rapsimng.decide.sowing

rapsimng.decide.sowing provides functions to analyse APSIM Next Generation simulation outputs to support sowing decision under defined environments and management assumptions.

The package operates on APSIM outputs already loaded into R and focuses on transparent, reproducible evaluation of sowing window, establishment risk, and early crop development. It does not run APSIM simulations, interpret user intent, or make automatic recommendations.

Installation

Currently on Github only. Install with:

remotes::install_github('byzheng/rapsimng.decide.sowing')

Overview

Farmers and advisors often ask:

What is the optimal sowing window and establishment risk for my paddock?

This package supports that question by analysing long-term APSIM simulations to describe:

  • optimal sowing window
  • establishment risk and early crop development
  • yield and risk across sowing dates
  • trade-offs between early and late sowing

Instead of returning a single “best” date, the package provides a decision landscape that highlights strengths, weaknesses, and uncertainties for each sowing option.

Scope

  • APSIM Next Generation outputs only

  • Input: a tidy data.frame of simulation results

  • Analysis at sowing date × environment × management level

  • Deterministic, reproducible (no optimisation or AI)

Not in Scope

  • Running or modifying APSIM (rapsimng)
  • Interpreting user intent (agrillm)
  • Economic optimisation or automatic recommendation
  • Black-box ranking or hidden decision rules

Input Data

The package expects a data.frame (or tibble) where each row represents a simulation outcome for:

  • sowing date (or sowing window factor)
  • year
  • environment/management factors

Typical required variables:

  • sowing date identifier
  • year
  • yield
  • establishment success/failure
  • optional: phenology stages, frost/heat indicators

The data can come from any source (e.g. rapsimng, database export, CSV), as long as structure is consistent.

Decision Context

The package assumes a decision context such as:

  • fixed location/environment (e.g. Wagga Wagga)
  • defined sowing window (e.g. 1–15 May)
  • long-term climate variability (e.g. 30 years)

What the Package Provides

1. Sowing Window Performance

  • mean / median yield by sowing date
  • establishment risk by sowing date
  • interannual variability

2. Risk Assessment (optional)

  • establishment failure risk
  • frost/heat exposure during early development
  • failure risk (probability below threshold yield)

All risk definitions are explicit and stored as metadata.

3. Sowing Window Robustness

  • performance across sowing dates within a window
  • sensitivity to sowing timing

4. Trade-off Analysis

The package highlights trade-offs such as:

  • early vs late sowing
  • high yield vs high risk
  • stable vs variable performance

Outputs

Core Outputs

  • sowing window performance summary
  • risk metrics per sowing date/window
  • assumption metadata

Decision-Support Outputs

  • filtered candidates meeting criteria
  • explicit reasons for inclusion/exclusion
  • trade-off summaries

Example:

Sowing in early May has high mean yield but high establishment risk in dry years.
Sowing in late May has lower yield but more stable establishment and minimal frost exposure.

Visual Outputs (optional)

  • yield distributions by sowing date/window
  • yield vs risk trade-off plots
  • sowing window performance comparisons

Design Philosophy

This package follows three principles:

  1. Transparency
    All assumptions, thresholds, and metrics are explicit.

  2. Reproducibility
    Same input data always produces the same results.

  3. Separation of concerns

    • rapsimng → APSIM interaction
    • rapsimng.decide.* → decision analysis
    • agrillm → intent and explanation

Relationship to Other Packages

  • Depends conceptually on: rapsimng
  • Independent of: APSIM runtime
  • Callable by: agrillm

Key Insight

This package does not select a sowing date.
It explains how sowing windows perform and trade off under uncertainty.

Status

Early-stage design focused on:

  • sowing window suitability under environment and management
  • extensibility to other decision domains