Leo Lymburner

An Evaluation of the Use of Atmospheric and BRDF Correction to Standardize Landsat Data

Normalizing for atmospheric and land surface bidirectional reflectance distribution function (BRDF) effects is essential in satellite data processing. It is important both for a single scene when the combination of land covers, sun, and view angles create anisotropy and for multiple scenes in which the sun angle changes. As a consequence, it is important for inter-sensor calibration and comparison. Procedures based on physics-based models have been applied successfully with the Moderate Resolution Imaging Spectroradiometer (MODIS) data. For Landsat and other higher resolution data, similar options exist. However, the estimation of BRDF models using internal fitting is not available due to the smaller variation of view and solar angles and infrequent revisits. In this paper, we explore the potential for developing operational procedures to correct Landsat data using coupled physics-based atmospheric and BRDF models. The process was realized using BRDF shape functions derived from MODIS with the MODTRAN 4 radiative transfer model. The atmospheric and BRDF correction algorithm was tested for reflectance factor estimation using Landsat data for two sites with different land covers in Australia. The Landsat reflectance values had a good agreement with ground based spectroradiometer measurements. In addition, overlapping images from adjacent paths in Queensland, Australia, were also used to validate the BRDF correction. The results clearly show that the algorithm can remove most of the BRDF effect without empirical adjustment. The comparison between normalized Landsat and MODIS reflectance factor also shows a good relationship, indicating that cross calibration between the two sensors is achievable.

Rapid, high-resolution detection of environmental change over continental scales from satellite data – the Earth Observation Data Cube

ABSTRACT The effort and cost required to convert satellite Earth Observation (EO) data into meaningful geophysical variables has prevented the systematic analysis of all available observations. To overcome these problems, we utilise an integrated High Performance Computing and Data environment to rapidly process, restructure and analyse the Australian Landsat data archive. In this approach, the EO data are assigned to a common grid framework that spans the full geospatial and temporal extent of the observations – the EO Data Cube. This approach is pixel-based and incorporates geometric and spectral calibration and quality assurance of each Earth surface reflectance measurement. We demonstrate the utility of the approach with rapid time-series mapping of surface water across the entire Australian continent using 27 years of continuous, 25 m resolution observations. Our preliminary analysis of the Landsat archive shows how the EO Data Cube can effectively liberate high-resolution EO data from their complex sensor-specific data structures and revolutionise our ability to measure environmental change.

Mapping of mangrove extent and zonation using high and low tide composites of Landsat data

Monitoring mangrove health and distribution requires reliable methods that can be undertaken rapidly and at a resolution that optimises costs and accuracy. The Landsat record has been used for this purpose, but its application has been limited by the capacity to provide accurate results that distinguish mangrove from adjoining communities. The Australian Geoscience Data Cube provides a framework for exploring the Landsat record from 1987 onwards, and as pre-processing has already been undertaken there are efficiencies gained using this resource. Using the Data Cube, we exploited the differential spectral signature of mangrove under high tide and low tide conditions at Darwin Harbour, Australia, a relatively stable mangrove ecosystem, using image composites that combined imagery corresponding to the highest 10% and lowest 10% of tides. By applying the automated RandomForest classification technique, we demonstrate the capacity to accurately determine the extent of mangrove zones. Classification identified five mangrove zones: (1) seaward margin dominated by Sonneratia alba, (2) Rhizophora zone dominated by Rhizophora stylosa, (3) tidal flat dominated by Ceriops tagal, (4) landward salt flat and (5) marginal hinterland. Image composites that included high and low tide images achieved the best outcomes with kappa co-efficient of 0.81 and overall accuracy of 82%.

Applying machine learning methods and time series analysis to create a National Dynamic Land Cover Dataset for Australia

The National Dynamic Land Cover Dataset (DLCD) classifies Australian land cover into 34 categories, which conform to 2007 International Standards Organisation (ISO) Land Cover Standard (19144-2). The DLCD has been developed by Geoscience Australia and the Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES), aiming to provide nationally consistent land cover information to federal and state governments and general public. This paper describes the modeling procedure to generate the DLCD, including machine learning methodologies and time series analysis techniques involved in the process.

Digital earth Australia – unlocking new value from earth observation data

Abstract Petascale archives of Earth observations from space (EOS) have the potential to characterise water resources at continental scales. For this data to be useful, it needs to be organised, converted from individual scenes as acquired by multiple sensors, converted into “analysis ready data”, and made available through high performance computing platforms. Moreover, converting this data into insights requires integration of non-EOS data-sets that can provide biophysical and climatic context for EOS. Digital Earth Australia has demonstrated its ability to link EOS to rainfall and stream gauge data to provide insight into surface water dynamics during the hydrological extremes of flood and drought. This information is supporting the characterisation of groundwater resources across Australia’s north and could potentially be used to gain an understanding of the vulnerability of transport infrastructure to floods in remote, sparsely gauged regions of northern and central Australia.

Dynamic Land Cover Dataset version 2: 2001-now…a land cover odyssey

Understanding how land cover responds to natural and anthropogenic drivers is critical as increasing population, climate fluctuations and competing land uses place increased pressure on both natural and food/fibre production systems. The Dynamic Land Cover Dataset (DLCD) Version 2 is a series of biennial land cover maps that uses the ISO 19144-2 land cover classification scheme. The Moderate Resolution Image Spectroradiometer (MODIS) Enhanced Vegetation Index (EVI) time series [1] are used to characterize greenness dynamics observed at 250-metre scale. These greenness dynamics are used to generate a series of 9 land cover maps. Version 2 of the DLCD provides a series of land cover maps updated on an annual basis to enable resource managers, decision makers and biophysical modelers to track the change in land cover on a systematic basis.

Creating multi-sensor time series using data from Landsat-5 TM and Landsat-7 ETM+ to characterise vegetation dynamics

The Landsat series of satellites provide the longest contiguous earth observation record of the Earths surface. This provides the unique capacity to track changes in vegetation over multiple decades. This paper illustrates how standardized Landsat data can be combined to create a time series of sensor independent observations. The impact of side-lap and cloud frequency on observation frequency are also examined with reference to two adjacent path/rows of data in southern Australia. The generation of Landsat scale time series provides the opportunity to track both subtle and dramatic changes in vegetation cover in much higher levels of detail than previously possible. However the approach presents new challenges associated with developing time series analysis techniques to characterize time series that have uncertain observation frequencies.

Generating Continental Scale Pixel-Based Surface Reflectance Composites in Coastal Regions with the Use of a Multi-Resolution Tidal Model

Generating continental-scale pixel composites in dynamic coastal and estuarine environments presents a unique challenge, as the application of a temporal or seasonal approach to composite generation is confounded by tidal influences. We demonstrate how this can be resolved using an approach to compositing that provides robust composites of multi-type environments. In addition to the visual aesthetics of the images created, we demonstrate the utility of these composites for further interpretation and analysis. This is enabled by the manner in which our approach captures the spatial variation in tidal dynamics through the use of a Voronoi mesh, and preserves the band relationships within the modelled spectra at each pixel. Case studies are presented which include continental-scale mosaics of the Australian coastline at high and low tide, and tailored examples demonstrating the potential of the tidally constrained composites to address a range of coastal change detection and monitoring applications. We conclude with a discussion on the potential applications of the composite products and method in the coastal and marine environment, as well as further development directions for our tidal modelling framework.