How Energy Availability Is Calculated (and Why It's Harder Than It Looks)

A coach tells you: “Eat 2,200 calories. Don’t go below 30 kcal per kilogram of fat-free mass per day of energy availability. Track it.”

So you log your food. You read your watch’s calorie burn. You check your smart scale. You divide and subtract. The answer is 31.4.

Are you safe? Or have you just summed up three separate 15-percent errors into a number that looks precise but isn’t?

Energy availability looks like a simple subtraction and division. In practice, every input has meaningful measurement noise, and the aggregate uncertainty is usually wider than the decision boundary. This post breaks down where the noise comes from and how to think about your own EA numbers without being falsely reassured or falsely alarmed.

This is a spoke in the energy availability and RED-S pillar. Start there for the clinical framing. This post is about the math.

The Formula

The definition is:

EA = (EI − EEE) / FFM

• EI: Energy intake in kilocalories per day.
• EEE: Exercise energy expenditure in kilocalories per day — the kcal specifically burned during exercise, not total daily expenditure.
• FFM: Fat-free mass in kilograms.

The research literature uses EA expressed as kcal per kg FFM per day. The threshold most often cited — from Loucks and Thuma’s 2003 work on luteinizing hormone pulsatility disruption and reinforced across later reviews — is around 30 kcal/kg FFM/day. Below that threshold, reproductive and metabolic disruption becomes measurable in controlled conditions. Above 45 kcal/kg FFM/day is often described as “optimal” for active athletes.

The thresholds are population-level research findings. Individual thresholds vary, and the research itself acknowledges that EA is a continuous variable with probabilistic rather than deterministic effects. A runner at 32 is not automatically safe; a runner at 28 is not automatically in crisis. The thresholds are anchors, not verdicts.

Where the Noise Enters

Each of the three inputs has known error characteristics. Here’s what the literature says about each.

Energy intake: 10–30 percent under-reporting

Self-report dietary intake is one of the most-studied measurement problems in nutrition science. Multiple methods have been used to validate intake logs against objective measures (doubly labeled water studies, direct observation, metabolic chambers). The consistent finding: free-living adults under-report energy intake by 10 to 30 percent on average, with higher under-reporting among women, overweight individuals, and people with disordered eating patterns.

Athletes are not exempt. A 2018 review by Capling and colleagues focused specifically on athlete dietary assessment methods and found under-reporting of similar magnitude. Contributing factors include forgotten snacks, under-estimation of portion sizes, skipped logging on busy days, and (more problematic in the RED-S context) deliberate under-reporting in people with restrictive eating patterns.

Modern food-logging apps improve consistency but don’t eliminate the bias. Barcode scanning, photo-based logging, and restaurant-chain databases help with foods that match. They don’t help with the home-cooked meal where the olive oil was eyeballed, the chicken portion was estimated by sight, and the dessert was “just a bite” of something someone else cooked.

The practical implication: if your logged intake is 2,200 kcal, your actual intake is plausibly 2,200 to 2,800, with the upper end of that range more likely. For EA calculations at the 30 kcal/kg FFM threshold, a 20 percent under-report on an athlete eating 2,200 kcal shifts computed EA by roughly 8 kcal/kg FFM/day at typical body sizes. That’s larger than the margin between “low” and “adequate” EA. It’s the single largest source of error in most self-computed EA.

What helps: logging consistency over time (under-reporting tends to be systematic rather than random, so trends are more informative than absolutes), occasional high-rigor logging periods where you weigh and photograph everything for 3–5 days, and honest acknowledgement that the number is an underestimate.

Exercise energy expenditure: 10–25 percent error

Wearable-estimated calorie burn during exercise comes from combinations of heart rate, motion, user-provided weight, and proprietary models. Validation studies — Shcherbina et al. (2017), Pasadyn et al. (2019), and subsequent reviews — have compared popular wearables against indirect calorimetry and ECG references during a range of exercise modalities.

The summary finding across these studies: wearables estimate heart rate during steady-state aerobic exercise reasonably well, and from there derive calorie estimates that are within 10 to 25 percent mean absolute error relative to indirect calorimetry. For intermittent exercise (intervals, strength training, CrossFit-style sessions), error increases, and some devices over- or under-estimate in systematic directions depending on their model assumptions.

GPS watches with heart rate tend to perform better on steady-state endurance than strap-less wrist-based devices. Chest straps improve HR accuracy, which improves calorie estimates proportionally. Strength training is consistently the worst case for wrist-based calorie estimation — heart rate during heavy lifting doesn’t map linearly to energy expenditure, and models built on aerobic assumptions systematically under-estimate by 15–30 percent in some cases and over-estimate in others.

The practical implication: for an athlete training aerobically, wearable EEE is approximately right but not precise. For an athlete with a heavy strength component, the EEE estimate could easily be 200–400 kcal off per week in either direction.

What helps: using the device as a trend indicator rather than an absolute truth, cross-checking against a simple MET-based estimate (a rough reasonableness check), and accepting the error band when comparing EA across weeks.

There’s more on training-load measurement in how wearables measure stress and strain and adaptive training intelligence.

Fat-free mass: method-dependent bias

FFM measurements come in several grades of accuracy.

DEXA (dual-energy X-ray absorptiometry) is the most accurate field-practical method for FFM. Typical precision is ±1–2 percent within the same device. Clinic access, cost, and radiation exposure limit frequency, but one DEXA per year is a reasonable anchor for most athletes.

Hydrostatic weighing and air displacement plethysmography (BodPod) achieve similar accuracy to DEXA for total FFM but don’t provide regional breakdown.

Bioelectrical impedance analysis (BIA) is the method most consumer smart scales use. BIA is sensitive to hydration status, recent meals, body temperature, and device construction. Validation studies show ±3–6 percent error for consumer BIA scales compared to DEXA, with day-to-day variability that can swing FFM estimates by 1–2 kg in either direction depending on whether you stepped on after a salty meal, a workout, or first thing in the morning.

Calipers and skinfold equations require skilled operators and produce body fat percentage estimates with ±3–5 percent error compared to DEXA. They work better over time when the same operator measures the same sites, and they’re the most accessible method for athletes without clinic access.

Ultrasound body composition is increasingly available and can match DEXA accuracy when done well, but the operator effect is significant.

The practical implication: if you’re using a smart scale for FFM, a single morning reading has meaningful noise. An FFM of 52 kg could plausibly be 50 to 54 kg based on hydration and measurement error. That shifts EA by roughly 1–2 kcal/kg FFM/day for a typical intake minus EEE. Smaller than the intake and EEE errors, but non-trivial.

What helps: using a smart scale first thing in the morning in consistent conditions, averaging over 7–14 days, and anchoring to a DEXA reading at least annually. For deeper treatment, see body composition: DEXA vs smart scales vs calipers.

Aggregate Uncertainty

Stack these errors and you get a compound uncertainty. Consider an athlete with:

• Logged intake: 2,400 kcal/day (true intake plausibly 2,400 to 3,100)
• Wearable EEE: 700 kcal (true EEE plausibly 525 to 875)
• Smart-scale FFM: 52 kg (true FFM plausibly 50 to 54)

The point estimate of EA is (2,400 − 700) / 52 = 32.7 kcal/kg FFM/day.

The plausible range, taking the error bands pessimistically and allowing them to compound, runs from about (2,400 − 875) / 54 = 28.2 to (3,100 − 525) / 50 = 51.5. That’s a very wide band for a single-day number.

Of course the errors don’t all conspire maximally in the same direction every day. Across weeks, some errors average out. The intake bias tends to be consistent within a person (you under-report systematically by roughly the same fraction), which means the trend direction of EA over weeks is more reliable than the absolute level. The noise on EEE and FFM is more day-to-day random.

Rolling averages are the correction. A 14-day rolling EA is much more informative than a single-day EA. A three-month trend direction — is my EA drifting down while my training is up? — is the most informative signal you can extract from these inputs.

What Serious Practitioners Do

Sport dietitians working with RED-S concerns rarely rely on a single-day EA number. They do some combination of:

• Multi-day high-rigor food logs (3–7 days of weighed and photographed intake) to calibrate the athlete’s habitual under-reporting fraction.
• Multi-method EEE estimation combining wearable data, training load data, and session-by-session MET estimates to triangulate.
• DEXA for anchor FFM, BIA or caliper for between-DEXA trend.
• Rolling multi-week EA estimates rather than daily values, often presented as ranges rather than point estimates.
• Clinical correlates — menstrual function, labs, fatigue reports, performance — as equally weighted signals alongside the computed EA.

This is a workflow, not a number. A single-day EA computation from consumer apps produces a number, but the professional assessment is built on a pattern of evidence.

What Surveillance Tools Can Do

Omnio’s surveillance does not compute a precise daily EA and use it as a primary flag. The reason is the error analysis above — a flag triggered by “your EA was 28 yesterday” would be firing on measurement noise more often than on real underfueling.

Instead, the surveillance uses EA as one supporting signal among several. The primary flagging logic, as described in the energy availability pillar and the confounders post, looks for compound biomarker trajectories over weeks, gated against cycle phase, illness, alcohol, travel, and training spikes. A sustained downward trend in computed rolling EA adds to the case for a pattern being real when other signals (HRV, RHR, sleep, weight trajectory) are moving together. It is never the sole flag.

This is deliberate. A flag-on-number approach would be noisy enough to erode trust. A flag-on-trajectory approach is more likely to surface patterns that a clinician would recognize as worth investigating.

Omnio is a hosted data-aggregation and pattern-recognition platform. Nothing in its logic claims diagnostic accuracy for RED-S, and the EA surveillance is feature-flagged in shadow mode until the empirical false-positive rate is measurably below target.

Practical Takeaways

If you’re computing EA yourself:

1. Treat single-day numbers as directional, not categorical. Use 1–2 week rolling averages.
2. Accept that your logged intake is an underestimate. The trend is more informative than the absolute.
3. Anchor FFM with a DEXA at least annually. Use a smart scale for trend, not truth.
4. Cross-check wearable EEE against simple MET calculations for your main training modalities.
5. Don’t obsess over the number. Look at the trend and the trajectory.

If your computed EA is below 30 on a single day:

Don’t panic. Look at the last 2 weeks. Look at the confounders. If the pattern is persistent, look at your training load, your weight trajectory, your menstrual or hormonal signals (if applicable), and your subjective experience. If multiple signals converge toward underfueling and cannot be explained by confounders, talk to a sport dietitian.

If your computed EA has been below 30 for multiple weeks:

Add a higher-intake day or two. Watch the data. If the signals (HRV, RHR, sleep, mood) don’t recover, if your weight is still trending down, or if you have amenorrhea, fatigue, or stress-fracture history, see a clinician.

If you have any of the hard signals — missed periods, stress fractures, persistent unexplained fatigue, restrictive eating causing distress — see a clinician now, regardless of the EA number. The hard signals outrank the computed EA in clinical relevance.

Putting It Together

Energy availability is a simple formula on paper and a challenging measurement in practice. Intake is systematically under-reported, exercise energy expenditure has 10–25 percent error, and fat-free mass estimates vary with method and hydration. A single-day EA computation carries error bars wide enough that categorical classification is unreliable.

Multi-week rolling averages, trend direction, and corroborating biomarker signals are the useful signal. Surveillance tools that treat EA as one trajectory among many, gated by confounders, are defensible. Tools that claim precision from a single number are not.

For the broader framing, return to the energy availability and RED-S pillar. The biomarker signatures post covers what the downstream wearable signals look like. The confounders post covers what else can produce similar signatures. For the nutrition context, see nutrition intelligence and your calorie target is wrong.