TLDR: This research paper introduces deep learning models, particularly Temporal Convolutional Networks (TCN), to accurately forecast Perennial Ryegrass growth in Ireland using historical grass height data. The TCN model demonstrated superior performance over traditional methods and other deep learning models, offering a cost-effective and practical solution for farmers to optimize grassland management and enhance sustainable dairy farming practices, especially in the face of climate change.
Grasslands are vital for our planet, serving as the second-largest terrestrial carbon sink and supporting immense biodiversity. In Ireland, the dairy sector is a cornerstone of the economy, but it faces ongoing challenges in maintaining profitability and sustainability. A key factor in sustainable dairy farming is effective grassland management, which heavily relies on accurate forecasts of grass growth.
Traditionally, predicting grass growth has depended on complex mechanistic models. These models require extensive data collection on various parameters like climate, nitrogen availability, and soil conditions, making them often impractical for many farms. This highlights a critical need for more efficient and data-light alternatives that can still provide reliable predictions.
A recent research paper, “Applying Time Series Deep Learning Models to Forecast the Growth of Perennial Ryegrass in Ireland,” by Oluwadurotimi Onibonoje, Vuong M. Ngo, Andrew McCarren, Elodie Ruelle, Bernadette O’Brien, and Mark Roantree, addresses this challenge by proposing deep learning models for forecasting grass growth. The study focuses on univariate datasets, meaning they primarily use historical grass height data, significantly reducing the data acquisition overhead compared to multivariate approaches.
The researchers explored several deep learning models, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Multi-Layer Perceptron (MLP), and Temporal Convolutional Networks (TCN), alongside the traditional AutoRegressive Integrated Moving Average (ARIMA) model as a baseline. The goal was to identify which model could most accurately predict the growth of Perennial Ryegrass, a common grass type in Ireland.
The study utilized a comprehensive dataset collected weekly by Teagasc, the Animal & Grassland Research and Innovation Centre, in County Cork, Ireland. This dataset spans an impressive 34 years, from 1982 to 2015, covering 1,757 weeks of grass growth height. County Cork experiences a Marine West Coast Climate, characterized by moderately warm summers and mild winters. The data clearly showed that grass growth typically peaks from mid-spring to summer (April to August), with May and June recording the highest average heights.
After extensive experimentation and hyperparameter tuning, the Temporal Convolutional Network (TCN) emerged as the top performer. It achieved the best results in forecasting Perennial Ryegrass growth in Cork, demonstrating high accuracy with an RMSE of 2.74 and an MAE of 3.46. This performance significantly surpassed that of the other models, including ARIMA, LSTM, GRU, and MLP, in terms of RMSE and MAE.
Interestingly, the study found that simpler architectures often performed better. For TCN, MLP, LSTM, and GRU, a single-layer architecture was generally optimal. The ideal input sequence length varied slightly by model, with TCN and MLP performing best using data from the preceding two weeks, while LSTM and GRU benefited from the preceding three weeks. While TCN required longer training times due to its complexity, its superior accuracy highlights its potential.
The implications of this research are substantial for the Irish dairy sector. Accurate grass growth forecasts can empower farmers to make better decisions regarding resource utilization, optimize grazing management, and reduce their reliance on costly external feed. This is particularly crucial given the anticipated impacts of climate change, which are expected to introduce greater variability in grass growth patterns, demanding more reactive and flexible farming practices.
Also Read:
- Enhancing Shipping Container Monitoring with Adaptive Machine Learning
- Streamlining Attention for Time Series Forecasting with Entropy Equality
By providing a cost-effective and robust method for forecasting grass growth, this study contributes significantly to advancing sustainable dairy farming practices in Ireland. The insights gained into model behavior and optimal configurations enhance the reliability of these forecasts, paving the way for more informed agricultural decisions. For more detailed information, you can read the full research paper here.
