Satellite-based peatland mapping: Potential of the MODIS sensor

https://doi.org/10.1016/j.gloplacha.2006.07.019Get rights and content

Abstract

Peatlands play a major role in the global carbon cycle but are largely overlooked in current large-scale vegetation mapping efforts. In this study, we investigated the potential of the Moderate Resolution Imaging Spectroradiometer (MODIS) to capture the extent and distribution of peatlands in the St. Petersburg region of Russia by analyzing the relationships between peatland cover fractions derived from reference maps and ∼ 1-km resolution MODIS Nadir BRDF-Adjusted Reflectance (NBAR) data from year 2002.

First, we characterized and mapped 50 peatlands from forest inventory and peat deposit inventory data. The peatlands represent three major nutritional types (oligotrophic, mesotrophic, eutrophic) and different sizes (0.6–7800 ha). In addition, parts of 6 peatlands were mined for peat and these were mapped separately. The reference maps provided information on peatland cover for 1105 MODIS pixels. We performed regression analysis on 50% of the pixels and reserved the remainder for model validation. Canonical correlation analysis on the MODIS reflectance bands and the peatland cover fractions produced a multi-spectral peatland cover index (PCI), which served as the predictor in a reduced major axis (RMA) regression model. The results suggest a high potential for mapping peatlands with MODIS. The RMA regression models explained much of the variance in the PCI (r2 = 0.74 for mined and r2 = 0.81 for unmined peatlands). Model validation showed high correlation between observed versus predicted peatland cover (mined: r = 0.87; unmined: r = 0.92). We used the models to derive peatland cover estimates for the St. Petersburg region and compared the results to current MODIS land cover maps.

Introduction

Peatlands constitute one of the most widespread wetland types in the world. The most significant regions in terms of absolute extent of peatlands are in Europe, including the former Soviet Union, and North America, in particular above 45° N (Charman, 2002). Despite their importance in the global carbon and hydrological cycle, and their significance as wildlife habitat, the global distribution and extent of peatlands remains uncertain (Maltby and Immirzi, 1993).

Peatlands are often referred to as organic wetlands, because of their characteristic layer of peat, which is plant detritus accumulated under anaerobic, water-logged conditions. Peat represents a major pool of organic carbon that amounts to about one-third to a half of the global soil carbon, which is almost the equivalent to the global atmospheric carbon pool (Charman, 2002). Total carbon stored in northern peatlands alone has been estimated to be about 455 Pg C (Gorham, 1991) with a current uptake rate in extant northern peatlands of 0.07 Pg C/yr (Clymo et al., 1998). At the same time, peatlands represent a major source of methane (CH4) (Matthews and Fung, 1987) and dissolved organic carbon (DOC) (Freeman et al., 2004). Since atmospheric greenhouse gases are linked with the global climate system, global information on size and distribution of peatlands is of fundamental importance to climate change research. Currently, no clear consensus exists about the net effects of and feedbacks on future climate (IPCC, 2001).

For the last decades, global climate and biogeochemistry models largely depended on data derived from preexisting maps and atlases. Some of the most commonly used global data sets on wetland distribution were compiled by Olson and Watts (1982), Matthews and Fung (1987), and Aselmann and Crutzen (1989; later revised by Stillwell-Soller et al., 1995). These data sources rely on extensive historic in situ measurements and represent the best available information at the time. Nevertheless, their accuracy has not been rigorously assessed. As peatlands change over time in the process of their natural evolution and under the impacts of natural disturbances (e.g. fires), and human activities (e.g. peat mining, drainage and conversion to agricultural and forest land), the old maps likely become even less accurate.

Peatlands, particularly in the boreal region, tend to lack tree cover and represent distinct vegetation types (Botch and Masing, 1983, Gorham, 1991) that can be identified on satellite imagery. Since ground information on peatlands is often limited or even lacking in remote regions (Sheng et al., 2004), satellite remote sensing could provide a valuable tool for monitoring peatlands, especially in the northernmost latitudes. Several local and regional studies have mapped peatlands with high and medium spatial resolution sensors such as Landsat TM (30 m) and SPOT HRV (20 m) (Markon and Derksen, 1994, Poulin et al., 2002). However, mapping vast regions such as Northern Eurasia at fine spatial resolutions does not appear practical because of the lack of cloud-free imagery for many areas (DeFries et al., 1997, Krankina et al., 2004b). Further, there are considerable costs and logistical difficulties involved with handling such high data volumes, limiting the repeatability of such studies.

In comparison, the medium to coarse resolution sensor MODIS provides consistent and frequent observations of global land cover and land cover change processes at essentially no cost to the user (Townshend and Justice, 2002). Since the first global satellite based land cover map produced by DeFries and Townshend (1994) with data from the advanced high-resolution radiometer (AVHRR), substantial advancements have been made towards the development of comprehensive global vegetation and land cover data sets (Friedl et al., 2002, Justice and Townshend, 2002, Hansen et al., 2003). Nevertheless, peatlands are largely overlooked in large-scale land cover mapping efforts.

This study tested the capability of the MODIS sensor to map peatland cover proportions in a taiga landscape of the East-European plain. Results presented here may provide useful information for future mapping algorithms and therefore may promote the development of global land cover maps where peatlands are adequately represented.

Section snippets

Study area

The St. Petersburg region in Russia (Fig. 1) was selected as a test site, because of its location in one of the most significant peatland regions of the world. The abundance of peatlands in the St. Petersburg region is representative for northwestern Russia and is also close to the overall peatland proportion of the whole Russian Federation (8%) (Kobak et al., 1998). The study area occupies about 8 million hectare of flat terrain that rests on ancient sea sediments covered by a layer of moraine

MODIS

The MODIS instrument provides near-daily repeated coverage of the earth's surface with 36 spectral bands and a swath width of approximately 2330 km. Seven bands are specifically designed for land remote sensing with a spatial resolution of 250 m (band 1–2) and 500 m (band 3–7) (Table 2). In addition, MODIS science teams are producing a suite of higher level radiation products (e.g. systematically atmospherically corrected surface reflectance, surface temperature and emissivity, bidirectional

Approach

The broad swath of the MODIS instrument allows for frequent observation of the earth's surface over broad geographic regions, facilitating regional and global analyses of land cover and change processes. Such temporal and spatial coverage, however, comes at the expense of spatial grain size: pixels 1 km in size typically include several types of land cover, especially in heterogeneous landscapes. Discrete land cover classification systems require that a single label be assigned to each cell,

Results

Our results (Fig. 2, Fig. 3) show a strong relationship between MODIS NBAR data and observed fractional peatland cover. The reduced major axis (RMA) regression models explained much of the variance in the peatland cover index (PCI) for unmined (r2 = 0.81) and mined (r2 = 0.74) peatlands. While the unmined PCI model includes all seven MODIS bands, the mined model is based only on five bands (Table 1). In the latter case bands 3 and 6 were discarded by forward stepwise regression (p < 0.01). Fractional

Discussion and conclusion

The observed relationships between surface reflectance measured by the satellite sensor MODIS and percent peatland cover information from forest inventory data indicate good potentials for MODIS-based peatland mapping. The predictions of percent peatland cover based on these empirical relationships suggest that a map of percent peatland cover could be derived with adequate accuracy for the St. Petersburg region. To develop global or continental maps, i.e. for Northern Eurasia, further research

Acknowledgements

This research was conducted at Oregon State University and at USDA Forest Service PNW Research Station with support from the Land Cover/Land-Use Change Program of the National Aeronautics and Space Administration (grant number NAG5-11250 and NNG04GB81G). We greatly appreciate the reviews and thoughtful comments of Robert Kennedy and an anonymous reviewer.

References (40)

  • I. Aselmann et al.

    Global distribution of natural freshwater wetlands and rice paddies, their net primary productivity, seasonality and possible methane emissions

    Journal of Atmospheric Chemistry

    (1989)
  • P.M. Atkinson et al.

    Mapping sub-pixel proportional land cover with AVHRR imagery

    International Journal of Remote Sensing

    (1997)
  • M.S. Botch et al.

    Mire ecosystems in the USSR

  • Botch, M.S., Kobak, K.I., Krankina, O.N., Spycher, G., in preparation. Peatlands in the St. Petersburg Region. In:...
  • D. Charman

    Peatlands and Environmental Change

    (2002)
  • R.S. Clymo et al.

    Carbon accumulation in peatlands

    Oikos

    (1998)
  • W.B. Cohen et al.

    Landsat's role in ecological applications of remote sensing

    BioScience

    (2004)
  • W.B. Cohen et al.

    Modelling forest cover attributes as continuous variables in a regional context with Thematic Mapper data

    International Journal of Remote Sensing

    (2001)
  • W.B. Cohen et al.

    MODIS Land Cover and LAI Collection 4 Product Quality across Nine Sites in the Western Hemisphere

    IEEE Transactions on Geoscience and Remote Sensing

    (2006)
  • R.S. DeFries et al.

    NDVI derived land cover classifications at a global scale

    International Journal of Remote Sensing

    (1994)
  • Cited by (26)

    • The potential for modelling peatland habitat condition in Scotland using long-term MODIS data

      2019, Science of the Total Environment
      Citation Excerpt :

      There have been several other studies that classified peatland vegetation types, rather than condition. Generally, these attempted to build high resolution models of vegetation types in relatively small areas (e.g. Mehner et al., 2004; Knoth et al., 2013; Harris et al., 2015; Middleton et al., 2012), however Pflugmacher et al. (2007) attempted mapping across a larger geographical region for the St Petersburg region in Russia using a sub pixel proportional cover approach. They trained a MODIS-based model on mapped peatland sites of different site nutrition types that were either mined for peat or not and were able to build a reasonably accurate model.

    • Remotely sensed MODIS wetland components for assessing the variability of methane emissions in Indian tropical/subtropical wetlands

      2018, International Journal of Applied Earth Observation and Geoinformation
      Citation Excerpt :

      Being an indispensable source of multi-temporal and spatial measures of land surface (Chen at el., 2014), remote sensing (RS) can generate the huge data pertaining to spatial and temporal distribution of wetlands over inaccessible and large areas. Numbers of studies have been done worldwide to evaluate the spatio- temporal distribution of wetlands (Pflugmacher et al., 2007; SAC, 2011; Chen et al., 2014; Garg, 2015) deploying various RS data and techniques which can further be used to assess the significance of wetland areal extent in regional and global CH4 emission studies. Temperature fluctuations in wetlands chiefly affect the CH4 emissions by controlling the rate of CH4 production by methanogens and release via wetland surface and vascular plants.

    • Comparison and assessment of coarse resolution land cover maps for Northern Eurasia

      2011, Remote Sensing of Environment
      Citation Excerpt :

      Agreement between the field data and the two analyzed land cover data sets, the AVHRR Global Land Cover Characterization Database and the MODIS C3 Land Cover Product, was 22% and 11%, respectively. Other comparison studies have focused only on wetlands (Krankina et al., 2008; Pflugmacher et al., 2007). Cross comparisons between global land cover maps help identify areas of potential high map uncertainty (Herold et al., 2008; See & Fritz, 2006), but independent validation studies are needed to reveal the sources of disagreement and provide local and regional scale estimates of classification accuracy.

    View all citing articles on Scopus
    View full text