Spatiotemporal

Predicting change and attributes in complex systems in space and time by integrating diverse data and knowledge.

The ability to characterise complex systems and to predict their change through space and time, is a key requirement for identifying new resources and adapting to change which will offer a global comparative advantage to CSIRO. This vertical activity aimed to meet demands for limited resources while moving towards a sustainable future, improved inference, prediction, and forecasting for spatio-temporal complex systems (e.g, agricultural systems, environmental systems, mineral systems or pest outbreaks) in space and time. Spatio-temporal data is characterised by sparsely distributed data observations in space and time such as (concentration of chemical elements, multispecies data, observations from weather stations from BOM, petrophysical measurements). These sparse measurements may be complemented by large volumes of remotely sensed data.

Challenges

  • Spatio-temporal encoding to ensure generalisation and transferability
  • Approaches to handle data sparsity, noise or errors in data, uncertainty, and data integration given the diversity of data types and scales at which data is collected
  • Learning spatio-temporal covariance structures from complex S-T data sets

Use Cases

There are several different areas of CSIRO science than benefited from the outcomes of this activity. Two main directions have been identified including the distribution of species from sparse observations, the estimation of 3D models from a combination of domain knowledge and local data.

An example of the first use case is robust prediction of pest/species distributions with high transferability into non-analogue space (Pigs, Fruit fly, Mosquitos). From a domesticated animal perspective, deriving animal behaviours, food intake, and feed efficiency from wearable sensors or imaging can result in production, health and welfare benefits. An example of the second use case is predicting soil geochemistry from sparse but rich point data with support from auxiliary array data to optimise sampling campaigns. This involves projecting geological domains into 3D space from 1D drill core measurements.