Overview of Farmadapt

1
Overview of Farm-adapt

• Sutton Bonington
• 2nd June 2004

2
Purpose

• How can farmers, at least financial cost
• reduce nitrate emissions?
• reduce GHG emissions?
• adapt to climate change?
• adapt to CAP reform?
• How will farmers, assuming profit maximisation
• adapt to reduced water resources and increased
  irrigation demand?

3
Farmer decisions

• Crop mix and animals owned
• Timing of operations
• Labour (employed, casual and contract)
• Machinery owned
• Investment in buildings
• Investment in reservoir and irrigation capacity

4
Implementation of core model 1

• Farm level model, simulates single or many farms
• captures the many options available to farmer
• Mixed integer linear programming (MIP) model
• maximises Farm net margin (? profit)
• subject to constraints
• available land
• labour machinery requirement to grow crops
• integers labour, machinery, switches

5
Implementation of core model 2

• The model is has not been parameterised by
  fitting to data
• known policy constraints, e.g. minimum set-aside
  land area
• assume farmer is following current good
  practice, published guidelines, e.g. fertiliser
  recommendations, animal diets
• several piecewise linear approximations
• uses published estimates, e.g. field workrates

6
Implementation of core model 3

• Inputs to model from many sources, varies
  depending on context
• other models, written by others
• crop growth models
• soil nitrogen models (nitrate/nitrous oxide
  emissions)
• ruminant digestion models (methane emissions)
• hydrology models (irrigation availability)
• survey data (e.g. actual farm crop yields)
• historical or estimates of future prices

7
Implementation of core model 4

• Uses Xpressmp optimization software
• libraries for matrix generation and optimization
• multiple ways to access libraries
• native language (primitive Pascal dialect),
  compiled to byte code, runs on virtual machine
• C, C, Java, Visual Basic
• different levels of access, raw matrix to total
  abstraction
• all model code written by us

8
How can farmers?

• Single farm
• Weekly time step
• Annual model
• Profit maximizing
• Clairvoyant farmer has perfect foresight
• What farmers should do, policy making

9
Robustness and uncertainty

• We want to recommend farm plans that are robust
  to uncertainty in inputs and to natural
  variability
• not only applicable to average farm in average
  year
• sensitivity analysis
• Monte Carlo simulation using output from input
  models, compare fixed farm plans with adaptation
  to each set of inputs

10
How will farmers? 1

• Multi-farm, allows trading of resources
• Weekly time step
• Multi-annual model
• Profit maximizing
• Smart farmer uses knowledge from previous
  years
• What farmers will do

11
How will farmers? 2

• Each year has a two step process
• Farmer makes planning decisions at the start of
  the year using expected crop yields, irrigation
  requirements, and water availabilities calculated
  from previous four years
• Annual actual outcome determined by farm plan,
  actual crop yields, irrigation requirements and
  water availability

12
Multi-annual 1 climate change

• Fifty years of crop yields and irrigation
  requirements for each period (baseline, 2020s
  2050s) from crop model
• Bootstrap sample from each period to create a
  pseudo-sequence 1983-2079
• Run model over period, investments in each year
  carried over to next year
• Repeat with new sample x100

13
Yield sampling for pseudo-sequence
1) Crop yield data
50x Baseline Yields
50x 2020 Yields
50x 2050 Yields
2) Bootstrap sample
3) Pseudo- sequence
1983
2020
2050
2079
4) Model run
1987
14
Multi-annual 2

• Samples with sequences of dry years effect on
  irrigation investment cropping
• Compare intra-period with inter-period variation
• Sampling methodology means probability of
  outcomes can be estimated

15
Model selection and Farm-adapt

• Not a regression or fitted model, AIC/BIC etc.
  cannot be directly applied
• Model over-complex?
• too flexible (? over-fitted?)
• which bits to remove?
• Many alternative sub-optimal solutions
• almost as good/more realistic?
• how similar/different to optimal solution?