INLA

INLA performs Bayesian inference using integrated nested Laplace approximations to fit spatial models for analysis of spatial variation and joint spatial–genetic effects in agricultural field trials.

Key Features:

• Bayesian Framework: Implements Bayesian inference via integrated nested Laplace approximations (INLA) for parameter estimation and uncertainty quantification.
• Spatial Modeling Capabilities: Fits independent row and column effects, separable first-order autoregressive (AR(1)) models, and Gaussian random field (Matérn) models to field trial data.
• Matérn Model Flexibility: Uses the Matérn Gaussian random field to provide interpretable parameters and accommodate flexible trial designs across multiple years and locations.
• Joint Spatial and Genetic Modeling: Supports joint modeling of spatial variation and genetic effects, including incorporation of genome-wide markers to improve genetic effect estimates.
• Simulation and Real-data Applications: Applied in simulation studies and fitted to empirical datasets such as wheat and tree breeding data to evaluate performance under varying spatial variation.
• SPDE Approximation: Approximates Gaussian random fields via stochastic partial differential equations (SPDE) for computational tractability.
• R Implementation: Provided as an R package enabling integration with R-based statistical workflows.

Scientific Applications:

• Field trial spatial analysis: Modeling spatial variation in agricultural field trials to improve precision of trait and yield estimates in breeding programs.
• Genetic effect estimation: Estimating genetic effects and incorporating genome-wide markers in wheat and tree breeding studies to enhance selection accuracy.
• Longitudinal and multi-location studies: Analyzing multi-year and multi-location experiments to separate spatial variation from temporal and environmental effects.

Methodology:

Performs Bayesian inference using integrated nested Laplace approximations, fits spatial models including independent row/column effects, separable AR(1), and Matérn Gaussian random fields, approximates GRFs via stochastic partial differential equations (SPDE), and is implemented as an R package.