Monitoring fire-driven forest dynamics over large areas using passive and active remote sensing

Forests are an integral part of the Earth’s biosphere, hosting a large proportion of global biodiversity and serving as a major stabilising factor in climate and weather regulation. To ensure their protection, nature reserves and national parks have been established in many parts of the world. Nevertheless, forests are increasingly under stress from the consequences of climate change and other human-driven pressures. In Australia, as in other regions of the world, wildfires are becoming more frequent and severe. Despite the evolutionary adaptation of Australian native vegetation to wildfires, concerns have been raised regarding the resilience of local forests to the changing fire regimes, and their ability to sustainably recover from fire. Satellite-based remote sensing has been shown to be a reliable and practicable solution for mapping and monitoring forests over large areas. Passive sensors such as those on the Landsat satellites collect multispectral information with a moderate spatial and temporal resolution, providing global wall-to-wall coverage that reaches back several decades. However, Landsat has well-documented limits in monitoring vertical forest structure. As an active remote sensing technology, lidar can help to fill this gap. Its signal can penetrate the canopy layer and gather information on the vertical distribution of plant material. Combining the advantages of both sensor types has been shown to enable more accurate mapping of forest structure across large areas. In particular, the advent of spaceborne lidar sensors such as the Global Ecosystem Dynamics Investigation (GEDI) has boosted the use of sensor fusion approaches that leverage passive and active remote sensing data for forest structure monitoring. This thesis evaluates the utility of Landsat time series (LTS) and GEDI for mapping forest spectral and structural characteristics following fire disturbance events. The first research question examined whether conservation efforts can be associated with forest resilience to fire. Specifically, it explored if forest tenure and protection status are related to post-fire recovery of forests. LTS satellite data was used to evaluate spectral recovery duration following fire disturbance for 25.4 Mha of forested land in southeast Australia. Results show that protected forests on public tenure spectrally recovered on average 0.4 years faster than those on privately held land. However, other factors such as climatic and topographic variables were found to have a far greater impact on forest recovery. As spectral recovery does not necessarily equate to structural recovery, subsequent research questions focus on how GEDI data can be used in conjunction with LTS. While GEDI is superior to Landsat in measuring attributes of forest structure, it is constrained by a relatively short scheduled mission duration (2019 to 2023) and its design as a monotemporal sampling tool. This thesis proposes a novel approach to overcome these constraints, based on the use of bi-temporal GEDI data. The method leverages footprint-level lidar observations of forest structure from two points in time, using airborne lidar datasets from the pre-GEDI era to simulate a first set of GEDI observations, complemented by a second set of observations provided by the spaceborne GEDI instrument for the same footprint locations, respectively. Throughout this thesis, this concept is referred to as ‘bi-temporal GEDI’ data. To validate this approach, the second research question compared real and simulated GEDI observations in undisturbed Australian sclerophyll forests. The results confirm that real and simulated GEDI observations, despite certain differences, are generally compatible for combined use in bi-temporal monitoring approaches. Additionally, recommendations were made for the settings of the GEDI simulator and specifications of ALS data used in the simulation. Furthermore, the research determined the main sources of error, namely steep slopes, dense canopy cover and dead standing trees. Highly irregular horizontal forest structure was found to pose a challenge, an issue related to GEDI’s geolocation uncertainty. To facilitate the novel integration of bi-temporal GEDI observations into established LTS-based forest monitoring approaches, the third research question explored relationships between the two data sources, focussing on the post-fire vegetation response across a range of structural and spectral metrics. One year after the examined fire event, spectral indices showed a moderate to strong decline to between 46.1 and 77 % of pre-fire levels, while most structural metrics demonstrated an even more substantial decline. An exception was canopy height which only dropped to 82.7 % of pre-fire levels. Increased fire severity led to a more pronounced post-fire decline across several spectral and structural metrics. Similarly, greater forest height was found to be associated with a larger post-fire decline across some spectral and structural metrics. Furthermore, the findings suggest the preferential use of GEDI full power beam observations to increase the utility of derived structural metrics. Finally, in the fourth research question, the novel concept of integrating bi-temporal GEDI into LTS-driven forest change monitoring was implemented. This study evaluated whether employing bi-temporal GEDI data resulted in improved model performance compared to using monotemporal GEDI. The findings show that leveraging bi-temporal GEDI data reduces the RMSE of canopy height change predictions by 0.94 m in undisturbed forest, and 0.49 m in recently disturbed forest. Similar trends were observed for other structural change metrics. Furthermore, the results demonstrate that in forest types with less dense canopy cover (<80 percent), canopy height change predictions from bi-temporal GEDI-driven models yielded a 1.51 m reduction in RMSE. Overall, the results provide evidence that Landsat-based models that were trained with bi-temporal GEDI data were superior in modelling vertical forest structure change, compared to more traditional approaches that were restricted to the use of monotemporal GEDI. The approach of fusing bi-temporal GEDI observations with multispectral LTS data as presented in this thesis generated promising results, outperforming other more established methods. It can be applied to enhance the monitoring of forest structure in temperate, subtropical, and tropical forests (i.e., areas covered by GEDI’s orbit between 52° N and S) where patches of historical ALS data are available. Leveraging bi-temporal GEDI data might be beneficial for biodiversity mapping, fire behaviour modelling or carbon budget monitoring, as well as studying the impacts of anthropogenic factors on forests.