Abstract
Grassland vegetation covers one third of the world’s surface and is responsible for several key environmental functions such as the storage of carbon and water or the provision of habitats for flora and fauna. These are vital for the enhancement of biodiversity, slope stabilization, the availability of fodder and the possibility of local recreation. Climatic changes, however, are expected to affect all parts of the world and in particular mountainous regions such as the European Alps. At the same time, management procedures have changed over the past years. A plant’s response to both natural and human-induced alteration is quantifiable by sensing the light reflected by the vegetation in diverse ranges of the electromagnetic spectrum. Nowadays, multiple optical sensors are able to directly, proximally and remotely retrieve changes occurring to vegetation. Numerous studies dealt with either monitoring vegetation at single time steps or on the same scale, the combination of multiple sensors detecting on diverse scales has become of major interest in the last few years. Boosting factors for multisensor and multiscale research are technological advancements such as the continuous availability thanks to new sensors, lowered sensor energy consumption, enhanced data transfer capabilities and new standards and advanced infrastructure for the access, storage and processing of data. However, working with multiple sensors requires handling a large amount of diverse datasets; therefore, unified standards and higher processing capacities are required, which, in turn, depend on smart infrastructures and new tools.
In this thesis, different optical sensors acquiring data in diverse spatial scales are combined in order to allow a detailed and continuous monitoring of dynamic changes as well as stress periods of alpine grassland vegetation. For this purpose, I analyzed point data gathered from a spectroradiometer and spectral reflectance sensors as well as spatially extensive imagery from repeated digital photography and satellite data from Sentinel-2. The thesis is divided into chapters treating (i) the automatization of data handling from environmental databases in South Tyrol to reduce computational effort for processing large and diverse datasets, (ii) the retrieval of short-term changes in the optical signal of vegetation and (iii) the observation of stress phases on alpine meadows and pastures.
In order to assess the aptitude of data storage, access and processing standards for a multisensor approach including different spatial and temporal scales I designed two software packages which are able to collect and process data available in standardized databases (Chapter III). These environmental databases entail data from optical sensors (Monalisa Network), meteorological stations (hydrological office) as well as remotely sensed raster data (Rasdaman multidimensional array) which are utilized in the following chapters. The developed software solutions promoted the standardization of datasets facilitating the researchers' handling of diverse data sources. The software solutions clearly accelerated data handling and processing, especially in terms of earth observation time series, and led to further software and front-end developments.
I evaluated the comparability of the Normalized Difference Vegetation Index (NDVI) of multiple optical sensors to detect short-term events such as snow cover and harvests on alpine grasslands (Chapter IV). I found that the different sensors emit quite different optic signals while their NDVI trend remains similar. The biggest changes in the signal are during short-term events. However, events are not equally detectable for all the grassland sites due to location, vegetation composition, management practice as well sensor positioning and time gaps. In order to assess the NDVI dynamics in regard to the sensor and to allow a combined approach, I calculated a derivative index based on the NDVI (Chapter V) with a moving window approach examining the NDVI behavior before and after a specific date and normalized for each station. In this way, I was able to determine potential site-specific thresholds allowing the identification of short-term dynamics by the optical sensor, and link them to influence variables such as snow and harvests.
Finally, I evaluated the detectability of stress periods in alpine grasslands (Chapter VI by thoroughly assessing a drought period in 2018. I included spectral signals acquired for both a pasture and a hay meadow, and evaluated the impact of drought phases on the optical signal as well as on the above ground biomass and Leaf Area Index differentiated by year and site. I calculated metrics for two drought types, namely agricultural and meteorological droughts. The signal differs greatly in dependence of the intensity of the drought type. The hay-meadow proved to be much less affected than the pasture, which resulted in obvious differences in the VIs between both years and sites. Surprisingly, this is not equally true for all sensors. While Phenocam and SRS spectral signatures clearly follow the drought phase, the VIs from Sentinel-2 are less evident in this respect, probably due to the sensor specific angle of observation and time gaps. Simultaneously, the behavior diverges depending on the management practices, and we detected that the soil water content decrease during drought periods correlates with optical signatures while, at the same time, the biomass and LAI are less affected by the drought periods whereas the plant water content considerably changed during periods of elevated droughts.
I was able to show that vegetation indices do not behave equally for monitoring grassland dynamics and stress but are very much dependent on site-specific as well as sensor-specific attributes. Furthermore, I proved that optical sensors are able to detect drought events on extensively used alpine pastures much better than on regularly managed alpine meadows. This detectability, however, depends on the sensor configuration and geometric exposure to the site and especially the sensor’s temporal resolution evidencing the importance of combining multiple data sources in a unified approach.