Just Give Me A Rainfall Forecast Connecting models to endusers

1
Just Give Me A Rainfall Forecast!Connecting
models to end-users

• Peter McIntosh1, Mike Pook1, James Risbey1 and
  Shaun Lisson2
• 1CSIRO Marine and Atmospheric Research
• 2CSIRO Sustainable Ecosystems
• and
• 1,2CSIRO Wealth from Oceans Flagship
• with help from Melissa Rebbeck (SARDI)

2
What is POAMA predicting?
3
ENSO and IOD years (Meyers et al. 2007)
4
Birchip La Niña/-IOD
5
Birchip La Niña/-IOD
6
Wheat Modelling at Birchip

• APSIM
• computer model of wheat plant growth
• uses daily climatological data
• run continuously for 116 years (1890-2005)
• three soil types clay, silt and sand
• three management decisions
• sowing date (every 8 days from 14 Mar to 28 Jul)
• starting N (25 to 250 kg/ha in steps of 25kg)
• topdressing N (0 to 150 kg/ha in steps of 25kg)
• find the best management each year

7
Management in El Niño/-IOD years
Sowing date
GM ()
Topdress N
Start N
8
GM outcomes in El Niño/-IOD years
9
ENSO and IOD years (Meyers et al. 2007)
10
(No Transcript)
11
Question

• What is the value of a perfect forecast of
  ENSO/IOD taken over all types of years?

12
Birchip clay mean rainfall and gross margin
13
GM (/ha) in ENSO/IOD categories
1890-1998, woodlands_clay_reset_Jan1_v2,
categories from paper (fixed), frost adjusted,
GMgt0
14
Valuing forecasts

• 109 year APSIM simulation on Birchip clay
• management sowing date, start N, topdress N

15
Summary

• Forget about forecasting ENSO/IOD state?
• Or learn to use the information better
• Forecasting in categories loses skill through CV
• particularly true for model forecasts (limited
  years)
• Can we substitute ensemble members?
• Should we aim to forecast seasonal rainfall
  amount?
• Distribution is King!
• We dont yet know how to connect perfect
  forecasts to management decisions
• What is the best strategy when forecasts are
  imperfect?

16
What does a good forecast look like?

• Daily (or weekly) rainfall for the season!
• Timing of the break (but is this really useful?)
• Chances of frost (and when)
• Windiness (evaporation)
• Ultimately P-E
• Chance of extreme heat at end of season
• Accurate 10 day rainfall forecast for topdressing
  (and harvesting?)

17
Rainfall Distribution

• 30mm on one day in a month is much more useful
  than 1mm every day
• Realistic rainfall distributions will be crucial
  in using models in Yield Prophet
• The models must get the rainfall mechanism right!

18
Model Synoptics Comparison with Obs.
1970-2000