Off-grid IoT: keeping a field node alive on solar and a battery

A field node with no mains power usually fails for a boring reason. The radio was specified correctly, the sensor was calibrated, and the enclosure was sealed. The node still went silent in its second winter because the energy budget was sized on the annual average instead of the worst low-sun period. The assumptions behind a solar and battery sizing, not the electronics, decide whether the node is still reporting in its second winter.

What an off-grid energy budget actually contains

An off-grid node is three components that have to be sized as one system: a solar panel, a battery, and a load that is mostly asleep. The panel collects energy during daylight. The battery carries the node through the night and through the cloudy days that follow. The load, dominated by the radio, only draws meaningful current for the few seconds it is awake to measure and transmit LoRaWAN.

The arithmetic is simple enough to do on one page. Take a node that draws 3.3 V at an average of 20 mA while awake, wakes for 10 seconds every 15 minutes, and sleeps at 10 uA between wakes. Ten seconds out of every 900 is a duty cycle of about one in ninety, so the awake current contributes roughly 0.22 mA on average. Add the 10 uA sleep current and the average is about 0.23 mA, which at 3.3 V is under 0.8 mW, or about 0.02 watt-hours per day. A single 18650 lithium cell rated at 3 Ah holds about 11 watt-hours, so in theory that node could run for well over a year on one charge. The energy-optimisation literature for off-grid agricultural IoT works through exactly this budget: panel, battery and duty-cycled load sized as one system Design of an energy-optimised IoT system for off-grid agriculture: Solar-powered monitoring and irrigation control.

The theory is not what kills nodes. What kills nodes is that the panel must replace what the battery loses every day, and the panel output in winter is much lower than the nameplate rating suggests. A 5 W panel in mid-summer at high noon can deliver close to its rating. The same panel on a short overcast winter day can deliver a small fraction of that Solar power. When the energy removed from the battery each day is bigger than the energy the panel puts back, the battery drains a little further every cycle until the node stops reporting.

The honest number is the winter one

The single most common sizing error is using an annual average irradiance figure. An annual average hides the season. A node that balances its budget on the annual average will, by construction, run a deficit through winter and a surplus through summer. If the deficit is not covered by battery headroom sized for the worst winter week, the node dies at the exact time of year when many agricultural monitoring jobs matter most, frost nights, irrigation scheduling after seasonal rain, and low-rainfall warnings.

This is where the installer enters the picture. The person who sizes the node is often the same person who sells it, and there is a quiet incentive to use the optimistic number. The nameplate panel wattage, the manufacturer's battery capacity, and the node's datasheet current are all best-case figures. None of them is the number that decides survival. The number that decides survival is the measured or conservatively estimated winter yield, derated for the site.

Derating is not optional

Three deratings matter, and they compound. The first is solar geometry and weather: winter days are shorter, the sun is lower, and overcast spells can run for days. The second is soiling. A panel lying flat or at a low angle on a farm in a dry, dusty region collects dust, and a dusty panel loses a meaningful share of its already reduced winter output. The third is temperature and ageing on the battery. Lithium cells deliver less capacity when cold, and heat accelerates their ageing; a capacity figure measured at a comfortable room temperature overstates what the battery will deliver on a cold night or after two summers in a black enclosure Design of an energy-optimised IoT system for off-grid agriculture: Solar-powered monitoring and irrigation control.

None of these is exotic. They are the difference between a node that is still reporting in its third winter and one that quietly stopped in its first. The field evidence in agricultural IoT deployments is full of the second case, though it is rarely written up, because a silent node does not send the log entry explaining why it went silent.

Duty cycle is the lever, but it has a cost

The radio is the only component the designer can trade directly. A low-power wide-area radio that transmits once an hour instead of once a minute sends 60 times fewer messages, and the sleep current of a well-chosen radio is small enough that the node's lifetime becomes a battery and panel problem, not a radio problem.

The trade is not free. A longer reporting interval means a longer gap before a problem is visible, and in a duty-cycled network, silence is the observable signal. If a node was meant to warn of a freezing pump or a failing moisture reading, an interval stretched too far turns the warning system into a system that discovers failures late. The designer should set the interval from the operational question being asked, then size the battery and panel to the worst winter interval, not the other way around.

What the backhaul actually demands

The energy budget also has to answer where the data goes. Low-power wide-area networks cover the range question well: the LoRa Alliance Smart Agriculture describes LoRaWAN as suited to agriculture because it connects sensors from a central point out to roughly 15 km, and because the sensors are battery powered. The real-world constraint is that coverage maps and catalogues are not terrain. Research on LoRaWAN gateway placement in difficult terrain found that planning with generic, site-independent channel assumptions changed the number of gateways needed and the reliability delivered Ray Tracing-Based LoRaWAN Gateway Placement for Reliable Connectivity in Amazonian Regions; the same lesson applies to a farm that sits behind a hill or in a valley.

A node that cannot reach a gateway is an energy problem with a different name. It will keep waking, transmitting into nothing, and draining its battery until the installer drives out to find it. The energy budget should include the realistic cost of a failed uplink, either by confirming coverage before deployment or by accepting that a node in a dead zone will burn its battery faster than the panel can replace it.

A worked check for a South African site

Put the pieces on one site. A frost-monitoring node in a winter-rainfall region of the Western Cape runs through its hardest period in June and July, when days are short, the sun is low, and overcast frontal weather can persist. If the budget uses a mid-summer yield to size the panel, the node enters winter with a structural deficit. The correct check is a winter week with the panel yield derated for soiling and the battery capacity derated for cold, and the interval set so the node still reports through that week with margin to spare.

The South African context sharpens this because the same country contains a dry Karoo site where soiling is the dominant loss and a coastal site where winter cloud is the dominant loss. A single national specification cannot serve both. The sizing has to be done against the site's worst season, and the site's worst season has to be named by someone who has looked at the site in that season, not at a spreadsheet of annual averages.

What the buyer should actually ask

A buyer commissioning off-grid monitoring should ask four questions, in order. What is the worst week of the year for this site's solar yield, and what number is being used for it? What is the battery capacity after cold and ageing derating, not before? What is the longest gap the operation can tolerate between readings, and is the duty cycle set to that gap or to a marketing claim about battery life? And what happens when the node stops reporting, who notices, and how long before someone drives out?

The answers sort the honest designs from the optimistic ones. The node that survives is not the one with the biggest panel or the longest advertised battery life. It is the one whose energy budget was written against the worst winter week, whose installer checked the site in that week, and whose buyer asked the questions above before signing. That is the entire difference between a monitoring network that is still alive in its third season and a collection of silent boxes that cost more to retrieve than they ever returned in data.