Peter Scarth

Bringing an ecological view of change to Landsat-based remote sensing

When characterizing the processes that shape ecosystems, ecologists increasingly use the unique perspective offered by repeat observations of remotely sensed imagery. However, the concept of change embodied in much of the traditional remote-sensing literature was primarily limited to capturing large or extreme changes occurring in natural systems, omitting many more subtle processes of interest to ecologists. Recent technical advances have led to a fundamental shift toward an ecological view of change. Although this conceptual shift began with coarser-scale global imagery, it has now reached users of Landsat imagery, since these datasets have temporal and spatial characteristics appropriate to many ecological questions. We argue that this ecologically relevant perspective of change allows the novel characterization of important dynamic processes, including disturbances, longterm trends, cyclical functions, and feedbacks, and that these improvements are already facilitating our understanding of critical driving forces, such as climate change, ecological interactions, and economic pressures.

Prediction and validation of foliage projective cover from Landsat-5 TM and Landsat-7 ETM+ imagery

The detection of long term trends in woody vegetation in Queensland, Australia, from the Landsat-5 TM and Landsat-7 ETM+ sensors requires the automated prediction of overstorey foliage projective cover (FPC) from a large volume of Landsat imagery. This paper presents a comparison of parametric (Multiple Linear Regression, Generalized Linear Models) and machine learning (Random Forests, Support Vector Machines) regression models for predicting overstorey FPC from Landsat-5 TM and Landsat-7 ETM+ imagery. Estimates of overstorey FPC were derived from field measured stand basal area (RMSE 7.26%) for calibration of the regression models. Independent estimates of overstorey FPC were derived from field and airborne LiDAR (RMSE 5.34%) surveys for validation of model predictions. The airborne LiDAR-derived estimates of overstorey FPC enabled the bias and variance of model predictions to be quantified in regional areas. The results showed all the parametric and machine learning models had similar prediction errors (RMSE < 10%), but the machine learning models had less bias than the parametric models at greater than ~60% overstorey FPC. All models showed greater than 10% bias in plant communities with high herbaceous or understorey FPC. The results of this work indicate that use of overstorey FPC products derived from Landsat-5 TM or Landsat-7 ETM+ data in Queensland using any of the regression models requires the assumption of senescent or absent herbaceous foliage at the time of image acquisition.

Determining Forest Structural Attributes Using an Inverted Geometric-Optical Model in Mixed Eucalypt Forests, Southeast Queensland, Australia

The Montreal Process indicators are intended to provide a common framework for assessing and reviewing progress toward sustainable forest management. The potential of a combined geometrical-optical/spectral mixture analysis model was assessed for mapping the Montreal Process age class and successional age indicators at a regional scale using Landsat Thematic data. The project location is an area of eucalyptus forest in Emu Creek State Forest, Southeast Queensland, Australia. A quantitative model relating the spectral reflectance of a forest to the illumination geometry, slope, and aspect of the terrain surface and the size, shape, and density, and canopy size. Inversion of this model necessitated the use of spectral mixture analysis to recover subpixel information on the fractional extent of ground scene elements (such as sunlit canopy, shaded canopy, sunlit background, and shaded background). Results obtained fron a sensitivity analysis allowed improved allocation of resources to maximize the predictive accuracy of the model. It was found that modeled estimates of crown cover projection, canopy size, and tree densities had significant agreement with field and air photo-interpreted estimates. However, the accuracy of the successional stage classification was limited. The results obtained highlight the potential for future integration of high and moderate spatial resolution-imaging sensors for monitoring forest structure and condition

Land condition monitoring information for reef catchments: a new era

Land condition monitoring information is required for the strategic management of grazing land and for a better understanding of ecosystem processes. Yet, for policy makers and those land managers whose properties are situated within north-eastern Australias vast Great Barrier Reef catchments, there has been a general lack of geospatial land condition monitoring information. This paper provides an overview of integrated land monitoring activity in rangeland areas of two major Reef catchments in Queensland: the Burdekin and Fitzroy regions. The project aims were to assemble land condition monitoring datasets that would assist grazing land management and support decision-makers investing public funds; and deliver these data to natural resource management(NRM) community groups, which had been given increased responsibility for delivering local environmental outcomes. We describe the rationale and processes used to produce new land condition monitoring datasets derived from remotely sensed Landsat thematic mapper (TM) and high resolution SPOT 5 satellite imagery and from rapid land condition ground assessment. Specific products include subcatchment groundcover change maps, regional mapping of indicative very poor land condition, and stratified land condition site summaries. Their application, integration, and limitations are discussed. The major innovation is a better understanding of NRM issues with respect to land condition across vast regional areas, and the effective transfer of decision-making capacity to the local level. Likewise, with an increased ability to address policy questions from an evidence-based position, combined with increased cooperation between community, industry and all levels of government, a new era has emerged for decision-makers in rangeland management.

Geometric correction and accuracy assessment of Landsat-7 ETM+ and Landsat-5 TM imagery used for vegetation cover monitoring in Queensland, Australia from 1988 to 2007

A range of programs exist globally that use satellite imagery to derive estimates of vegetation-cover for developing vegetation-management policy, monitoring policy compliance and making natural-resource assessments. Consequently, the satellite imagery must have a high degree of geometric accuracy. It is common for the accuracy assessment to be performed using the root mean square error (RMSE) only. However the RMSE is a non-spatial measure and more rigorous accuracy assessment methods are required. Currently there is a lack of spatially explicit accuracy assessment methods reported in the literature that have been demonstrated to work within operational monitoring programs. This paper reports on the method used by the Statewide Landcover and Trees Study (SLATS) to georegister and assess the registration accuracy of Landsat-5 Thematic Mapper (TM) and Landsat-7 Enhanced Thematic Mapper (ETM+) imagery in Queensland, Australia. A geometric baseline with high accuracy (a statewide mean RMSE of 4.53 m) was derived by registering Landsat-7 ETM+ panchromatic imagery acquired in 2002 to a database of over 1600 control points, collected on the ground using a differential global positioning system. Landsat-5 TM and Landsat-7 ETM+ imagery for 12 selected years from 1988 to 2007 was registered to the baseline in an automated procedure that used linear geometric correction models. The reliability of the geometric correction for each image was determined using the RMSE, calculated using independent check points, as an indicator of model fit; by analysing the spatial trends in the model residuals; and through visual assessment of the corrected imagery. The mean RMSE of the statewide coverage of images for all years was less than 12.5 m (0.5 pixels). Less than 1 percent of images had non-linear spatial trends in the model residuals and some image misregistration after applying a linear correction-model; in those cases a quadratic model was deemed necessary for correction. Further research in the development of automated spatially explicit accuracy assessment methods is required.

THE ROLE OF INTEGRATED INFORMATION ACQUISITION AND MANAGEMENT IN THE ANALYSIS OF COASTAL ECOSYSTEM CHANGE

This book chapter represents a synthesis of the work which started in my PhD and which has been the conceptual basis for all of my research since 1993. The chapter presents a method for scientists and managers to use for selecting the type of remotely sensed data to use to meet their information needs associated with a mapping, monitoring or modelling application. The work draws on results from several of my ARC projects, CRC Rainforest and Coastal projects and theses of P.Scarth , K.Joyce and C.Roelfsema.

A method for mapping Australian woody vegetation cover by linking continental-scale field data and long-term Landsat time series

ABSTRACT There is a significant need to provide nationwide consistent information for land managers and scientists to assist with property planning, vegetation monitoring applications, risk assessment, and conservation activities at an appropriate spatial scale. We created maps of woody vegetation cover of Australia using a consistent method applied across the continent, and made them accessible. We classified pixels as woody or not woody, quantified their foliage projective cover, and classed them as forest or other wooded lands based on their cover density. The maps provide, for the first time, cover density estimates of Australian forests and other wooded lands with the spatial detail required for local-scale studies. The maps were created by linking field data, collected by a network of collaborators across the continent, to a time series of Landsat-5 TM and Landsat-7 ETM+ images for the period 2000–2010. The fractions of green vegetation cover, non-green vegetation cover, and bare ground were calculated for each pixel using a previously developed spectral unmixing approach. Time series statistics, for the green vegetation cover, were used to classify each pixel as either woody or not using a random forest classifier. An estimate of woody foliage projective cover was made by calibration with field measurements, and woody pixels classified as forest where the foliage cover was at least 0.1. Validation of the foliage projective cover with field measurements gave a coefficient of determination, R2,of 0.918 and root mean square error of 0.070. The user’s and producer’s accuracies for areas mapped as forest were high at 92.2% and 95.9%, respectively. The user’s and producers’s accuracies were lower for other wooded lands at 75.7% and 61.3%, respectively. Further research into methods to better separate areas with sparse woody vegetation from those without woody vegetation is needed. The maps provide information that will assist in gaining a better understanding of our natural environment. Applications range from the continental-scale activity of estimating national carbon stocks, to the local scale activities of assessing habitat suitability and property planning.

Remotely-sensed analysis of ground-cover change in Queensland’s rangelands, 1988–2005

A dynamic reference-cover method and remotely-sensed ground cover were used to determine the change in the state of ~640 000 km2 of rangelands in Queensland at a sub-bioregional scale between 1988 and 2005. The method is based on persistence of ground cover in years of lower rainfall and objectively separates grazing effects on ground cover from those due to inter-annual variation in rainfall. The method is applied only to areas where trees and shrubs were not cleared. An indicator of rangeland state was derived, at Landsat-TM pixel resolution, by subtracting automatically-calculated reference ground cover from actual ground cover and then spatially averaging these deviations across the area of each sub-bioregion. Landscape heterogeneity may affect reference cover but, because it is stable over time, change in mean cover deficit between sequences of dry years reliably indicates change due to grazing. All 34 sub-regions analysed had similar or increased levels of seasonally-adjusted ground cover at the end of the analysis period, which was either 2003 or 2005. Allowing for possible landscape heterogeneity effects on assessed condition, the Einasleigh Uplands bioregion was comparatively in a better state and those analysed parts of the Mulga Lands bioregion in poorer state at the first assessment in 1988. Most sub-regions of the Cape York Peninsula, Brigalow Belt North, Desert Uplands, Gulf Plains and Mitchell Grass Downs bioregions lay between these two end-states. Simulated levels of pasture utilisation based on modelled pasture growth and statistically-based grazing pressure supported the results of this regional assessment of land condition. The dynamic reference-cover method will allow the Queensland Government to monitor future grazing effects on rangeland ground cover between sequences of drier years – quantitatively and efficiently across the entire state. The method can potentially be adapted to other rangeland jurisdictions where a suitable multi-temporal database of remotely sensed ground cover exists. The results from further analyses of remotely sensed ground cover will be reported through the Australian Collaborative Rangelands Information System.

Calibration of multiple Landsat sensors based on pseudo-invariant target sites in Western Queensland, Australia

The Statewide Landcover and Trees Study (SLATS) has used both Landsat-5 TM and -7 ETM+ imagery to monitor short-term woody vegetation changes throughout Queensland. In order to analyse long-term vegetation change, time-based trends that are an artifact of the sensor system must be removed. Although the calibration trends of TIM and ETM+ are well described, information on the calibration of the older Landsat-2 and -5 MSS sensors is relatively scarce. This paper describes the use of three pseudo-invariant target sites in western Queensland to achieve an operational calibration of TM and ETM+ data. Following the success of the current study, these targets will be used in the near future to achieve sensor specific calibrations for the Landsat-2 and -5 MSS SLATS data

Spectral Mixture Analysis for Ground-Cover Mapping

Monitoring of ground-cover is an important task for land management since it has been linked to indicators of soil loss, biodiversity, and pasture production. Ground-cover is an indicator adopted by Queensland natural resource and catchment management groups. However, accurate spatial estimation of ground-cover is confounded by varying cover types, cover greenness and soil colour.