Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Upscaling models, downscaling data or the right model for the right scale of application?

Upscaling models, downscaling data or the right model for the right scale of application?

Plant epidemiological models are used in a range of applications, from detailed simulation models that closely follow pathogen infection and dispersal, to generic template-based models for rapid assessment of invasive species. There is increasing interest in applying small scale models - e.g., based on tissue, organ or whole plants - using remotely collected daily data, to generate regional risk information (e.g., maps). The assumption made is that such small scale models “scale-up” appropriately to regional, continental or even global scale. However, these models are often constructed using locally collected, hourly data. By necessity data available are often at much coarser scale, both temporally and spatially, than the data used to develop the model. Computational requirements increase considerably when more detailed models that require fine resolution data (if available) are applied to large areas, while small scale models often add little useful information at these scales and may lead to error propagation. Ideally, detailed models should be used at small temporal and spatial scales and less detailed models used for larger temporal and spatial scales. This paper presents examples of different approaches for changing scales - including upscaling models, downscaling data, and developing new models - and the issues that these approaches create or solve, along with ideas about how we can ensure that the scale of model and data match the desired application.

Adam H. Sparks

July 31, 2018
Tweet

More Decks by Adam H. Sparks

Other Decks in Science

Transcript

  1. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Where do you put the uncertainty?

    View full-size slide

  2. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Upscaling models, downscaling
    data or the right model for the
    right scale of application?
    A/Prof Adam H Sparks, Prof Karen Garrett,
    Prof Chris Gilligan, Prof Andrew Nelson, Dr
    Keith Pembleton

    View full-size slide

  3. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Multi-scale influence of
    weather (data) on pathogens
    & disease development
    (models)

    View full-size slide

  4. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Data are common
    Correlative models
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  5. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Data becoming more
    common
    High-res processes
    Better understanding
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  6. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Uncommon data
    Uncommon models
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  7. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    GCM outputs, climatic
    averages
    Mainly correlative models
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  8. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    How do you handle this?

    View full-size slide

  9. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    APSIM#
    #Agricultural Production Systems sIMulator
    *

    View full-size slide

  10. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    RICPEST (and WHEATPEST)
    *

    View full-size slide

  11. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    EPIRICE
    *

    View full-size slide

  12. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    SimCast
    *

    View full-size slide

  13. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  14. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Generate the Data

    View full-size slide

  15. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    GEOSIMCAST
    *

    View full-size slide

  16. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Modify the Model

    View full-size slide

  17. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    SimCastmeta
    *

    View full-size slide

  18. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    What’s gained or lost?

    View full-size slide

  19. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    What’s Changing?

    View full-size slide

  20. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  21. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  22. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  23. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  24. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  25. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081

    View full-size slide

  26. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Resolution of Predictor Variables
    (often weather data)
    monthly HOURLY
    Extent of Predictor Variables
    plant GLOBAL

    View full-size slide

  27. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    Acknowledgements
    My co-authors for the conversations, their inputs and ideas that inspired me
    Ross Darnell - Data61, CSIRO, Australia
    Emerson Del Ponte - Universidade Federal de Viçosa, Brazil
    Greg Forbes - CIP, Peru
    Robert Hijmans - UC Davis, USA
    Serge Savary - INRA, France
    Laetitia Willocquet – INRA, France

    View full-size slide

  28. CRICOS QLD00244B NSW 02225M TEQSA:PRF12081
    @adamhsparks

    View full-size slide