Integrated evaluation of sensor data streams and field observations

Many phenomena in our environment are extremely complex and cannot simply be captured via sensors. Additional field observations are often indispensable. This project used two case studies to examine how sensor data and field observations can be combined and evaluated.

  • Portrait / project description (completed research project)

    Dropdown Icon

    Understanding complex phenomena in our environment requires more than capturing and analysing large quantities of data via sensors. Additional field observations are often indispensable. By means of two case studies, the WeObserve project has combined and conducted an integrated evaluation of, first, sensor data that is limited in terms of accuracy, and second, field observations that are precise but often only selectively available.

    In the first case we developed a dual remote recording system of this type in order to carry out a comprehensive analysis of soil erosion events in the Alps.

    In the second case we integrated pre-existing sensor and remote field observation data to gain a better understanding of the mass movement of migratory birds.

  • Background

    Dropdown Icon

    Analysing sensor data streams is essential for better understanding many phenomena in our environment. However, because automatic evaluation of these data streams is limited, their complex interconnections often remain hidden. WeObserve developed new processes for combining a large quantity of sensor data with precise but selective field observations whenever available and analysing them in an integrated way. Depending on the application domain, the selective information increasingly originates from non-specialists, who document relevant observations and complement existing data sources.

  • Aim

    Dropdown Icon

    The aim of the project was to develop processes for collecting, combining, and evaluating comprehensive sensor data and selective remote field observations. Two case studies

    (a) achieved a better understanding of erosion processes and

    (b) improved the forecasting of spatial and temporal migration patterns of bird species.

  • Relevance/application

    Dropdown Icon

    Developing new methods of recording, integrating and analysing heterogeneous data streams and selective field observations will lead to new knowledge in the environmental sciences. The first case study will improve our understanding of alpine ecosystems, which are of great relevance to biodiversity and the water balance. The second case study will enable us to use better knowledge of annual bird migrations to answer questions about the responsible use of wind energy and how migratory birds spread diseases.

  • Results

    Dropdown Icon

    In both applications, we generated novel insights valuable for an academic audience, practitioners, and an interested public alike. The developed methods, based on Machine Learning and Deep Learning techniques, provide workflows how to integrate heterogeneous datasets and advance current possibilities to enable new applications and large-scale analyses. Our main results include:

    (a) Soil erosion: By employing different models, we were able to show an increase of soil erosion in Swiss Alpine grasslands in recent years. In particular, for the Urseren valley we concluded that sites affected by soil erosion have increased by between 156% and 201% from 2000 to 2016. Furthermore, we analysed ten Grassland hill-slope sites across Switzerland with respect to causal factors for a common erosion type. Our investigation yielded that, in general, information on the slope, its orientation as well as terrain roughness were most relevant for occurrence of erosion, with additional site-specific factors. We demonstrated that our fully-automated neural network approach allows scaling erosion site identification to unprecedented extents.

    (b) Bird Observations: We evaluated several approaches to modelling species-wise bird observations in Central Europe. By considering the fraction of each species, we addressed inherent biases in Citizen-Science data, like a strong correlation of observations to population density. The results explain a significant amount of variation of month-wise abundances of multiple species based on the environment.
    Bird Migration: By including weather information, we extended Gaussian Process models to improve the prediction of migration velocity. As these results are typically difficult to visualise and interpret, we also developed a visualisation for simulated migration paths.
    As an unprecedented feature, the simulated paths make use of the uncertainty of the model. These new tools facilitate studying bird migration and abundance in more detail.

  • Original title

    Dropdown Icon

    weObserve: Integrating Citizen Observers and High Throughput Sensing Devices for Big Data Collection, Integration, and Analysis