Estimating Pasture Biomass from Top-View Images: A Dataset for Precision Agriculture
Qiyu Liao, Dadong Wang, Rebecca Haling, Jiajun Liu, Xun Li, Martyna Plomecka, Andrew Robson, Matthew Pringle, Rhys Pirie, Megan Walker, Joshua Whelan · 2025
Read Estimating Pasture Biomass from Top-View Images: A Dataset for Precision Agriculture by Qiyu Liao, Dadong Wang, Rebecca Haling, Jiajun Liu, Xun Li, Martyna Plomecka, Andrew Robson, Matthew Pringle, Rhys Pirie, Megan Walker, Joshua Whelan (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-22916.txt.
Introduction
Estimating pasture biomass helps farmers set stocking rates well. Qiyu Liao leads a team with ten co-authors. In 2025 they release a dataset for that task. It holds top view pasture images from Australia. Nineteen locations and multiple seasons are covered.
Each image shows a small marked quadrat. Ground measures pair with each image. Biomass is sorted into green, dead, and legume parts. Vegetation height is also recorded. A vegetation index from optical sensors is included. The set holds over a thousand annotated images. It is hosted as a Kaggle competition. The task is biomass estimation from images. Precision grazing management is the target use.
Catalog record
| CREATOR | Qiyu Liao, Dadong Wang, Rebecca Haling, Jiajun Liu, Xun Li, Martyna Plomecka, Andrew Robson, Matthew Pringle, Rhys Pirie, Megan Walker, Joshua Whelan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22916
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Estimating Pasture Biomass from Top-View Images: A Dataset for Precision Agriculture Qiyu Liao, Dadong Wang, Rebecca Haling, Jiajun Liu, Xun Li, Martyna Plomecka, Andrew Robson, Matthew Pringle, Rhys Pirie, Megan Walker, Joshua Whelan Abstract Accurate estimation of pasture biomass is important for decision-making in livestock production systems. Estimates of pasture biomass can be used to manage stocking rates to maximise pasture utilisation, while minimising the risk of overgrazing and promoting overall system health. We present a comprehensive dataset of 1,162 annotated top-view images of pastures collected across 19 locations in Australia. The images were taken across multiple seasons and include a range of temperate pasture species. Each image captures a 70cm × 30cm quadrat and is paired with on-ground measurements including biomass sorted by component (green, dead, and legume fraction), vegetation height, and Normalized Difference Vegetation Index (NDVI) from Active Optical Sensors (AOS). The multidimensional nature of the data, which combines visual, spectral, and structural information, opens up new possibilities for advancing the use of precision grazing management. The dataset is released and hosted in a Kaggle competition that challenges the international Machine Learning community with the task of pasture biomass estimation. The dataset is available on the official Kaggle webpage: https://www.kaggle.com/competitions/csiro-biomass 1 1 footnotetext: Corresponding authors: {qiyu.liao}@csiro.au 2 2 footnotetext: Data Availability and Acknowledgement: The Image2Biomass Pasture Innovation Challenge dataset is derived from data originally captured under ‘B.GSM.0010 - Tools for real time biomass estimation in pastures’ . The authors acknowledge the support of FrontierSI (previously known as the Cooperative Research Centre for Spatial Information) and matching funds provided by the Australian Government for this project. The annotated dataset discussed here is publicly available under a Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license through the Kaggle competition platform. 1 Introduction Accurate estimation of pasture biomass is essential for profitable and sustainable grazing management. It supports agricultural productivity while preserving ecosystem integrity. Currently, producers are still heavily reliant on subjective visual assessments to optimise feed allocation efficiency. Inefficiencies resulting from suboptimal management practices can result in significant economic losses and contribute to environmental degradation, including soil erosion, biodiversity loss, and carbon depletion [ 1 ] . Research shows that improved pasture management through accurate biomass estimation can boost farm profitability by up to 10% in Australian beef and sheep enterprises. Based on reported pricing in 2018, this translates to gross margin increases of approximately $96 per hectare for sheep and $52 per hectare for cattle operations [ 2 ] . In addition to economic benefits, optimized grazing pressure also creates a twin benefit of improved environmental outcomes [ 2 ] . The challenge of biomass measurement is global in scope. Grazing systems cover roughly 25% of the Earth’s land surface (approximately 3.4 billion hectares) [ 3 ] and manage about 50% of Australia’s landmass (380 million hectares) [ 4 ] . These vast grassland ecosystems play a vital role in carbon sequestration, water cycle regulation, and biodiversity conservation. Traditional biomass measurement methods [ 5 , 6 , 7 , 8 , 9 , 10 ] are typically destructive, time-consuming, expensive, or require specialized expertise, limiting their widespread adoption in sustainable grazing systems. Recent advances in computer vision and artificial intelligence offer transformative opportunities to overcome these limitations. By automating the analysis of pasture images, AI models can estimate biomass more precisely and even at the species level. This reduces the need or effort required for manual calibration while providing an enabling data layer for assessment of pasture genotypic composition and phenological stages – allowing for more accurate derivation of quality attributes that support both livestock nutrition and ecosystem health. Put simply, accurate, real-time biomass estimation at a species level could facilitate adaptive grazing strategies that help to maintain optimal pasture conditions, prevent overgrazing, and support natural regeneration cycles. This paper introduces a novel dataset specifically designed to advance AI-driven pasture biomass estimation using top-view imagery (shown in Figure 1 )). This contribution addresses a critical gap in publicly available, professionally annotated datasets that integrate visual, spectral, and structural information for pasture assessment. This enables the development of technologies that support both agricultural productivity and environmental sustainability. Figure 1: Sample images from the training set and their labels. Images cropped to show the internal section of the 70 x 30cm quadrants, and with dry weight of biomass fractions shown (Green, Dead, Clover). 2 Related Work 2.1 Traditional Biomass Measurement Methods Current approaches to estimating pasture biomass have significant limitations that restrict their practical applications. Destructive methods such as clipping and weighing (cut-dry-weigh method) provide high accuracy through direct measurement but are labor-intensive, time-consuming, and impractical for large-scale or frequent monitoring [ 5 ] . Non-destructive mechanical methods include the rising plate meters, which measure compressed pasture height but require calibration for different pasture types and show variable accuracy with pasture density and moisture content [ 6 ] , and capacitance meters that measure electrical capacitance of vegetation but require complex, pasture-specific calibration with inconsistent results across different conditions [ 7 ] . Optical and spectral methods have gained attention because of their potential to provide rapid, nondestructive assessments. Active Optical Sensors (AOS), originally developed for cropping industries to infer nitrogen levels, measure vegetation indices like NDVI by directing light beams onto the canopy and calculating optical reflectance [ 8 , 9 , 11 ] . Research has shown that combining NDVI measurements with height data from LiDAR or mechanical sensors can significantly improve the accuracy of biomass estimation, addressing limitations such as NDVI saturation at high biomass levels, typically when Leaf Area Index (LAI) > 3, and height measurement errors caused by lodging or sparse canopies [ 10 ] . However, the primary challenge remains the development of robust calibration methods suitable for diverse pasture systems [ 2 ] . Remote sensing approaches using satellite and drone imagery, combined with machine learning, enable large-scale biomass estimation. However, they require technical expertise, significant computational resources, and extensive ground-truth calibration data [ 12 ] . 2.2 Computer Vision Approaches and Existing Datasets Recent advances in computer vision have shown promising results for agricultural applications, with deep learning models demonstrating success in crop yield prediction and vegetation analysis. However, pasture systems present unique challenges including species diversity, structural complexity, and environmental variability that make them more challenging than monoculture crops. The most relevant prior work is the GrassClover dataset by Skovsen et al. [ 13 , 14 ] , which provides 435 images with biomass annotations for grass-clover mixtures. However, the Image2Biomass Pasture Innovation Challenge dataset presented here has several important differentiators compared to the GrassClover dataset. This dataset is larger in scale, offering 1,162 annotated images versus 435, and has additional metadata (species identification, seasonal context, location details, height measurements, and NDVI data). In terms of temporal coverage, the GrassClover dataset is concentrated between May and October, whereas our dataset spans the entire year. Geographically, it is restricted to two locations in Denmark, while our dataset covers 19 sites across four Australian states, representing a wider range of climatic conditions. Additionally, the Image2Biomass Pasture Innovation Challenge dataset includes separate annotations for dead matter. Finally, while the GrassClover images were captured using professional cameras under controlled lighting conditions, our dataset uses diverse