|dc.description.abstract||The climate in the Mediterranean area is characterized by hot, dry summers and cool winters. The area has shown climate change in the past (last centuries) and is vulnerable for climate change in the future as result of the location of the area in the climate system. The climate affects the ecosystem and the response of vegetation through time is closely related to this.
Vegetation in the Mediterranean area is subject to changes depending on environmental factors and human influence. The objective is the application of object-based change detection to determine temporal patterns in vegetation and to explain these using static environmental factors. Static environmental factors are factors which do not change over time, such as geological units, aspect, slope and elevation.
Temporal patterns in vegetation are assessed by the variables biomass, Leaf Area Index and cover fraction. These patterns are assessed by the application of regression analyses and image segmentation on hyperspectral images of 2003 and 2008. For the 2003 image field data of 2005 is used (field data of 2003 was not available) and for the 2008 image field data of 2008 and 2009 is used.
Predictive vegetation maps are generated by the application of ridge regression resulting in the following R² values: biomass 2003: 0.58 and 2008: 0.55, Leaf Area Index 2003: 0.40 and 2008: 0.44 and cover fraction 2003: 0.35 and 2008: 0.27. Biomass is the best predicted variable followed by LAI and cover fraction. Biomass and LAI can well be estimated by a combination of hyperspectral images and field data. The R² values for cover fraction are low compared to the others, meaning that the predictions of this variable are more uncertain.
Object-based image analyses took place by the application of segmentation to obtain the spatial structure and variability present in each variable separately. Segmentation is applied on predictive vegetation variable maps by using three methods and ten different object scales (scale 1 till 10). As expected the highest R² is reached by the methods where the predictive vegetation variable maps from both acquisition years are used (R² biomass: 0.66, LAI: 0.42 and cover fraction: 0.41).
The optimal segmentation scale varies between the vegetation variables (biomass scale 4, LAI scale 8 and cover fraction scale 3). These differences can be explained by the spatial structure and variability present inside the predictive vegetation variable map (heterogeneity in the data).
Temporal patterns assessed by change detection show that 55% of the natural vegetation areas have an increasing change pattern for biomass. LAI and cover fraction mainly show a decrease (LAI 39% and cover fraction 35% of the natural vegetation areas). This does not immediately mean that these temporal patterns are wrong, because an increase in biomass do not have to result in increasing values for cover fraction and/or LAI.
Temporal patterns in vegetation variables are compared with static environmental factors. There are some patterns visible inside a specific category of an environmental factor (geology and elevation), but it is not certain that this is the case over longer period or that it only dominates in the temporal patterns between 2003 and 2008. To give a more certain conclusion about change patterns on its own and in relation to environmental factors, further investigation over a longer timescale is required.||