GIS Colloquium – Talks (Summer Term 2017)
Asher Yair Grinberger
Mon, April 3, 2.15 pm (Ort: INF 348, Raum 015)
Correctly identifying the contextual units that influence geographic phenomena is a fundamental issue within spatial analysis. As ‘true casually relevant’ activity-related geographic contexts vary at the level of the individual by time, space, and activity, they are prone to errors of misspecification, which affect the validity of results. The uncertainty related to this issue is reduced today when widely available high-resolution mobility data is used to reconstruct individual activity spaces. Yet, as activity is mediated by spatial cognition, knowledge, and preferences, delineating these spaces using only objective spatio-temporal constructs may hamper the effort, i.e. considering spaces based only on physical accessibility and not on their behavioral relevance may constrain the extent to which uncertainty may be reduced. To establish this argument, this presentation would rely on three studies: a field experiment studying changes in activity patterns within a tourist attraction when visitors are exposed to different spatial information and geographical layouts; a model predicting visit probabilities within a road network via the integration of non-Euclidean time-space constructs into models of Probabilistic Time Geography; a procedure formalizing topologies of time-space consumption behaviors represented by movement trajectories as well as delimiting the activity spaces related to them. The first study stresses the need to consider the cognitive-behavioral element when delineating geographical contexts, while the latter two studies constitute the means for this end.
Mon, April 24, 2.15 pm (Ort: INF 348, Raum 015)
The first part of the talk will present the 3D geoinformation research group at the Delft University of Technology and cover current developments in 3D GIS in the Netherlands, such as standardisation of 3D geoinformation and applications. The second part will focus on the PhD research of the presenter. The research comprised several topics around the level of detail (LOD) in 3D city models. For example, it developed a method to benchmark the influence and value of different LODs in different spatial analyses, and the influence of the propagation of positional error. The method is supported by procedurally generated (synthetic) 3D data that may be automatically degraded thus simulating acquisition errors.
Prof. Dr. Christine Pohl
Mon, April 24, 2.15 pm (Ort: INF 348, Raum 015)
Natural disasters caused by geophysical, climatological, meteorological and hydrological events harm societies and create large damage. Within the disaster management cycle the four phases disaster event, response, recovery and risk reduction greatly benefit from geographical data. The observations of location, frequency and magnitude of the event as well as the information on the destruction of natural resources, infrastructure and settlements often originate from satellite sensors of different types. Remote sensing is a major source of relevant information to support the assessment of hazards as well as the mitigation of the disaster impact. Satellites provide images with different spectral coverage and spatial resolutions on a regular basis. Especially since the availability of Sentinel-1, -2 and -3 imagery large volumes of data are acquired and archived. This enables a very high temporal resolution creating new possibilities in time series and multitemporal image analysis but also poses new challenges. The data is free of charge and can be accessed through different portals. In the process of analyzing the multisensor data there are various options to add value to the data. There is a great benefit from observations with different scales and spectral bands. The combination of different sensors and the monitoring over time delivers valuable insights in the disaster extent and damaged objects, which would not be available from one source alone. The presentation introduces the aspects and benefits of multimodal and multitemporal remote sensing for hazard and disaster monitoring. It provides an overview on existing processing and fusion strategies and illustrates the results using various examples.
From Clouds to Crops: A Journey Through Various data Assimilation Showcases in Environmental Science - ENTFÄLLT / CANCELLED
Dr. Oliver Sus
Tue, May 2, 10.15 am (Ort: INF 348, Raum 015)
Data assimilation, or more generally phrased machine learning, has been gaining popularity in academia and industry in recent years. Today, affordable computing power and the immense availability of accessible data are fueling a data science boom. With the start of the EU's Copernicus programme, highly resolved satellite data will further push this development in the geosciences. Such recently emerging sources of spatial data provide new challenges and opportunities in geographic information technology. In this talk, I will summarize my own research activities in applied data science or assimilation throughout the last 10 years. To quantify the carbon balance of croplands, I have assimilated carbon flux and remotely sensed NDVI data (MODIS) into an ecosystem model. I trained the same model for an improved simulation of plant hydraulics and drought resistance through corroboration with transpiration data, and investigated statistical relationships between variables of forest inventory data. I finally applied an artificial neural network and a variational data assimilation algorithm within a cloud model to retrieve climatologies of cloud parameters from space (AVHRR, MODIS). These attempts are a mixture of success stories and failures, and both of which shall be given attention in this presentation. I will conclude with a brief outlook on the potential of the Copernicus programme to spark research activities and business ideas around the globe.
Mon, May 8, 2.15 pm (Ort: INF 348, Raum 015)
Constant monitoring of tropical forests is important to increase our knowledge on effects of climate change. However, mapping aspects like forest structure, degradation and deforestation is time-consuming and expensive. Innovative high-tech approaches like Terrestrial Lidar Scanning (TLS) and Hyperspectral Sensing from Unmanned Aerial Vehicles (UAV) may revolutionize the way we monitor our forests. In the last few years, we conducted a number of campaigns to investigate the use of innovative technologies for mapping the plant traits of tropical forest. These campaigns consisted of UAV data acquisition, TLS measurements, or combinations of both techniques. UAV based sensing allows for multi-scale observations and fills the gap between ground based sampling and satellite based observations. Our in-house developed Hyperspectral Mapping System observes both forest structural information, which can be derived from the 3D point cloud data, and the biochemistry of the tree canopy which can be characterized from the hyperspectral data cube. The structure of the forest is assessed with the TLS, where our research focusses on the 3D reconstruction of individual trees by using Quantitative Structural Models (QSMs). These can further be used to determine individual trees biomass and e.g. branch architecture. Upscaling these procedures to the plot level allow less biased biomass estimates than the currently used allometric equations. Finally, some first results of the newly acquired RiCopter UAV based Lidar system will be presented. This system will allow us to upscale our laser scanning activities over larger areas, preserving the high accuracy and point density required for accurate reconstruction of the tree shapes. I will show examples of our work across the tropics, with illustrations from field campaigns in Indonesia, Ghana and Guyana and describe our methods to derive individual tree-level plant traits and plot-level estimates of biomass.
Prof. Bo Huang
Mon, May 15, 2.15 pm (Ort: INF 348, Raum 015)
Recent years have witnessed a dramatic increase in our ability to collect data from various mobile devices, sensors and the Internet. This has created an opportunity to capture human activities, which were, however, difficult to obtain. In this presentation, I will give a brief introduction to big data including an analysis of their pros and cons. Subsequently, I will discuss our recent work on using big data to detect urban structure and urban vibrancy and to estimate population-weighted air pollution exposure. Gathering the dynamic distribution of population, our studies are expected to gain more insights into the human behavior in relation to space and time, thereby facilitating the development of more informed spatial plans and designs.