Most irrigation systems in the ground today run on timers. A controller fires valves according to a schedule that someone set weeks, months, or years ago. It waters Tuesday at 4am for 12 minutes whether the soil needs it or not.
This approach has one advantage: simplicity. And one critical flaw: it has no idea what's actually happening in the soil.
That number comes from decades of field research. The EPA WaterSense program cites 50% waste as typical for poorly maintained systems. Even well-maintained timer systems lose 25-35% because they fundamentally cannot adapt to changing conditions.
Where Does the Water Go?
The waste doesn't happen in one place. It's the accumulation of several independent inefficiencies:
A timer-based system can't fix any of these because it doesn't know they're happening. It has no sensors, no weather data, and no model of how water moves through your soil.
The Three Levels of Irrigation Intelligence
Not all "smart" irrigation is created equal. There's a spectrum:
| Timer-Based | ET-Adjusted | Physics-Based (AI) | |
|---|---|---|---|
| Knows soil moisture? | No | Estimated | Measured + Modeled |
| Adjusts for rain? | No | Yes (shutoff) | Yes (predictive) |
| Zone-by-zone? | Fixed runtimes | % adjustment | Individual optimization |
| Handles uncertainty? | No | No | Stochastic optimization |
| Checks hydraulics? | No | No | Full pipe network model |
| Typical efficiency | 50-55% | 70-75% | 85-90% |
What Physics-Based Optimization Actually Does
Here's what happens when a physics-based optimizer plans tomorrow's irrigation:
1. It models water movement through your actual soil
Using the Richards equation (the standard PDE for unsaturated flow), the optimizer simulates how water moves through each zone's soil profile. It knows that sandy loam drains in hours while clay holds water for days. It accounts for root depth, slope, and how the soil properties change with depth.
2. It predicts crop water demand
Using the Penman-Monteith equation with actual weather forecast data (temperature, humidity, wind, solar radiation), it calculates exactly how much water each zone will lose to evapotranspiration over the next 24-48 hours.
3. It finds the optimal schedule
Using gradient-based optimization (the same math behind machine learning), it searches for the schedule that minimizes a multi-objective cost function:
- Crop stress: Keep soil moisture above the stress threshold
- Water use: Don't apply more than plants can use
- Energy cost: Minimize pumping energy
- Hydraulic feasibility: Don't run more zones than the pipe network can handle
4. It checks if the schedule is physically possible
Most scheduling tools ignore a critical question: can the pipe network actually deliver the water? If too many zones run simultaneously, pressure drops and sprinklers under-perform. The optimizer runs a hydraulic simulation (Newton-Raphson solver on the pipe network) and adjusts the schedule to ensure every zone gets adequate pressure.
5. It hedges against forecast error
Weather forecasts are uncertain. Instead of trusting a single forecast, the optimizer simulates dozens of possible weather outcomes and finds a schedule that works well across all of them. It specifically penalizes the worst-case scenarios (using CVaR - Conditional Value at Risk from financial mathematics) to ensure your turf stays healthy even if the forecast is wrong.
The key insight: Timer-based irrigation answers "when should the valves be open?" Physics-based irrigation answers "what do the plants actually need?" and works backward to the valve schedule.
The Numbers: A 50-Acre Golf Course
Let's make this concrete. For a 50-acre golf course in a semi-arid climate:
| Timer | AI-Optimized | |
|---|---|---|
| Daily water use | ~250,000 gal | ~140,000 gal |
| Annual water cost | ~$175,000 | ~$98,000 |
| Annual savings | - | ~$77,000 |
| Turf quality | Acceptable (overwatered areas, dry spots) | Consistent (zone-by-zone optimization) |
| Superintendent time | Manual adjustments weekly | Dashboard monitoring |
At $3.50/1,000 gallons, the savings pay for the system in under 18 months. After that, it's pure savings.
Why Now?
Three things have changed that make physics-based irrigation practical today:
- Sensor costs have dropped. Reliable soil moisture sensors that cost $500 five years ago now cost under $100.
- Computing is cheap enough. Running a Richards equation solver on a small edge computer was impossible a decade ago. Now it takes seconds.
- Weather forecast ensembles are freely available. NOAA's GEFS provides 30-member ensemble forecasts at no cost, enabling stochastic optimization.
The physics has been understood for decades (Penman-Monteith was published in 1965, van Genuchten in 1980). What's new is the ability to run these models in real-time, at the edge, for pennies of compute.
See what this means for your property
Run your numbers through our water savings calculator, or get in touch for a site assessment.
Try the Calculator Get a Site AssessmentFurther Reading
- Soil Moisture Sensor Interpreter - Understand what your sensor readings mean
- ET Calculator - See how much water your landscape loses each day
- How Differentiable Physics Powers Smart Irrigation - The technical deep-dive