Open Access Repository

Development of remote sensing products to investigate the impact of tropical cyclones on natural vegetation communities in the wet-dry tropics of northern Australia

Staben, GW 2021 , 'Development of remote sensing products to investigate the impact of tropical cyclones on natural vegetation communities in the wet-dry tropics of northern Australia', PhD thesis, University of Tasmania.

Full text not available from this repository.

Abstract

Understanding ecological changes in native vegetation communities often requires information over long time periods. The influence of fire on woody vegetation structure in the Northern Territory has been well studied, however limited work has been undertaken to understand the impact of tropical cyclones, which can dramatically alter vegetation structure. Woody vegetation structure has been identified as an important metric for monitoring trends in biomass, primary productivity and biodiversity. Both cover and height have been identified as important structural attributes required for ecological studies investigating long term changes and trends. Satellite remote sensing products have the potential to provide information on cover and height at suitable spatial and temporal scales. Obtaining estimates of cover and height from satellite sensors requires the development of predictive models, often developed by relating field measured data to satellite imagery. The aim of this thesis is to develop remote sensing products measuring cover and height enabling the assessment of the impact of tropical cyclones on natural vegetation communities in the wet-dry tropics of northern Australia. To achieve the aim of the study, the following objectives were identified to develop mapping products that quantified and characterised woody vegetation structure based on cover and height.
The first objective was to develop and assess the suitability of using digital aerial photography to obtain woody vegetation biophysical parameters. This was done to enable validation of a Landsat satellite-based model predicting woody foliage projective cover, developed using field data collected over the state of Queensland. In this study, quantitative measurements obtained from digital aerial photography were compared to woody biophysical parameters measured from 1 ha field plots. There was a strong relationship (R2 _ 0.85) between all field measured woody canopy parameters and aerial derived green woody cover measurements, however only foliage projective cover (FPC) was found to be statistically significant. The results of this study show that accurate woody biophysical parameters can be obtained from aerial photography from a range of woody vegetation communities across the Northern Territory.
The second objective was to develop a predictive model to estimate vertical tree canopy structure. A machine learning algorithm, random forest regression, was used to predict canopy height from a single date Landsat-5 TM scene, using training data derived from a 1 m canopy height model produced from LiDAR. The overall accuracy of the model was expressed by an R2 of 0.53 and RMSE of 2.8 m. The model was also applied to Landsat-7 Enhanced Thematic Mapper Plus (ETM+) resulting in an R2 of 0.49, RMSE of 2.8 m. The model was applied to Landsat imagery over the years 1988 to 2016. This study demonstrated that canopy height can be predicted from Landsat imagery. The robustness of the model across a range of vegetation communities and three different Landsat sensors illustrated that the approach could be successfully used to explore changes in woody vegetation height through time.
The third objective was to develop a predictive model to estimate a range of structural metrics characterising tree canopy structure from Sentinel-2 MSI and Landsat-8 OLI satellite sensors. Models were developed at 10 m, 20 m and 30 m spatial resolution for Sentinel-2 models, which enabled comparisons with the Landsat-8 results. To address limitations identified in the second objective, models were developed from seasonal composites (annual and dry season) for the respective satellite sensors, using training datasets captured across the Northern Territory. Of the seven models H99 (representing maximum canopy height) had the strongest relationship for both Sentinel-2 and Landsat-8 with R2 values ranging from 0.7 to 0.81, and RMSE% between 22.9 and 33.8. Model accuracy was found to improve as spatial resolution decreased, with models produced at 30 m recording the highest overall accuracy. This study developed robust models predicting important forest structural metrics from Sentinel-2 and Landsat-8 satellite sensors, providing new insights into vertical tree canopy structure across an area covering 355,500 km2 in the Northern Territory.
The fourth objective was to combine annual estimates of canopy cover and height (H99) from Landsat satellite imagery to produce a structural classification product for a 30 year period (1988-2017). The structural mapping product was then used to investigate the dynamics of woody vegetation in a region (_ 11,500 km2) impacted by severe tropical cyclone Monica in 2006. Landsat estimates of woody foliage projective cover (FPC) were validated and corrected for bias using estimates of FPC obtained from aerial photography, prior to being converted to canopy cover using a generalised model developed from field data. Independent datasets obtained from field data and LiDAR were used to validate the Landsat CC and height estimates. It was estimated that a total area of 3,551 km2 was substantially impacted by cyclone Monica (2006). In 2017 it was estimated that an area of 70 km2 was still severely impacted. The proportion of each structural class was used to gain insight into the dynamics and recovery of woody vegetation post cyclone Monica. The results show that recovery is occurring across the region, however the dynamics observed between the structural classes suggest that the region is still recovering 11 years after the cyclone.
This thesis has developed methodology to quantify woody vegetation structure using digital aerial photography, Landsat and Sentinel-2 satellite sensors. It enables the spatial and temporal assessment of historic (three decades) woody vegetation structure. The mapping products developed in this thesis have the potential to map and investigate woody vegetation (historic and contemporary) change across northern Australia.

Item Type: Thesis - PhD
Authors/Creators:Staben, GW
Keywords: Remote Sensing, Machine learning, Tropical cyclone impacts, Native vegetation, Forest structure.
Copyright Information:

Copyright 2021 the author

Item Statistics: View statistics for this item

Actions (login required)

Item Control Page Item Control Page
TOP