Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatio-Temporal Graph Neural Networks for Dairy Farm Sustainability Forecasting and Counterfactual Policy Analysis

Surya Jayakumar, Kieran Sullivan, John McLaughlin, Christine O'Meara, Indrakshi Dey

TL;DR

The paper tackles county-scale dairy sustainability forecasting by integrating herd-level operational data into a spatio-temporal graph neural network framework. It introduces a VAE-based data augmentation step to mitigate sparsity, PCA-derived pillars (Reproductive Efficiency, Genetic Management, Herd Health, Herd Management) to build an interpretable composite sustainability score, and an attention-enhanced STGNN to capture spatial dependencies and temporal dynamics across Irish counties. It provides uncertainty quantification via Monte Carlo simulations and evaluates counterfactual policy interventions, demonstrating that targeted improvements yield sustained gains in sustainability forecasts and that the STGNN outperforms baseline models in stability and interpretability. The work offers a practical, policy-relevant tool for benchmarking and scenario planning, with demonstrated headroom for lower-performing regions and a clear path to incorporating direct environmental indicators in future extensions.

Abstract

This study introduces a novel data-driven framework and the first-ever county-scale application of Spatio-Temporal Graph Neural Networks (STGNN) to forecast composite sustainability indices from herd-level operational records. The methodology employs a novel, end-to-end pipeline utilizing a Variational Autoencoder (VAE) to augment Irish Cattle Breeding Federation (ICBF) datasets, preserving joint distributions while mitigating sparsity. A first-ever pillar-based scoring formulation is derived via Principal Component Analysis, identifying Reproductive Efficiency, Genetic Management, Herd Health, and Herd Management, to construct weighted composite indices. These indices are modelled using a novel STGNN architecture that explicitly encodes geographic dependencies and non-linear temporal dynamics to generate multi-year forecasts for 2026-2030.

Spatio-Temporal Graph Neural Networks for Dairy Farm Sustainability Forecasting and Counterfactual Policy Analysis

TL;DR

The paper tackles county-scale dairy sustainability forecasting by integrating herd-level operational data into a spatio-temporal graph neural network framework. It introduces a VAE-based data augmentation step to mitigate sparsity, PCA-derived pillars (Reproductive Efficiency, Genetic Management, Herd Health, Herd Management) to build an interpretable composite sustainability score, and an attention-enhanced STGNN to capture spatial dependencies and temporal dynamics across Irish counties. It provides uncertainty quantification via Monte Carlo simulations and evaluates counterfactual policy interventions, demonstrating that targeted improvements yield sustained gains in sustainability forecasts and that the STGNN outperforms baseline models in stability and interpretability. The work offers a practical, policy-relevant tool for benchmarking and scenario planning, with demonstrated headroom for lower-performing regions and a clear path to incorporating direct environmental indicators in future extensions.

Abstract

This study introduces a novel data-driven framework and the first-ever county-scale application of Spatio-Temporal Graph Neural Networks (STGNN) to forecast composite sustainability indices from herd-level operational records. The methodology employs a novel, end-to-end pipeline utilizing a Variational Autoencoder (VAE) to augment Irish Cattle Breeding Federation (ICBF) datasets, preserving joint distributions while mitigating sparsity. A first-ever pillar-based scoring formulation is derived via Principal Component Analysis, identifying Reproductive Efficiency, Genetic Management, Herd Health, and Herd Management, to construct weighted composite indices. These indices are modelled using a novel STGNN architecture that explicitly encodes geographic dependencies and non-linear temporal dynamics to generate multi-year forecasts for 2026-2030.
Paper Structure (37 sections, 31 equations, 14 figures, 4 tables)

This paper contains 37 sections, 31 equations, 14 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology for Data Pipeline
  3. Dataset Description and Preprocessing
  4. Synthetic Data Generation with Variational Autoencoder (VAE)
  5. Encoder and Reparameterization
  6. Decoder and Reconstruction
  7. Training Objective
  8. Synthetic Sample Generation and Dataset Augmentation
  9. Robustness and Validation of Synthetic Data
  10. Sustainability Pillar Identification using PCA
  11. Linking PCA Components to Sustainability Pillars
  12. Environmental Interpretation of Pillars
  13. Robustness of Pillar Structure
  14. Weighted Pillar-Based Sustainability Scoring Framework
  15. Robustness and Validation of the Scoring Framework
  16. ...and 22 more sections

Figures (14)

  • Figure 1: Concept diagram of the entire framework
  • Figure 2: STGNN Architecture
  • Figure 3: Training and Validation Accuracy of VAE
  • Figure 4: PCA feature loadings illustrating the relative contribution of each variable to the four pillars.
  • Figure 5: Training $R^2$ comparison for five models: STGNN, RNN, LSTM, FFNN, and GKR.
  • ...and 9 more figures