Agro-Volta

Inspiration

Agrivoltaics offers a way to combine food production and renewable energy on the same land, but solar panels also change the crop microclimate. Changes in temperature, humidity, radiation, and soil moisture can affect how much and when crops need irrigation. We were inspired to build a system that doesn't simply monitor these changes, but understands them and turns them into actionable irrigation decisions.

What it does

Agro-Volta is an IoT and AI-powered irrigation decision-support system for agrivoltaic farming.

Monitors the open-field and under-panel microclimates simultaneously.

Measures temperature, humidity, pressure, soil temperature, soil moisture, light, and spectral conditions.

Calculates reference evapotranspiration (ET₀) and crop water requirements.

Uses Growing Degree Days (GDD) to account for crop development.

Uses Bi-LSTM to forecast soil moisture 24 hours ahead.

Uses SHAP to explain which environmental factors influence the prediction.

Converts the predictions and crop-water information into three practical actions:

IRRIGATE

DELAY

HOLD

Aims to reduce unnecessary irrigation while maintaining suitable conditions for crop growth.

How we built it

We built the system around an ESP32-S3-based IoT sensor platform.

The hardware integrates:

BME280 for air temperature, humidity, and pressure

DS18B20 for soil temperature

Capacitive soil-moisture sensors

BH1750 sensors for light measurements

AS7341 for spectral measurements

MicroSD for local data storage

Wi-Fi/MQTT for data communication

The system collects synchronized data from the open-field reference zone and agrivoltaic zone.

On the software side, we developed a pipeline combining:

IoT sensing → data processing → ET₀ → GDD → soil-moisture forecasting → SHAP → irrigation decision

The AI component compares Bi-LSTM performance with simpler forecasting approaches, including a persistence baseline, rather than assuming that a more complex model is automatically better.

Challenges we ran into

One of our biggest challenges was understanding that agrivoltaic irrigation cannot simply use conventional open-field assumptions.

Other challenges included:

Building a reliable multi-sensor IoT node.

Calibrating soil-moisture and light measurements.

Handling noisy and missing sensor data.

Synchronizing measurements from two experimental zones.

Integrating physical agricultural models with machine learning.

Designing an irrigation decision system rather than stopping at prediction.

Making the AI explainable instead of treating it as a black box.

Balancing research-level methodology with hardware and time constraints.

Accomplishments that we're proud of

We are particularly proud that Agro-Volta goes beyond being "just an automatic irrigation system."

Our system connects:

Renewable energy → Agrivoltaic microclimate → Agricultural physics → IoT → AI → Explainability → Irrigation decisions

We also designed the experiment around a dual-zone comparison, allowing us to quantify how the agrivoltaic environment differs from an open-field reference.

Most importantly, the AI doesn't simply say:

"Soil moisture tomorrow = X."

It attempts to answer:

"Should we irrigate, delay irrigation, or hold—and why?"

What we learned

We learned that building an agricultural AI system requires much more than training a neural network.

We learned the importance of:

Experimental controls and counterfactual comparisons.

Sensor calibration and data quality.

Agricultural physics before machine learning.

Strong baseline models for evaluating AI.

Crop development and phenology in irrigation modelling.

Explainability when AI decisions affect agricultural operations.

Designing research around measurable scientific questions, rather than simply adding more technology.

The biggest lesson was that prediction alone is not enough. The real value comes from converting reliable predictions into decisions that can be evaluated in the real world.

What's next for Agro-Volta

Our next step is to move from the prototype and modelling framework toward full field validation.

Grow and evaluate cowpea in both experimental zones.

Collect crop phenology, growth, biomass, and irrigation data.

Validate the GDD-based crop-development framework against observed crop stages.

Validate AI soil-moisture forecasts under real agrivoltaic conditions.

Evaluate irrigation recommendations against conventional scheduling.

Quantify irrigation reduction while monitoring crop performance.

Develop automated pump/valve control for closed-loop irrigation.

Expand the system to multiple crops and different PV configurations.

Build a scalable dashboard for real-time monitoring, historical analysis, AI explanations, and irrigation recommendations.

Our long-term vision: make agrivoltaic farming not only a way to produce food + electricity, but a system where renewable energy, environmental sensing, AI, and precision irrigation work together to use every drop of water more intelligently.