Tropical forests are the largest of the global forest biomes and play a crucial role in the global carbon, hydrological and biochemical cycles. Increasing demand for resources rapidly increases the pressure on tropical forests. As a result tropical regions have been undergoing rapid changes in forest cover in recent decades. These changes are the second largest contributor of greenhouse gas emissions in the atmosphere. Spatially and timely consistent detection of tropical deforestation and forest degradation is fundamental to reliably estimate greenhouse gas emissions, and to successfully implement climate mechanisms like reducing emissions from deforestation and forest degradation (REDD+).
To assess historical and future changes in forest cover, satellite remote sensing at medium resolution scale constitutes a powerful tool. Reviewing satellite-based optical and Synthetic Aperture Radar (SAR) efforts for tropical forest monitoring revealed that operationalised optical-based approaches exist, but frequent cloud cover limits their applicability in the tropics. SAR remote sensing has also demonstrated its capability, but the observation frequency of SAR imagery and appropriate time series methods are limited. Research has indicated there is potential for multi-sensor approaches to overcome the limitations of the single-sensors, but so far developments are restricted to mapping approaches. This thesis addressed the need for advancing multi-sensor methods that combine time series imagery from medium resolution SAR and optical satellites to improve tropical forest monitoring. The main scientific contributions include the introduction of three novel SAR-optical approaches, two of them capable of exploiting the full observation density of time series. Furthermore, an approach for multi-model land cover dependent SAR slope correction was proposed.
Chapter 2 introduced an approach for feature level fusing of multi-temporal L-band SAR and optical forest disturbance information. Using Landsat and ALOS PALSAR imagery from 2007 and 2010, we applied the approach to map forest land cover and to detect deforestation and forest degradation of a persistently cloud covered mining region in Central Guyana. By making use of the complementarities of Landsat and ALOS PALSAR, we demonstrated the reduction of Landsat (cloud cover, Landsat 7 scan line corrector error) and PALSAR data gaps (SAR layover and shadow in mountainous area) to a negligible amount.
Chapter 3 described a practical approach for multi-model land cover dependent slope correction of SAR images that can handle a wide range of terrain and topographic conditions. We corrected ALOS PALSAR images of two topographically complex sites in Fiji (study site of Chapter 4 and 5) and Brazil and showed that the remaining slope effects for the multi-model case are marginal for all land cover types. Particularly, this improves the detection of forest degradation and biomass changes. Considering the large change in the L-band backscatter signal caused by the removal of forest, however, remaining slope effects are already sufficiently small after applying a single-model approach already.
Chapter 4 presented a novel multi-sensor time series correlation and fusion (MulTiFuse) approach that was applied to fuse Landsat NDVI and ALOS PALSAR time series. The fused Landsat-PALSAR time series was used in a change detection framework to detect deforestation at a managed forest site in Fiji for the period 01/2008 - 09/2010. We tested the impact of persistent cloud cover in the tropics by increasing the per-pixel missing data percentage of the Landsat time series in a stepwise manner. The results were evaluated against three-monthly reference data that covered the entire study area. For the Landsat-only case, a very strong decrease in spatial and temporal accuracies were observed for increasing Landsat missing data. This highlights the vulnerability of tropical forest monitoring systems that rely only on optical data. In contrast, the results for the fused Landsat-PALSAR case remained high with increasing missing data and were observed to be always above the accuracies for the Landsat- and PALSAR-only cases.
To address the need for near real-time monitoring systems at medium resolution scale, Chapter 5 introduced a Bayesian change detection approach to combine SAR and optical time series for near real-time deforestation detection. We applied the approach in a simulated near real-time scenario using Landsat NDVI and ALSO PALSAR time series already used in Chapter 4. Once a new image of either of the two time series was available, the probability of deforestation was calculated immediately and deforestation events were indicated. These near real-time capabilities are essential to support timely action against illegal forest activities. Spatial and temporal accuracies for the fused Landsat-PALSAR case were consistently higher than those of the Landsat- and PALSAR-only cases, even for increasing Landsat missing data.
With these studies we demonstrated the potential of SAR-optical time series approaches to use the historical and upcoming streams of medium resolution optical and SAR satellite image time series for improving forest monitoring in the tropics.
|Qualification||Doctor of Philosophy|
|Award date||19 Jun 2015|
|Place of Publication||Wageningen|
|Publication status||Published - 2015|
- satellite imagery
- satellite surveys
- tropical forests
- forest monitoring
- time series