The Earth’s peatlands are experiencing widespread environmental pressures, caused by climate change as well as anthropogenic forces. Some of the main dangers include increased frequency and severity of droughts, a higher wild fire frequency, drainage, and peat mining [1
]. Another threat to the peatlands in Australia is created by the trampling and grazing of feral animals, which results in compaction and drainage and eventual desiccation of peatlands [1
]. Little is known about the response of peatlands to these disturbances [2
]. Yet peatland ecosystems play a vital role in local and global environmental processes.
Peatlands are wetlands with a low mineral content (<35%) and an organic soil layer of at least 30 cm depth [6
]. This organic layer is sustained by a high-water table which creates anaerobic conditions and maintains a low decomposition rate. These conditions make peatlands a major global carbon sink: covering barely 3% of the Earth’s surface, boreal and subarctic peatlands function as a sink for 15–30% of the global soil carbon, resulting in a net cooling effect on the Earth’s climate [6
]. Apart from their global climatic importance, peatlands fulfill a hydrological function in catchments, acting as water flow regulators. With an exceptionally high water holding capacity, peatlands efficiently store surface water, thereby filtering out sediment and moderating runoff [1
In Australia, 0.14% of the total land mass is covered by peatlands [10
]. This includes coastal peatlands and peatlands in the sub-alpine and alpine regions of south eastern Australia that occur c. 1000 m Above Sea Level (a.s.l.) [11
]. The sub-alpine or alpine peatlands are either montane mires or Sphagnum
dominated communities, geographically constrained to poorly drained areas and therefore at risk of ecological collapse [1
, a large colonial bryophyte, is considered to be an important contributor to the resilience of sub-alpine and alpine peatlands, as it can easily survive under extremely nutrient-poor conditions and can produce resistant organic matter which further promotes peat accumulation [1
mires are also an important habitat for several wildlife species including endangered frogs, such as the Corroboree frog [10
]. In Tasmania on the Central Plateau, Sphagnum
often forms the understory of the endemic conifer pencil pine (Athrotaxis cupressoides
). Pencil pines are endemic to Tasmania and are mainly concentrated on the broad Central Plateau, where they develop on open sites [13
]. They are a long-lived species, growing up to 1000 years in age and contain seed productions of 5–6 year intervals. The first seed production generally sets in after the pencil pine has reached an age of 100 years [13
There are numerous records of peatland damage from wildfires in Australia that result in the loss of peat soils and modify their drainage [16
]. Both Sphagnum
and pencil pine are known to be fire sensitive species (Figure 1
). Post-fire conditions of Sphagnum
bogs often show slow recovery, or conversion into grass and fern dominated lands [18
]. Pencil Pine are very sensitive to fire, grow slow and episodically produce seed and typically reproduce colonially [21
]. Because of these properties, pencil pine is rarely found to be naturally recovering after fire damage and has undergone a range of declines through the Holocene, and particularly since European colonization 200 years ago [21
Two examples of severe peat fires in the last century are the 1961 fire in the Tasmanian Central Plateau and the 2003 fire in the Australian Alps, Victoria. The Tasmanian fire of 1961 was human-ignited and over a period of five months it smoldered in approximately 60% of the Tasmanian Central Plateau, reducing 20,000 ha of peat soils into mineral soils [13
]. The Victorian 2003 fire was preceded by three years of drought, leaving the peatlands in the Alps dehydrated and susceptible to fire [16
]. As a result, almost all the alpine, subalpine and montane mires and fens in Victoria were burnt, causing a loss of approximately 15% of the functional plant communities [16
Clarkson et al. [10
] stated that the relatively small peatland extent of Australia and low levels of research funding have contributed to limited peatland restoration programs and research publications. However, with Sphagnum
as the dominant contributor to continuing accumulation of organic matter in peatlands, recovery of Sphagnum
is essential for subalpine and alpine peatland communities [16
]. In Tasmania, rehabilitation of Sphagnum
has never been extensively trialed, thus, post-fire peatland management is of increasing importance here [19
To make peatland rehabilitation methods more efficient and effective, an improved knowledge of the response of peatlands to disturbances is required [1
]. Current physical mapping and monitoring techniques of Australia’s peatlands, such as described by Whinam et al. [18
] and Clarke and Martin [23
], are often labor intensive, time consuming, and can further damage peatlands by stepping on this delicate vegetation. Additionally, the use of sample areas (quadrats and transects) generally covers only 1–4% of the peatland being studied, hence, are vulnerable to sampling bias. Alternatively, conventional remote sensing techniques such as satellites and aerial photography are inefficient for peatland monitoring due to the fragmented nature and typically small areas (<1–2 ha) of individual peatlands in Australia [11
]. Moreover, to capture the detailed information of peatlands such as species composition and vegetation health, sub decimetre resolution imagery is required, which traditional remote sensing techniques cannot provide. Ultrahigh spatial resolution (<10 cm/pixel) images are also required to capture the micro-topography of peatlands, which is essential for the monitoring of hydrological pathways and peat bog volumes.
Surveying the fragmented areas of Australia’s peatlands can be achieved by capturing ultra-high spatial resolution imagery from unmanned aerial systems (UAS, UAVs, or drones). A small UAS is ideal for mapping areas of <10 ha with an ultra-high resolution where 1–2 cm/pixel is feasible for the typical areas that Sphagnum
mires cover. Also, UAS have previously proven their suitability for environmental mapping and monitoring of micro-topography and species composition including moss beds [24
]. To map and monitor Sphagnum
mires over a larger geographical area that contains hundreds of individual mires, it would be necessary to select a representative sample of Sphagnum
mires. The spectral information provided by the high-resolution images from a UAS, has the potential to provide a better insight into different health states of the Sphagnum
moss. Additionally, digital surface models (DSMs) can be created by generating very high-resolution 3D point clouds with photogrammetric software and computer vision techniques. High resolution DSMs, can be used for hydrological modelling to provide detailed information about flow directions of surface water and suitable locations for rehabilitation methods within each Sphagnum
mire, such as channel blocking [10
Images captured by a UAS are typically geotagged with the camera location at the time of image capture, the position being supplied by the onboard global navigation satellite system (GNSS). However, with a potential error of several meters, these positions do not have a geometric accuracy that is compatible with the ultra-high resolution UAS imagery [26
]. Furthermore, DSMs generated from UAS imagery with low accuracy geotagging and lack of high accuracy ground control can be artificially tilted or have a more complex form of geometric distortions, such as doming or twisting, resulting in inaccurate hydrological models. To create a reliable DSM, a more accurate method of image georeferencing is required. A commonly applied method to remove these distortions is the use of ground control points (GCPs). The markers are laid out in the mapped area and their geographic coordinates are measured using a differential GPS (DGPS) real time kinematic (RTK) system with an accuracy of approximately 2 cm in the horizontal and 4 cm in the vertical directions. The markers are later manually identified in the images and used during structure from motion (SfM) and multiview stereopsis (MVS) algorithms. This enhancement of the photogrammetric model accuracy corrects many model distortions and translates the model into an accurate real-world co-ordinate frame. However, the use of GCPs during fieldwork is time-consuming and labor intensive, especially for areas that are not easy to access [27
]. Furthermore, for vulnerable areas such as peatlands, manual placement, and retrieval of GCPs can damage the delicate vegetation within the study site. Minimizing the amount of GCPs will minimize the amount of damage caused by trampling.
A few studies have documented the distribution of GCPs in relation to DSM accuracy [28
]. Harwin et al. [29
] found that accuracy mostly decreased in vertical direction when reducing the amount of GCPs, and highlighted the importance of precise control. Gindraux et al. [30
] stated that DSM accuracy is further influenced by factors such as camera focal length, flying height, and image quality. Along with varying sizes and complexity of the topography of study areas, comparisons between studies are difficult to make. Tonkin and Midgley [28
] came to the conclusion that for an area of irregular topography, the use of four or more GCPs is acceptable, with a vertical RMSE of 0.064 m, compared to a vertical RMSE of 0.059 with 101 GCPs. The spatial distribution of the GCPs was highlighted, where vertical errors increased significantly after 100 m distance between GCPs with a flight altitude of 280–330 m above sea level (ASL).
The objective of this study is to test the viability of hydrological modelling of peatlands covering approximately 1 ha and to determine the amount of GCPs that are required to create reliable DSMs.
4.1. DSM Accuracy Assessment
Five different Sphagnum
dominated study sites in northern Tasmania were surveyed with a UAS, and DSMs were generated using different georeferencing scenarios. Overall, the vertical errors that were computed by Photoscan for the All_GCP scenarios (16-20 GCPs) were similar to findings of other studies that used ground control for UAS derived DSMs and found to be in the range of ~0–5 cm [28
]. For the Four_GCP scenarios, the errors remained similar to the All_GCP scenarios for the Heath and Jack’s lagoon sites, but increased to exceed 10 cm for the remaining scenarios. Additionally, doming in the DSMs occurred in the Four_GCP scenarios for Eagle Valley and Basin. It should be noted that the topography of Eagle Valley was naturally doming (altitude was highest towards the center of the site), which could have been an underlying reason for the doming effect that occurred for Eagle Valley Four_GCP. As previously mentioned, Tonkin and Midgley [28
] found errors below 10 cm when using four ground control points, whereas the study conducted by Clapuyt et al. [26
] revealed larger errors when using four corner points, with a mean of 0.31m, compared to a mean of 0.02m when using 15 GCPs, for a similar sized area. Clapuyt et al. [26
] came to the conclusion that georeferencing errors increase proportionally with distance between GCPs, resulting in larger errors and weak reproducibility with irregularly scattered GCPs. Additionally, James et al. [41
] stated that control precision, i.e., the effect of underestimation or overestimation of control point accuracy, is equally important as the GCP distribution. Harwin and Lucieer [27
] commented that variation in terrain should be considered with GCP distribution, where steeper terrain requires a higher GCP density.
A few hypotheses are presented that could explain the overall higher accuracy performance of Heath and Jack’s Lagoon, compared to the other sites, Basin, Eagle Valley, and Flat Valley. First, the DSMs of Heath and Jack’s lagoon differed from the other sites, as they were relatively square in shape, whereas the areas of the other sites were an elongated rectangle. Additionally, Basin, Eagle Valley, and Flat Valley were surrounded by steep terrain that was (partially) captured by the UAS and pencil pine trees were distributed over the area. These factors increased the amount of DSM variation in vertical direction and might have added a complexity factor to mapping the topography.
This study has highlighted that for reliable DSM production GCPs are required. Whilst using as few as four GCPs did produce accurate results for some trial sites, this method did not work in all cases (see Table 2
) and thus cannot be used confidently. An alternate method for removing the need for GCPs was tested during this study. That is, for one site (Basin) we collected imagery on two separate field trips, under different lighting conditions (one late in the evening, one during the afternoon). The afternoon imagery contained GCPs and was processed and aligned based on camera reference with Photoscan to create an orthophoto and DSM. The late evening imagery was then imported into the Photoscan job and aligned with the already controlled afternoon imagery. The original afternoon images were then disabled, and the orthophoto and DSM created based on the now accurately aligned evening imagery.
The aim of this methodology was to effectively control the second image dataset based on the first, if this were successful it would mean that once there is an accurately controlled model of a site, subsequent image collections would not need control as they could be matched to the earlier dataset. This method requires the scenes to have sufficient elements that remain unchanged over time, such that features can be matched between the two datasets. It was hypothesized that this would work for the peatlands as they contain a lot of rocks that will not change over relatively short time frames.
The method of aligning new images to a controlled model was not successful. Whilst the orthophoto produced with this methodology had an accuracy of 1–2 pixels in comparison to the fully controlled orthophoto, there were issues with the DSM. The DSM exhibited a significant artificial slope (>1°) similar to the Sky_GCP scenarios (see Table 2
). Thus, for the purposes of this study, which relies on accurate DSMs for hydrological surface modelling, this method also fails to eliminate the need for GCPs.
Therefore, future steps will focus on efficient ways for georeferencing areas that are to be monitored over time. Prospects for future approaches include using objects that stay consistent over time—e.g., exposed rocks—that can function as permanent GCPs. This could be achieved by painting markers on the rocks and collecting their position with a RTK DGPS. However, this method would still require extensive field campaigns for any new sites in which markers would have to be established and measured with a RTK DGPS. Also painted markers will have to be upkept to ensure they remain visible for future campaigns, and of course it will be necessary to ensure that the method used to mark the rocks does not introduce any harmful substances into the environment. Alternatively, onboard, accurate RTK GPS units and post processed kinematic (PPK) solutions are now becoming a populate and more cost-effective option of geotagging UAS imagery. It is yet to be seen if they can achieve a high absolute accuracy in a repeatable fashion and thus if they would be a viable option for creating accurate DSMs of Sphagnum peatlands.
4.2. Hydrological Models
The results of the hydrological modelling provide a good qualitative indication of flow accumulation in Sphagnum
dominated peatlands and are in line with the study by Lucieer et al. [33
], where the same TauDEM tools were used to simulate snowmelt in Antarctic moss beds. The topography of the study area of Lucieer et al. [33
] is similar to these study sites, as they both consist of low vegetation cover and rocks, with a complex hydrological micro-topography. For the Monte Carlo simulation, a vertical error had to be defined. In reality, this error spatially varies as it is related to geometry of the camera [33
]. Additionally, the accuracy assessment demonstrated the variability of the DSM vertical error between study sites and scenarios. However, as this study aimed to compare the performance of the DSM regarding the different scenarios, all other variables were kept constant.
The results from the Basin site showed that if the artificial slope present in the DSM is too large it is not possible to derive reliable hydrological maps. However, for the Heath site, the DSM generated without any ground control did not deviate significantly from the reference model, possibly because the Sky_GCP scenario contained artificial slopes (~0.35°) that were much less than the natural slope of the site (~6–7°). Small differences for the Heath site in streamlines became visible between the flow accumulation map of the GCP and Sky_GCP scenarios, which are mainly caused by the difference in size between the DSMs. The Heath reference DSM was built with photos covering a larger area than the Sky_GCP scenario. As the hydrological model input values of surface water are defined as the size of each pixel (10 cm), a DSM with a larger area results in higher accumulation values and can influence the flow direction.
This study has demonstrated the best methodology for the creation of accurate hydrological models of Sphagnum mires. This methodology will now be used to map and monitor the hydrology of Sphagnum mires in a representative selection of sites throughout the effected landscape as part of an ongoing restoration program. The ultra-high resolution DSM data is essential to detect the micro-topographic changes in the damaged mires and also to model the potential locations for restoration intervention.
The hydrological models could be further improved by including vegetation variables, such as water infiltration. However, because the Sphagnum sites that were visited during this study were largely damaged by fire, their water holding capacity is uncertain, hence this factor will be difficult to include. An overall objective of this project is to inform peatland rehabilitation, therefore the next step will focus on quantifying hydrological changes when artificial restoration interventions (e.g., dams) are added to the hydrological surface models with the aim to increase water availability to areas with damaged Sphagnum. Another approach that could be of significant importance to peatland rehabilitation, is the use of spectral signatures gained from high-resolution multispectral UAS imagery to identify different health states of Sphagnum, which is planned for future field campaigns.
The objective of this study was to generate reliable digital surface models (DSMs) of Sphagnum dominated peatlands from UAS, for the purpose of creating hydrological surface models, thereby assessing efficient ways for accurate georeferencing. This has been achieved by generating three different scenarios for five study sites in northern Tasmania. Sites Heath and Jack’s Lagoon presented the lowest vertical errors. The errors of All_GCP and Four_GCP scenarios were within a range of 0.03 m and for the Sky_GCP scenarios, errors remained below 0.5 m. For these two sites, artificial slopes in the DSMs were found for the Four_GCP scenarios, however these did not exceed 0.05°. DSMs without ground control showed a strong increase in artificial slope for both sites, reaching up to around 0.8°. The DSMs from the sites Basin, Flat Valley, and Eagle Valley resulted in significantly larger vertical errors for the Four_GCP scenarios and often exceeded 10 cm. Slopes for the Sky_GCP scenarios often exceeded 1°. Additionally, the slope assessment revealed doming of the Four_GCP scenarios for sites Basin and Eagle Valley. Basin, Eagle Valley and Flat Valley differed from Heath and Jack’s Lagoon by shape (elongated rectangle vs. square) and complexity of topography, where for the first three sites the trees and steep rocky ridges increased the amount of variations in the vertical axis of the DSMs.
After the DSM accuracy assessment, hydrological surface models were created for Heath and Basin, using the reference scenario with all ground control points (16–20) and the Sky_GCP scenario, where no ground control points were used. After a visual comparison with the orthophoto, the reference scenarios suggested to provide good qualitative representations of flow accumulation and topographic wetness. The artificial slope of the DSM for Heath for the Sky_GCP scenario (0.35°) was too small to cause significant changes in the hydrological models. For Basin, the Sky_GCP scenario contained an artificial slope that resulted in a false representation of water accumulation in the topographic wetness index model. The use of regularly distributed GCPs are necessary to generate a reliable topography from UAS for Sphagnum dominated peatlands. However, for areas where Sphagnum is badly damaged and further damage needs to be minimized, the use of at least four GCPs around the edges and potentially an additional GCP in the center of the area, would suffice. DSMs that are generated without any ground control method are not reliable enough to be used for hydrological surface models, hence the use of GCPs is strongly advised. Further research should demonstrate whether other approaches could be used to easily facilitate accurate georeferencing of UAS imagery.