Remote Sensing and GIS Techniques for Terrestrial Carbon Inventory

Remote sensing is a technique that holds great potential for long-term monitoring of changes in area and carbon stocks. This chapter discusses the application of different techniques for different project types in terms of feasibility and reliability; highlights uncertainties, cost and required technical capacity; describes the application of geographical information systems (GIS) methods for carbon inventory for different projects; and also assesses the role of remote sensing and GIS techniques for long-term carbon inventory.

Data from remote sensing are those acquired by sensors, which may be in the form of optical devices, radar or lidar on-board satellites or cameras equipped with optical or infrared films installed in aircraft. The data can be thought of as an image representing the ground. These data can be used to provide estimates of land cover and area although, being essentially interpretation of images, it is usually necessary to validate the data against ground truth to know how accurate the interpretations are (IPCC 2006). Remote sensing is a powerful tool in that it covers large areas and enables inventories to be made at a low cost per unit area. Analysis of satellite imagery is the most practical approach to monitor vegetation cover over large areas periodically as a routine (DeFries et al. 2005). If sample plots with adequate ground-based measurements are insufficient to support a proper inventory over a large area, it is necessary to use auxiliary variables correlated with land-use variables. One such variable can be obtained by remote sensing or GIS (Lappi and Kangas 2006). Another benefit is the savings since field measurements are one of the most expensive components of sampling-based land-use inventories for large areas (Tomppo 2006).

Data from remote sensing can be classified either by visual analysis of the imagery or by digital, computer-based methods. The strength of remote sensing is its ability to provide spatially explicit information repetitively. Archives of remote sensing data also span several decades and can therefore be used to reconstruct land cover and land use as a time series. Remote sensing is particularly useful in obtaining estimates of area under different types of land cover and land-use categories as described in Chapter 8. Furthermore, remote sensing can assist in identifying and defining relatively homogeneous areas that can guide in sampling (see Chapter 10 for information on sampling design and size of samples).

The challenges for remote sensing are in interpretation, which is the process whereby images or data are translated into meaningful information on, for example, land cover and land use. A common obstacle is the presence of clouds, aerosols and haze, which limit the availability of data to varying extent, depending on the sensor used. A radar, which actively sends out a signal that bounces back from the surface, is not limited by these factors whereas passive sensors that rely on the actual reflectance from the surface are hampered by such obstructions. Another obstacle is the difficulty in distinguishing between different land-use or cover types that give very similar signals. Another concern when comparing data over long period of time is that remote sensing systems may have changed over time in terms sensors, bandwidth or maintenance.

14.1 Implications for Carbon Inventory

A carbon inventory requires estimates of biomass stocks, and remote sensing helps in generating information or data required for such estimates or for validating the estimates made by other means. Estimates of biomass stock are based on features of vegetation such as coverage and canopy.

One of the crucial issues in using data from remote sensing is the accuracy of estimates. The concern also extends to assessment of carbon inventories. Remote sensing, attempts to correlate a spectral signature with a specific land use, and it is necessary to define how closely the interpretation matches reality (UNFCCC 2006). Accuracies of 80-95% are achievable with high-resolution imagery (where every pixel covers only a small area) in discriminating between forest areas and non-forest areas. However, to detect biomass and hence the carbon content, such high accuracy is much harder to attain - the most reliable estimates of carbon stocks are those based on field measurements.

Carbon is the main component of biomass vegetation and is invisible. Therefore, it is necessary to focus on features of vegetation to estimate carbon stocks. These features can be age of the vegetation (see, e.g. Zheng et al. 2004), tree diameter (see, e.g. Drake et al. 2003), intensity of chlorophyll activity or biomass density (see, e.g. Tan et al. 2007).

14.2 Data from Remote Sensing

Remote sensing is the process of obtaining information about an object, area, or phenomenon through the analysis of data acquired by instruments not in direct contact with the object being investigated. Reading, for example, is a remote sensing process. Eyes act as sensors that respond to the light reflected from a page in a book. The data that the eyes acquire are impulses corresponding to the amount and pattern of light reflected from the dark and light areas on the page. These data are analysed or interpreted by the brain to enable a person to explain the dark areas on the page as a collection of letters forming a word (Lillesand and Kiefer 1994).

At present there are about 800 satellites operating with the aim of collecting information on a variety of environmental topics, such as the atmosphere, snow, oceans and vegetation. The satellites move in different types of trajectories, orbits or paths, such as geostationary or polar satellites. Most of the sensors placed on satellites for earth observations are on polar-orbiting satellites at a height of 450-900 km, usually synchronized with the Sun for visibility since the optical sensors need light to collect information.

A remotely sensed image is a pixel-by-pixel measurement of reflected or emitted energy from the Earth's surface (Brown 1997). A few of the most commonly used types of remote sensing data are aerial photography, satellite imagery using visible and/or near infrared bands, satellite or airborne radar imagery and lidar. Combinations of different types of remote sensing data may very well be used for assessing different land-use systems or areas and thus for estimating carbon stocks. These combinations can include interpretations of two data sets to increase accuracy and the use of two or more bands to produce indices. One set of such useful indices comprises vegetation indices using both visual and infrared bands. A system based on remote sensing to track land-use conversions can include many combinations of sensors and data types at a variety of resolutions.

There are several important criteria for selecting remote sensing data and products for terrestrial carbon inventory (IPCC 2006):

1. Adequate land-use system stratification scheme Stratification of the project area has to be robust and clear to be able to distinguish between them. The stratification should be of adequate spatial resolution to enable use of remote sensing.

2. Appropriate spatial resolution If broad categories or distinct land-use differences are sought, such as forested and non-forested land, low-resolution remote sensing might be adequate, compared to a detailed categorization of different agricultural land that requires high resolution.

3. Appropriate temporal resolution Estimating land use changes in boreal forest systems might require data that span over decades, whereas for estimating changes in grassland, data for even a single year may be sufficient. Seasonality of the vegetation is an important factor since peak vegetation period is usually the best time for inventory of terrestrial carbon.

4. Availability ofhistorical assessment Often the limitation of conducting a remote sensing survey is the availability of historical data. In that sense the future is promising, since more, readily available, sensors and products are being developed.

5. Transparent and consistent methods applied in data acquisition and processing Since carbon inventories are performed frequently and require monitoring over time, the methods that are used have to be repeatable.

6. Consistency in data and availability over time The products used should be consistent over time for the same reason as stated in point five above.

Different remote sensing sensors are receptive to energy from diverse parts of the electromagnetic spectrum, such as visible, near-infrared, infrared or thermal.

Sensors collect parts of the wavelength into different bands or information sets, which means that over a particular pixel, several bands are produced for the same area, which can be used to construct indices for numerous features such as vegetation types. The process is described later in this chapter. Different features such as forest, bedrock, soil or cropland have different reflectance and it is the spectral differences between these features that enable the user to classify the image into different land-use types (Brown 1997). More information on the elements of photographic systems related to remote sensing can be found in textbooks such as that by Lillesand et al. (2004).

Classification of land use with remotely sensed data can be achieved visually or digitally; the latter essentially means computer-based analysis. Each approach presents advantages and disadvantages. Visual analysis allows for human inference through the evaluation of overall characteristics of the image. Usually, this is done by analysing the contextual aspects of the image. Digital classification, using computer hardware and software, allows for several manipulations to be performed with the data, such as merging of different spectral data (Fuentes et al. 2006) and adding information from ancillary data from object-oriented methods (Bock et al. 2005), which can help to improve modelling of biophysical ground data, such as tree diameter, height, basal area, biomass, time of flowering and harvest or disturbances such as drought, diseases or fire. Digital analysis allows immediate computation of areas associated with different land-use categories and has developed rapidly over the past decades; given the concurrent development related to computers, the necessary hardware, software and also the satellite data are now readily available at low cost.

Aerial photography Analysis of aerial photographs can reveal differences in land-use or land-cover system such as agriculture, grassland, forest tree species and forest structures from which the relative distribution and tree health may be judged. In agriculture, similar analyses can show crop species, crop stress or tree cover in agroforestry systems (Fig. 14.1). The smallest spatial unit that can be seen depends on the type of aerial photos used, but for standard products it is often as small as one metre (IPCC 2006).

Optical satellite images Complete national or regional land-use and land-cover analyses may be facilitated by satellite images. This section describes passive satellite data in the visible and near-infrared spectra. Passive sensors rely on reflectance of solar energy from the surface back to the sensor or detector. This energy is captured in the visible, near- and middle-infrared portion of the electromagnetic spectrum (~0.4-2.5 |im). Digital multispectral remote sensing data record spectral information in a number of wavelengths referred to as bands. Information up to 10 bands per pixel or unit of land can be recorded. Hyperspectral data, consisting of 100-200 bands of information, are available but require special processing methods (see, e.g. Tamas and Lenart 2006). Green vegetation exhibits a unique signature characterized by strong reflectance in the green and infrared portions of the electromagnetic spectrum and strong absorbance in the red and some mid-infrared regions (see Fig. 14.2). Variations in internal cell structures of leaves, absorption levels of chlorophyll and variations in leaf water content make it

Fig. 14.1 An aerial photograph with 4 m resolution over Julita, in mideastern Sweden (~2 x 2.5 km). The photo is an orthophoto, meaning that it has been geometrically corrected. Agriculture, deciduous forest and shelterbelt trees can be detected as well as individual houses and gardens. The area in the lower left corner is part of Lake Oljaren. (Courtesy of Lantmateriet Gavle 2007. Medgivande I 2007/437)

Fig. 14.1 An aerial photograph with 4 m resolution over Julita, in mideastern Sweden (~2 x 2.5 km). The photo is an orthophoto, meaning that it has been geometrically corrected. Agriculture, deciduous forest and shelterbelt trees can be detected as well as individual houses and gardens. The area in the lower left corner is part of Lake Oljaren. (Courtesy of Lantmateriet Gavle 2007. Medgivande I 2007/437)

possible to distinguish between different types of vegetation (Patenaude et al. 2005). Various indices, such as normalized difference vegetation index (NDVI), have been designed to optimize these spectral signatures of vegetation.

Time series can be obtained for any area of interest since the satellite passes over it continuously and regularly. Reliable optical data going back to the early 1990s can be accessed and used for interpretations with good confidence for assessing changes in land use (DeFries et al. 2006). Because satellites circle the globe in different orbits at different heights and speeds, the interval between consecutive passes over one particular geographical area differs between satellites. The images usually generate a detailed mosaic of distinct categories, but matching them to proper land cover and land-use categories commonly requires ground reference data from maps, field surveys or other available information.

The smallest unit to be identified depends on the spatial resolution of the sensor and the scale of work. The most common sensor systems have a spatial resolution of 20-30 m. At a spatial resolution of 30 m, units as small as 1 ha can be identified. Data from higher resolution satellites are also available (IPCC 2006). Several satellite products are presented in Table 14.1 with a range of resolution from 0.6 to 1,100 m, from fine and high to coarse and low resolution. Many of the satellite products can be accessed through official web sites.

Fig. 14.2 An optical 20 m resolution SPOT multispectral image (228/309) over Western Orissa, India, on 29 December 1994 (~60 x 60 km). The image consists of information from all bands, from visible green and red to near-infrared, also called a false colour composite image, which makes green vegetation appear red, a feature that is important for carbon monitoring

Fig. 14.2 An optical 20 m resolution SPOT multispectral image (228/309) over Western Orissa, India, on 29 December 1994 (~60 x 60 km). The image consists of information from all bands, from visible green and red to near-infrared, also called a false colour composite image, which makes green vegetation appear red, a feature that is important for carbon monitoring

Aerial Jpg
Table 14.1 Examples of passive satellite images. (From UNFCCC, 2006.)

Satellite (sensor)

Resolution (m)

Time coverage

Cost

Source

NOAA (AVHRR)

1,100-8,000

1978-

Free

http://edc.usgs.gov

EnviSAT (MERIS)

300-1,200

2002-

$525/scene

http://envisat.esa.int/

Terra (MODIS)

250

2000-

Free

http://edc.usgs.gov http://glcf.umiacs.umd.edu/data/ gimms/

Landsat (MSS)

60

19721992

Free to $375/scene

http://www.spaceimaging.com http://edu.usgs.gov http://glcf.umiacs.umd.edu/data/ gimms/

Landsat (TM)

25

1982-

Free to $625/scence

http://www.spaceimaging.com http://edu.usgs.gov http://glcf.umiacs.umd.edu/data/ gimms/

Landsat (ETM+)

15

1999-

Free to $800/scene

http://www.spaceimaging.com http://edu.usgs.gov http://glcf.umiacs.umd.edu/data/ gimms/

SPOT (VGT)

20 (10)* 2.5

1986-

$1,200-10,125/scene

http://www.spotimage.fr/home http://www.spot.com

Terra (ASTER)

15

1999-

$145-580/scene

http://edc.usgs.gov

IKONOS

4 (1)*

2000-

$16-56/km2

http://www.spaceimaging.com http://glcf.umiacs.umd.edu/data/ gimms/

Quickbird

2.4 (0.6)*

2001-

$5,000-11 500/scene $16-45/km2

http://www.digitalglobe.com http://glcf.umiacs.umd.edu/data/ gimms/

*Available in panchromatic, which means it is in black and white

*Available in panchromatic, which means it is in black and white

The relationships established between vegetation biomass and hence carbon stocks and the data from optical sensors with low resolution are generally weak (Rosenqvist et al. 2003). Many of the early or pre-2000 satellite products are low to medium in resolution, a fact to be taken into account when seeking historical information from optical satellite imagers on carbon stocks. Radar images Unlike optical satellites, which rely on solar illumination, radar, an acronym for radio detection and ranging, are active microwave sensors that emit energy to survey the Earth (Lillesand et al. 2004). A major advantage of a radar system is that it can penetrate clouds, aerosols and water vapour (Fig. 14.3). Radar also acquires data during the night, whereas passive and optical products cannot do so. Therefore, the system is called an active satellite system since it sends out a signal that hits the surface and returns to the sensor. This makes radar perhaps the only reliable source of remote sensing data in many areas of the world with frequent cloud cover, such as the tropics.

The most common type of radar data is the so-called synthetic aperture radar (SAR) sensor system that operates at microwave frequencies, such as RadarSAT. By using different wavelengths and different polarizations, SAR systems may be able to distinguish between land-use systems, for example, forest and non-forest, or the biomass content of vegetation. The radar system transmits either horizontally (H) or vertically (V) polarized electromagnetic (EM) energy and then receives either of these polarizations. The frequency transmit/receive configuration of radar data is typically stated by a three-letter code: the first letter designates the band of the radar and the last two letters state the polarization configuration. Four combinations are in use, namely HH, HV, VH and VV. L-VH radar, for example, means an L-band system that transmits vertically and receives horizontally in terms of polarized EM energy (Kasischke et al. 1997). At present radar has some limitations when biomass is high due to signal saturation, that is the signal does not change beyond a certain chlorophyll level.

Fig. 14.3 Two radar images, namely synthetic aperture radar (SAR) and phased array type L-band synthetic aperture radar (PALSAR) showing deforestation in Amazon from 1996 to 2006. Grey indicates forested area and black represents deforested area. (Courtesy of Japan Aerospace Exploration Agency's Advanced Land Observing Satellite (JAXA/ALOS) )

Fig. 14.3 Two radar images, namely synthetic aperture radar (SAR) and phased array type L-band synthetic aperture radar (PALSAR) showing deforestation in Amazon from 1996 to 2006. Grey indicates forested area and black represents deforested area. (Courtesy of Japan Aerospace Exploration Agency's Advanced Land Observing Satellite (JAXA/ALOS) )

In recent years, SAR backscatter has been increasingly investigated for use in inventorying forests because radar wavelengths penetrate the vegetation and, in that sense, provide direct feedback about vegetation structure and biomass (Patenaude et al. 2005). The most commonly used systems are the C-band (wavelength ~5 cm), L-band (~24 cm) and P-band (~70 cm) (Kasischke et al. 1997; Igarashi et al. 2003; Lucas et al. 2006). The short wavelength, C-band, is sensitive to small components of the canopy such as leaves and twigs whereas the other two bands penetrate deeper and are predominantly sensitive to larger branches and trunks.

One limitation of radar SAR systems is their sensitivity to surface topography, which limits their general application to flat or gently undulating terrain (Rosenqvist et al. 2003). The main external factor controlling the sensitivity of the SAR signal to biomass is the structural properties of the forest or vegetation and its underlying surface. Given equal biomass, the radar backscatter response will be significantly different for a forest composed of sparsely distributed large trees and one composed of dense stands of young, small trees (Patenaude et al. 2005). Examples of available data collected using active radar remote are listed in Table 14.2.

In response to the demand for technical and scientific input to the work on climate change, carbon inventory and the Kyoto Protocol, the Kyoto & Carbon initiative was initiated by Japan Aerospace Exploration Agency (JAXA) as part of the Advanced Land Observation Satellite (ALOS) in 2000. One of the focal points was to utilize ALOS phased array type L-band synthetic aperture radar (PALSAR) to support the type of information needs on a regional scale associated with carbon estimations (ALOS 2006). As of now (early 2007), the products generated from this effort are made available to general users about 6 months after initial distribution to the Kyoto & Carbon Science team.

Lidar Light detection and ranging, or lidar, uses the same principles as radar. The lidar instrument transmits light to the target; the transmitted light interacts with the

Table 14.2 Examples of satellite images taken with active radar systems. (From ALOS 2006; Rosenqvist et al. 2003; Lillesand et al. 2004; UNFCCC 2006.)

Satellite (sensor)

Resolution (m)

Time

coverage

Cost

Source

ENVISAT (SAR,

25-150

2002

$150-1,000/scene

http://envisat.esa.int/

ASAR MERIS)

ERS-1 (SAR)

25-150

1991

2000

$150-700/scene

http://www.esa.int

ERS-2 (SAR)

30

1995

$150-700/scene

http://www.esa.int

JERS-1 (SAR)

18-100

1992

1998

$100-1,000/scene

http://www.Eorc. nasda.go.jp/JERS-1

ALOS (PALSAR)

9-157

2006-

Free to $250/scene

http://earth.esa.int/ dataproducts/

RadarSAT 1

8-100

1995

Free

http://www.rsi.ca

(SAR, ASAR)

RadarSAT 2

3-100

2004-

http://www.rsi.ca

(SAR, ASAR)

target and is changed by it. Some of this light is scattered and reflected back to the instrument, which analyses the light. The change in properties of the light enables some properties of the target to be determined. The time taken by the light to travel to the target and back to the lidar is used to determine the distance to the target (IPCC 2006). Lidar resolution or footprint size may vary from 0.25 to 25 m (Drake et al. 2003; Rosenqvist et al. 2003), which makes lidar more detailed than the other technologies described so far. Products of lidar come in a wide range, although most of them are commercial, and therefore not as readily available as those from optical and radar data. Lidar is at present not available from satellite platforms, which limits its use even more (Patenaude et al. 2005). Irrespective of the type of lidar instrument used, the general approach has been to use some physical attribute a forest canopy (Nssset 2002), such as canopy height (Kimes et al. 2006), tree height and stem volume (Holmgren et al. 2003) and canopy elements in three dimensions (Lovell et al. 2003) to estimate biomass, particularly above-ground biomass (Lim and Treitz 2004). Lidar is the youngest technology of gathering data among the remote sensing ensemble, which can be used in characterizing vegetation. Lidar products have demonstrated a strong relationship between physical features such as tree height, stem volume, biomass and canopy closure (Drake et al. 2003; Rosenqvist et al. 2003).

The limitation of lidar is that it is a fairly sophisticated technology, which requires trained experts for analysis not available everywhere. Also, at present, it is too expensive to be applied over large areas (Skutsch et al. 2007). Another limitation is lidar's inability to distinguish between species of trees in a forest. As wood density and hence biomass varies between tree species of similar height and age, estimates of biomass using lidar alone may be less accurate (Rosenqvist et al. 2003).

Laser The concept of laser is yet another active system, where a spot on the Earth's surface is illuminated by a laser beam and the distance to the spot determined. Laser as a remote sensing technique is considered a promising tool for monitoring changes in vegetation changes such as avoided deforestation (Joanneum et al. 2006). The geosceince laser altimeter system (GLAS) sensor on-board the Ice, Cloud and Land Elevation satellite - Icesat - was primarily aimed at monitoring mass balance of the polar ice sheet but has also proved useful in assessing vegetation. GLAS produces a series of spots, ~70 m in diameter, which are collected in a telescope 1m in diameter, which gives high-resolution data for analyses. The satellite was launched in 2002 but started operating in 2004.

Project Management Made Easy

Project Management Made Easy

What you need to know about… Project Management Made Easy! Project management consists of more than just a large building project and can encompass small projects as well. No matter what the size of your project, you need to have some sort of project management. How you manage your project has everything to do with its outcome.

Get My Free Ebook


Responses

  • Ricardo
    What is terrestrial based technique in GIS?
    12 months ago
  • toini himanen
    What is terrestrial based data acquistion method in gis?
    9 months ago

Post a comment