Monitoring Forest Dynamics using Time Series of Satellite Image Data in Queensland, Australia

Santosh Bhandari (2011). Monitoring Forest Dynamics using Time Series of Satellite Image Data in Queensland, Australia PhD Thesis, School of Geography, Planning & Env Management, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
s4125261_PhD_final_Thesis.pdf Final PhD Thesis application/pdf 5.10MB 16
Author Santosh Bhandari
Thesis Title Monitoring Forest Dynamics using Time Series of Satellite Image Data in Queensland, Australia
School, Centre or Institute School of Geography, Planning & Env Management
Institution The University of Queensland
Publication date 2011-03
Thesis type PhD Thesis
Supervisor Professor Stuart Phinn
A/Prof. Clive McAlpine
Dr. Tony Gill
Total pages 179
Total colour pages 19
Total black and white pages 160
Subjects 05 Environmental Sciences
Abstract/Summary Incremental, cyclic and periodic changes in vegetation structure and condition are complex and continuous phenomena affected by multiple factors. Knowledge of the extent and type of such change at specific spatial and temporal scales is critical for resource management, policy making and ecosystem research. Remote sensing change detection methods are one of the only viable and spatially explicit options for monitoring the changes over large spatial extents repetitively. Bi-temporal change detection methods, however, do not account for shorter periodicity variation in vegetation structure and condition, such as phenological changes, inter-annual climatic variability and other changes of a cyclic nature. Therefore, the change information produced by those methods may not truly characterise the trend in vegetation structure and condition occurring within a specified period. A dense time series of satellite images, with images collected at regular, short period intervals will be capable of accounting for seasonal variation, and time series analysis methods could provide information on trends in vegetation properties along with the phenological properties of the vegetation communities. Although significant research in land cover change detection has been carried out, the majority of these works have focused on bi-temporal change detection methods. Only a limited number of works have addressed the issue of phenology and long term trend analysis from satellite images. Most of those time series analysis works are, however, confined to coarse spatial resolution remote sensing data such as Advanced Very High Resolution Radiometer (AVHRR) (≥1 km2 pixels over > 106 km2) to study the phenomena at regional to hemispheric and global scales. There is a paucity of research into evaluating trends of vegetation properties using medium spatial resolution data such as Landsat TM/ETM+ and MODIS, which produce information useful for resource management at local to regional scales (10 – 100 km’s). This work developed a method for using time series of Landsat Thematic Mapper (TM)/ Enhanced Thematic Mapper Plus (ETM+)and Moderate Resolution Imaging Spectroradiometer (MODIS), data sets to characterise the phenological properties and long term trends of structural properties in sub-tropical woodlands and forests of Queensland, Australia. The research had three main components: (1) developing an understanding of image time series analysis procedure; (2) preparing and evaluating a ready-to-use time series product; and (3) characterising vegetation phenology and long term trends in forest structure variables in forest communities of the Barakula State Forest area in Queensland, Australia. In the first stage, a framework of time series image analysis process was developed through a literature review. Essential processing steps required for image time series analysis process were identified and available tools and techniques were reviewed. Identification and derivation of suitable image derived variables to represent the phenomenon under study was one of the important steps. Foliage projective cover (FPC) was identified as a suitable variable to represent vegetation structure and condition for this study. Geometric and radiometric correction of images and gap filling, analysis methods and validation of the results were other important steps identified. Based on the framework, a ready to use Landsat image time series (LITS), a sequence of Landsat TM images with observations on every 16 days was developed for the period of five years commencing July 2003. A common point comparison (CPC) method was used for geometric correction and 6S radiative transfer code was used for atmospheric correction assuming a fixed AOD of 0.05. The Spatial Temporal Adaptive Reflectance Fusion Model (STARFM) algorithm was used to create the synthetic Landsat TM scenes to fill the gaps due to unavailable images and cloud-cover over the study area using available Landsat TM scenes and MODIS Nadir Bi-directional reflectance Distribution Function (BRDF) Adjusted Reflectance (NBAR) imagery. The ability of LITS to measure attributes of vegetation phenology was examined by: (1) comparing the reflectance of predicted images with reference images; and (2) comparing LITS generated normalised difference vegetation index (NDVI) and MODIS NDVI (MOD13Q1) time series. The pixel based reflectance comparison showed a good agreement between predicted and reference images, with a R2 >0.8 for all bands. Comparison between vegetation phenology parameters estimated from LITS generated NDVI and MODIS NDVI showed no significant difference in their trends, and less than 16 days (composite period of MODIS data used) difference in key seasonal parameters, including start- and end-of-season dates in most of the cases. The results showed that the LITS could be used to monitor vegetation phenology and trends using time series analysis techniques. The impact that viewing and illumination geometry differences had on MOD13Q1vegetation index values, and their subsequent ability to map changes arising from phenology and disturbances in a number of forest communities in Queensland, was examined next. MOD13Q1 Normalised Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) were compared to normalised NDVI and EVI (NDVInormalised and EVInormalised), which were derived from the reflectance modelled from BRDF/Albedo parameter product (MCD43A1) using fixed- viewing and -illumination geometries. Time series plots of the vegetation index values from a number of pixels representing different forest types and known disturbances showed that the NDVInormalised time series was more effective at capturing the changes in vegetation than the NDVI. MOD13Q1 NDVI showed higher seasonal amplitude, but was less accurate at capturing phenology and disturbances compared to the NDVInormalised. The EVI was less affected by variable viewing and illumination geometry in terms of amplitude, but in terms of phase shift was more affected than NDVI. The combined effect of sun zenith angle seasonality and the timing of green up and senescence appeared to cause the shift of both normalised VIs to earlier dates in the time series, compared to NDVI and EVI. The study showed that there were significant impacts of viewing and illumination geometry variations, which a user is required to recognise and take into account before using the products for phonological studies. Finally, the LITS was used to examine phenological properties and long term trends in forest structural properties, mainly using foliage projective cover (FPC) in a number of forest communities in the study area. FPC time series were derived from LITS using a multiple regression method. Image derived woody FPC maps produced by the Queensland Department of Environment and Resource Management (DERM) were used to train the regression model. A number of phenological metrics were derived from the FPC time series and used to characterise forest communities. Negative-trend, positive-trend and no-change areas were identified from the FPC time series using the Mann-Kendall trend test (p=0.05). The results showed that the FPC generated from LITS had better agreement with total FPC though the reference data used to train regression were image derived woody FPC. Phenological metrics, particularly those showing amplitude and base information, were able to be used to characterise the dynamics of forest communities in the study area. Accuracy assessment of the trend map using the reference data generated by visual image interpretation showed an overall accuracy of 84 %. Prolonged drought during the study period and fires were identified as potential causes for the very high proportion (40%) of negative trend area compared to 4% area of positive trend in forest areas. The time series analysis approach used and the results of this study have some important contributions to the use of Landsat TM/ETM+ and MODIS time series to monitor structural properties and condition of sub-tropical forest and woodland communities in eastern Australia, and others similar environments around the world. The approach used to derive FPC time series from LITS enables forest managers and researchers to examine phenological and trend changes at spatial scales directly relevant to local to regional land management scales. The FPC time series could also be used to separate the woody and non-woody components of FPC, i.e. to canopy and understorey vegetation, to improve the accuracy of FPC estimates. The approach may also be applicable in other environments around the world if modified appropriately. The findings about the impact of viewing and illumination geometries on MODIS Vegetation Index (VI) series also have a broader significance, particularly to the study of phenology using satellite image time series. Future research should focus on examining the impacts in different environments and testing whether the regular or normalised VIs profile more accurately represents phenology of vegetation, as measured on the ground.
Keyword Landsat TM/ETM+, MODIS, time series, foliage projective cover, phenology, trend, NDVI, EVI, Australia, vegetation communities
Additional Notes Colour Pages 42, 44, 76, 80, 84, 85, 86, 88, 89, 104, 106, 110, 117, 131, 132, 145, 147, 151, 153 Landscape Pages 44, 80, 88, 89, 100, 104, 142

Citation counts: Google Scholar Search Google Scholar
Created: Wed, 19 Oct 2011, 19:03:25 EST by Mr Santosh Bhandari on behalf of Library - Information Access Service