funcrop

Functional Data Analysis for Crop Variety Trials

funcrop integrates B-spline basis functions with linear mixed models to analyse functional traits (grain-fill curves, NDVI profiles, stay-green trajectories) in crop variety trials. It supports scalar-on-function regression, 2D functional surfaces, and multi-environment trial (MET) analyses with genotype-by-environment (GxE) interaction.

Key Features

• Scalar-on-function regression -- relate yield to the entire grain-fill trajectory via a coefficient function beta(t)
• Multi-environment FDA -- FA1/FA2/US/diagonal GxE structures on functional coefficients
• Genomic prediction -- VanRaden G-matrices, pedigree A-matrices, blended H-matrices
• 4 estimation backends -- mgcv, lme4, ASReml-R, bayesreml

Installation

# From GitHub
remotes::install_github("biometryhub/funcrop")

At least one backend must be installed:

Backend	Install	Notes
mgcv	Ships with R	Default. Open-source.
lme4	install.packages("lme4")	Open-source. GLMM support.
ASReml-R	vsni.co.uk	Commercial licence. FA, AR1, genomic vm().
bayesreml	install.packages("bayesreml")	Open-source. Full posterior inference.

Quick Start

library(funcrop)
data(sim_grain_fill)

# Stage 1: Fit variety-specific grain-fill profiles
profiles <- fit_functional_profiles(
  data      = sim_grain_fill,
  time_col  = "time",
  value_col = "grain_weight",
  id_col    = "plot_id",
  group_col = "variety",
  n_knots   = 6,
  engine    = "auto"
)

# Stage 2: Scalar-on-function regression (yield ~ grain-fill curve)
yield_vec <- setNames(
  sim_grain_fill[, .(y = mean(yield)), by = variety]$y,
  sim_grain_fill[, .(y = mean(yield)), by = variety]$variety
)

sof <- scalar_on_function(
  primary_trait       = yield_vec,
  functional_profiles = profiles,
  engine              = "auto"
)

# Inspect the coefficient function beta(t)
plot(sof, which = "coef")

Backend Selection

funcrop auto-detects the best available backend (mgcv > lme4 > ASReml > bayesreml). Override with the engine argument or set a session default:

# Check available backends
funcrop_engines()

# Set default for the session
funcrop_default_engine("asreml")

# Or specify per call
profiles <- fit_functional_profiles(..., engine = "lme4")

See vignette("choosing-backends") for a detailed comparison.

Core Functions

Function	Purpose
fit_functional_profiles()	Stage 1: variety-specific functional curves
scalar_on_function()	Stage 2: coefficient function beta(t)
fit_fda_joint()	Single-stage joint scalar-functional model
fit_2D_functional()	2D functional surfaces (time x depth)
fit_fda_met()	Multi-environment FDA with GxE
bspline_basis()	B-spline basis construction
make_Zspline()	Mixed model reparameterisation
make_genomic_matrix()	VanRaden genomic relationship matrix

Mathematical Foundation

The package implements P-spline mixed models where the smoothing penalty is estimated as a variance ratio. For variety g at time t:

f_g(t) = X_0(t)' beta + Z(t)' alpha_g, alpha_g ~ N(0, sigma^2_u I)

where X_0 spans the unpenalised null space (population mean curve) and Z spans the penalised range space (variety-specific smooth deviations).

A full mathematical specification (24 pages, LaTeX) is available in the project repository as funcrop_mathematical_specification.pdf.

Citation

citation("funcrop")

Licence

GPL (>= 3)

Acknowledgements

Funded by the Analytics for the Australian Grains Industry (AAGI) project. Developed at the University of Adelaide.