Could Solar Power and Weather-Smart Drones Solve Farm Sensors’ Battery Life Problem?

Researchers Nada Ahmed, Asmaa N. Ghareeb, Yasser Fouad and Enas Selem tested the design through computer simulations, finding a substantial increase in the time before the first sensor failed. Their results reveal how much that improvement depends on sunlight, network density and the decisions built into the system.

A farm sensor running out of battery can leave a gap in the information growers need to understand their fields, and replacing batteries across a large or hard-to-reach site can become a demanding job. A study published in Scientific Reports, titled "EH-SWADS: energy-harvesting smart weather-aware drone sink for agricultural WSNs," explores a system that combines solar-powered sensors, drone-based data collection and weather prediction to keep agricultural monitoring networks working longer.

Researchers Nada Ahmed, Asmaa N. Ghareeb, Yasser Fouad and Enas Selem tested the design through computer simulations, finding a substantial increase in the time before the first sensor failed. Their results reveal how much that improvement depends on sunlight, network density and the decisions built into the system.

How sunlight and a drone help sensors stay connected

EH-SWADS builds on the team's earlier SWADS architecture by adding solar energy harvesting to sensors with rechargeable batteries. Small photovoltaic panels replenish some of the energy spent on sensing, processing and communication, reducing dependence on the charge available at installation.

Sensors are organised into groups, with a selected "cluster head" gathering nearby readings and forwarding them to a collection point. A drone serves as a mobile collection point, shortening communication distances, and a fixed ground station provides an alternative when weather conditions trigger a handover.

A weather-prediction model known as an LSTM uses sequences of meteorological readings to guide that choice. Reinforcement learning helps assign the demanding cluster-head role by considering battery charge, distance, communication workload, available solar energy and previous responsibility for the job.

What the simulations actually found

The researchers simulated 200 sensors across a 100-by-100-metre field for 20,000 communication rounds, using a real, changing weather record. Each sensor began with 0.75 joules of energy. The matched system without solar harvesting experienced its first sensor failure at round 841. EH-SWADS delayed that point to round 6,458, an increase of approximately 668%, or about 7.7 times as many rounds before the first failure.

Total successful data delivery reached 885,202 packets with EH-SWADS, compared with 457,574 without harvesting, giving the solar-powered design roughly 1.93 times the cumulative throughput. EH-SWADS delayed the first failure by approximately 43% against LEACH, 32% against HEED and 9% against MCSOC, an approach that already accounts for energy harvesting. These comparisons used simplified implementations, including a reduced version of MCSOC's search process, which limits how broadly the rankings can be applied.

Weather-responsive collection contributed to the result: keeping the collection point fixed and disabling weather-triggered handovers reduced the first-failure point from round 6,458 to 4,392. Reported delays were similar, and the delay model mainly represented radio propagation rather than the full range of delays a working farm network would encounter.

EH-SWADS recorded more lost packets in absolute terms, reflecting greater transmission activity rather than a higher failure probability for each attempt. The main comparison used a fixed 5% packet error rate; the clear benefit was a longer productive life and more data delivered.

Why the extra learning features were not the main breakthrough

Solar charging was the main reason the sensors kept working longer, and adding more complicated rules to the learning system did not always improve its performance. An early version gave too much importance to small differences in the energy sensors collected when sunlight was weak. It could choose a sensor with a nearly empty battery to lead its group simply because that sensor had done the job less often.

The researchers corrected these problems by measuring harvesting against a fixed sunny-weather reference and linking the rotation reward to remaining battery charge.

The corrected version performed roughly 4.4% worse in one randomised run and 5.2% better in another than a simpler reward formula that ignored the extra harvesting and rotation terms. Those tests showed no clear advantage from the added terms, even though correcting them removed the substantial disadvantage seen earlier.

The weather model achieved 97.97% accuracy across 10,010 test samples. Performance was uneven: it recognised rain, snow and sun with recall above 98%, compared with about 31% for fog and 74% for drizzle, two categories representing less than 1% of the test set.

Experiments that randomly corrupted 15% or 30% of weather labels unexpectedly improved simulated lifetime. The researchers attributed this to interruptions in sustained adverse-weather sequences within the model. This finding does not establish that inaccurate forecasts help real deployments; consistently missing an approaching storm could produce a very different outcome.

What still stands between the model and a working farm

EH-SWADS reached half-network failure at round 6,591, just 133 rounds after losing its first sensor. Attempts to reduce frequent handovers and adjust activity during energy shortages did not resolve that pattern, leaving its cause open for investigation.

Under constant sunny conditions, the first failure occurred at round 13,322, compared with 5,604 under cloudy skies, 3,328 in rain and 2,751 in storms. The storm scenario combined no solar harvesting with greater communication costs and a higher packet error rate. The 668% improvement measured with 200 sensors fell to approximately 22% with 500 sensors and 31% with 1,000. Denser networks offered shorter communication routes and more neighbouring sensors, reducing dependence on the benefit provided by harvesting.