Spec sheet battery life numbers for underground sensor nodes are almost always wrong — not through dishonesty, but because they're measured in conditions that don't resemble deployment reality. Ambient temperature, event frequency, radio relay distance, and the on-board compute load at trigger time collectively determine actual consumption in ways that a bench test at 25°C with a synthetic event cadence can't predict. We've now accumulated 18 months of continuous deployment data across three distinct geology types and roughly 140 individual node-deployments. The picture is more complicated than the spec sheets suggest, but it's also more actionable.
The three deployment environments
Our three test geology classes were selected to span the range of thermal and mechanical conditions our nodes are likely to encounter in operational use. The first was a dry sandstone environment in the high desert Southwest — temperature range roughly 8–28°C at depth, low humidity, high thermal stability day-to-night. The second was a granite formation in a mountainous region, colder baseline (2–14°C) with higher humidity and more pronounced seasonal variation. The third was a clay-soil tunnel environment with active groundwater infiltration — temperature stable at 11–13°C year-round but with humidity at or near 100% for extended periods and significant risk of partial submersion during weather events.
Node configurations were held as constant as possible across sites — same hardware revision, same firmware build, same nominal duty cycle parameters. The intention was to isolate geology-driven environmental variables from hardware variation. We didn't fully achieve that in year one because we made two firmware updates mid-deployment that affected sleep-mode power draw, but we've tracked those changes in the dataset and the results below account for them.
What the data actually showed
Spec-projected battery life for our nodes at the nominal duty cycle was 22 months. Actual median across all deployments came in at 16.4 months — a 25% shortfall. The distribution matters more than the median: the best-performing nodes in sandstone hit 21 months, close to spec. The worst-performing nodes in the clay environment terminated before 11 months. That 2:1 spread between best and worst case is operationally significant. If you're designing a 12-month deployment cycle based on spec numbers, you'll be pulling nodes that still have life left and missing nodes that are already dead.
Temperature was the dominant factor across all sites but not in the way we initially modeled it. Cold temperatures reduced battery capacity as expected — lithium chemistry performance degradation below 10°C is well-documented — but the bigger effect was the thermal cycling amplitude. Nodes in the clay environment, despite having the most stable ambient temperature, showed accelerated capacity degradation that we eventually traced to repeated brief temperature drops during flood events. Cold water infiltration dropped node case temperature by 4–6°C within minutes, then slow thermal recovery over hours. This high-rate cycling caused measurable capacity fade not present in the more stable sandstone and granite environments.
Radio transmit power: the dominant variable we underweighted
Our original power budget model treated radio transmit events as a known quantity based on trigger frequency and packet size. What it didn't adequately account for was relay distance variability. In planned deployments, we pre-survey the mesh topology and know the expected hop distance for each node. In operational contexts, nodes sometimes end up in positions where the nearest relay is farther than planned — because the relay was disabled, because geology attenuated the signal more than expected, or because the deployment geometry changed after initial placement.
A node that has to boost transmit power to reach a relay at the edge of its range uses 3–5x the energy per transmit event compared to a node in its designed hop geometry. We observed this effect across roughly 12% of deployments — nodes where the as-deployed relay distance was more than 40% over the planned distance. Those nodes showed battery life reductions of 30–45% relative to similarly-deployed nodes with clean hop topology. The fix is adaptive power control with feedback, which we've implemented in the current firmware revision. But the lesson from 18 months of data is that radio power is the budget line item most sensitive to deployment execution quality.
Clay geology: the worst-case and what to do about it
The clay environment data deserves its own section because it was consistently the most problematic. Beyond the thermal cycling issue described above, clay environments present an additional challenge: electrochemical migration. In high-humidity, ionically-active soil environments, even well-sealed enclosures eventually develop micro-ingress at cable entry points and sensor ports. We observed corrosion-related current leakage in 8 of 47 clay-environment deployments — not catastrophic failures, but steady parasitic draws of 5–15 mA that shortened effective life by 2–4 months.
Our response has been a combination of improved conformal coating on PCB traces near entry points and a more aggressive ingress-protection rating requirement for any node targeted at high-humidity environments. We've also moved to IP68-rated connectors at all external interfaces, which added cost and assembly complexity but eliminated the leakage issue in our most recent clay-site deployments. The tradeoff is real — IP68 connector systems add roughly 18% to per-node hardware cost — but the battery life recovery more than justifies it in the clay context.
Revised planning numbers for deployment teams
Based on 18 months of data, we now use environment-adjusted battery life estimates rather than a single spec number. For dry, thermally-stable environments (sandstone class): plan for 19 months, pull at 17 for margin. For granite or similar cool-stable environments: plan for 17 months, pull at 15. For high-humidity or thermally-cycling environments (clay class, flood-risk corridors): plan for 13 months, pull at 11. These are median-based numbers — some nodes will go longer, and if you're operating where a dead node before recovery is a mission-failure condition, you should pull earlier still.
The more important operational change is pre-deployment topology verification. Before any deployment, we now require a radio survey that maps actual path loss to every intended relay and flags any node position where the as-planned hop distance exceeds the design range by more than 20%. Nodes that would exceed that threshold get repositioned or the mesh topology is redesigned. This step adds 2–3 hours to a deployment campaign but has consistently been the highest-ROI change we've made to our field procedures.
Where the next 18 months goes
The dataset we've built is useful for what it is — a retrospective on three geology classes with consistent hardware. What it doesn't tell us is how newer battery chemistries perform in these environments. We're currently qualifying a lithium-iron-phosphate variant that has substantially better thermal cycling characteristics than the standard lithium-ion cells we've been using. Early bench data suggests the LFP chemistry should close most of the gap we see in the clay environment. We'll have field validation data by late 2026.
We're also working on a node-level state-of-health telemetry channel that would allow battery state estimation from surface, enabling condition-based pull scheduling rather than fixed-interval replacement. The challenge is doing this without a meaningful increase in standby power draw — the telemetry itself can't cost more energy than the information value it returns. That tradeoff is still being worked out.