DEFINITION
Land cover is defined as the observed physical cover including the vegetation (natural or planted) and human constructions that cover the earth's surface. Land cover includes water, ice, bare rock, and sand surfaces. (TEMS, FAO 2001)
Many networks and research institutes do not distinguish between land cover and land use. Land use, which concerns the purpose or function for which the land is being used, should be considered separately from land cover type.
UNITS
Land cover class, surface area of land cover (ha).
CLASSIFICATION SYSTEMS
Although Earth observation satellites and in situ devices produce extensive data on land cover, the source, accuracy, spatial resolution, and land cover classification systems of these land cover data sets vary widely. The multiple definitions and thresholds for a particular land cover type, such as forests, result in very different representations of forest cover in global land cover maps. The lack of consistency with regard to land cover classification system used by various Space Agencies and other research institutes is particularly problematic. (GCOS 2003) Products such as AFRICOVER, IGBP DISCOVER, CORINE land cover, and the University of Maryland land cover product exist as independent datasets.
Harmonization is a “bottom-up” process of emphasizing similarities and reducing inconsistencies between existing definitions of land cover classification systems to allow for direct comparisons and compatibility between various land cover datasets. (Herold 2006a) Harmonization efforts should first harmonize the parameters used for description of land cover classifiers, and then, once applied to systems and legends, the individual criteria used for creating land cover categories should be harmonized.
The use of a single classification system, such as the Land Cover Classification System (LCCS), by research institutes and agencies dealing with land cover would improve the quality and continuity of data. FAO and UNEP designed LCCS as a comprehensive hierarchical system to enable compatibility with existing classification schemes. The LCCS is undergoing approval to become a standard of the International Standards Organization (ISO).
MEASUREMENT VARIABLES & METHODS
Variables:
Spectral radiance (for remote sensing) – the particular way in which a given type of land cover reflects and absorbs light. The units are watts per steradian per square meter (W·sr-1·m-2).
Temporal signature (for remote sensing) – various land cover types produce distinct temporal backscatter signatures that can be used to determine both type and extent of a land cover.
Spatial signature (for remote sensing) – the spatial signature, or particular spatial configuration of objects, can be used to identify land cover type from satellite imagery.
Additional variables for in situ monitoring?
Methods:
Field observations (in situ)
Field mapping (remote sensing)
EXISTING IN SITU INPUT MEASUREMENT METHODS & STANDARDS
Regular collection of in situ data is needed for monitoring of land cover, vegetation migration, and related phenomena, and is also used as ground truth for validation of land cover and land cover change measurements by satellites. In situ data will also be necessary to the development of internationally-agreed protocols for land cover and land cover change observations and products.
Field observations
Discuss sampling method per hectare.
The following are among the networks that collect in situ land cover data for validation purposes:
- International Long Term Ecological Research Sites (ILTER) - 195 T.Sites
- Terrestrial Ecosystem Monitoring Sites (TEMS) - 146 T.Sites
- IGBP Land Cover Validation Confidence Sites - 413 T.Sites
- EOS Land Validation Core Sites - 31 T.Sites
- SAFARI 2000 Validation Sites - 20 T.Sites
- FLUXNET Network - 266 T.Sites
- BIGFOOT Network - 19 T.Sites
- GLC 2000 Validation Sites
(GOFC-GOLD)
EXISTING SATELLITE INPUT MEASUREMENT METHODS & STANDARDS
Variables
Spectral radiance
Spectral radiance is the primary variable used to determine land cover type from remote-sensing data. (NASA Glossary) Pixel values in commercially available imagery register the radiance of the surface in the form of digital numbers (DN), which are calibrated to fit a certain range of values. In the case of ETM+ imagery, these radiance values are scaled to numbers between 0 and 255. Conversion of DN back into absolute radiance is necessary for comparative analysis of several images acquired at different times. Conversion to radiance values allows for a more accurate comparison of images across rows, paths, and dates. (Varlyguin 2001)
Digital numbers registered by the remote sensing device are used to calculate spectral radiance (Lλ) according to the following equation (USGS 2001):
Lλ = LMINλ + ( (LMAXλ – LMINλ) / (QCALMAX) ) * QCAL
Where,
- QCALMIN = 1, QCALMAX = 255 and QCAL = Digital Number.
- The LMINλ and LMAXλ are the spectral radiances for band 6 at digital numbers 1 and 255 respectively. (Van)
This information from remote sensing instruments is generally presented as a digital thematic map in raster format, with pixels ranging in size from 500m to 1000m. Global land cover maps to date include data from AVHRR, SPOT-Vegetation, and MODIS, and future maps will use MERIS data. Given a sequence of registered multispectral and multitemporal images, a classification process assigns a land cover type label to each pixel. Classification procedures often function by comparing the vector of pixel-based observations to a database of examples of such observations drawn from the land cover types represented in the legend. (Strahler 2006)
Temporal signature
Temporal backscattering signals can be used to identify land cover type and extent. Radar remote sensing images have proven effective for this purpose in part because of radar is not affected by solar illumination and cloud cover. (Panigrahy 1997)
Spatial signature
Land cover type can be extracted from satellite images through the spatial configuration of objects in an image. Objects in an image can be extracted and then grouped into spatial configurations though an automated or manual classification process. (Scott 2005)
Land cover remote sensing activities (global and regional)
Multispectral and multitemporal global land and regional land cover data sets are currently produced by a range of space agencies and research institutes at medium resolutions (250m-1km) for determining land cover type, and fine resolutions (10-50m) for determining type and detecting land cover change. (UNFCCC 2006)
Global:
International Geosphere-Biosphere Programme (IGBP) provides a quantitative understanding of the Earth’s past climate and environment.
The High Resolution Data Exchange Project is a joint project of the CEOS and IGBP, which is focused on testing the utility of multi-sensor data acquisition, and a multi-agency international coordinated system of remote sensing observations. It is building a dataset containing several hundred SPOT, MOS, JERS-OPS, ERS-1, and IRS data at selected global change study sites around the world.
Land Use and Land Cover Change (LUCC) project is a program element of the International Geosphere-Biosphere Programme (IGBP) and the International Human Dimensions Programme on Global Environmental Change (IHDP).
Global Land Cover Facility (GLCF), based at the University of Maryland, develops and distributes land cover data with a focus on determining the location, extent, and drivers of land cover changes around the world.
DISCover was developed under the auspices of IGBP and provided 1km resolution land cover products (derived from data from the Advanced Very High Resolution Radiometer (AVHRR)).
Regional:
CORINE is a land cover database produced by the European Environmental Agency for the 25 EC Member States and other European countries; it includes 44 land cover and land use classifications. CORINE is also concerned with developing procedures for collating, standardizing, and exchanging data on the environment in the EC Member States.
Pan-European Land Cover Monitoring project (PELCOM) is a Europe-wide land cover product from AVHRR at 1 km resolution with a limited number of classes.
Africover (FAO) is a georeferenced database of land cover data for the whole African continent.
Regional Multiresolution Land Characteristics (MRLC2001) is funded by the US Geological Survey (USGS) and the US Environmental Protection Agency (EPA); the project was focused on creating a general, consistent, and seamless 30m land cover data set by the year 2000.
The National Land Use Mapping Project and Dynamic Monitoring Project, conducted by the Chinese Academy of Sciences (CAS), has led to the establishment of the National Land Use and Land Cover Database of China through the use of Landsat TM. (Skole 1998)
Current & future measurement instruments
Low- to medium resolution (250m – 1km) instruments
Measurement goals: coarse resolution of land cover types, land cover change; ideal for wide, global coverage.
Revisit rate: 6 times per year
Spectral resolution: multi to hyper
Technology type: imager, radiometer, spectrometer
Examples: AVHRR (1982-2000); MODIS (2000 – 2010+); VIIRS (~2010 - )
Fine resolution (10-30m) instruments
Measurement goals: land cover types, land cover change
Revisit rate: 6 times per year
Spectral resolution: multi to hyper
Technology type: high resolution imagers, radiometers, radars, and lidars.
Examples:
Imagers: SPOT, Landsat, IRS series
Radiometers: AVNIR-2, PRISM on ALOS, AVHRR/3, MERIS, VEGETATION
Radars: ERS, Envisat, Radarsat, ASAR, SAR (RADARSAT), and PALSAR
Lidar:
(CEOS Land monitoring, CEOS Multipurpose Imagery, NASA ESTIPS)
The table below summarizes the instruments and missions used for remote sensing of land cover and those instruments that will be in use in the coming years.
Instruments for detecting land cover type and land cover change
Historic / Current & near-term / Through 2010 / Future / Other contributing remote sensors / Potential new missions / Spatial sampling frequency / Temporal sampling frequencyAVHRR
LandSat 1-7 / AVIRIS / Synergistic multispectral optical + multi-frequency polarimetric radar / 30m and 250m-1km / weekly
SPOT 1-5 / AIRSAR
AVNIR / MODIS
GLI
ASTER
MERIS
AVNIR-2 / LDCM
NPOES
VIIRS
Prep. Prog. / NPOES
VIIRS
(Ciais et al.)
Existing measurement methodologies
NASA’s Strategic Plan for U.S. Climate Change Science proposes a framework for data requirements and a technical approach for standardized monitoring of four main types of land cover products, discussed in further detail below:
- Global, 1km annual land cover type – required for global climate, hydrologic, and biogeochemical modeling, and representation of significant land cover change at annual time scales.
- Global, 30m decadal land cover type – required for meso-scale climate and ecological studies.
- High-resolution 30m interannual land cover change – required for assessing ecological changes in response to climate variability and human activities, and for quantifying changes in carbon stocks.
- Continuous Fields representations – required for more accurate land surface parameterizations at subpixel scales, and for examining long-term changes in vegetation components. (Masek)
Global, 1-km Annual Land Cover Type
Data requirements:
1. A repeatable classification algorithm that can be applied uniformly across all regions of the Earth.
2. Use of the highest spatial resolution achievable for global land cover maps.
3. Annually updated maps to identify land cover change. Since the classification error rate is much higher than the annual rate of land cover change (and consequently changes observed are often due to algorithm errors or changes in training), a consistent and repeatable classification system is needed.
4. The highest classification accuracy possible. Accuracies associated with specific classes should not be less than 65 percent correctly classified, and classification accuracies shouldn’t vary widely due to geographic location.
5. A statistically rigorous validation strategy that assesses overall classification accuracy and accuracy within classes.
Technical Approach:
• Input data algorithms must be processed to minimize variations between and within sensors.
• To support of supervised classifications, high-resolution training data sets are needed; creation of such datasets requires protocols for geographic and ecological sampling, minimum patch size, quality assessment, and procedures for detecting land cover change in any given patch.
• Use of a validation strategy that uses a probability-based sample design with adequate samples to estimate overall accuracy and class-specific accuracy at continental, or if feasible, regional scales.
(Masek)
Global, Decadal, 30m Land Cover Type
Data requirements:
1. Based on a flexible and hierarchical land cover classification scheme with categories relevant for assessing a wide range of environmental applications. In particular, attention should be devoted to classes that are poorly represented in coarse-resolution representations, and those classes reflecting human land use (e.g. urban types, agricultural types, impervious surfaces).
2. A spatial resolution of 30m with temporal updates every 5 or 10 years.
3. Overall and regional accuracies exceeding 90 percent at the highest level of aggregation.
4. Validation should be based on the use of a probability-based sampling strategy.
Technical approach:
• The use of computer-assisted methods enables a cost-effective approach to creating accurate, high-resolution imagery.
• Validation must be statistically rigorous. Finding suitable sources of validation can be problematic; high resolution satellite imagery and aerial photography may be costly but are useful. (Masek)
Global Continuous Fields
Data requirements:
1. The use of explicit physiognomic-structural definition sets that are easily incorporated into FAO Land Cover Classification System and that enable the derivation of a mutually exclusive and exhaustive land cover classification.
2. (Modular) vegetation trait definitions that allow for their direct incorporation into global, continental and regional scale biogeochemical, hydrological and other natural resource and ecological modeling exercises.
3. An algorithm that yields the highest accuracy possible.
4. Annual or more frequent monitoring for those VCF layers suitable for change monitoring, and five year intervals for layers not likely to exhibit change.
6. Spatial resolution of at minimum 500 meters to permit large area monitoring of key vegetation change dynamics (e.g. deforestation).
7. Quality assessment mechanisms for each observation or pixel.
8. Validation protocols for both VCF layers and derived change products.
Technological approach:
• A supervised algorithm to ensure repeatability. Tree-based algorithms meet key criteria of repeatability, transparency, and a high level of accuracy.
• Training data should be derived from high-resolution data sets (5-50 meters) for calibrating the algorithm.
• Vegetation train definitions used should be compatible with FAO’s Land Cover Classification System.
• Probability-based sample designs for assessing product accuracy should be based on the direct observation or measurement of the respective vegetation trait. (Masek)
Inter-annual Land Cover Change and Disturbance