A collective audience of some 110 delegates joined us, drawn from over 60 organisations, all with an interest in the role and opportunity of digital approaches in supporting environmental decision making.

This event marks the culmination of the ‘CDE’ programme, the ‘Constructing a Digital Environment’, which commenced way back in 2018. Digital Environmental Science has developed apace since those early days and the future opportunities are bright. The Royal Society was a fitting venue to conclude our SPF CDE.

@SPFDigiEnv

@NERCscience

#DUEC

This Constructing Digital Environment (CDE) mini-demonstrator project has allowed the WCO team to collaborate with data providers, data users and data archiving centres to identify, investigate and trial approaches to increase the FAIR-ness of marine observation data. Focussing on the observatory’s L4 CTD (conductivity, temperature, depth) data, the project has developed a range of intuitive data visualisation and exploration tools that seamlessly access PML and BODC datasets.

In the addition to the toolsets that are now available to support the public exploration of the WCO’s CTD data (here…) the project released a reusable Reference Architecture that documents the project’s design approach and provides a valuable resource to guide the development of the CDE communities’ other FAIR near real-time marine environmental datasets. 

Keywords: FAIR; Data and data management; Metadata; Marine; Visualisation

In this webinar, Tom will present recent advances and applications of neural processes (NPs) in environmental sciences. NPs are versatile deep learning models which can tackle a diverse array of environmental prediction problems, including sensor placement, downscaling, forecasting, and infilling missing satellite data. This versatility is enabled by modelling flexibility: NPs can ingest arbitrary sets of observations of point-based or gridded modalities, predict at arbitrary locations, and quantify prediction uncertainty. However, the flexibility of NPs can make them poorly suited to small-data settings, and open questions remain about how to optimise their performance. To answer these questions and accelerate research, Tom is developing an open-source Python package for environmental NP modelling, DeepSensor, which will be presented in this webinar.

Further information and details, this and the wider webinar series is online at https://digitalenvironment.org/webinars/cde-webinar-series-upcoming/#andersson.

In this talk, we will survey state-of-the-art approaches to the analysis of imaging and sensing data from the biological and medical sciences. Our focus will be on understanding what problems have robust solutions, which problems are still very much open, and what types of approach are showing promise on the most difficult problems. Links will be drawn to imaging and sensing in the environmental sciences, with areas of potential for greater collaboration identified.

Further information and details, this and the wider webinar series is online at https://digitalenvironment.org/webinars/cde-webinar-series-upcoming/#styles .

Also co-leading the Hackathon event were Matt Pritchard and colleagues from the STFC JASMIN team. JASMIN is NERC’s globally-unique data analysis facility, providing storage and compute facilities to enable data-intensive environmental science. Delegates were walked through migrating existing modelling approaches onto first the interactive shell, then the batch environments in JASMIN, plus the use of the interactive Jupyter Notebook environment. Also covered was how models can make use of and draw upon the extensive data holdings of the CEDA archives.

Overall the hackathon was a great success and highlighted the value of bringing a group of digitally-enabled scientists together with expert instructors to move forward their research techniques. We would like to see far more of this sort of offering in the future. Particular thanks go to the instructor teams.

In this talk Rossella introduces Data Learning, a field that integrates Data Assimilation and Machine Learning. Data assimilation (DA) is a techniques used to develop digital twins, considering the uncertainties of both model and data. However, these approaches are linear and are often incapable of fully representing real world situations. Machine learning (ML) shows great capability in approximating nonlinear systems and extracting meaningful information from high dimensional data. Integration of ML with DA increases the reliability of prediction by including information in real time and with a physical meaning.

Further information and details, this and the wider webinar series is online at https://digitalenvironment.org/webinars/cde-webinar-series-upcoming/#arcucci .

Another genre of generative AI perhaps less well known is that of image and media creation. Where this has touched the news is around ‘deep fake’ images that are increasingly hard to distinguish from human-generated output.

It is one thing to have a textual summary of a topic, or a meaningful exchange with a chatbot, there is however something of deeper connection with the human psyche to see images, and in fact works of art, produced by AI. There are a number of generative AI-based image creation tools out there, images.ai was an early entrant, and other tools such as NightCafe and starry.ai are also popular. Adobe Firefly is another recent addition, and there are many more.

OpenAI, the creators ChatGPT have released an AI image generator called DALL-E, which will create new imagery from simple text descriptions. With a focus on what Digital Environment can mean, and to discussions above here – we had a go and the results were striking.  One can use natural language to describe a scene and then the AI tool will interpret this for us, thus:

A oil painting of a robot digging the soil in woodland with butterflies

We’re not sure what that last robot is up to!!!