Towards operational monitoring of a northern wetland using geomatics-based techniques
Introduction
Wetlands are considered an integral part of the global ecosystem as they prevent or reduce severity of floods, feed groundwater aquifers and provide a unique habitat for flora and fauna (Mitsch & Gosselink, 1993). Because of this, many wetlands around the world are protected and monitored by various agencies. Further, the importance of wetlands is recognised by international treaties, such as the Ramsar Convention on Wetlands. Their dynamic hydrological characteristics and frequently complex terrain means wetlands are often difficult to monitor in situ. In addition, many wetlands are situated in remote locations with limited access and may cover extensive areas, as is the case with the wetlands of the Peace–Athabasca and Mackenzie deltas in northern Canada, the Amazon River floodplain in Brazil and the Okavango Delta in Botswana. Because satellite remote sensing provides a large spatial view, combined with a relatively high return frequency, it is a very cost-effective tool for wetland monitoring. Remote sensing has been used for wetland mapping since the launch of ERTS-1, the first satellite of the Landsat MSS series, in the 1970s (Best & Moore, 1979, Wickware, 1978). While the 80-m spatial resolution of Landsat MSS was often considered too crude for mapping vegetation types where the spatial mix was high (Gammon et al., 1979, Wickware & Howarth, 1981), today, the spatial resolution of optical satellite imagery, such as SPOT-4, SPOT-5 and Landsat TM (10–30 m), and very high resolution optical sensors, such as IKONOS and Quickbird (between 2.5–4.0 m for the multispectral scenes and 0.6–1.0 m for the panchromatic scenes), makes them ideal for capturing small features. Unfortunately, increases in resolution capabilities also increase the costs of imagery collection.
The objective of this paper is to illustrate how multi-sensor and multi-platform remote sensing techniques can be used in the operational monitoring of spatio-temporal changes in a northern wetland. Existing and time-tested techniques are used to generate spatially distributed baseline data on a timely-basis. The importance of continuous monitoring is demonstrated by the usefulness of the time-series data. Moreover, these generated data are combined in an easy and innovative way to evaluate the hydro-ecological relationships in the study area.
The most notorious limitation of optical imagery is that its wavelengths are short enough (0.4–3.0 μm) to be scattered by clouds and dense smoke, and can therefore only be used for monitoring under clear-sky conditions. Because atmospheric scattering decreases with increasing wavelength, the longer wavelengths of SAR, such as the C-band sensors (∼5 cm microwaves) on the Canadian Radarsat-1 and European ENVISAT satellites, are able to penetrate clouds and are not limited to clear weather monitoring. SAR sensors also have the potential to penetrate vegetation cover and detect sub-canopy conditions which can be very beneficial for wetland monitoring. Moreover, since microwave backscatter is influenced by the roughness and the dielectric constant of the target, SAR sensors are ideal for hydrological applications.
For example, although the presence of water (high dielectric constant) enhances the radar signal, a smooth water surface acts as a specular reflector of microwaves and the energy is reflected away from the sensor, resulting in a very low backscatter signal. Most other natural targets, such as vegetation canopies, are heterogeneous with a high roughness and the microwave energy is scattered diffusely. For most dry natural land surfaces, a portion of the energy is scattered back to the sensor, generating an intermediate backscatter signal. The presence of standing water beneath the vegetation cover produces increased backscatter since the radar signal is enhanced and reflected off the water surface and then re-directed back to the sensor by the vegetation trunk. This type of corner reflection, or double bounce effect, is described by Richards et al., 1987a, Richards et al., 1987b and Wang et al. (1995) for L-band SAR sensors, which have longer wavelengths (∼25 cm microwaves) and better vegetation penetration capabilities when compared with C-band SAR sensors. C-band microwaves are also subject to the double bounce effect as long as the vegetation attenuation by volume scattering is not too large.
In spite of the fact that L-band SAR may be better suited for detecting standing water beneath dense vegetation canopies, there are currently no active satellites that operate in the L-band spectrum. Imagery from formerly active sensors or shuttle missions has been successfully used to map inundated wetland areas (Hess et al., 2003, Pope et al., 1997, Wang, 2004a). Research has also shown that C-band Radarsat SAR is able to detect standing water beneath fully foliated shrubs (Töyrä et al., 2001) and forest cover (Townsend, 2001). On the other hand, Radarsat SAR has difficulties separating inundated shrub or forest from other dry vegetation or non-vegetated areas, a problem which Töyrä et al. (2001) solved by combining the SAR data with optical imagery, and which Townsend (2001) averted by masking out non-forested areas. Arzandeh and Wang (2002) illustrated how grey-level co-occurrence matrix (GLCM) texture measures can also be used to increase the accuracy of Radarsat-based wetland delineation. The classification accuracy increased from 22% (Kappa coefficient) using Radarsat image alone, to 74% using a combination of texture measures. Landsat or SPOT images have also been used to delineate maximum flood extents (Hudson & Colditz, 2003, Wang, 2004b) and to map areas of open water (Frazier & Page, 2000, Pietroniro et al., 1999), but have the distinct disadvantage of not being able to detect water beneath vegetation canopies.
Mapping wetland vegetation is a difficult task because of its high spectral and spatial variability. Schmidt and Skidmore (2003) point out that while there have been successful cases of mapping forest type or major physiognomic classes (such as shrub cover versus grass cover), the differentiation among various grasses and herbs remains difficult. Landsat TM and SPOT imagery are often used to map general vegetation classes (May et al., 1997), while hyperspectral data (Schmidt & Skidmore, 2003) and the new very high resolution sensors (Sawaya et al., 2003, Wang et al., 2004) are evaluated for more detailed vegetation applications. Kite and Pietroniro (1996), Ozesmi and Bauer (2002) and Pietroniro and Leconte (in press) provide good reviews of remote sensing applications in hydrology and wetland mapping.
From an operational management perspective, a single image showing the vegetation types or surface water extent of an area simply provides a map for the area. However, if a number of these maps are generated on a regular basis, various relationships between the vegetation and other parameters can be explored and changes can be monitored. Some examples could include flood frequency and duration, drainage patterns, vegetation change and vegetation flood resistance/preference. When combining such information with elevation data, even more relationships can be investigated. Elevation data can be interpolated into a Digital Elevation Model (DEM), which may be used to estimate the water levels required for overbank flooding into wetland basins and riparian environments, and the spatial extent of flooding at various water levels. These elevation data may also be used as input in hydrodynamic models (Pietroniro et al., 2001) and combined with temporal flood and vegetation maps to examine the relationships between vegetation types, elevation and flood duration. Such elevation data could verify the flood maps and vice versa, and both data sets could be used to validate various hydrological models. For example, Townsend and Walsh (1998) developed a floodplain inundation model for the Roanoke River in North Carolina using a DEM with 1 m vertical resolution and digital hydrography data together with known flood elevations that could be tied in to river position and floodplain location. They used a series of L-band J-ERS, C-band ERS-1 and Landsat TM imagery to map the inundated areas and found that flood outlines generated by the GIS model were similar to the radar derived outlines. Townsend and Walsh (1998) also noted the optical Landsat TM images were less successful at identifying inundated vegetation than the two radar image types. Cobby et al. (2003) used airborne scanning LiDAR to retrieve ground elevations as well as top-of-vegetation elevations for a 17 km stretch of the River Severn flood plain in UK. They illustrated how these data could be used to improve a two-dimensional river flood model by providing model bathymetry as well as vegetation heights for parameterization of model friction. Accurate flood mapping of low relief wetlands requires detailed and accurate elevation data since a relatively small change in water level can result in large changes in inundated areas. Dense wetland vegetation can often obscure the inundated region, thereby reducing the chance of extracting highly detailed and accurate ground measurements.
This study was conducted in the PAD, located in northeastern Alberta, Canada (Fig. 1). The delta is a large wetland complex that was formed on the western end of Lake Athabasca where the Athabasca River joins the Peace River in the north through three connecting channels (Fig. 1). The size of the delta varies depending on where the boundary is drawn, but the total delta area mapped by this study is 5600 km2. The delta is located within Wood Buffalo National Park (WBNP) and has been deemed a wetland of international significance by the Ramsar Convention. The PAD geology is characterised by Precambrian Canadian Shield granites and gneisses to the north and east, Precambrian Athabasca Sandstone to the southeast and Devonian limestone and gypsym to the west (Bayrock & Root, 1972). Aside from the river levees and outcrops of the Canadian Shield in the northeastern region, the delta is comprised of large areas of alluvial sediment deposits where the relief seldom exceeds 1 m above the major delta lakes (PAD-TS, 1996). The PAD possesses a Boreal forest climate, or more specifically, a cold, snowy forest climate with cold, moist winters and cool, short summers. 4 (2002) points out that the limited extent of peatland, typical of the Boreal forest region, and the high degree of marshes, meadows and savannahs, actually makes the delta similar to the Continental Prairie wetland region of Canada. The average date of Peace River freeze-up and break-up (measured between 1973 and 1992) is November 21 and April 28, respectively (Conly & Prowse, 1998). The hydrology of the delta is complex, with water flowing north into the Peace River during most of the year, but occasionally shifting southward depending on the relative water levels in Peace River, Lake Athabasca and the connected delta lakes (Lake Claire and Mamawi Lake). Numerous smaller wetland basins are disconnected from the river network and rely on large overland floods to replenish and maintain their productivity. The discharge in the Peace and Athabasca Rivers usually peaks in late April or early May during ice break-up and once more in late May or early June due to upstream snowmelt runoff (Prowse & Lalonde, 1996). The large overland floods are generally not caused by these high water flows, but rather by occasional ice-jams that may form in the rivers during break-up (Prowse & Lalonde, 1996). These ice-jams produce a backwater effect that causes the water to spill over the river levees and into the wetland basins.
The hydrology of the PAD has received much attention since the construction of the W.A.C. Bennett Dam and Williston Reservoir on the Peace River (approximately 1100 km upstream of the delta) in 1968 (PAD-TS, 1996). There was concern that these control structures would reduce the flood frequency, and thereby the productivity, of the wetland complex. After the construction of the dam, the delta experienced major floods in 1972, 1974, 1996 and 1997. There were no major overland floods during the 22 year period between 1974 and 1996. Since the ecology of the PAD is driven by the hydrologic regime, there have also been many studies assessing the effect of flood frequency on the vegetation patterns (PAD-PG, 1973, PAD-TS, 1996, Timoney, 2002). The results in PAD-PG (1973) implied that reduced flooding and declining water levels would allow willows to encroach into the wetland basins and replace the productive emergent vegetation. Prowse et al. (1996) related the 1974–1996 absence of large overland floods to reduced occurrences of break-up ice-jams due to a combined effect of flow regulation along the Peace River and a decline in spring snowpacks caused by climatic variability. Timoney, 1996, Timoney, 2002, on the other hand, has suggested that the supposed reduction of flood frequency has no statistical basis and that wetland ecology is much more resistant to change than previously believed. Although there is still considerable scientific debate on the frequency of flooding and the role water plays in the encroachment or senescence of vegetation, it has become clear that an effective means of assessing these changes both spatially and temporally is required. The large size, challenging terrain and the remoteness of the PAD, along with the lack of detailed and accurate topographic information presents an ideal opportunity for employing remote sensing as a monitoring tool.
The authors have conducted a number of remotely sensed flood, vegetation and elevation studies in the PAD (Pietroniro et al., 1999, Töyrä et al., 2001, Töyrä et al., 2002, Töyrä et al., 2003), as have others (Adam et al., 1998, Dirschl et al., 1974, Jaques, 1989, Jaques, 1990, Jaques, 1998, Terrain Resources Ltd., 1995, Wickware, 1978, Wickware & Howarth, 1981). Jaques, 1989, Jaques, 1990 used a series of Landsat MSS imagery from 1974, 1975, 1976, 1977, 1979, 1981, and 1985 to map the changes in flood outlines. These flood outlines were based on open water or areas with sparse vegetation since Landsat MSS cannot detect water beneath dense vegetation. Based on the results, Jaques (1989) generated a map showing areas with open drainage, restricted drainage and severely restricted drainage for the PAD. Jaques (1998) used Landsat MSS imagery from 1976 and 1989 to map the wetland complex into open water, productive vegetation (emergents, grasses and sedges, meadows and meadows with sparse willow cover), and unproductive vegetation (bedrock outcrops, meadows with dense willow, willow, deciduous, and coniferous vegetation). Despite these studies, however, there has been no systematic approach to wetland monitoring which makes use of existing satellite technology to generate and combine these data to develop an accurate and effective monitoring and management system.
Section snippets
Methodology
As part of an overall geomatics-based management framework for the PAD, the flood extent, vegetation and topography were mapped using remotely sensed data.
Flood mapping
Classification accuracies obtained by Töyrä et al. (2001) were used as a guideline to estimate and understand accuracies of the flood maps since the same study area, methodology and data types were used in this study. Table 3 lists the classification accuracies that can be expected for the various image combinations. Waves on water surfaces often increase the radar backscatter, sometimes to the point where open water is confused with dry vegetated areas. Radarsat S1 and S2 images are especially
Conclusions and recommendations
A time-series of flood maps was generated for the PAD using combinations of SAR and optical satellite imagery. The derived flood maps provided useful quantitative estimates of the flooding extent and the subsequent drying of the delta. In 1996 and 1997, flood waters drained and flooded vegetation was replaced by non-flooded land. A positive correlation between the extent of total flooded area and water levels was found for the dates when most of the flooded areas were openly connected. This
Acknowledgements
This work could not have been completed without the help and support of many people. The authors would like to acknowledge Northern River Ecosystem Initiative (NREI) and Canadian Space Agency (under the Government Related Initiatives Program) for funding the satellite remote sensing study and BC Hydro for funding the LiDAR study. Wood Buffalo National Park provided field logistics and Kelly Best (National Water Research Institute), Tom Carter (National Water Research Institute), Krysha Dukacz
References (61)
- et al.
Dual-season mapping of wetland inundation and vegetation for the central Amazon basin
Remote Sensing of Environment
(2003) - et al.
Flood delineation in a large and complex alluvial valley, lower Pa'nuco basin, Mexico
Journal of Hydrology
(2003) - et al.
Detecting seasonal flooding cycles in marshes of the Yucatan Peninsula with SIR-C polarimetric radar imagery
Remote Sensing of Environment
(1997) - et al.
Extending satellite remote sensing to local scales: Land and water resource monitoring using high-resolution imagery
Remote Sensing of Environment
(2003) - et al.
Spectral discrimination of vegetation types in a coastal wetland
Remote Sensing of Environment
(2003) - et al.
Modeling floodplain inundation using an integrated GIS with radar and optical remote sensing
Geomorphology
(1998) - et al.
Multisensor hydrologic assessment of a freshwater wetland
Remote Sensing of Environment
(2001) - et al.
Understanding the radar backscattering from flooded and non-flooded Amazonian forests: Results from canopy backscatter modeling
Remote Sensing of Environment
(1995) - et al.
RADARSAT flood mapping in the Peace–Athabasca Delta, Canada
Canadian Journal of Remote Sensing
(1998) - et al.
Texture evaluation of Radarsat imagery for wetland mapping
Canadian Journal of Remote Sensing
(2002)
Geology of the Peace–Athabasca River Delta Region, Alberta
Landsat interpretation of Prairie lakes and wetlands of eastern South Dakota
Two-dimensional hydraulic flood modeling using a finite-element mesh decomposed according to vegetation and topographic features derived from airborne scanning laser altimetry
Hydrological Processes
Temporal changes to the ice regime of a regulated cold-regions river
Analysis of C-band SIR-C radar backscatter over a flooded environment, Red River, Manitoba
Landscape classification and plant succession trends
Commercial implications of topographic terrain mapping using scanner airborne laser radar
Photogrammetric Engineering and Remote Sensing
Water body detection and delineation with Landsat TM data
Photogrammetric Engineering and Remote Sensing
Accuracy evaluation of Landsat digital classification of vegetation in the Great Dismal Swamp
Ecology of water-level manipulations on a northern marsh
Ecology
Remote sensing applications in hydrological modelling
Hydrological Sciences
A comparison of LANDSAT thematic mapper and SPOT multi-spectral imagery for the classification of shrub and meadow vegetation in Northern California, U.S.A
International Journal of Remote Sensing
Satellite remote sensing of wetlands
Wetlands Ecology and Management
Peace–Athabasca Delta water management works evaluation. Appendix B, biological assessment. Peace–Athabasca Delta Implementation Committee
Cited by (157)
UCL: Unsupervised Curriculum Learning for water body classification from remote sensing imagery
2021, International Journal of Applied Earth Observation and GeoinformationCitation Excerpt :Radar data have the advantage of capturing the information in almost every weather and day-night condition. However, the prominent features of vegetation (Huang et al., 2018), waves (Töyrä and Pietroniro, 2005; Marti-Cardona et al., 2013), sand (Martinis et al., 2018) and radar shadows produced by landscape features (Giustarini et al., 2013) deterrent the effective separation of water from the land surface. Therefore, the extraction of water bodies from remotely sensed data is more effective from optical imagery than from radar data (Schumann et al., 2009).
A new lake classification scheme for the Peace-Athabasca Delta (Canada) characterizes hydrological processes that cause lake-level variation
2021, Journal of Hydrology: Regional StudiesSmart solutions for smart cities: Urban wetland mapping using very-high resolution satellite imagery and airborne LiDAR data in the City of St. John's, NL, Canada
2021, Journal of Environmental ManagementA methodological approach for mapping Tunisia's lower coast's risk of submersion: The case of the coastal sabkhas of Sidi Khlifa and Halq El Minjel (Central-East Tunisia)
2020, Journal of African Earth SciencesCitation Excerpt :LiDAR data are available as a scatter plot. The various research projects on the topographic analysis of wet and low areas of LiDAR data deal with different methods of interpolation of “terrain” points in DTM: Triangulated Irregular Network « TIN » (Werbrouck et al., 2011), Nearest Neighbor « NN »; (Bater et Coops, 2009), krigeage (Töyrä et Pietroniro, 2005), spline (Cavalli et al., 2008), Inverse Distance Weighting « IDW » (Rosso et al., 2006). The airborne altimetry laser technology also allows extremely dense topo-bathymetric surveys with decimetric accuracy.
- 1
Tel.: +1 306 975 5723; fax: +1 306 975 5143.