Integration of earth observation data for improved forest resource management

The world’s forests cover a total of 32% of the Earth’s surface. They are a vital habitat for flora and fauna, play a key role in mitigating the effects of climate change, and provide fundamental economic opportunities in rural areas. Therefore, the sustainable management of forests to meet the requirements in order to balance all user’s needs is of upmost importance. In this thesis, nutrient deficient coniferous forests have been mapped in the Republic of Ireland using machine learning algorithms and a time series early warning system has been created. The time series can be used to identify the age at which the spectral response from a nutrient deficient forest is separate from a healthy forest.

This research has identified that most of the literature for estimat?ing carbon emissions from forest fires does not incorporate the burn severity or apply a different Burning Efficiency (BE) per severity when calculating the carbon released. Additionally, most of the literature of BE are from site preparation burns, prescribed burns, or slash and burns and may not be useful proxies for wildfire burns due to the differing characteristics. Furthermore, the IPCC recommended BE val?ues are lacking as there are none for wildfires and the recommended ”other” temperate forest fire BE values are based on studies from the tropics. This work has highlighted these issues while also illustrating a refined methodology, based on the recommended methodology by the IPCC, to include burn severity and best estimates of BE for future research.

Finally, utilising field data, LiDAR data, and machine learning algorithms, estimates of a range of forest parameters for the Slieve Blooms in the Republic of Ireland have been produced. These estimates are rarely accompanied by the uncertainty associated with the estimates or what parameters affect the uncertainty. An investigation of the ef?fects of the value of k and the size of an area when estimating the vari?ance of forest parameter estimates for multiple pixel AOIs has been carried out. The results are as the value of k increases, the variance decreases and as the size of an AOI increases, the variance decreases. These results are similar to those previously reported which illustrates the applicability of this particular variance estimation technique in another study area. It also illustrates that this variance estimation technique is not independent of areal size.