The turf industry has embraced various technologies to improve irrigation efficiency. Hand-held moisture meters have replaced pocketknives. Mower-mounted sensors help turf managers detect soil moisture variability, while in-ground sensors relentlessly monitor changes over time. Now, machine-learning optimization—a type of AI that powers GreenKeeper Insight—can be used to accurately predict soil moisture changes days in advance.
Recently, a good friend and GreenKeeper user called BS on this claim. While it sounds outlandish, the data doesn’t lie. GreenKeeper learns how your turf uses water and can accurately predict when wilt will occur.
How Can AI Help Predict Soil Moisture?
Turfgrass systems lose water through the process of evapotranspiration (ET) and gain water through natural precipitation and irrigation. These processes are governed by many factors, including weather conditions, shade, site slope/exposure, soil texture, irrigation system design, grass species, mowing height, and more. Managing these variables is complex, yet experienced turf managers develop an intuitive understanding of their site over time.
Hand-held moisture meters have accelerated this learning process. These tools precisely measure soil moisture content—a key metric that can be tracked, recorded, and managed. Measuring moisture when turf begins to wilt helps determine a site’s specific wilt point. Taking readings throughout the day reveals how quickly a site dries under different weather conditions. Measuring again the next morning demonstrates how irrigation affects soil moisture content.
Hand-held moisture meters don’t just help humans learn—they help machines learn, too. GreenKeeper App already has access to site-specific information, including grass species, soil texture, and hourly weather data, to estimate ET. In fact, hourly ET is estimated using an ensemble of 40 different ET models. By pairing these data with daily soil moisture measurements and irrigation runtimes, the machine-learning algorithm in GreenKeeper Insight can identify additional factors affecting soil moisture fluctuations. These insights are then combined with forecast ET—GreenKeeper predicts reference ET up to 14 days into the future—to model soil moisture changes with impressive accuracy.
Validating GreenKeeper Insight on a Golf Course Green
The GreenKeeper Insight algorithm was firest validated on the sixth green at the Country Club of Lincoln. This bentgrass putting green, maintained to the highest standards, sits on a modified push-up root zone in the heart of Lincoln, NE.
Each morning, we measured the average soil moisture using 12 to 18 readings from a Spectrum TDR 350 with 3-inch tines in Sand mode. We also recorded the previous night’s irrigation runtime and entered both values into GreenKeeper App’s Water Resource Planner. (Afternoon hand watering was not accounted for in our model.) The experiment was conducted from August 4 to September 6, 2024.
In the initial model training phase (gray rows in the data table), the Water Resource Planner used standard estimates for variables like crop coefficient and irrigation output to predict soil moisture for the following day. Users can manually adjust some of these factors in the Settings section of the Water Resource Planner.
After several days of consistent data entry, GreenKeeper Insight began optimizing its predictions using machine learning (white rows in our dataset). The model continued to improve in accuracy with regular soil moisture measurements.
How Accurate Were the Predictions?
After just 10 days, GreenKeeper Insight dramatically improved the accuracy of soil moisture predictions. Between August 14 and September 6, the average difference between estimated and actual soil moisture was only 0.5% volumetric water content (VWC). On most days, the difference was less than 0.3%. In comparison, before GreenKeeper Insight optimization, the average difference was 1.9% VWC over the same period.
One standout day was August 14:
-
- Measured soil moisture: 25.5%
- GreenKeeper Insight-predicted soil moisture: 25.2% (only 0.3% off)
- Standard model prediction (without AI): 19.8% (off by 5.7%)
Another key observation occurred on September 3 after the model ran for 120 hours (5 days!) without adjustment.
-
- August 29 measured soil moisture: 17.8%
- September 3 measured soil moisture: 19.4%
- GreenKeeper Insight-predicted soil moisture: 19.4% (exact match)
- Standard model prediction: 14.8% (off by -4.8%)
There were two outlier days (August 21 and September 5) when measured soil moisture was higher than projected by both models. Possible causes include:
- Significant afternoon hand watering the previous day added unexpected moisture.
- A few unusually low readings skewed the measured average one day, resulting in an over-correction the next day
GreenKeeper Insight Optimization is Not Static
GreenKeeper Insight continuously evolves. Key variables—such as sun angle, shade presence, day length, and effective root depth—change throughout the season. The AI adapts by prioritizing recent soil moisture data for model optimization.
To trigger machine learning, consistent data entry is required. A single measurement without other recent data points is ignored. Regular sampling ensures the model remains optimized.
What’s Next for GreenKeeper Insight?
Machine learning in GreenKeeper Insight is set to revolutionize multiple agronomic models within GreenKeeper. Our goal is to scale the Water Resource Model across all irrigation zones on your course or field.
We are also working on:
- Integration with Performance Tracker – Let your staff input data effortlessly.
- Connections to SPIIO and Soil Scout – These in-ground soil moisture sensors will automate data collection and expedite training.
- Expanding AI-driven predictions – GreenKeeper Insight will soon help predict:
- Clipping volume trends
- Soil organic matter changes
- PGR rate recommendations
- Performance metrics (green speed, firmness, etc.)
Data Entry Discipline: A Small Effort with Big Rewards
Incorporating data collection into your daily routine doesn’t have to be time-consuming. Start simple and small, but stay consistent.
For example:
- Measure clipping volume, average soil moisture, and green speed from one representative green daily.
- Choose the first green you mow each morning—when mower buckets are clean and reels are sharp—for consistency.
- Enter these values into GreenKeeper, along with the previous night’s irrigation runtime.
- These two minutes of data collection allow GreenKeeper to predict soil moisture changes, optimize irrigation needs to prevent wilt,t rack fertilizer removal through mowing, and schedule PGR applications, mowing, and rolling more efficiently
It’s better to spend 2 minutes measuring one area daily than 40 minutes measuring all greens just a few times per week.
GreenKeeper App: Smarter Turfgrass Management
GreenKeeper App helps turfgrass managers save time, save money, and make more confident decisions. Our software now has a labor management feature – GreenKeeper WhiteBoard – to assign jobs, track labor, and customize work programs.
Start your subscription today at GreenKeeperApp.com.