also known to be tolerant of either well-drained or poorly drained soils, and adaptable to a range of soil types (https://www.australianplants.com/plants.aspx?id=1113, accessed 30 March 2020). Before 2010, however, M. linariifolia had not been recorded, nor observed, in areas that were frequently inundated.

This study has documented the phenology of wetland vegetation, especially the encroachment of woody vegetation in a rush-dominated wetland as an early indicator of a regime shift to a drier successional state (extended El Niño event), and the dynamic nature of the wetland vegetation following inundation through a La Niña event. The measure of this wetland’s ecosystem resilience is its capacity to undergo an extended dry phase without loss of its ecological character. Periodic drying out of wetlands is a natural alternative state that wetland plant species can adapt to, where woody vegetation may establish during a long dry phase. However, continued surveillance monitoring may further identify whether the Melaleuca will continue to flourish after an extended wet phase.

As many as 90% of restorations in the USA, Australia, and Europe are not monitored beyond visual examination been attributed to a lack of resources; i.e. time and money (Huddart et al. 2016). Using additional resources, such as engaging CS as a supplementary tool, could potentially alleviate this knowledge gap. This study contributes another case study of a participatory tool to the broader scientific community to improve efficiencies to increase the frequency of timing of data validation points. By drawing on the universal ownership of mobile devices for georeferenced observations along with high-resolution imagery, this inclusive method of supplementing CS data supports a better understanding of plant phenology and mapping for threatened plant species, noxious non-native species, and water level mapping.

Previous studies suggest that the two biggest challenges for CS are related to the quality of the data and how to engage and retain citizen participation in the longer term (Kosmala et al. 2016; Fritz et al. 2017). Other studies have shown that CS observations are spatially biased to easily accessible sites (Tiago et al. 2017). To address these issues, we implemented the continuous monitoring of data quality (with photograph validation) and feedback to volunteers to improve quality through training over time. Communication of the project was an essential component of any CS project, with regular contributions to newsletters, presentations and websites concerning gaps in location records. This study found that ground truthed observations as part of the CS programme were an ideal method to track changes in the encroachment of woody vegetation. Given adequate professional support, this community-based wetland vegetation monitoring provided data reliable enough to augment professionally collected data, and increase the opportunities, confidence, and skills of community members.

The method used in this study has practical implications for researchers and practitioners interested in harnessing the power of public participation for the ground truthing and validation of wetland mapping. Key learnings of the project included the importance of two-way communication. For example, by publicising the science at citizen science days, scientists received essential feedback and recruitment of community members. Also given their eagerness for information, providing regular feedback loops to keep CS updated about the project through CS days, community newsletters, and through enthusiastic key leaders that can train, organise and encourage volunteers. Such feedback is critical to a project’s success (de Vries et al. 2019).

Ultimately, this project showed that monitoring the phenology of vegetation in a peat wetland that is driven by changes in climate, can be supplemented by the inclusion of a CS programme. Further, CS efforts created more insightful awareness of environmental issues, connected the local community, scientists and land managers to make scientific resources more efficient to assist in the continuous surveillance monitoring of the local wetland vegetation to collect high-frequency, longitudinal data.

The data that support this study will be shared upon reasonable request to the corresponding author.

No potential competing interest was reported by the authors.

This study was funded by DPE, partly under the Thirlmere Lakes Research Program.