Seasonal Agricultural Forecasting 

Main contacts for Initiative

Phil Alderman and Stefan Niemeyer, Alex Ruane, Jonas Jägermeyr, Meridel Phillips, Christian Branneon, Cynthia Rosenzweig, and Ronan Trepos

AgMIP Seasonal Forecasting Community

Awareness of ongoing efforts and operational products in pre-competitive space.

As noted by Van der Velde et al. (2018), not many publications on crop yield forecast performance — What are the best metrics to examine?

  • Correlation/variance explained?
  • Performance in anomalous years? Ability to capture particular extremes?
  • Relative change or absolute change?
  • Lead time metrics?
  • False alarm / false positive / hit rate?

Community engagement could establish best practices:

  • Use of latest information on farm environment (e.g., soils and management)
  • Use of overlapping weather/climate forecasts
  • Identification of data and model improvement priorities (consider remote/field observations)
  • Harmonization and multi-model tools
  • Extension of tools to additional applications and time horizons
  • Connection to user communities
  • Engagement and utilization of big data tools
  • Treatment of uncertainty
  • Stress tests, attribution, and scenario analysis

Iterative model development 

Forecasts can be improved via:

  • Model/framework improvement
  • Longer/improved statistical records
  • Configuration improvement

Open system would allow for community-wide improvement of pre-competitive information

  • Do we know where line is between pre-competitive and competitive space?
  • Where can methodological advancements, configuration data, model improvements, and new observational datasets more efficiently benefit all forecasts?

 

 

 

 

Summary

Yield forecasting needs to draw upon the best of direct observations, remote sensing, and crop modeling.

The AgMIP community is interested in developing and applying advanced modeling tools:

  • Facilitate knowledge exchange
  • Accelerate model and configuration improvement
  • Determine best inputs into big data applications
  • Engage with extensions and connected applications

Preliminary work is underway to build advanced modeling system in the US with aim for remote sensing assimilation, probabilistic forecasts, and iterative configuration improvement.