Elements of Change 1995: Session I Appendix 1

In global change studies, the investigation of Earth surface properties needs to be dynamic in approach. However, past applications of remote sensing techniques have been more synoptic and provided current status of Earth surface and atmospheric conditions. Recently, a concerted effort has been undertaken to better document changes in Earth surface properties, especially in terrestrial ecosystems undergoing rapid changes.

Current issues related to land use change and changes in land cover are applying dynamic modeling and field studies to assess process level changes in association with changes in land cover related to environmental and land use changes. These studies are limited in space because site level studies are constrained by the number of sites that can be investigated. The linking of multi-temporal remote sensing studies and process level studies in order to develop a method of assessing regional and global dynamics to changes in climate or land use will be needed.

In the savanna and steppe regions of the world, land surface and ecosystem properties have not been well documented. These regions are important because of their areal extent and the human population which depends on them. These ecosystems are sensitive to changes in rainfall, temperature, and land management practices which determine the extent of woody and grass species cover. Changes in land use practices are continuing to modify these ecosystems due to population pressures and intensification of land use practices (e. g., grazing intensity, cropland expansion, agroforestry expansion, fuel wood extraction, etc.). These changes in land use and environmental factors which alter the extent of woody vegetation in the savanna and steppe regions influence the energy, hydrological and biogeochemical fluxes within these ecosystems. Changes in these fluxes influence local and global environmental changes, such as modifying carbon stocks, and changing evapotranspiration fluxes and forage availability and quality in the region. The two areas this group focused on are undergoing rapid land use change: the cerrado region of Brazil and the Eurasian steppe.

Changes in land use and environmental factors which alter the extent of woody vegetation in the savanna and steppe regions influence the energy, hydrological and biogeochemical fluxes within these ecosystems.

Detection of these changes during the past 30 years is possible with the remote sensing data and field observations available.

Many of these changes in woody land cover can take several decades to manifest themselves over large areas; however, fine scale changes can be detected with repeated observations. Short-term changes can be detected in situations where human activities related to grazing, fire, cropland conversion, or reforestation take place. The areal extent of these changes is usually large enough to detect using AVHRR-level remote sensing techniques. The initial test areas are the cerrado region of Brazil and the steppes of Asia.

In these areas, anthropogenic and natural processes modifying the woody extent will be identified and studied. Fire, grazing, and cropland conversion are key land use practices which contribute to changes in woody cover in these regions. Human activities can accelerate or reduce the rate of woody cover change through different applications of fire, grazing, and cultivation. Natural changes in climate patterns leading to drought or intensification of rainfall can modify the vegetation community. Climate variability in these systems is very large; a 30 -year analysis of the coefficient of variance of rainfall can exceed 50%. This characteristic variability results in a highly dynamic vegetation response and hampers our ability to detect changes solely due to human activities.

An integrated, multi-technique approach is proposed for the study of changes in woody cover extent in the savanna and steppe regions of the world. This integrated approach will incorporate multi-temporal, multi-spatial monitoring of the land surface using remote sensing techniques (AVHRR, MSS, SPOT, TM, SAR, aerial photographs, etc.). Modeling analysis of ecological features of these systems will be used to assess current changes in ecosystem characteristics that can be used to evaluate current dynamics and future changes in woody dynamics of these systems. Field studies of processes influencing changes in land cover and measuring the rate of land cover change will be incorporated to supply more detailed information on factors affecting the rate of change in these systems. These field studies will incorporate information dealing with social-economic-political sectors as well as environmental features.

Remote sensing techniques can be applied to woody cover detection at several scales of analysis. At the fine spatial scale of resolution we are limited by the temporal resolution of analysis. However, synoptic, subdecadal to decadal analysis of land cover over the region can provide the needed information to determine the synoptic status of woody cover in the region. Ancillary data from field studies can then be applied to determine the ecological changes in the system that can be used for extrapolation studies.

Coarse spatial resolution analysis can be applied with higher temporal frequency and be used to document large scale changes in these systems. New techniques with unmixing of end-member analysis, spectral cluster analysis, and BRDF techniques can be used to determine different aspects of the land surface that would indicate changes in the woody extent and other changes in land cover features related to changes in grazing and fire intensity.

Spectral classifiers operate by grouping together those pixels with similar reflectance values over the various sensor bands. Since agricultural areas and savanna/steppe regions have distinct spectral characteristics, especially if time of scene selection is chosen to optimize these differences, this technique is able to distinguish cover classes such as cropland and rangelands. Integration of the resulting clusters will be based both on the spectral statistics of the spectral bands as well as their spatial structure.

The high temporal frequency of AVHRR-like instruments can also be used to analyze for causal factors of change to the region. Fire is an example of an event that can quickly reduce woody cover. Malingreau et al. (1993) developed a conceptual framework describing the mechanics of a fire information system. The only method of monitoring fire activity is through the use of remote sensing. Nearly all fire detection is done using the NOAA-AVHRR (Advanced Very High Resolution Radiometer) 3 micron channel (1-km resolution). High resolution instruments are used to account for the average fire scar size and this is used in the algorithm for estimating area burned represented by a detected hot spot. GIS methods are extensively used in the application of models for estimating fire effects including the release of greenhouse gases and particulate matter from detected fires. Layers within the GIS framework must include total fuel loading, weather, appropriate fire model, and algorithms for estimating fuel consumption and emissions production (Scholes et al., 1995).

There are several limitations of AVHRR for use as the driver of a fire information system:

An integrated, multi-technique study approach is proposed, incorporating multi-temporal, multi-spatial monitoring with remote sensing, as well as modeling analysis and field studies.

These limitations must be accounted for through compensating for the instrument bias. Some of the limitations of AVHRR will be overcome with the launch of the Moderate Resolution Imaging Sensor (MODIS) as part of the Earth Observing System (EOS) platform (Kaufman et al. 1994). This system will considerably enhance the capability to measure smoke, clouds and fires by measuring their properties in additional channels that are more suitable (e. g., the blue channel for smoke measurements or the channel at 2.2 nm for cloud drop size measurements). The channel at 3.75 nm will be used for fire detection and will have a much higher temperature of saturation (500°K). This should allow separation of smoldering and flaming processes.

These limitations may also be addressed through the use of the international SAR sensors (e. g., ERS-1, ERS-2, RADARSAT, ENVISAT). Radar can directly observe land surfaces independent of cloud and smoke cover and time of day. Thus it may be possible to use AVHRR or MODIS data for fire detection and supplement these measurements with the detailed monitoring of the SAR for quantifying landscape change.

The angular distribution of radiation scattered by the Earth surface also contains information on the structural and optical properties of the surface. Potentially, this information may be retrieved through the inversion of surface bidirectional reflectance distribution function (BRDF) models. In contrast to vegetation index methods, inversions produce quantitative estimates of biophysical parameters that do not depend on empirical relationships. While previous inversion work has largely been limited to use of data from ground-based radiometers, recent work (B. H. Braswell and J. L. Privette, personal communication) has shown that accurate inversions are possible with 1.1 km data from the AVHRR. Compared to other environmental satellite sensors, AVHRR employs a large field of view which results in relatively diverse angular measurements. This is advantageous to the inversion problem. Moreover, AVHRR samples each Earth target at least twice each 24 hours.

The availability of remote sensing data and the increasing availability of land surface statistical information has improved our ability to model terrestrial ecosystem dynamics over greater spatial extents.

Since data from multiple AVHRR sensors are typically ingested each day, the potential for collecting a sufficient number of cloud-free samples during a relatively short period (~10 days) is significant. For example, LAI and leaf optical properties were accurately retrieved from data collected during the FIFE experiment. About 3 to 6 cloud-free samples, collected during an 11-day period, were used in each inver sion. By combining samples collected over adjacent land targets, Braswell successfully inverted the SAIL model with the PATHFINDER AVHRR data from Africa. Once model parameters are retrieved, the models may be used in forward mode to estimate albedo and fAPAR for any solar angle. Currently, techniques are being used to deconvolve AVHRR data collected over heterogeneous areas such that inversions may be possible at 180 m resolution (D. Baldwin, personal communication). Upon launch of MODIS (an AVHRR successor) and MISR (which collects samples at nine different view zenith angles per pass), global inversions may routinely be possible at relatively high resolution (100-300 m).

Ecosystem models can provide information on terrestrial ecosystems that cannot yet (or in the future) be measured remotely . However, knowledge of the spatial extent of various land surface properties that can be used to parameterize ecosystem models is currently limited. The availability of remote sensing data and the increasing availability of land surface statistical information has improved our ability to model terrestrial ecosystem dynamics over greater spatial extents and to verify the behavior of these models within a limited extent.

The current modeling capability of most ecosystem models does not need specific information on spatial patterning of woody and herbaceous vegetation in these regions. Aggregated spatial information is adequate for most modeling needs. In special applications more detailed pattern analysis is needed to simulate certain climate and mass fluxes from these systems. Land use information within these regions is important for simulations of current dynamics of these ecosystems, and the future changes in these systems.

Prediction of the ecosystem response to changes in land use and environmental conditions can be simulated assuming current trends of land use and climate.

Current knowledge of ecosystem dynamics relative to climate variability and land use patterns is limited to relatively few sites. Extensive data on land cover and land use patterns are available, but work is needed to incorporate and collaborate these data within a process-level modeling system and a remote sensing framework.