Improving Early Detection of Winter Damage on Golf Greens

A research team led by Ce Yang at the University of Minnesota has developed a new artificial intelligence (AI) framework that significantly improves early detection of winter damage in turfgrass, an issue that complicates maintenance planning for golf courses and other managed turf systems (1). The study, published in the journal Computers and Electronics in Agriculture, introduces a combined data-generation and image-segmentation pipeline designed to identify faint, early-stage injury that existing tools routinely miss (1).

Turfgrass systems are commonly found in golf courses, sports fields, and home lawns (2). These systems are particularly ideal in hotter climates because of their ability to provide a cooling effect (2). In addition to being visually appealing, turfgrass systems also have numerous ecosystem benefits, including reducing the force of flowing water, which results in more water being trapped in the soil (3). This, in turn, helps fill the groundwater reserves (3). Other benefits of turfgrass systems include helping to prevent erosion, replenish air, and support bioremediation (3).

Over the winter months, however, turfgrass systems normally see some damage in the form of subtle discoloration or low-contrast patches. In this study, the researchers proposed a new method that could identify early-stage injury to turfgrass systems and explained why their method improves upon existing techniques.

Yang and her team used a conditional deep convolutional generative adversarial network (cDCGAN) to generate synthetic vegetation index (VI) maps, including NDVI and NDRE. Vegetation indices provide more stable representations of plant health than raw spectral data, helping reduce noise and increase the diversity of training data sets (1). By generating high-fidelity synthetic VI imagery, the team compensated for limited annotated field data, which is an ongoing barrier in precision turfgrass management.

Apart from cDCGAN, the researchers also used a Transformer-based segmentation model equipped with a newly designed adaptive attention decoder (AAD) to great effect. The decoder refined the multi-scale features, enabling the model to better recognize the irregular shapes and weak signals associated with early winter injury (1). According to the researchers, this targeted architectural improvement is critical for distinguishing true damage from minor fluctuations in healthy turf (1).

Using golf course putting greens in central Oregon, the researchers tested their model. Their system achieved an 82.47% mean Intersection over Union (mIoU), 97.85% accuracy, 85.62% recall, and an F1-score of 88.30%, which outperformed established deep learning architectures such as U-Net and DeepLabV3+ (1).

For golf course superintendents and turf managers, earlier and more reliable detection could translate directly into operational and financial benefits. Winter injury can delay spring green-up, disrupt tournament schedules, and require costly re-sodding or renovation (1). Tools that detect damage even days or weeks earlier offer opportunities to intervene before injuries spread or worsen (1).

The researchers emphasized that the framework is designed with practical deployment in mind (1). Using VI maps allows the system to work with data collected from commonly used multispectral cameras, and the synthetic augmentation helps reduce the need for time-intensive manual labeling. This combination could make advanced damage detection accessible to facilities that lack large historical data sets or specialized imaging expertise (1).

The study also includes ablation experiments showing that both the synthetic VI augmentation and the AAD independently boost performance, but they deliver the greatest benefit when combined. The authors state that this demonstrates the value of pairing domain-specific data generation with targeted model improvements (1).

As climate variability increases the frequency and severity of winter stress events, the turfgrass industry faces mounting pressure to adopt more predictive, data-driven management tools. The authors stated that the framework they built and tested could improve turfgrass management.

“Overall, this research presents a problem-driven framework that integrates targeted data augmentation with an improved segmentation architecture, offering a robust and accurate solution for early detection of winter damage in precision turfgrass management,” wrote the authors in their article (1).

References

1. Li, X.; Sanaeifar, A.; Padilla, N.; et al. Winter Damage Diagnostic Modeling Based on Synthetic Vegetation Indices from UAV-based Multispectral Imaging. Com. Elect. Agric. 2026, 243, 111334. DOI: 10.1016/j.compag.2025.111334
2. Braun, R. C.; Mandal, P.; Nwachukwu, E.; Stanton, A. The Role of Turfgrasses in Environmental Protection and Their Benefits to Humans: Thirty Years Later. Crop Sci. 2024, 64 (6), 2909–2944. DOI: 10.1002/csc2.21383
3. U.S. National Park Service, The Benefits of Sustainably Managed Turf. NPS.gov. Available at: https://www.nps.gov/articles/000/benefits-of-sustainably-managed-turf.htm (accessed 2026-01-13).