Abstract
Ground-penetrating radar (GPR) is a non-intrusive geophysical observation method based on propagation and reflection of high-frequency electromagnetic waves in the shallow subsurface. The vertical cross-sectional images obtained allow the identification of thickness and lithologic horizons of different media, without destruction. Over the last decade, several studies have demonstrated the potential of GPR. This paper presents a review of recent GPR applications to peatlands, particularly to determine peat stratigraphy. An example study of acquisition and comparison of peatland soil thickness of a fen-dominated watershed located in the James Bay region of Quebec, using (1) a meter stick linked to a GPS RTK and (2) a GSSI GPR, is given. A coefficient of determination (r2 ) of 56% was obtained between the ordinary krigings performed on data gathered using both techniques. Disparities occurred mainly in the vicinity of ponds which can be explained by the attenuation of GPR signal in open water. Despite these difficulties – the higher time required for analysis and the error margin – it seems more appropriate to use a GPR, instead of a graduated rod linked to a GPS, to measure the peat depths on a site like the one presented in this study. Manual measurements, which are user-dependent in the context of variable mineral substrate densities and with the presence of obstacles in the substrate, may be more subjective.
I Introduction
The representation of shallow subsurfaces is of interest in several geophysical sciences such as hydrology, geomorphology and hydrogeology. Indeed, knowledge of geological layers is essential to locate potential drinking-water sources and other natural resources and highlight the risk areas for construction and natural disasters, to name a few examples. Moreover, several stratigraphic and rock studies have led to substantial discoveries about the history of the Earth and its landforms (Neal, 2004).
Invasive techniques such as excavations are time-consuming and expensive, provide discrete information, and often cannot be made because study sites are protected (Neal, 2004). In the 1970s, useful geophysical, non-invasive, techniques such as ground-penetrating radar (GPR, ground-probing radar, surface penetrating radar, subsurface radar, or impulse radar) were developed to characterize type of material, subsurface processes, internal structures and dimensions (Annan, 2003; Neal, 2004; Van Dam et al., 2002). Daniels (1996) and Reynolds (1997) highlighted that pulsed electromagnetic waves were first used in the mid-1920s. GPR, which is a non-destructive geophysical method, produces vertical cross-sectional images of the shallow subsurface, based on propagation, reflection and scattering of high-frequency electromagnetic waves within it (Benson, 1995; Beres and Haeni, 1991; Gawthorpe et al., 1993; Mellet, 1995). The resulting image is similar in style to seismic reflection profiles (Davis and Annan, 1989). The geological expertise required to interpret textural variations in radar data often limits the application to experienced users (Franke, 2012).
Improvement in the instrumentation and comprehension of the effect of electrical properties on wave propagation have allowed for an increase of the potential applications (Annan, 2003; Cassidy, 2008). Several studies have showed the effectiveness of GPR in different fields of applications (geomorphology, archaeology, sedimentology, hydrogeology, stratigraphy, civil engineering, etc.). Studies conducted about the use of GPR on peatlands are not numerous in the literature. However, such research may be useful to evaluate water storage in a watershed.
The objectives of this paper are: (1) to present a review of recent ground-penetrating radar studies on peatland stratigraphy; and (2) to introduce a case study of the use of GPR to determine the peat soil thickness of a north boreal peatland in Quebec (Canada).
II Recent peatlands stratigraphic studies using ground-penetrating radar in the literature
In the last 20 years, GPR has seen several developments and advances that have the potential to have an impact on how GPR is used for landform studies (Worsfold et al., 1986). The use of GPR is a good alternative to manual sampling which has a considerable uncertainty (Doolittle and Butnor, 2009; Rosa et al., 2009). The aim of this section is to include recent references of the use of GPR on peat deposit profiling. Table 1 presents ground-penetrating radar key references in geomorphology and hydrology and examples of application for each area.
Geomorphological and hydrological ground-penetrating radar key references.
Several studies have focused on the identification of the transition between mineral and organic soils (Dallaire and Garneau, 2008; Doolittle and Butnor, 2009; Lowry et al., 2009; Plado et al., 2011; Rosa et al., 2009; Sass et al., 2010). The location of this transition is generally useful for many environmental studies (stratigraphy, gas content, identification of groundwater paths, age of peat, etc.) or to estimate the volume of harvestable peat. GPR is an effective way to measure this parameter and the success depends on the high differences in the materials, the penetration depth and noise that produce minor reflections (Moorman et al., 2003). The strong electromagnetic (EM) wave reflection results from the sharp reduction in the volumetric moisture content between the peat and the underlying mineral soil (Comas et al., 2005a). Problems may arise when the contrast between subsurface layers decreases (Munroe et al., 2007).
In the literature, the GPR measurements are usually collected on peatlands using systems equipped with antennas with frequency varying between 100 and 200 MHz (for large peat depths) and 300–400 MHz (low peat depths), which provide a good compromise between depth coverage and resolution (Comas et al., 2005a; Kettridge et al., 2008). The spacing between traces can be up to 0.5 m (Comas et al., 2005a, 2005b; Kettridge et al., 2008). The sampling time window varies in different studies between 500 and 1000 ns depending on antennae frequency and assuming a dielectric constant (∊r) that is invariant with depth (Comas et al., 2005a, 2005b, 2011; De Oliveira et al., 2012). GPR surveys are often combined with other available methods to meet the specific aims of the studies. To help the interpretation and to validate the data obtained from GPR surveys, peat cores can be extracted or manual sampling can be done along the transects collected (Kettridge et al., 2008; Plado et al., 2011; Rosa et al., 2009).
The average wave velocity is often determined from (1) the time of propagation of the wave out of the reflection recorded from the peat-mineral soil contact observed in common-midpoint surveys (CMP) or (2) the measurement of the two-way travel time at invasive sampling locations where the peat-mineral soil contact was precisely measured (Comas et al., 2004, 2005a; Lapen et al., 1996; Slater and Reeve, 2002; Theimer et al., 1994; Warner et al., 1990). The processing steps are often limited to application of a time-varying gain (to equalize the amplitude of reflections with depth), a filter (‘dewow’ or band-pass, to remove the low-frequency noise), a static correction (to eliminate time delay between trigger and recording) and another static correction to account for the topography of these basin-scale surveys (Comas et al., 2005a; Kettridge et al., 2008).
Rosa et al. (2009) showed that 30 direct core samples are necessary to adequately calibrate the velocity of electromagnetic waves. If this number of measurements is performed on a given site, the GPR proves to be a good method to estimate peat thickness, with possible adaptations to take into consideration different geological settings and EM wave propagation (Rosa et al., 2009). GPR can provide a continuous image of the peat-mineral contact and yield a more accurate peat-volume estimate than interpolated manual measurements. Moreover, in areas with dense wooded heath vegetation and open pools, GPR measurements are difficult to take, thereby creating data gaps in this transect (Comas et al., 2005a). GPR can penetrate up to 10 m in peat soils, with approximate 0.25 m resolution (Comas et al., 2005b), mapping peat thickness at a high spatial resolution (Worsfold et al., 1986). Resolution is affected by greater antenna frequencies and depths (Plado et al., 2011).
The use of GPR on peatlands studies is particularly challenging. Compared to other geological media, the velocities are low (0.033–0.049 m/ns) and the relative permittivities high, due to high peat water content (Plado et al., 2011; Rosa et al., 2009). The peat heterogeneity may induce differences in pore-size distribution or other physical properties (Rosa et al., 2009). For example, clay or root layers or the varying degree of decomposition of the peat can excessively attenuate EM wave propagation and reduce its depth of penetration (Theimer et al., 1994). Given that the EM velocity can vary both vertically and laterally, the heterogeneity of peat may lead to distortion in radar reflection profiles (Neal, 2004).
GPR profiles, combined with coring at selected locations, are generally characterized by the presence of areas of EM wave scattering (loss of coherent reflectors) at shallow depths (Comas et al., 2005b). The wave velocity (v) shows small variations between the different microhabitats. Drier microhabitats show higher v values than wetter. The differences are attributed to peat porosities, biogenic gas content and water-table depth (Kettridge et al., 2008). Ice and snow could have an impact on the permittivities (Plado et al., 2011). The moisture-content changes within the peat are often responsible for numerous strong GPR reflections. For example, lake sediments can be associated with a distinct change in physical properties and high electric conductivity and an absence of reflectors (Comas et al., 2005a; Plado et al., 2011). On a GPR survey conducted on a fen, the transition from wooded heath vegetation to open-pool areas may lead to chaotic distortions (Comas et al., 2005a). The conclusions of many studies lead to the conclusion that GPR provides additional information, but cannot serve as an independent method (Plado et al., 2011).
III Case study: peat thickness determination in north boreal Quebec, Canada
Peatlands are recognized as important carbon sinks, biodiversity-rich ecosystems, and environments with substantial water storage (Charman, 2002; Holden, 2005a, 2005b, 2005c; Payette and Rochefort, 2001; Rydin and Jeglum, 2006). In Quebec, they represent nearly 12% of the total land cover, and are predominantly located in the James Bay region where they can account for 20–40% of some watershed areas (Payette and Rochefort, 2001). Observations indicate that many of the North Boreal peatlands in Quebec are subject to aqualysis, an enlargement and a coalescence of ponds to the detriment of vegetative strips that collapse and degrade (Dissanska et al., 2009; Payette, 2008; Tardif, 2010). Although they are located upstream of large hydroelectrical power plants, the hydrology and physiography of these ecosystems are poorly documented in the literature. Since physiographic data are prime inputs to hydrological modelling studies, there is a need to develop high-resolution digital elevation models (DEMs) to support inflow forecasting models and improve our understanding of the soil and surface structure of peatlands.
This paper presents the acquisition and the comparison of peat depth using manual measurements combined with GPR profiles. The derived data, combined with hydrometeorological data, when incorporated into mechanistic or conceptual models will allow for a better understanding of the hydrological behaviour of small drainage basins dominated by highly aqualysed fens.
IV Study site
The study site is a small drainage watershed dominated by a highly aqualysed fen (54°06’868” N, 72°30’083” W) located between Hydro-Québec LG-4 and LA-1 dams in North Boreal Quebec (River La Grande drainage basin). This site was chosen for its high proportion of surface water pools, small size, easily identified outlet and ease of access (Figure 1). The structured surface of this peatland appears to be typical of many of those found at this latitude. The peatland is formed by two oblong regions (named south and north), made of a series of ponds and strips parallel to each other and perpendicular to the general slope (i.e. the predominant flow direction), ultimately flowing in a larger pond just upstream of the outlet (Figure 1). There is a sphagnum sward in the upstream portion of the north section, making the latter about 1.5 times longer than the south section. Visually, upstream and downstream portions of the watershed have slopes much steeper than the mid-section, where the boundaries seem visually difficult to define. The bedrock consists mainly of tonalite and gneiss. Glacial deposits and rock outcrops form the hills (Dissanska et al., 2007). The predominant peatland vegetation in this region consists of small trees (Picea mariana (Mill.) BSP and Larix laricina (Du Roi) K. Koch), heaths (Kalmia polifolia Wangehn., Andromeda glaucophylla Link, Vaccinium oxycoccos L. and Chamaedaphne calyculata (L.) Moench) and sedges (Carex sp., principally). We find a dominance of bryophytes on swards and a significant presence of Sphagnum fuscum (Schimp.) and Sphagnum rubellum Wilson on strips and hummocks. Total precipitation in 2009 was 786 mm.
Location of the study peatland (54°06’868” N, 72°30’083” W).
V Methodology
1 Peat depth surveys
We performed depth surveys using two methods of measurements: (a) a meter stick (accuracy ± 0.5 cm) linked to a dGPS; and (b) a GSSI GPR system.
a Meter stick linked to a dGPS
To obtain positions (X,Y,Z) on the peatland, we used a dGPS (Trimble model 5800; http://www.trimble.com). The fixed GPS receiver and radio transmitter were installed on a geodetic control mark with known coordinates. We used the geodetic control mark 89KB053, located about 1 km away from the peatland.
To ensure proper communication with the ratio transmitter and receiver of the mobile GPS, the fixed GPS receiver must be configured using a notebook (Trimble TSC2; http://www.trimble.com). Subsequently, it can be positioned correctly. When the mobile GPS was located far from any obstacle on XY, we used a maximum displacement of 3 m from the radio transmitter to get an accuracy of 1 cm or less. At lower resolution, the range can be up to several kilometers. The accuracy in Z varies between 2 and 5 cm.
In the case of this study, we used the RTK mode (Real Time Kinematic) to eliminate signal interference. This method decomposes the mobile-receiver distance into wavelengths, providing real-time corrections and positioning (Bottom et al., 1997).
We sampled a total of 251 depths, with the area stratified by pond string and even spacing and points then selected randomly. We measured depths by pushing a 3 m graduated metal rod down until it struck the hardened layer of mineral soil. The sounding points were located using a dGPS, making it possible to construct a georeferenced database to be subsequently analyzed using a GIS.
b GPR profiles
The GPR was composed of one 400 MHz Shielded GSSI antenna and electronic transmitters (one enclosure contains both transmitter and receiver). This antenna system can provide profiles between 0 and 4 m deep (Geophysical Survey System, Inc., GSSI SIR-3000 Model; http://www.geophysical.com).
In March 2009, a total of 32 GPR transects were acquired on site. The total length of all transects was 7.08 km. During the survey, there was 30–50 cm of snow on the ground. The GPR data were georeferenced using a GPS RTK (data taken in real time, to obtain X,Y,Z coordinates).
A trench was dug on a sward of the north section beside a GPR measurement. The trench allowed for visual identification and measurement of the thicknesses of the different peat/soil horizons, thus providing a validation of the GPR signal acquired at that location. The same procedure was performed on a pond by digging a hole and measuring the thicknesses of different layers. In the trench on the sward strip, total depths of 70 cm of snow, 30 cm of frozen peat (ice) and 150 cm of saturated peat were measured above the mineral soil. In the pond hole, thicknesses of 50 cm of snow, 40 cm of ice, 200 cm of water and 10 cm of saturated peat were identified.
The interpretation of raw and unfiltered signals, at these locations, allowed for the distinction of changes in the penetration velocity of the electromagnetic waves, corresponding to the transitions between the different layers. The Radan Data Analyzer v.6.5 software (Geophysical Survey System, Inc.; http://www.geophysical.com) supplied with the GPR was used. The following equation was used to calculate the velocity of wave penetration (mm/ns) in the various substrates:
where v is the velocity (mm/ns), D the distance (mm) and t the time (ns). Velocities of 200 mm/ns in the snow, 150 mm/ns in the frozen peat, 33 mm/ns in the water and 35 mm/ns in the saturated peat were identified and used.
After this validation, GPS RTK data were analyzed using a GIS (ArcGIS v.9.2, ESRI; http://www.esri.com) to form a data line (polylines). The 32 transects were divided into 50 sections and turned into points, from which an attribute table was generated. The GPR profiles located closest to the points corresponding to the meter stick surveys were selected for comparison (n = 148) (Figure 2).
Watershed land cover with location of the meter stick and GPR surveys for peat depth measurement.
Each transect was inserted into the Radan Data Analyzer v.6.5 software (Geophysical Survey System, Inc.; http://www.geophysical.com). The 148 profiles selected from the aforementioned methodology were interpreted. Different filters were tested on profiles to reduce the noise but their application was inconclusive and, therefore, the raw profiles were analyzed. The depths of the lithological layers were determined with the penetration velocities found by validation.
c Topographical survey
A total of 1416 GPS RTK points were picked around the ponds, at the limit between water and strip. A lot of points were also taken randomly on the site, allowing for the determination of the watershed limits.
2 Geostatistical interpolation
a Physical characteristics of the peatland drainage basin
A panchromatic image acquired on 08/08/2009 by GeoEye (resolution 0.50 m) was integrated into a GIS (ArcGIS v.9.2, ESRI; http://www.esri.com). The latter was georeferenced using the positions (X,Y,Z) obtained by dGPS. This procedure helped to define the watershed boundary of the study site.
b Interpolation and data elevation model (DEM)
The dGPS data (X,Y,Z) were converted into vector data and integrated into an ArcMap project. Interpolation by ordinary kriging was performed (Govaerts, 1997). The kriging estimator was a weighted average of the observed values z(x
i) used to estimate the variable identified at a specific location x
0 where there are no measured values (equation 2):
where λi are the weights of the estimator that minimize the variance of the estimation error (ordinary kriging weights). In ordinary kriging, the weights are estimated under the assumption of constant variable mean within a neighborhood and of the constraint that weights must add up to one (Govaerts, 1997).
A Gaussian semivariogram model was applied to the anisotropic experimental semivariogram. A cross-validation was performed using a jackknife resampling approach and the Root Mean Square (RMS) was used to assess the performance of the interpolated topographies of the DEM (surface and subsurface). The calculation of RMS is performed using equation (3):
where a corresponds to the difference between the measured coordinate and that produced by the interpolation of the points, and n is the number of points.
The subsurface topography of the site was plotted using the same process, but this time interpolated from data acquired by the two methods of measurements: (1) depth surveys with a graduated meter stick; and (2) GPR transects. The depths obtained by these methods were subtracted from the Z coordinates obtained by dGPS. This produces a mineral base elevation relative to sea level. The difference between Zgpr and Zms, EZ
, calculated from equation (4), was also interpolated the same way:
where every Z and EZ
are in centimeters. The latter interpolation was intended to highlight areas where a method overestimated the depth of peat compared the other and vice versa. A relative bias was also calculated from equation (5):
where n corresponds to the number of sampling points. The kriging characteristics (semivariograms, cross-validations and predicted errors) can be found in Table 2.
Characteristics of ordinary kriging performances.
The results of kriging were cookie-cut to the shape of the watershed and converted into vector data. Two TINs (Triangular Irregular Network), generated by processes of triangulation representing the relief of a medium, were created. Five transects were drawn on TIN, each comprising 100 surface points and depth surveys from an area that included a margin of 30 m on each side of the transect. Once these coordinates were obtained, it was possible to produce multilayered profiles.
VI Results
This section presents the quantitative physical characteristics and the different kriging results of the study watershed.
1 Peatland land cover
Table 3 and Figure 2 quantitatively and qualitatively describe the land cover characteristics of the peatland watershed obtained by GPS RTK surveying and validated by the GeoEye orthoimage. The drainage basin occupies 125,490 m2 with 71.9% forest, 19.1% peat soil and 9.0% water. The north section is the largest part of the peatland at 49.0%, followed by the south section at 36.8%, and finally the region of the large pond and outlet covering 14.2%. Surface water covers 31.0% of the peatland area. Of that percentage, 43.0% is located in the south section and 29.0% in the north section. The large pond near the outlet has the same water surface area as the north section, on a smaller area, thus making the peatland aqualysis largest downstream.
Land cover of the study watershed.
2 Surface and subsurface topographies obtained by two measuring methods
Figure 3 shows the surface topography of the peatland drainage basin obtained by ordinary kriging (RMS = 0.28 m). The relief ranges from 439.20 m (downstream) to 448.82 m (upstream) above sea level with an average slope of 1.77%. The steeper areas of the watershed are located near the outlet with slopes ranging from 5 to 6%. The elevation of the peatland subsurface, calculated from the data acquired by the graduated meter stick and GPR, varies from 437.35 to 444.01 m (average slope of 1.83%) and 438.00 to 443.13 m (average slope of 1.22%), respectively. All slopes were calculated in the direction of the mean flow or –135° relative to north (Figure 3).
Topography of the peatland watershed obtained by ordinary kriging (RMS = 0.28 m). The dotted line corresponds to the transects on which the surface slope was calculated.
The comparison to ordinary kriging, performed using data acquired by the GPR and meter stick surveys, is shown in Figure 4. The manual sampling, with 103 extra points, has a slightly lower RMS: 0.59 m instead of 0.61 m. Figure 5 shows the relationship between the two methods of depth measurement. The coefficient of determination (r2 ) was 0.56 and in both cases the average is 108.70 cm. On average, for peat depths between 100 and 150 cm, both methods are equivalent. Nevertheless, the GPR data do not exceed 210 cm. The relative bias of GPR measures compared to meter stick surveys is 0.26 (n = 146). For values above 175 cm, the relative bias is –0.26 (n = 19) and 0.34 for values below 175 cm (n = 127).
Comparison of peat depths derived using ordinary kriging of data gathered by GPR (RMS = 0.61 m) and meter stick surveys (RMS = 0.59 m).
Relationship between peat depths obtained by GPR and meter stick surveys (r2 = 0.56).
Ordinary kriging was performed using the results obtained by equation (1) (Figure 6). Three categories were identified (EZ < –10.87 cm; –10.87 cm < EZ < 10.87 cm; EZ > 10.87 cm), where the difference in depth EZ = 10.87 cm corresponds to an error threshold of 10%. This threshold was calculated from the average depth of data obtained by the GPR and graduated meter stick (the average is identical in both cases). The largest differences appear to be around the ponds. The meter stick survey values are generally larger than the GPR data, particularly within the larger pond and where the north section’s ponds are situated. The two peat depth measuring methods seem similar where the peat is thin and the slope is more abrupt.
Spatial representation of the difference between the peat depths obtained by GPR and meter stick surveys (10% error). Ez represents the meter stick depth subtracted from the Georadar depth, in centimeters.
3 Multilayered traverse profiles and peatland DEM
Figure 7 shows five stratigraphic profiles. On each graph, the surface and subsurface layers calculated from GPR data and graduated meter stick can be seen. Peat thickness seems more important in the downstream portion of the watershed and in the south section (generally between 150 and 280 cm) when compared to the upstream portion of the peatland (between 70 and 100 cm). The peat depth seems thinnest on the swards. Comparison of depths curves shows a high variability and few outliers (transverse transect) in the case of GPR data. In addition, this method suggests a smaller range of values of peat depths (0–210 cm). Nevertheless, this restriction may be related to the choice of the antenna or the device programming.
Five peatland multilayered profiles showing the surface and subsurface topographies obtained by GPR and meter stick surveys.
VII Discussion
GPR has proved to be a good alternative to manual sampling to estimate peak thickness (Doolittle and Butnor, 2009; Rosa et al., 2009). Section II showed that the successes of GPR surveys are affected by the high differences in the materials (Moorman et al., 2003). For fens such as the one in our case study, i.e. with open pools, Comas et al. (2005a) highlighted the fact that GPR measurements are difficult to take during the ice-free season and hence winter surveying is preferable. However, Plado et al. (2011) mentioned that the GPR profiles taken in the winter can be problematic to interpret as the snow and ice can generate differences in permittivities. Finally, strong EM wave reflections, resulted by moisture and porosity changes for example, can cause problems to identify the transition between mineral and organic soils, which can also lead to uncertainties (Comas et al., 2005a; Munroe et al., 2007). On the other hand, manual sampling can definitively help to calibrate the velocity of electromagnetic waves (Rosa et al., 2009) and to validate the data obtained (Kettridge et al., 2008; Plado et al., 2011; Rosa et al., 2009). The following section will discuss the comparison of the two measuring methods and the peat matrix thickness spatial variation through the ecosystem studied.
1 Comparison of the two measurement methods
Two peat depth measurement methods were compared in this article: acquisition of GPR profiles and depth surveys with a graduated meter stick.
Figures 4 –6 show a smaller range of depth values derived from GPR profiles (maximum depth = 210 cm). The relative bias for all sampling points (n = 146) suggests that the meter stick data overestimate thickness by 26% relative to measures obtained by GPR. For peat depths below 175 cm, this bias becomes greater (34%). Above this threshold, it appears that GPR overestimates thickness relative to a meter stick by 26%. The depths obtained by both methods are similar upstream of the north section ponds and in the sward between the two sections and the largest pond, locations corresponding to average depths of 120 cm. In the swards, the depths obtained by GPR seem more important. Near ponds, the meter stick generally gives deeper values in the north section and in the downstream portion of the site (larger pond area). The GPR depths are more important in the south section.
Wave propagation through a medium depends on the sources of attenuation encountered (Dallaire, 2010). When attenuation is high, the signal penetration is reduced (Cassidy, 2008; Davis and Annan, 1989). The wave frequency and electric conductivity of a material, like a saturated soil (peat, for example), may attenuate the signal (Cassidy, 2008; Dallaire, 2010). This limit may have affected our field validation. Indeed, the interpretation of all profiles was based on velocities (mm/ns) validated by the observations made in the trenches. The uncertainty associated with this calculation could be extrapolated on all data. Jol and Bristow (2003) reported that a first interpretation of the profiles should be made prior to visual observations in the trenches in order to preserve objectivity when interpreting.
The choice of the antenna frequency proved adequate. Indeed, the antenna was in direct contact with the ground and the box protected the receiver from the solar radiation. Its range (0–4 m) was appropriate for the survey. An 800 MHz antenna provided a higher resolution but a restricted range, and vice versa for the 250 MHz antenna.
A coefficient of determination of 0.56 was obtained between the two types of measurements. These two techniques differ greatly. The GPR is an electromagnetic, geophysical, non-invasive technique (Bano, 2000; Gawthorpe et al., 1993; Healy et al., 2007; Neal, 2004). Peat depths are extracted following interpretation of geophysical profiles. They are directly related to the observer’s experience, the signal’s sharpness and the amount of interference encountered. In addition, the accuracy is ±25 cm depth, which is not negligible. It was more difficult to interpret the GPR signals and to extract the depths from areas that were covered with water, such as ponds (Dallaire, 2010). Indeed, the signal is strongly attenuated when the water table is near the surface (Cassidy, 2008). The water is the medium with the largest dielectric constant (David and Annan, 1989). The minerotrophic peatlands are more difficult to characterize than ombrotrophic peatlands, because the number of ponds is larger (Dallaire, 2010). Several interferences are detected at these levels (Comas et al., 2004). Indeed, the GPR profile nearest to the meter stick measurement was sometimes impossible to interpret. A point located further away had to be selected, which may have led to greater differences in the vertical structure of the profiles that were compared.
Manual surveys are often used to validate GPR data in the literature (e.g. r2 = 0.94 by Hänninen, 1992; r2 = 0.89 by Holden et al., 2002). They are also imparted with considerable uncertainty (Doolittle and Butnor, 2009; Rosa et al., 2009). The difference between the two methods in this study (r2 = 0.56) can be explained by numerous caveats in the field protocol: the sampler’s strength may vary, the flexibility and the possible obliquity of the graduated rod penetrating the peat layer may all be sources of error (Jol and Smith, 1995; Worsfold et al., 1986). The difference can also be the result of spatial variability (distance between comparison points up to 3 m). It may be influenced by microtopography. The rod can hit obstacles in the peat column (branch, rock, etc.). Authors have highlighted that the nature of the contact between peat and mineral layers can lead to additional uncertainty (Jol and Smith, 1995; Worsfold et al., 1986). Indeed, if the mineral substrate beneath the peat layer is unconsolidated, like clay or sand, the rod can penetrate beyond the peat layer.
The meter stick used in this study had a length of 3 m. Nevertheless, this limit did not affect the data, because only one measurement reached 3 m. It is important to consider that the GPR and the manual data were collected in March and October, respectively. Freeze-thaw cycles regulate partly the expansion and contraction of peat (Fritz et al., 2008). Changes of the order of a few millimeters may be generated between winter and spring depths.
In reviewing the performance of both methods, it seems more appropriate to use a GPR to take depth data on fens. This instrument constitutes an objective technique to measure peat depths and to take large data sets in a short period. Nevertheless, we have to consider a margin of error associated with the interpretation of the profiles. These uncertainties could be attenuated by the realization of electric conductivity measurements, pH and apparent permittivity tests (Dallaire, 2010). A larger number of validation points, using trenches in different places, would allow one to calculate a propagation velocity mean in each substrate and to verify the nature of bedrock. The use of a graduated meter stick to do manual surveys may in some cases be seen as suffering from greater subjective interpretation (Jol and Smith, 1995; Worsfold et al., 1986).
According to the cross-validation statistics presented in Table 1, the RMS data obtained by the meter stick is slightly lower (RMS = 0.59 m) than those resulting from the GPR (RMS = 0.61 m). It is, however, important to consider the larger number of points sampled by the meter stick. Nevertheless, the uncertainty associated with the geostatistical interpolation method is also important. Fitting a semivariogram manually in ArcGIS includes a certain margin of error. Furthermore, ordinary kriging involves a risk of errors in non-sampled areas. It provides an accurate estimate of the measurement points, but on the other hand produces significant discontinuities in the vicinity of these points (Govaerts, 1997).
2 Spatial variation of the peat matrix thickness
The thickness of the peat matrix is greater downstream and in the south section than in the upstream portion of the study site. It is important to consider the differences between both methods. The peatland subsurface slope from the meter stick surveys is more pronounced (1.83%) than the subsurface slope determined from GPR data (1.22%). These findings make sense considering the greatest range of values for the same distance.
The deepest peat areas are located in the downstream portion of the site and at both places where the ponds have a larger area. These sites have the lowest topographic levels. Indeed, the slope is less pronounced (1.06–1.09%). The classic model of accumulation of peat over a long period seems to be applicable. Filling by lake mud, coal and ligite is a major mechanism of peatland formation (Payette and Rochefort, 2001; Rydin and Jeglum, 2006). The stages of peat depend on vegetation bounding, topography and drainage basin (Payette and Rochefort, 2001). In recent years this mechanism may have been replaced by a dynamic coalescence of ponds that may evolve towards shallow lakes. This hypothesis should be verified by paleo-environmental reconstructions methods, such as palynological studies (e.g. Payette and Rochefort, 2001).
Another question can be raised from the difference in thickness between the north and south sections. An elevation formed by rocks and hummocks is located between the two sections that may be the remnant of a boundary between what used to be two independent peatlands, which have grown and merged with time. A large sphagnum sward forms half of the north section (from upstream to downstream), and the peat thickness is shallower there. This pattern could be the result of a paludification process in which peatland extends laterally following a change in water regime (e.g. Payette and Rochefort, 2001), the peat matrix here gradually invading the forest edge of the upstream site.
VIII Conclusion
At the beginning of this study, two basic questions were raised. (1) Do methodologies used to collect GPR data on peaty soils to characterize the stratigraphy yield similar results? (2) More specifically, is there a significant difference between peat depth measurement using a GPR and a manually driven meter stick?
In the literature, the number of papers reporting the use of GPR on peatlands has greatly increased. GPR can penetrate up to 10 m (±0.25 m) in peat soils (Comas et al., 2005b), allowing the possible creation of high spatial resolution maps (Worsfold et al., 1986). However, GPR surveys are challenging on peatlands as there are numerous areas conducive to EM wave scattering (Comas et al., 2005b). Indeed, the moisture content fluctuations within the media, the porosity differences, the water-table depth, the biogenic gas content, and the presence of root or clay layers could be responsible for numerous strong GPR reflections (Comas et al., 2005a; Plado et al., 2011). Many authors have highlighted the fact that to obtain a good modeling of an environment, with peatlands as an example, GPR data have to be combined with other methods to fill in the gaps, to validate the data (Comas et al., 2005a; Neal, 2004; Plado et al., 2011) and to reduce the uncertainties associated with interpolation (Rosa et al., 2009).
The two techniques examined for measuring peat depth have a coefficient of determination (r2 ) of 56%. Disparities between the two methods occurred mainly in the vicinity of ponds, this observation may be caused by the difficulty of interpreting the GPR profiles taken over water. Indeed, the signal is attenuated when the water table is near the surface (Cassidy, 2008), due to its high dielectric constant (Davis and Annan, 1989). However, it seems more appropriate to use a GPR to take peat depth measurements on a site like the one introduced in this study. This technique allowed for the gathering of an important number of profiles in a short period. Nevertheless, the interpretation time and the error margin must be considered. The use of a graduated rod in the context of variable mineral substrate densities and with the presence of obstacles such as woody debris within the peat layer may be more subjective (Jol and Smith, 1995; Worsfold et al., 1986). Multilayered transects showed that the lowest points of the study site are those with the greatest thicknesses of peat. These locations also correspond to those located near the largest pools, indicating that the likely mechanism of peat accumulation in the past may have been lake infilling. This assumption could possibly be validated by paleo-ecological studies. These data, in addition to future stratigraphical, ecological and paleo-ecological studies, will serve as the basis of a conceptual physiographic model to be used in a hydrological model, allowing for the advancement of knowledge on the small watersheds dominated by minerotrophic fen ecosystems and further understanding of hydrological processes in the middle boreal north of Quebec.
Footnotes
Acknowledgements
We would like to thank Serge Payette (Université Laval), principal investigator of the CRD-NSERC grant (Ecohydrology of minerotrophic peatland of the La Grande River Watershed, Northern Quebec: water cycle, CO2 and CH4 monitoring), as well as the remote sensing team of Monique Bernier and Karem Chokmani of the Institut National de la Recherche Scientifique, Eau Terre Environnement (INRS-ETE). We are grateful to the anonymous reviewers who provided detailed comments and suggestions which improved this article.
Funding
This research was financially supported by the Collaborative Research and Development (CRD) program of the Natural Science and Engineering Research Council (NSERC) of Canada, Hydro-Québec and Ouranos.