Environmental Data Science, and its Transformative Potential. 5 th September 2017 Gordon Blair and Graham Dean

Size: px

Start display at page:

Download "Environmental Data Science, and its Transformative Potential. 5 th September 2017 Gordon Blair and Graham Dean"

Emery Wilkins
5 years ago
Views:

1 Environmental Data Science, and its Transformative Potential 5 th September 2017 Gordon Blair and Graham Dean

2 Structure An Introduction to Environmental Data Science [GSB] Overview of some statistical methods [GD] Extreme Value Methods Changepoint Analysis Machine Learning Pointers to some computational resources and data science services [GD] Cluster Based Processing Data Science Services Code Sharing and Versioning Collaborative Development and Dissemination

3 What is Environmental Data Science? The development of data science principles and techniques for sense making and decision support related to the natural environment Focus on methodological innovation Importance of integration

4 Challenges of the Area The data challenge The modelling challenge The complexity challenge The spatial/ temporal challenge The uncertainty challenge The cross-disciplinary challenge

5 Scope of Environmental Data Science

6 A Research Roadmap Data Science of the Natural Environment: A Research Roadmap Gordon S. Blair 1,2, Peter Henrys 2, Amber Leeson 1, John Watkins 2, Emma Eastoe 1, Susan Jarvis 2 and Paul Young 1,3 1 Data Science Institute, Lancaster University, Lancaster, LA1 4WA, UK 2 Centre for Ecology and Hydrology, Lancaster Environment Centre, Lancaster, UK 3 Pentland Centre for Sustainability in Business, Lancaster University, Lancaster, UK About to be submitted to Frontiers of Environmental Science

7 A Research Roadmap (continued) Challenge 1: To encourage and enable the required level of cultural shift towards open science, that is towards a science that is more collaborative and integrative through open approaches to data, models and knowledge formation, and also towards a science that is more transparent, repeatable and reproducible. Challenge 2: To build on the benefits of cloud computing, but offer levels of abstraction (and associated services) that are much better suited to the domain of science, including high-level support for running complex, integrated modelling in the cloud. Challenge 3: To address complexity more fundamentally and explicitly and, in particular, seek data science techniques that recognise and resolve key issues of complex systems including feedback loops, inter-dependent variables, extremes and also to detect and manage emergent behaviour. Challenge 4: To provide techniques and frameworks to both reify uncertainty in scientific studies and also reason about the cascading uncertainties across complex experiments, e.g. in integrated modelling frameworks and in ensemble approaches. Challenge 5: To seek adaptive techniques driven by considerations of uncertainty and also the goals of a scientific study, including adaptive approaches to sampling or gathering of data (including in real-time in an Internet of Things), and also in adaptive modelling approaches. Challenge 6: To seek approaches that deal with epistemic uncertainty in environmental modelling, noting the important links with dealing with emergent behavior in complex and irreducible phenomena. Challenge 7: To seek novel data science techniques and, in particular, innovative combinations of data science techniques that can make sense of the increasing complexity, variety and veracity of underlying environmental data, exploiting also multiple data sets including real-time streaming data. Challenge 8: To seek innovations in modelling by combining process models with data-driven or stochastic modelling techniques and also seeking ways of assimilating a range of data sources more generally into steering model executions. Challenge 9: To incorporate sophisticated spatial and temporal reasoning, including reasoning across scales, as an integral aspect of environmental data science and not something that is just provided through separate tools such as GIS tools. Challenge 10: To discover new fundamentally new modes of working, methods and means of organisation that enable the required level of cross-disciplinary collaboration as required to address the key grand challenges of earth and environmental sciences and, more specifically environmental data science in its contribution to these grand challenges.

8 The Importance of Virtual Labs

9 Ongoing Initiatives Data Science Institute at Lancaster Senior Fellowship in Digital Technology and Living with Environmental Change

10 Extreme Value Theory Characterising Rare Events

11 Extreme Value Methods Motivating Questions In order to answer questions like: What is the probability of: Wave Heights: no offshore structure damaged in a storm? Temperature: a heatwave exceeding 40 C at collection of sites? River Flows: total insurance loss from a flood exceeding 1B?

12 Extreme Value Methods Use Cases Intergovernmental Panel on Climate Change Evaluating flood risk to railway infrastructure Construction of offshore flood defences National Flood Resilience Review

13 National Flood Resilience Review Joint work with JBA Consulting Probability of an extreme flow at a gauge somewhere in a year

14 Changepoints Understanding Changes in a Process

15 Changepoints Motivating Questions In order to answer questions like: Finance: can we detect changes in the volatility of the stock market? Storm surge: are winter seasons becoming stormier? Climate change: are temperature changes being detected at a number of gauging stations? Changes in the mean Changes in the variance Changes in the mean and variance

16 Changepoints Use Cases Credit card fraud detection Changepoint detection for acoustic sensing signals Check reliability of oceanographic hindcast data sets Changes in chromosome structure

17 Optimising Exercise Regime Significance Magazine Article, 2015 Kaylea Haynes, STOR-i Alumni

18 Machine Learning Recognising Structure in Data

19 Machine Learning Motivating Questions In order to answer questions like: How to we identify: Species Recognition: How do we recognise an animal species from audio data? Environmental anomalies: How do we recognise an oil spill from imagery data? Behaviour Classification: How can we infer specific types of behaviour from sensor data?

20 Machine Learning Use Cases Identification of Species Using Echolocation Identification of Oil Spills Changes in Land Cover Use Species Distribution Modelling

21 Machine Learning Final Year U/G Project Animal Behaviour Elliot Wright, B.Sc. (Hons) Computer Science, 2016

22 Cluster Based Processing Apache Hadoop & Spark

23 Data Science Services Amazon AWS & Hortonworks

24 Code Versioning and Sharing Github & Bitbucket

25 Collaborative Development R-Shiny & Jupyter

ENVRIPLUS GENERAL INTRODUCTION. Ari Asmi ENVRIplus director. H2020 Project Project Number:

ENVRIPLUS GENERAL INTRODUCTION Ari Asmi ENVRIplus director CONTENTS! Strategy and aims! Factsheets! Structure! Products CHALLENGE VISION KEY STRATEGY DIRECTIONS OF ENVRIPLUS! Primary customers are Research