What 400 Turfgrass Researchers Are Working on in 2024

I analyzed over 200 abstracts presented at last month’s 2024 ASA-CSSA-SSSA Annual Meeting, and used natural language processing to show what turfgrass researchers are talking about. Here’s what I found.

Top turfgrass research areas in 2024

The research community is working on a wide range of subjects, but here are the top 5 themes:

1. Remote sensing and precision management · 13% of abstracts

Advanced imaging and remote sensing technologies, machine learning, precision through data, predictive modeling, and high-throughput analytics

2. Athletic field and golf surface properties · 8% of abstracts

Surface performance of athletic fields and golf courses, sports field safety, biomechanical data, traffic, hybrid systems, and synthetic surfaces

3. Light and shade · 7% of abstracts

Shade and light quality, impacts of foliar shade, light intensity, and R:FR ratios

4. Soil amendments and organic matter · 7% of abstracts

Organic amendments, compost and biochar, soil health, microbial activity, water retention, carbon sequestration, and nutrient mineralization

5. Fungicides and disease management · 6% of abstracts

Fungicide efficacy, application timing models, integrated strategies to control disease, synthetic and biological fungicides, resistance management

Behind the data

The ASA-CSSA-SSSA Annual Meeting is the largest academic conference for turfgrass science in North America, where university and industry scientists share their latest work. The turfgrass science division (known as C-5) is part of the Crop Science Society of America (CSSA). Each presentation includes a title and abstract, and they’re all publicly available in the conference program.

The conference platform has an API that makes it easy to fetch all abstracts in a few minutes. I got a list of the C-5 abstract ids, and then grabbed the associated titles, abstracts, and authors.

# Get all C-5 paper ids
program_data = requests.get("https://scisoc.confex.com/scisoc/2024am/meetingapi.cgi/Program/2161").json()
paper_ids = [p.split('/')[1] for p in program_data['ChildList_Paper']]

# Fetch each paper's data
papers = []
for pid in tqdm(paper_ids):
    paper = requests.get(f"https://scisoc.confex.com/scisoc/2024am/meetingapi.cgi/Paper/{pid}").json()
    papers.append({
        'title': paper.get('Title', ''),
        'abstract': paper.get('Abstract', ''),
        'authors': paper.get('EntryCitation_apa', '').split('(')[0]
    })
    time.sleep(0.5)  # Be nice to the server

This year there were 201 unique abstracts from 391 authors.

With a little data cleaning, I created a dataset that looks like this:

I used topic modeling to discover the topic clusters in this collection of abstracts, without having to read them all myself. We could easily do this for thousands of abstracts if we wanted to.

Discovering research clusters

First, I turned these unstructured texts into embeddings. This basically means I turned them into numerical representations with hundreds of dimensions. The embeddings capture the semantic meaning of each abstract, and I used this information to determine how similar abstracts are. I used SPECTER, an embedding model developed specifically for scientific papers.

BERTopic is the algorithm that I used to cluster topics. The T in BERT stands for transformers, which is also the T in ChatGPT. I set min_cluster_size=4 so each topic has to have at least 4 abstracts, and ngram_range=(1, 3) so that 1 to 3 word phrases are included in topic representation. This can capture more meaningful concepts like “dollar spot” instead of just individual words.

The model generated 20 topics. The table below shows topic numbers, count of abstracts in that topic, and the proportion of total abstracts. The Representation column is the top 10 words that represent that topic. The last column is the CustomName I gave to the topic based on the key words. Topic -1 is a catch-all for abstracts that don’t fit well into any of the other topics.

Some topics are well-defined and reflect major research priorities like remote sensing [accuracy, imagery, precision, sensing] and surface properties of golf and athletic fields [traffic, gs3, athletic, surface]. The model captured 3 different breeding topics, including genomic tools, mutation protocols, and breeding for specific traits like salt tolerance. If combined, these breeding-related topics would represent one of the top research areas. Newer areas of study like bee lawns and carbon sequestration show up, even though these represented smaller portions of the total research.

The smaller topics toward the bottom (those with 4 abstracts) sometimes show unexpected word combinations. Like Topic 16 consisting of [tallgrass, prairies, kikuyugrass, thatch]. These abstracts happened to share certain words rather than representing a truly cohesive research theme. This can happen with topic modeling when applied to smaller document sets, and the groupings can be useful starting points to explore rather than distinct research themes.

Visualizing topics and words

Science is easier to understand with visuals. We can visualize the topics, documents, and words in different ways. The plot at the top of the page shows clusters of documents in a 2-dimensional space. Each dot represents one abstract, and dots that are closer together are abstracts about similar subjects. On desktop, you can hover over any point to read the abstract title.

The bar chart below shows how important and distinctive each word is to a topic. Higher scores (longer bars) indicate that the word is more uniquely associated with that topic. For example, in Topic 2, shade has the highest score, meaning it’s the most distinctive and important word for that topic. The other words like [light, night, foliar shade] have lower scores but are still important terms that define that topic. This topic was definitely about lighting conditions and turfgrass growth in shade.

Insights about turfgrass research trends

While the clustering isn’t perfect (about 70% of abstracts were classified with high confidence), it points to what is going on in turfgrass research.

• Remote sensing and computer vision applications represent the largest research category this year, aligned with tech adoption being a hot topic in turfgrass management
• Athletic field and golf surface properties remain a top research area, which makes sense given how important playing surface quality is to the turfgrass industry
• The conference had a special symposium about carbon accounting in turfgrass systems, and the model picked up on this as its own topic cluster
• There is significant research happening with the WinterTurf project about winter stresses on turfgrass, but those abstracts are distributed across other topics instead of forming their own cluster
• While herbicides and fungicides are always a focus, there’s interesting work happening on biostimulants and alternative approaches
• Breeding and genetic tools have become sophisticated—way beyond crossing plants and selecting the best ones

Most of this research is still in progress, so keep an eye on these topics in the coming years—they’ll shape how we manage turfgrass.