Apply artificial intelligence to perform climate projections

The student project will be to do both training and testing on historic meteorological data. E.g. to predict interesting variables that are not part of climate model simulations like sea wave heights, humidity, sun hours etc from variables that are part of climate simulations like temperature or sea level pressure.
Master

Future changes in wave height, wind climate, sunshine duration etc. have important implications for society and affect agriculture, water resources, risks for human beings and infrastructures due to wave and wind extremes and renewable energy production. However, the understanding of projected changes in these variables is limited relative to many other climate variables such as temperature and precipitation.

Over the recent years the field of artificial intelligence (AI), and in particular Artificial Neural Networks (ANN) techniques, has matured significantly and a wide range of new methods have been developed. Today, the methods can document exceptional classification and prediction performance and even outperform human beings on advanced tasks like image classification, speech recognition, geolocation (guessing geographic location of an image), and age estimation. The methods are expected to be important to all parts of society from self driving cars to medical diagnosing.

Goal

The student project will be to do both training and testing on historic meteorological data. E.g. to predict interesting variables that are not part of climate model simulations like sea wave heights, humidity, sun hours etc from variables that are part of climate simulations like temperature or sea level pressure. Train the statistical relations for observations from e.g. 1970 - 2000 and predict for the years 2001 - 2017. Compare predictions with real observations. In the project we want investigate spatiotemporal models (e.g. CNN + LSTM), but maybe a student project can limit to temporal modelling. For each grid point, compare traditional time series models and RNN/LSTM to perform such predictions.

Learning outcome

  • Insight into advanced techniques of machine learning
  • Working on a real world application
  • Collaboration with researchers in the topic of machine learning, specifically deep learning
  • Possibility to implement and research a novel approach

Supervisors

  • Michael Riegler
  • Hugo Lewi Hammer, HiOA
  • Morten Goodwin, UiA

Collaboration partners

  • HiOA -Oslo and Akershus University College of Applied Sciences
  • UiA - University of Adger