Remote sensing’s functional role in studies of land-use/land-cover change
Terrestrial ecology, the study of the land, depends on data derived from space-borne sensors. Not just limited to finite locations on the ground, terrestrial ecology integrates biospheric, atmospheric, and hydrologic processes at a variety of spatial scales. Increasing awareness of and appreciation for the insight that remote sensing techniques and data lend to these studies have expanded the role of satellite imagery in ecology. Indeed, remote sensing has been called essential "for addressing the role of ecological complexity in global processes" by Matson and Ustin. The power remote sensing offers ecological studies extends beyond the ability to map vegetation classes on the Earth's surface: functional variables can be derived from satellite data. The information ecology can then generate becomes integral to global change research. In particular, land-use/land-cover change research comprises both one of the main ecological applications of remote sensing and one of the primary foci of global change research. After a brief review of some fundamental ecological concepts, the role of remote sensing in studies of land-use/land-cover change will be considered.
Scaling structure and function in ecological studies
Structure and function comprise the two major components of ecological systems. Structure, concerned with pattern, deals with the physical aspects of land cover, such as vegetation-type distribution, canopy structure, and leaf and stem areas. Function describes the processes that transfer matter and energy in the system, such as gas fluxes and water exchange between the biosphere and atmosphere. Structure and function are linked. Due to the complexity of their linkages, however, ecological studies usually concentrate on one or the other. Yet to fully understand how ecosystems operate, the two need to be integrated. In general, variations in canopy structure, such as size and spacing, usually accompany functional differences among vegetation types. Landscape structure (e.g. surface roughness) can indicate variations in such processes as evapotranspiration and carbon dioxide (CO2) assimilation.
Scale presents another challenge facing ecological studies. Both structural attributes and functional processes represent entities with nonlinear components. Accordingly, data needs to be collected at a scale coarser than the plot level in order to understand large-scale dynamics. Remote sensing addresses this concern by providing data products with coarse spatial extent. The measurement taken by the sensor constitutes the result of interactions of electromagnetic radiation with surface constituents, including the soil and vegetation layers. With knowledge of the particular characteristics of canopy and landscape structures, remote sensing also allows the linkage of structure and function in ecological studies. For example, photosynthesis and carbon allocation processes are constrained by the amount of incoming radiation, which is affected by the canopy configuration. In turn, ecosystem structure emerges from the collective functioning of vegetation components. These connections will be further explored in later sections.
The global change of land use and land cover
As its own area of global change research, land-use/land-cover change focuses on the characteristics, causes, and consequences of shifts in vegetation and other types of land cover. This area of research recognizes that human activities affect most, if not all, "natural systems" and so incorporates anthropogenic disturbance into ecological studies. Land use is but one disturbance that affects any area on the Earth. Also, it occurs in the context of natural variability and disturbance. The issue of control makes land use unique: its rate is quicker, its magnitude wider, and its occurrence non-random. Land use alters land cover, thereby affecting ecosystem processes. Even without specifically altering land cover, land use affects ecosystem functioning through intensification of usage. Either type of change alters biophysical, biogeochemical, and hydrological states and processes. Further, disturbances (natural or anthropogenic) affect ecosystems differentially, depending on edaphic, physiological, structural, and climatic constraints, so figuring out the causes and consequences of change will help evaluate ecosystem function and influences on biogeochemical cycling at local, regional, and global scales.
NASA's Earth Science Division has developed the Land Cover Land Use Change (LCLUC) Program for the purpose of monitoring changes in land cover and understanding the consequences of land-cover and land-use change for the continued provision of ecological goods and services. Land-use/land-cover change could affect biogeochemical cycling, biophysical processes, biodiversity, trace gas and particulate fluxes, and coastal zone conditions. Accordingly, the program needs both basic and applied research in ecological structure and function. The status of land cover inherently affects the vitality of ecosystems by virtue of the ecological processes its structure supports. Research and data, then, need to examine the current distribution and conversion (both past and future) of cover types. Of special interest is the conversion of forests due to trace gas fluxes that occur in these types of systems.
Land use affects ecosystem processes in two ways beside vegetative composition: intensification and degradation. For example, intensification can alter hydrological and biogeochemical cycles through such activities as irrigation and fertilization in agricultural systems. Degradation occurs due to salinization, desertification, and erosion; it will alter processes in the course of an area converting to another cover type.
The roles of remote sensing in land-use/land-cover change studies
These coarse scale concerns, and the repeated coverage necessary for studying processes of change, require the use of remote sensing for the retrieval of appropriate data. Remote sensing measurements of spectral signatures provide data on spectral color, temperature, moisture content, and organic/inorganic composition as well as spatial properties of areal extent, geometry (size, shape, and texture), and position. NASA's LCLUC Program promotes several sensors—MODIS, Landsat MSS and TM, Spot Panchromatic and Multispectral, Lewis Hyperspectral, and Clark Panchromatic—as viable techniques for monitoring, quantifying, and forecasting land use and land cover. The most emphasis is placed on monitoring for land degradation.
The use of remote sensing into land-use/land-cover studies occurs in three types of analyses: 1) spatial assessments through vegetation mapping and classification; 2) productivity assessments through vegetation indices; 3) process studies with parameters specified by data derived from satellite imagery. The spatial assessment aspect of land-use/land-cover change requires repeated global inventories of land cover. Studies at the ecosystem and biome scales require canopy information, such as leaf area index (LAI) and fraction of absorbed photosynthetically active radiation (fAPAR), for understanding ecosystem processes, such as gas exchange and nutrient cycling. To understand the consequences of cover type change and land-use intensification, real data must be entered into realistic ecosystem process models. The accuracy of ecosystem process models is increased, then, by using remotely-sensed data to define or constrain plant allocation parameters. The data needed at large scales for a variety of modeling include vegetation cover type, structural parameters (e.g. effective LAI), and biophysical parameters (effective fAPAR and albedo). Improved spatial resolution of structure, such as live vegetation and litter, are needed to detect small changes. This article considers all three types of remote sensing applications.
Monitoring vegetation change through mapping
The most established application of remote sensing to ecological studies is mapping patterns of the Earth's surface. Vegetation mapping allows quantification of actual vegetation on the ground. This provides ecological studies with realistic information instead of potential natural vegetation maps based on observed vegetation relationships to climatic factors. This type of analysis is fundamental to the study of land cover and land use, and much land-use/land-cover research still focuses on this type of work. First, it provides the current spatial distribution of land cover and land use for any given area that has been imaged by a sensor. Much of the global land-use classification is done with 1-kilometer (km) and 8-km AVHRR data. Other sensors, especially Landsat TM and SPOT, are used for finer-scale classifications. Second, repeat coverage allows temporal studies of change. Three decades of satellite imagery allow decadal studies of land-cover change. These archives of satellite imagery also serve as a baseline for future monitoring and change assessments. Change detection is accomplished by creating algorithms to quantify the magnitude and direction of change. Overall the assessments of land-use/land-cover change can be quantitative and comprehensive. This type of spatial analysis can also help to target study areas.
Classification techniques and algorithms, such as multi-temporal comparisons and decision trees, have been developed for creating land-cover type maps from satellite images. Further development of classification continues for more specific identification of land cover types. Especially important are those classes representing conditions after landscape conversion, such as secondary growth tropical forests that grow in after deforestation events. Intensification or degradation cannot necessarily be derived from the spatial information on satellite imagery; other information is needed. When categories of intensity can be translated into a land-cover type, some satellite information is useful. For example, Landsat imagery can detect the presence of intensive logging, yet technique development is still required for analyzing intensification and long term degradation.
Beyond assessing spatial changes, land-cover distribution and conversion maps can be used in the development and verification of biogeochemical and biophysical models, as well as in analyses of the effects of spatial patterns and conversion histories on ecosystem structure and biodiversity. The satellite-derived vegetation maps can be compared to the results of models for model verification. They can also serve as inputs to the model by prescribing the ecosystem type in the parameterization of biogeochemical models. The vegetation distribution maps can be used to describe land surface characteristics, such as roughness, resistance to CO2 and water vapor exchange, or ecosystem components, such as (C:N) ratios. Through this type of modeling, the changes in land cover can be assessed since actual, instead of literature-derived, values are used.
Further studies can be completed by fusing remote sensing data with other data layers in a geographic information system (GIS). One modeling proposal for land-use/land-cover change studies integrates satellite and socioeconomic data into dynamic deforestation models to understand the characteristics and rates of deforestation, regrowth, and land-use transition. Another study considers alternative approaches to depict land-cover heterogeneity and change through regional and global biosphere-atmosphere models.
Empirical sources of function: Vegetation Indices
Beyond spatial description of change, scientists are interested in the amount of change. Vegetation indices offer quantitative information about vegetation productivity based on spectral information found in satellite imagery. Essentially, the indices serve as a surrogate for vegetation components. Generally, they are useful for continental- to global-scale models, which require coarse resolution inputs. Often, AVHRR has been used to calculate vegetation indices for monitoring land cover and vegetation phenology due to its global daily coverage. Indices can also be calculated from the spectral bands of any other sensor.
The most widely used vegetation index, the normalized difference vegetation index (NDVI), relates near infra-red to visible red reflectances (NIR-VIS)/(NIR+VIS) in order to take advantage of the differential reflectance characteristics of vegetation in these two spectra. The biological controls on this measure are foliage density and leaf chlorophyll content. NDVI, like most vegetation indices, relates a measure of "greenness", which is empirically related to vegetation structure and function, through variables such as LAI, vegetation cover, above-ground biomass, photosynthetic efficiency, fAPAR, and stomatal conductance. These variables, in turn, can be linked to large-scale ecosystem net primary production by using an efficiency factor model. NDVI, though, is very dependent on the reflectance characteristics of non-photosynthetic vegetation, such as woody stems, soil characteristics, and sun angle and viewing angle of the sensor. So, the index does not represent absolute ecophysiological attributes.
Multiple indices exist; they have all attempted to describe properly vegetative cover. NDVI was based on the simple ratio (SR) that exploited the differences between vegetation reflectances in the near infra-red and visible red spectra: (NIR/VIS). Other indices have been developed in an attempt to account for the variables that affect the calculation of vegetation content. The soil-adjusted vegetation index (SAVI) incorporates an adjustment factor (a) to the NDVI to account for the amount of exposed substrate since soil reflectance strongly affects NDVI: (NIR-VIS)/(NIR+VIS+a). Other indices—the optimized soil-adjusted vegetation index, the modified soil-adjusted vegetation index, and the transformed soil-adjusted vegetation index—specify SAVI's adjustment factor differently in an attempt to better account for substrate reflectance. Instead of relying on ratios, orthogonal indices depend on the existence of a "soil line" in spectral space; the most widely used is the greenness index, or green vegetation index defined by the tasseled cap method.
All of these vegetation indices are correlated with vegetation cover, although to varying degrees in different environments. New techniques for extracting ecological variables from satellite imagery include combining NDVI with texture analysis to constrain LAI better and performing multiple regression analyses directly on spectral bands instead of using vegetation indices. Most likely, these empirical methods will never be exact, so quantitative methods also have been explored.
Quantitative sources of function: Radiative Transfer Model Inversion
The problem of identifying the causes and consequences of land-use/land-cover change, part of the LCLUC Program mission, requires that the functional aspects of ecosystems be examined. This involves the incorporation of land-use change into models that can couple land use with biogeochemical, biophysical, and atmospheric dynamics. New techniques in quantitative data retrieval allow the application of remote sensing to expand from a role of mostly spatial description to one where functional relationships can be examined. The basic concept is to use the structural attributes that can be pulled from satellite imagery to relate directly to functional variables. These ecological variables can then be incorporated into models instead of literature values in order to make more realistic assessments of linkages between changes in canopy structure and biogeochemical processes. This technique holds promise for creating a general method of linking structure to function.
Biogeochemical models require the parameterization of plant carbon allocation either from knowledge of plant phenological properties or from proxies derived from remotely-sensed data that constrain aboveground carbon pools via variables such as LAI or non-photosynthetic vegetation index (NPVI). Accurate estimation of these variables will then strengthen biogeochemical models. These variables are derived from satellite imagery by the inversion of radioactive transfer models. The radioactive transfer models use physical process algorithms to explain how large-scale structural and biophysical attributes affect canopy reflectance. By iteratively changing these variables and comparing the resultant reflectance from the model to an image reflectance, the structural variables driving the reflectance in the image can be derived. The model parameters are then forward-integrated to yield bulk biophysical properties such as fAPAR and albedo.
This physically-based method also exploits the bi-directional reflectance distribution function (BRDF) of vegetation by expressly incorporating multi-view angle measurements into the process. This allows more accurate calculation of values. Vegetation reflects radiation anisotropically, depending on the leaf, canopy, and landscape structural and compositional characteristics and illumination and viewing angles. The multi-view angle measurements of vegetation BRDF will allow improved access to canopy structural characteristics (e.g. LAI) and the simultaneous retrieval of biophysical variables (e.g. fAPAR). The BRDF parameters are constrained either by specifying a possible range of values for a variable through theory and field measurements, or by finding relationships between variables that force them to systematically covary during model inversion. Leaf optical properties constrain the BRDF model inversions best since they are stable, and LAI and leaf angle distribution vary spatially, temporally, and within and among species. Studies of spectral behavior in AVHRR Bands 1 and 2, have found that green and senescent foliage should be treated differently in BRDF inversions.
Despite its success, the radioactive transfer model inversion technique faces its own challenges. The lack of reflectance information—usually a lack of observations (more than the free model parameters) and geometrically unique observations (sun-view angle)—limits its applicability.
BRDF inversion techniques aim for applicability to new environments. The method can also be applied to standard techniques for improved information extraction. Multi-view angle and geometric-optical inversions can help improve classification and mapping by backing out crown dimensions (geometry) and spacing (shadowing). Since these canopy characteristics are usually more stable than LAI or fAPAR, they will provide a more concrete signal of vegetation type.
The Earth Observing System
The sensors brought on line as part of NASA’s Earth Observing System aid the study of land-use/land-cover change immensely. NOAA's AVHRR sensor had been one of the few instruments capable of acquiring off-nadir radiance measurements with adequate repeatability for BRDF model inversions. New sensors, though, have multiple viewing capacities. The high spectral resolution provided by the new sensors allows for the review of the accuracy, and further detailing, of classification into different cover types and chemical composition of plant canopies (including photosynthetic pigment components and nitrogen status). The 20 optical channels and multi-angle viewing capability of MODIS revolutionizes measurements of key ecological variables. For example, there is a 250-meter global land-use/land-cover change product available from the red and infra-red bands on MODIS, which provides information about vegetation, as outlined above. The BRDF inversion techniques benefit from shortwave infra-red channels that have increased the number of unique optical channels. This sensor increases the spatial and spectral resolution, and the MISR instrument the multiple view angle data, over that already available from AVHRR.
Traditional and innovative techniques incorporate data derived from remote sensing into land-use/land-cover change studies. Vegetation classification and mapping continues to be improved. The data products from this method permit the continued monitoring of land use and land cover. Efforts to understand ecological consequences of land-use/land-cover change employ traditional (vegetation indices) and new (model inversion) techniques for deriving function from structure. These advancements only reinforce the importance of remote sensing as a component of ecological studies.
The increased interest in maintaining ecological goods and services while monitoring large-scale changes in land cover has increased the importance of remote sensing in ecological studies. The mandate to understand the nature and consequences of land-use/land-cover change requires advances in both basic and applied ecological research. Accordingly, research progress will both advance the intellectual status of ecological research and address societal needs.
- Asner, G.P., B.H. Braswell, D.S. Schimel, and C.A. Wessman. 1998. Ecological research needs from multiangle remote sensing data. Remote Sensing of Environment, 63:155-165.
- Asner, G.P., C.A. Wessman, and S. Archer. 1998. Scale dependence of absorption of photosynthetically active radiation in terrestrial ecosystems. Ecological Applications, 8:1003-1021.
- Asner, G.P., C.A. Wessman, and D.S. Schimel. 1998. Heterogeneity of savanna canopy structure and function from imaging spectrometry and inverse modeling. Ecological Applications, 8:1022-1036.
- Asner, G.P., C.A. Wessman, D.S. Schimel, and S. Archer. 1998. Variability in leaf and litter optical properties: Implications for BRDF model inversions using AVHRR, MODIS, and MISR. Remote Sensing of Environment, 63:243-257.
- Crist, E.P. and R.J. Kauth. 1986. The Tasseled Cap de-mystified. Photogrammetric Engineering and Remote Sensing, 52:81-86.
- DeFries, R. 1998. Characterizing land cover heterogeneity and land cover change from multisensor satellite data. Project Abstracts.
- Earth Sciences Portal, NASA Goddard Space Flight Center.
- Fischer, A., S. Louahala, P. Maisongrande, L. Kergoat, and G. Dedieu. 1996. Satellite data for monitoring, understanding and modelling of ecosystem functioning. In B. Walker and W. Steffen, eds. Global Change and Terrestrial Ecosystems. Cambridge University Press: New York. pp 566-591. ISBN: 0521578108
- Green, K., and D. Weinstein. Automated Change Detection Using Remotely Sensed Data. Pacific Meridian Resources. unpublished manuscript.
- Hansen, M., R. DeFries, C. Dimiceli, C. Huang, R. Solhberg, X. Zhan, and J.R.G. Townshend. 1998. Red and infrared space partitioning for detecting land cover change. IGARSS’98: 1998 IEEE International Geoscience and Remote Sensing Symposium Proceedings. Seattle, Washington. Vol. V: 2512-2514.
- Lawrence, R.L. and W.J. Ripple. 1998. Comparisons among vegetation indices and bandwise regression in a highly disturbed, heterogeneous landscape: Mount St. Helens, Washington. Remote Sensing of Environment, 64:91-102.
- Matson, P. A., and S. L. Ustin. 1991. The future of remote sensing in ecological studies. Ecology, 72:1917.
- NASA's Earth Observing System
- NASA's Earth Observing System Program Office.
- NASA's Land Cover Land Use Change Program.
- NASA's MISR Program
- NASA's MODIS Program
- Nielsen, A.A., K. Conradsen, and J.J. Simpson. 1998. Multivariate Alteration Detection (MAD) and MAF Postprocessing in multispectral, bitemporal image data: New approaches to change detection studies. Remote Sensing of Environment, 64:1-20.
- Quattrochi, D.A. and R.E. Pelletier. 1991. Remote sensing for analysis of landscapes: An introduction. In M.G. Turner and R.H. Gardner, eds. Quantitative Methods in Landscape Ecology. Springer Verlag: New York. 51-76. ISBN: 0387973265
- Skole, D. 1998. Case studies and diagnostic models of the interannual dynamics of deforestation in southeast Asia: Is the missing sink for carbon in land cover change. Project Abstracts.
- Spell, R. E., R. G. Kempka, J. K. Graves, and P. T. Cagney. 1995. Change Detection of Pacific Coast Estuaries and Bays Using Landsat Thematic Mapper. In Third Thematic Conference on Remote Sensing for Marine and Coastal Environments, Seattle, Washington.
- Wessman, C.A. and G.P. Asner. 1998. Ecosystems and problems of measurements at large spatial scales. In M. Pace and P. Groffman, eds. Successes, Limitations, and Frontiers in Ecosystem Science. Springer Verlag: New York. 346-371. ISBN: 0387984763
- Wolter, P.T., D.J. Mladenoff, G.E. Host, and T.R. Crow. 1995. Improved forest classification in the northern lake state using multi-temporal Landsat imagery. Photogrammetric Engineering & Remote Sensing, 61:1129-1143.
- Wulder, M.A., E.F. Ledrew, S.E. Franklin, and M.B. Lavigne. 1998. Aerial image texture information in the estimation of northern deciduous and mixed wood forest Leaf Area Index (LAI). Remote Sensing of Environment, 64:64-76.