Somewhere in the Pacific Northwest, water started slipping behind a stucco elevation about five years ago, probably at a window head flashing that nobody detailed quite right. Nothing showed on the surface. Paint held. Caulk aged gracefully. Inside the wall, the sheathing began the long, quiet conversion from structural material to compost. HomeAdvisor puts the homeowner-visible phase of this process honestly: hidden moisture "develops for years before any stain or bulge becomes visible." By the time a stain appears on the drywall, the project has left the repair schedule and entered the remediation schedule. For a 1,500-square-foot home, This Old House puts large repairs of this kind at up to $75,000.
Stucco is an honest cladding applied to a dishonest assembly. Cement stucco itself breathes; the trouble begins when water gets behind it and cannot get out, and engineered systems made it worse before they made it better. Older EIFS, the synthetic stucco systems installed before roughly 1990, were barrier systems: no drainage plane, no way out, and once sealant at a window or penetration failed, the wall became a terrarium. NOVA Engineering, a building-enclosure consultancy, is blunt that buildings wrapped in those older systems need evaluation before repairs, insurance claims, litigation, refinancing, or acquisition. They see this failure mode at the autopsy stage, which is where everyone sees it, because the wall looks fine until it does not.
Readers of this publication will have seen the early warnings and dismissed them, because that is what early warnings are for. NOVA's checklist runs like this: finish-coat cracks, failed sealant at windows and doors, dark staining below window sills, soft or spongy-feeling wall areas, interior water stains near windows. Each of these reads as cosmetic until you understand the assembly. A cracked finish coat on a barrier EIFS system is not a paint problem. It is a breached envelope with the evidence wearing camouflage.
Now the part the inspection industry will tell you if you ask twice. A professional stucco inspection, done to Exterior Design Institute and EIFS Industry Members Association standards, is a sequence: visual survey, infrared thermography inside and out, electronic moisture scanning, then invasive probing of the framing and substrate firmness testing. Inspection firms state the honest version outright: only the invasive probe confirms moisture content. Infrared is a screening tool. It maps temperature differences that may indicate wet material behind the cladding, and the map tells the inspector where to open the wall, instead of opening the whole elevation on a hunch.
Infrared thermography has done this job for decades, and it has always had the same bottleneck: the camera is the easy part and the reader is the hard part. A thermogram is an abstract field of temperature gradients, and distinguishing trapped moisture from thermal bridging, solar loading on a dark elevation, an HVAC exhaust discharge, or a reflective surface throwing the sensor off takes a trained thermographer. Certified people do not scale. Stucco homes do.
This is where the 2025 paper in MDPI Sensors gets interesting, because it attacks exactly the bottleneck. Researchers flew drones carrying visible-light and thermal infrared cameras over existing residential buildings, registered the two image streams through alignment, and trained a pipeline that extracts walls with a SegFormer model fine-tuned on a 900-image wall dataset, detects leakage from thermal anomaly regions annotated as ground truth, and maps cracks with KNet and UPerNet networks. The full study is open access. Nobody is claiming this replaces a thermographer. That is a narrower and better claim: the tedious anomaly-hunting across hundreds of thermal frames, the part that burns the most expensive hour in the workflow, can be automated to a first pass.
Skeptics of the method should note the lab numbers are real but borrowed. In a materials-science context, a ResNet50 model hit 94.75 percent accuracy, 95.60 percent precision, and 93.45 percent F1 detecting invisible internal damage in thermal images, which demonstrates that thermal imagery carries signal neural networks can beat human visual inspection on, even if the substrate was fiber-reinforced concrete and not your stucco. That paper is here. Separately, a VGG-16 study with class-activation mapping automated the detection and localization of dampness-related defects in buildings, and its authors opened with a sentence the whole field should have framed: despite buildings being one of the largest asset classes on earth, there is a general lack of research in automated condition assessment. Paper, and the arXiv preprint for those who prefer it.
Methodical people will now ask where this actually earns its keep, and the honest answer is scale. A drone plus a model does not replace the invasive probe on your one house. It changes the economics for the portfolio buyer screening forty buildings, the property manager triaging which walls deserve the two-thousand-dollar probe job, the insurer deciding whether the EIFS apartment complex gets renewed at this rate. The AI layer is a triage layer. Triage is unglamorous and it is where the money is.
Here is the math, with every assumption visible, because this is the part that should change a line item. Angi's 2026 figures put crack and sealant repair at $250 to $800, water-damage repair at $1,000 to $3,000, full resurfacing at $3,000 to $8,000, and EIFS work at $12 to $50 per square foot against $8 to $30 for conventional stucco. Catch the failure at the sealant stage and you are in the hundreds. Catch it at the stained-drywall stage, one 200-square-foot elevation in EIFS, and you are at $2,400 to $10,000 before anyone knows whether the sheathing survived. Miss it for five years and This Old House says $75,000. Between early and late intervention on one elevation, the spread is roughly ten to one, and that ratio is the entire business case for putting a thermal camera, with or without a drone, on every stucco transaction in a wet climate. The model does not need to be perfect. It needs to beat waiting for the stain.
Water damage rarely travels alone, which is why the bill accelerates. Angi's 2026 water-damage restoration data averages $3,863 and runs to $16,000, with mold removal at $1,200 to $3,800 layered on top. HomeAdvisor's 2025 to 2026 survey puts the average stucco repair project at $1,629, a number that reflects the repairs people actually commission, which are the small visible ones, not the rot they never found.
Twenty years of projects going sideways teaches you to price the false positive, so here is the case against the hype, stated plainly. Thermal cameras measure surface temperature, not moisture, and the failure modes of the tool read like a list of ordinary afternoons: sun loading on a south elevation, thermal bridging at every stud and slab edge, an HVAC exhaust plume, a reflective coating bouncing infrared noise into the sensor. A model trained on limited data will flag these as wet walls, and every flagged anomaly that is not wet buys the homeowner an expensive, destructive follow-up probe for nothing. That 2025 drone study trained on its own custom dataset, which means generalization across climates, cladding colors, and stucco textures is unproven. Nobody has published the study this field actually needs: the AI model against a certified thermographer, same homes, same day, scored head to head. Until that exists, AI thermal screening is an upgraded screening step, not a replacement inspector.
There are mundane frictions too, the kind that never appear in papers. Residential drone flights need FAA Part 107 compliance, and HOAs have opinions about buzzing quadcopters photographing elevations at dawn, which is when thermal contrast is best. Plan for both before you promise a client a drone survey.
What this analysis did not prove: the ten-to-one spread is an order-of-magnitude comparison across different scopes and sources, not a controlled study, so treat it as directional, not a quote. What that 2025 UAV paper demonstrates is automated facade defect detection on residential buildings; it does not benchmark against human thermographers and does not claim to. Those ResNet50 accuracy figures come from internal defect detection in a different material system and transfer as evidence that thermal imagery is machine-readable, not as stucco performance claims. No source here establishes what fraction of stucco homes in any market carry hidden moisture, so do not extrapolate from the failure mode to a prevalence rate.
If you are buying a stucco home older than about fifteen years in a wet climate, order a moisture survey with infrared thermography from a firm that does invasive probing, not just a general home inspection. General inspectors are not trained thermographers, and a visual-only stucco inspection is a confidence ritual. Read the report for the numbers: moisture-meter readings above 15 to 20 percent at windows, doors, and penetrations are the intrusion threshold on standard inspection checklists. Treat every infrared anomaly as a suspect, not a verdict: a flagged elevation earns a probe, and the probe earns the renegotiation or the walk-away, in that order. If you are a builder, the ten-to-one spread is your argument for specifying drainage-plane EIFS, never barrier EIFS, and for pre-drywall moisture verification on every stucco job, because the wall you can still open costs hundreds and the wall you cannot costs tens of thousands. Rot runs on a clock. The only decision is whether you read it early or pay for it late.