Key Projects:

Integration of Data Science into GNS

Grow integration of data science within research programmes and projects by encouraging uptake through base integration, seminars, interactive workshops and internships.

• Base integration project is an open program component that was established to bring data science into other projects around GNS. Funding has been set aside for data scientists to be able to support projects that want to use and leverage data science expertise. 

•  Our internship program is led by Rob Buxton, Yannik Behr, Tatiana Goded. GNS Science supports a small number of Data Science students every year. The students undertake to complete a small project over the summer break period (typically November to February). Previous projects have included the design and development of a prototype Volcano Rover – an autonomous platform for collecting data on volcanos and the redevelopment of code to process earthquake felt reports more efficiently.  
• Seminars and workshops are hosted by the AI & AA Team, including a weekly seminar timeslot where members of team, the wider GNS community and guest speakers have presented 30 – 40 minute presentations on topics related to Data Science.

Science to Operations:

Operationalise data science applications on internal and external cloud computing platforms for more effective distribution of data science tools

• Earthquake catalog generation pipeline to develop workflow-template for reproducible and operational science applications with an initial focus on earthquake catalog generation in partnership with Victoria University of Wellington.

Theory of Data Science:

Explore new machine learning and artificial intelligence methods in collaboration with peers, stakeholders, and end-users, with a focus on theoretical data science and machine intelligence

• Manuscript on review of graph models for earth science. Work is underway to review recent (post 2015) publications on graph data structures related to earth sciences. 
• Natural Language Processing for social science (Rob Buxton, Ceilia Wells, Christof Mueller). The AI Team collaborated with the GNS Social Science Team to investigate the possible use of Natural Language Processing techniques to automate a time consuming, previously manual exercise called “content analysis”. Natural Language Processing is ideal for this purpose since it is the science or discipline of creating machines that can understand written or spoken language. Content analysis is a process whereby media is analyzed to ascertain the underlying meaning.  
• Probabilistic Circuits (Florent Aden, Yannik Behr): Forecasts from AI/machine learning models often come without uncertainties . This makes it difficult for us to use these forecasts with confidence. Probabilistic Circuits (PCs) are machine learning models that combine the expressiveness of deep neural networks with a probabilistic view of the data that we want to model. As a result, PCs provide probabilistic forecasts that end-users can trust.  

Data Science Methodology transfer:

Application of established machine learning and artificial intelligence methods, including deep learning, surrogate/meta modelling, and natural language processing and graph analysis

• Ground-water meta-modelling with Neural Networks (Sapthala Karalliyadda, Conny Tschritter, Brioch Hemmings):  
  This research aims to support the metamodeling workstream of Te Whakaheke o Te Wai (TWOTW) research program by developing a neural-network model that mimics the groundwater age calculations using numerical models. This work will complement the metamodeling capabilities developed under TWOTW program. 
• Use geodetic and seismic data to detect and analyze slow slip events (Florent Aden):  
  We explore how a Data Science can help identifying and recognizing in seismic and InSAR data, spatio-temporal patterns related to ground deformation such as creeping fault of Slow Slip events. 
• Supervised or unsupervised segmentation of terrain data and terrain data generation (Christof Mueller): 
  GNS is working with a wealth of geospatial data (InSAR, LiDAR, Satellite Imagery, Aerial Photography, etc.). We extract knowledge about landforms and landform features from these data sets to better understand natural hazards, landform development and land-use, to give some examples. With an ever-growing amount of data, the manual interpretation of such data sets becomes an almost unsurmountable task. Like in other disciplines, such as medical imaging or image analysis in self driving cars, we work towards using deep learning machine intelligence methodology to augment the interpretation process of such data sets.