J.-P. Malingreau
Jean-Paul Malingreau is based in Ispra, Italy, at the Institute for Remote Sensing Applications of the European Communities' Joint Research Centre.
The development of better procedures for monitoring the state of and changes in the world's forests is considered a major requirement for supporting national policy as well as the emerging science of global change. For nearly 20 years, space observation techniques have contributed to local and national forest inventories. This article assesses the potential of current remote sensing approaches and technologies for answering emerging needs in the field of forest assessment. Indications of what could constitute a global forest information system and the challenges to its development are described. Concepts expressed here are derived in part from presentations and discussions at the World Forest Watch Conference, organized as part of the International Space Year and held in São José dos Campos, Brazil, in May 1992 (Malingreau, Da Cunha and Justice, 1992).
Historically, humans have had an ambivalent relationship with the forest; simultaneous needs for exploitation and conservation continue to colour appreciation of the role of the forest in local economies, national development, environmental protection, global change dynamics, etc. Needs for information about the forests are as varied as opinions and positions on the issues themselves. An additional complexity in the current context is related to the fact that most information requirements are naturally expressed on "human scales"; that is those which correspond most directly to visual observations or measurements on the ground. Forest classification schemes reflect such a perspective, since they are based on criteria such as floristic composition and structure. These parameters are obviously not directly measurable from space sensors, particularly those working at coarse ground resolutions.
One can thus, at the outset, question the relevance of orienting remote sensing analysis only towards the search for known ground features (e.g. a known forest class) by inversion of a spectral signal or otherwise. There may instead be novel information on the nature and dynamics of the forest cover imbedded in such data sets, information linked to the spatially integrating nature of the measurement. Research should obviously find here a privileged field of investigation. Today, most of the "common" needs expressed with respect to forest cover form a limited group of basic information items. Increasingly, however, sophisticated users, such as those investigating sustainable forestry practices or the role of forest biomass in biogeochemical cycles, will address more complex questions requiring novel data analysis approaches.
This leads to an additional point. The increasing complexity of the demand for forest-related data and the growing sophistication of remote sensing techniques have led to an ambiguous situation with respect to the technology. Local and national inventories are needed to answer very immediate management needs while, at the same time, pressing calls are being made for monitoring techniques aimed at detecting changes in the forests on a global scale. What is becoming increasingly clear, if not always well understood, is that it is very difficult to reconcile the two ends of the spectrum of requirements (Justice, 1992). Remote sensing techniques operated on continental and global scales can bypass locally significant data. The still necessary "anchoring" of global analyses in local validation exercises can rely on very different approaches from those used in "traditional" local assessment.
Space-based earth observation technology applicable to forest inventory and monitoring is fairly well known: the Landsat Multispectral Scanner, Thematic Mapper, the SPOT Multispectral and Panchromatic instruments, the Japanese Marine Observation Satellite and the United States National and Oceanic and Atmospheric Administration's Advanced Very High Resolution Radiometer (NOAA-AVHRR) sensors virtually form a complete list of operational earth observation instruments. The Synthetic Aperture Radar of the ERS- 1 satellite has recently been added to the panoply, although its application to vegetation studies is still at a development stage. Extensive literature on the use of satellites for forest surveys exists. The following sections discuss a series of issues regarding current and future uses of those sensors.
Low-resolution sensors
Low-resolution data, such as those provided by NOAA-AVHRR (1.1 km resolution), have found an increasing field of application in forest monitoring over the past few years. Indeed, they appear to provide a unique compromise between ground resolving power, vegetation-related information content, the frequency of acquisition and the breadth of geographical coverage. Currently, the AVHRR data appear to be the only practical solution to the acquisition of a so far, still elusive - global land coverage. While a host of problems still persist in such an approach, it appears certain that the AVHRR will provide the main operational source of data on forest cover at continental to global scales in the near future. Several large-scale projects have begun acquisition and analysis of "wall to-wall" AVHRR data sets. The TREES Project (CEC, 1991) is currently collecting a suitable AVHRR data set (1989-1992) for forest analysis of the entire tropical forest belt. Similar regional exercises are being undertaken by various national or international institutions (United Nations Environment Programme - Global Resource Information Database [UNEP-GRID], the National Aeronautics and Space Administration [NASA], Woods Hole Research Institute, etc).
Human visual analysis of spectral data is crucial in the interpretation of remote sensing images
The experience gained in the analysis of AVHRR data for forest monitoring and in the preparation of the above-mentioned large-scale projects highlights a number of important issues:
· AVHRR 1 km full-resolution data come from local receiving stations (more than 150), with the exception of limited on-board recorded data sets; global coverage therefore requires the networking of stations.· Substantial efforts are required to preprocess AVHRR data sets for geometric and radiometric corrections and calibration.
· For most forest mapping applications, a major operation in the AVHRR methodology is the search for acceptable images through a screening process involving the display and evaluation of each orbit pass to ensure that smoke, clouds or other obstructing factors are not present.
· The AVHRR analysis consists essentially of the local search for contrasts between vegetation forms in one or more spectral channels. Therefore, there is no methodology that can be universally applied, since reflectance characteristics, vegetation mosaics, seasonality effects, etc. differ widely from area to area.
· The seasonality of tropical vegetation is also an important classification criterion as well as a fundamental ecological characteristic. When captured by a time series of AVHRR data, seasonal variations are useful keys to the identification of vegetation formation.
· Additional clues to the detection and identification of deforestation processes can be obtained via the use of thermal channels that can detect the presence of fire (burning is a common way of clearing forest vegetation in many parts of the world). Active fires or burn scars are very apparent on the AVHRR data sets.
High-resolution sensors
Landsat MSS, Thematic Mapper and SPOT Multispectral instruments, the so-called high-resolution instruments (10 to 80 m resolution), are important data sources for forest mapping and change detection at the local and national level. Many countries have embarked on an intensive analysis of such products. Most of the analyses, outside research, are carried out visually on enhanced photographic products. Increasingly, wall-to-wall high-resolution data sets are being assembled.
The efforts of Brazil are noteworthy in this respect. By combining a relatively inexpensive reproduction technique (colour composite or even black and white single-channel prints) with labour-intensive analysis, exhaustive forest inventories of large regions can be obtained. There are now proposals to cover the tropical forest portions of other areas, including Africa and Southeast Asia, with "historical" MSS data in the framework of NASA's Landsat Pathfinder activities. The Forest Image Catalogue of Europe, presented at the World Forest Watch Conference is another application of the extensive use of Thematic Mapper data.
This trend towards large-scale applications of high-resolution data sets has several implications for the future. It implies that classification schemes will have to take into account not only the information needs but also the capabilities of the remote sensing technique in terms of class or attribute identification. Moreover, if high resolution is to become the norm in vegetation monitoring, the pricing of the data will have to decrease significantly or more resources will need to be made available if assessment with adequate frequency is to be affordable. Systematic and exhaustive coverage of large forested areas will require that special attention be given at the satellite-management level to maximizing the probability of acquiring the full data set with the required frequency.
High-resolution data also play an important role in the validation of low-resolution data. A multilevel analysis must be performed in which information obtained at a more detailed level helps in validating that obtained at a coarser level. Inversely, the global perspective afforded by the AVHRR can help in the sampling design used for the selection of sites for higher-resolution analysis.
High-resolution data images can al so be used, as in the method employed by FAO in the Forest Resources Assessment Project 1990, as units of analysis in a worldwide change measurement exercise based on a combined stratified and proportional-to-size random sampling.
Three groups of forest parameters form the core of current requirements: vegetation classification; changes in vegetation cover; and surface physical characteristics.
Mapping vegetation by remote sensing
Mapping vegetation based on high-resolution data is the "classical" operation using remote sensing observations. It consists of separating spectral data provided by the satellite sensors into groups that can be associated with specific vegetation classes. Although these classes are often identified on the basis of floristic composition, structure and ecoclimatic zone (e.g. ombrophilous evergreen rain forest), remote sensing classification is based exclusively on a set of spectral measurements. Direct classification is accurate insofar as these measurements are unambiguously associated with a particular vegetation class. High accuracy is possible when distinguishing between the most contrasted situations (e.g. forest cover and clearings), but the level of accuracy is less certain when there is a higher complexity of the association of vegetation forms (e.g. in successional stages).
Given the enormous spatial variability of spectral characteristics of surfaces and the unique ability of the human brain to distinguish between similar but not identical spectral features, visual analysis of spectral data tends to be a more reliable, although slower, approach to the analysis of remote sensing images. In this approach, it is important to use the same individual or team of interpreters over time to ensure consistency. Outputs from such analyses are then incorporated in a geographical information system (GIS) where they are integrated with other information layers.
Classification problems are quite different when dealing with low-resolution data of the kind derived from the AVHRR instrument. In this case, there are fewer landscape clues and the analysis relies more on the reflectance or emittance values in a few spectral channels. The contrast between a forest canopy and other cover types determines the ease with which a formation can be identified. Given the large variability associated with this type of measurement, there is often a limited possibility of defining spectral signatures; spectral contrasts alone are then used. For example, large forest blocks isolated in the middle of an agricultural area can be clearly identified using brightness temperature contrasts, while primary forest blocks in the middle of secondary or less-structured vegetation will be seen as an area of lower, near-infrared reflectance. In this regard it is noteworthy that, with rising population pressure on forest resources, increasing areas of forested landscapes are likely to be characterized by mosaics of cover types rather than by "pure" formations.
|
Remote sensing defined By definition, remote sensing encompasses all the techniques necessary to obtain information about an object without actually being in physical contact with it. In practice, it chiefly concerns the technology related to the mapping and monitoring of the earth's features and resources from data collected by means of aircraft or satellites (space borne). To a large extent, remote sensing depends on measuring electromagnetic waves emitted and/or reflected from objects on the earth's surface. All objects reflect and radiate electromagnetic radiation and, thanks to their specific features and behavior with regard to different wavelengths within the visible and invisible spectra, all objects have a "fingerprint" and can be recognized on remote sensing data, as collected by various remote sensors. Remote sensors can be mounted on a large variety of air- or space-based platforms, operating at different altitudes and for different periods of time. Satellite remote sensing data are being used increasingly to map and monitor earth resources, especially over extensive areas. Satellite systems are usually classified according to their spatial resolution into environmental satellites (Meteosat, GOES, NOAA); medium-resolution satellites (Landsat MSS, IRS1, JERS1); and high resolution satellites (Landsat TM, SPOT, ERS-1, MAK6-M). Environmental satellites are best suited to frequent (daily or weekly) monitoring of relatively large areas, such as continents, subregions or countries. They are primarily used in meteorology and oceanography and, more recently, for monitoring vegetation conditions, principally of large pastures or forested areas, at scales from 1:10 000 000 to 1:2 000 000. Medium-resolution satellites are primarily represented by the Landsat, with its Multispectral Scanner (MSS) which has been in operation since 1972. They provide imagery at small to medium scale, 1:1 000 000 to 1:200 000, for land-use studies and have particular application to forestry. High-resolution satellites are more recent and have principally been used since the mid-1980s. They allow mapping at scales of up to 1:25 000 (e.g. SPOT). The electromagnetic waves that are received by the satellite sensors produce electrical signals which are subsequently transformed or processed into usable products (e.g. photographs or digitally coded image computer tapes). With the aid of appropriate tools such as analogue and digital processors as well as the human eye, these products can be analysed to gain an understanding of the type of earth resources present as well as of the quantitative and qualitative changes or variations (both spatial and temporal) of these resources. In the case of forestry, remote sensing can be useful in the identification and analysis of forest areas, i.e. their location and size, state of degradation and the level of human pressure visible through heavy deforestation, fires and agroforestry. With high-resolution satellites, certain physiognomic parameters related to various cover classes permit the discrimination of forest, woodland and shrubland, while floristic parameters permit the determination of broad-leaved, coniferous and mixed stands. Satellite remote sensing can also assist in forest management by providing information on accessibility, e.g. topography, paths or roads, and also permit yearly or even monthly monitoring of main forest stands and logging over very large areas such as provinces or countries. |
Detecting and measuring changes by remote sensing
An area of particular interest is the detection and measurement of the transformation of one type of forest cover into another or into an entirely different type of land use, e.g. deforestation. Again, the accuracy of change measurement using remote sensing depends on the ground resolution and the spectral discrimination between forms. In theory, change detection using spectral data does not require the preliminary identification of the constituting land cover types. However, this permits only the identification of the disappearance of one particular class of forest, without addressing the problem of identifying the replacement class (e.g. forest/non-forest determination). The interpretation of change in ecological or other terms requires the characteristics of the beginning and end terms to be known.
Remote sensing components of a forest assessment information system
|
Low-resolution component · Continuous and wall-to-wall coverage of all land surfaces with appropriate spectral channels, at a medium to coarse resolution (250 to 1 000 m) for a general monitoring of vegetation distribution and conditions. · The global system should be capable of distributing data to local users and of assembling a complete archive of all acquisitions (non-sampled) in a central location. The Advanced Very High Resolution Radiometer (AVHRR) experience could be used as a model. · The global analysis should to be validated by high-resolution data and field verification. · High-resolution systems should be on stand-by to focus on areas of active deforestation as detected by an "alarm system", based on low-resolution satellite and other conventional data sources. High-resolution component · A regular (every three to five years) comprehensive assessment of forest cover should be undertaken using high-resolution data sets, such as Landsat TM. Data from national and local mapping and monitoring activities should be integrated into the global assessment (e.g. via regional centres). This would require a level of agreement on a hierarchical classification amenable to progressive generalization. · High-resolution data sets should be made increasingly available; if commercialization can contribute to such dissemination, a price structure must be established that will permit the extensive and repeated applications required for forest monitoring. · A geographical information system (GIS) is a major component of the exercise leading to an advanced forest information system. |
Currently, it is still difficult to ensure that all types of forest cover change are measured or even detected by using remote sensing applications. Particularly difficult is the identification and measurement of the process of forest degradation. There is an urgent need for a better characterization of spatial patterns of change on various scales (stand, regional, subcontinental, etc.) and at various resolutions. A typology of forest/non-forest/patterns and of the mechanisms and features of their transformation over time would be highly valuable. It would, in addition, be a step towards defining a common framework for inventory and analysis.
The value of forest change information derived from remote sensing data will be greatly enhanced if it is integrated into a spatially explicit GIS. This will permit harmonization with data related to other landscape characteristics (e.g. soils, climate, river networks, field measurements of biomass and timber volumes) or societal characteristics (roads, population densities, migration patterns) which are important elements in the deforestation process. It also opens the door for predicting spatio-temporal trends.
Forest seasonality
Forest seasonality is a characteristic used explicitly in tropical forest classification, where it is mainly related to rainfall patterns (as opposed to temperature-controlled seasonality in temperate areas). While such classification is often used to determine whether a forest is seasonal or not, it generally does not reveal information on the intensity or even the facultative nature of the seasonality characteristic. Seasonality in spectral signals is also an important mapping criterion; yet time series of satellite data usually show gradients of seasonality more than a clear-cut distinction between evergreen and deciduous classes. Furthermore, satellite data taken over a long period show that inter-annual variability must be considered in the classification. The enormous spatio-temporal variability of the satellite signals is, for example, seen in parts of Southeast Asia and Africa where altitudinal and latitudinal gradients control the distribution of seasonal formations. The spectral distinction between a dry dipterocarp and a mixed deciduous forest of northern Thailand or Myanmar is, for example, notoriously difficult even on high-resolution data sets.
Time series of satellite-derived vegetation index data are now available for over ten years in the Global Area Coverage (GAC) product; they show a continuum of signal amplitudes as one goes from the typical "evergreen" to the more seasonal formations. This information would be relevant for a series of analyses; for example, seasonal flushes affect evapotranspiration, primary productivity and geochemical cycles. However, time series remote sensing data on forest canopies has been little exploited to date, be it for classification or process studies. The accurate evaluation of changes in forest canopies, for example those associated with climate change or degradation, will be possible only if the seasonal changes of the vegetation under "normal" circumstances can be identified.
The development of remote sensing techniques for forest monitoring shows various levels of maturity, depending on the instrumentation and the scale of application. While some aspects of high-resolution image analysis are fairly straightforward, those related to large-scale coverage or low-resolution image exploitation are still in the development stage. Progress over the past five years has shown, however, that prospects are broader than what had been previously imagined. For example, new dimensions are being opened by the growing availability of extensive coverage with high-resolution data sets (Landsat and SPOT) and by the extensive and repetitive coverage with low-resolution data sets (AVHRR). The temporal dimension of those analyses has been radically revised by recent investments in the processing and analysis of historical data sets (e.g. NASA's Landsat Pathfinder) and in the collection and evaluation of long-term frequent measurements over large areas (e.g. the improved and global land cover data sets of the International Geosphere-Biosphere Programme (IGBP)). There now appears to be the potential to move towards an effective satellite-based global forest monitoring system. To be effective, such a system would need to have the following main features:
· it would need to incorporate global and national objectives - analyses and results must be amenable to "nesting" on different scales;· validation would need to be a constant concern and be built into the operational procedure;
· results of the remote sensing analysis would need to feed into a more comprehensive forest information system;
· the system would need to have the potential to support practical forest planning and management as well as scientific research;
· it would need to include a continuous monitoring component providing an "alarm function" as well as stand-by capabilities to identify quickly areas of rapid change;
· it would need to include regular extensive coverage of all forest areas by high-resolution instruments for measuring change within an accuracy of 10 percent.
The remote sensing components of such a forest assessment system are summarized in the Figure.
The experiences of the past few years have divided the community of remote sensing specialists into two schools of thought with respect to global forest monitoring. One approach advocates a distributed monitoring system, with national assessments being generated by national forest services and fed into global syntheses at a central facility. The second advocates that a central agency generate the complete global assessment by independent means. Each approach presents advantages and disadvantages.
The first alternative has the key advantage of placing national forest inventories in the hands of the agencies responsible for forest management. An intimate knowledge of field conditions is obviously a guarantee of the pertinence of the remote sensing data analysis. However, this approach requires access to the appropriate monitoring techniques (satellite data, GIS, field facilities, etc.) and the capability to perform remote sensing work at "exchangeable" standards. Problems of data flow and the harmonization of disparate classifications can slow down significantly the aggregation of data in a central processing facility. Upgrading and harmonizing national forest monitoring approaches is clearly a long-term proposition which must be envisaged in the framework of the development of national capabilities.
The centralized approach would be a more immediate answer to the needs for data on regional, continental and global scales because it relies on technologies and procedures that could already be implemented in a coordinated - if not centralized - fashion. The potential for adoption of a single methodology and classification would facilitate the rapid analysis of satellite and ancillary data. Of course, the assistance of experts and forestry services who have the necessary field experience would still be required in the process of validation; the means of integrating national forest services into such a centralized effort has yet to be assessed. It is perhaps worth considering the experiences of other environment analysis systems for hints on how to proceed. For example, the meteorological community undertakes global analysis and modelling in highly centralized and specialized institutes while national meteorological offices continue their routine work at their own scales of interest.
The urgency of acquiring precise and up-to-date assessments of forest conditions around the world on a continuing basis is clearly having a strong pull on remote sensing techniques. Forests are widely dispersed, complex and dynamic systems that are amenable to the kind of observations provided by orbital sensors. Remote sensing techniques have already made significant contributions. Efforts must now concentrate on the integration of past experience and results into truly global and multilevel exercises.
Furthermore, the level of sophistication of the applications should progressively match the complexity of the host of challenges raised by attempts to delimit a path towards the conservation, management and sustainable exploitation of the world's forests. Although technology and applications are still lacking, enhancing institutional support and developing adequate scientific capability at all levels are also high on the priority list. The importance of the issues surrounding the use and survival of the world's forests warrants that efforts towards their proper monitoring be continued with redoubled commitment.
CEC. 1991. Tropical ecosystem environment observations by satellites. Special paper SP-1.90.31. Ispra, Italy, Joint Research Centre, CEC.
IGBP. 1992. Improved global data for land applications. A proposal for a new high resolution data set. Report 20. Stockholm, IGBP.
Justice, C.O. 1992. Satellite monitoring of tropical forests. A commentary on current status and international roles. In J.-P. Malingreau, R. Da Cunha & C. Justice, eds. Proc. World Forest Watch Conf. CEC EUR 14651 EN. Ispra, Italy, Joint Research Centre, CEC.
Malingreau, J.- P., Da Cunha, R. & Justice, C. 1992. Proc. World Forest Watch Conf. CEC EUR 14651 EN. Ispra, Italy, Joint Research Centre, CEC.
Malingreau, J.-P. 1992. Satellite-based forest monitoring. A review of current issues. In J.-P. Malingreau, R. Da Cunha & C. Justice, eds. Proc. World Forest Watch Conf. CEC EUR 14651 EN. Ispra, Italy, Joint Research Centre, CEC.
Myers, N. 1992. Future operational monitoring of tropical forests: an alert strategy. In J.-P. Malingreau, R. Da Cunha & C. Justice, eds. Proc. World Forest Watch Conf CEC EUR 14651 EN. Ispra, Italy, Joint Research Centre, CEC.
