Assessing the variability of air traffic controllers’ workload: A comparative perspective

Abstract

Controller workload remains the principal functional constraint on air traffic management system capacity, yet its measurement in operational settings is complicated by substantial inter-individual variability that aggregate traffic-based models fail to capture. This case study investigates whether physiological workload differs significantly between certified air traffic controllers (ATCs) exposed to identical operational conditions, a question with direct implications for both safety management and dynamic capacity planning. Three licensed tower (TWR) controllers were monitored during a standardised 50-min heavy-load simulation exercise. During the experiment, physiological stress indicators were monitored, specifically LF/HF, SDNN, and mean RR, in 5-min intervals measured by a single-lead ECG monitor with a sampling frequency of 1000 Hz, supplemented by a 3D accelerator for actigraphy. Simultaneously, photoplethysmographic (PPG) recording was performed in synchronization with a second single-lead ECG for control purposes. Despite identical traffic scenarios, statistically significant inter-individual differences were confirmed for all three parameters (p ≤ 0.002). Median LF/HF values differed by up to 74% between controllers, SDNN by up to 45%, and mean RR by 5.83%. These differences exceed the magnitudes typically reported between low- and high-workload conditions in within individual aviation studies, demonstrating that individual physiological reactivity, rather than traffic complexity alone, is a primary determinant of operational workload. The findings challenge the current practice of treating airport capacity as a fixed threshold and support the concept of dynamic, real-time capacity management informed by continuous individual physiological monitoring. Practically, the approach enables identification of stress-susceptible controllers, supports personalised shift scheduling, and provides a foundation for deploying operationally compatible wearable ECG monitoring during actual ATC operations.

Keywords

air traffic controller workload heart rate variability electrocardiogram simulation safety

Introduction

As stated in scientific literature focused on air traffic control, “Controller workload is likely to remain the single greatest functional limitation on the capacity of the ATM system”.^1,2 Yet despite decades of research, reliably measuring, predicting, and acting upon that workload in real time remains an open problem.^1,3 The present study addresses a gap that is underexplored: rather than asking how workload varies with traffic complexity, it asks how workload varies across individual controllers exposed to identical operational conditions. Demonstrating such inter-individual variability can have direct implications both for controller safety management and capacity planning.

Workload involves the allocation of mental resources to meet task demands and, like complexity, is not directly observable. However, it can be inferred from multiple sources, making it a multidimensional construct.² The literature organises workload measurement approaches into three broad categories: subjective measures, performance-based measures, and physiological measures.⁴ Each carries distinct strengths and limitations, and selecting the right approach depends on the operational context and the research question.

Subjective measures, most notably the NASA Task Load Index (NASA-TLX), have been widely used in ATC research since the 1980s.^5,6 NASA-TLX derives an overall workload score from a weighted average of six subscales: mental demand, physical demand, temporal demand, performance, effort, and frustration.⁶ Its main advantages are low cost and broad applicability. However, subjective ratings suffer from several well-documented limitations. They are retrospective: administering them during a task is intrusive, while waiting until completion risks recall decay.^7,8 Crucially, a user's perception of their own task performance can weigh heavily on all sorts of ratings, so workload scores could be higher or lower depending on whether they believed they completed the task successfully even when workload was in fact the same.⁹ These limitations have led many researchers to recommend supplementing or replacing subjective ratings with objective physiological measurements.^7,10 An additional layer of complexity arises from the finding that perceived workload is significantly related to individual factors such as gender and working experience among air traffic controllers,¹¹ meaning subjective scores conflate external task demands with individual differences, which is exactly the confound this study aims to isolate.

Performance-based measures such as number of aircraft handled per hour, separation errors, and communication frequency offer observable, objective proxies for workload. Some studies propose to derive workload assessment from a set of ATC activities, transforming subjective practical experience into a precise value.^12,13 However, mixed results have been obtained across previous studies, possibly because of the difficulty of combining data related to the changing distribution of traffic and the assessment of rapidly fluctuating psychophysiological variables.¹² Performance measures are also largely insensitive to the early stages of workload build-up, performance degradation typically appears only after cognitive resources are already severely taxed.^14,15

Physiological measures represent the most direct category of workload assessment, as advances in wearable sensing have enabled continuous, non-disruptive, real-time monitoring that can be standardised across individuals and studies.^15,16 Heart Rate Variability (HRV) measurement is a non-invasive method to evaluate relevant physiological changes in a human body, and HRV can reflect cognitive workload objectively.¹⁷ A 2022 scoping review of 39 ATC studies identified cardiovascular measures among the most consistently effective modalities,¹⁴ and a 2024 systematic review of 29 aviation studies confirmed HRV as a promising continuous workload detection tool, while noting significant variability in results across study designs and operators.¹⁸ The main limitation of physiological approaches is their susceptibility to confounding factors like respiration rate, posture, physical activity, and individual baseline autonomic tone all influence HRV independent of cognitive load.¹⁹ This is precisely why controlling for confounds (e.g., seated, stationary, well-rested participants under standardised conditions, as in the present study) is methodologically critical.

A critical but underappreciated dimension of ATC workload research concerns inter-individual variability. Mental workload depends not only on external task demands but on individual characteristics including experience, personality, coping strategies, and physiological state.^19–21 A study of 256 ATCs confirmed that perceived workload is significantly related to gender and working experience,¹¹ and while predictive models are increasingly common, most are trained on group-level patterns and fail to capture individual baselines.¹³ The present study targets this gap directly by comparing HRV responses of three controllers during the same simulated exercise under identical traffic loads, isolating the inter-individual signal while holding task demands constant.

The motivation is twofold. First, safety: if two controllers experience fundamentally different physiological workload under the same scenario, shift planning based on aggregate estimates will fail to protect the more vulnerable individual. Second, capacity: if workload is subjective and individual, airport capacity cannot be expressed as a single fixed number but must be conceived as a dynamic interval conditioned on real-time physiological monitoring. These concerns motivate a design that shifts the research question from workload versus traffic to workload versus individual.

Specifically, the aim of this study was to analyse inter-individual differences in HRV-based workload indicators (LF/HF, SDNN, and mean RR) among three certified TWR controllers during a standardised 50-min high-load simulation exercise. We hypothesised that, even under identical operational conditions, statistically significant between controller differences would emerge across all three physiological dimensions. Demonstrating individual differences in workload would then mean that capacity should not be a fixed number of movements of aircraft in a given area, but rather an interval that should change based on real-time workload measurements at specific ATCs. This would contribute to more effective traffic planning and optimization of aircraft control, which can have a direct impact on increasing the safety and operational efficiency of air traffic.²²

Preliminaries

Physiological indicators of workload

Understanding why ECG-derived HRV was selected for the present study requires situating it within the wider landscape of available physiological measurement approaches.

Electroencephalography (EEG) provides high temporal resolution and has been extensively used to detect workload-related changes in ATC tasks.²³ Radüntz et al.²³ demonstrated stable classification of ATC mental workload using their Dual Frequency Head Map method across multiple sessions. However, EEG carries substantial practical constraints in operational settings: electrode caps interfere with ATC headsets, create discomfort during extended sessions, and are sensitive to movement artefacts caused by head rotation and speech. Classifier accuracy also tends to degrade over days without retraining.^23,24

Functional near-infrared spectroscopy (fNIRS) has shown sensitivity to prefrontal workload increases and is more motion-tolerant than EEG,²⁵ but requires participants to remain relatively still and is incompatible with standard ATC headgear. EEG appears to be less sensitive to small changes in cognitive workload than fNIRS, though fNIRS could not discriminate between higher workload levels.²⁶ Eye-tracking offers non-contact measurement of pupil diameter, blink rate, and fixation patterns, all sensitive to workload changes,^10,13,27 but requires participants to face a fixed screen within a calibrated region of interest, which is incompatible with tower operations where the visual field is dynamic and three-dimensional. Electrodermal activity (EDA) reflects sympathetic arousal but is highly susceptible to motion, perspiration, and ambient temperature, and lacks the temporal specificity needed to track workload fluctuations within 5-min windows.¹⁰

Given the constraints of the ATC tower environment and the specific research question, ECG-based HRV analysis offers a uniquely suitable combination of properties. A single-lead ECG monitor can be worn under standard ATC clothing without interfering with headsets, freedom of head movement, or voice communication. Unlike EEG caps or eye-tracking goggles, it imposes essentially no cognitive or physical overhead on the operator. For ECG, reliable metrics include heart rate (HR) and heart rate variability (HRV) indices such as mean RR and LF/HF¹⁰ and these can be computed continuously without task interruption, which is a critical requirement in ATC where pausing or querying the controller is operationally impermissible.

ECG provides beat-to-beat data at high sampling frequencies (1000 Hz in the present study), enabling fine-grained temporal segmentation. The 5-min interval analysis adopted here is well-established in short-term HRV research and is sufficient to compute stable frequency-domain estimates of sympathovagal balance.²⁸ HRV metrics are computed according to internationally agreed standards,²⁸ enabling comparison across individuals, sessions, and studies, a property that questionnaire-based methods conspicuously lack. HRV measurement is a non-invasive method to evaluate relevant physiological changes in a human body, and HRV can reflect individuals’ cognitive workload objectively,¹⁷ which is central to the study's aim of detecting inter-individual differences that self-report measures would systematically obscure.

Heart rate variability

Heart rate variability (HRV) and other cardiovascular measures are widely used in workload research due to their sensitivity to autonomic nervous system activity. Metrics such as the low-frequency to high-frequency ratio (LF/HF), the standard deviation of normalized interbeat intervals (SDNN), and the mean of a selected series of interbeat intervals (mean RR) provide quantifiable information about the balance between sympathetic and parasympathetic nervous system activity, which is directly related to workload and stress.

A 2024 systematic review of 29 studies confirmed that HRV has emerged as a potential tool for detecting pilot mental workload during real flight operations,¹⁸ while a 2022 scoping review of ATC studies confirmed that cardiovascular measures frequently allow meaningful mental workload assessments.¹⁴ A study of commercial airline pilots in a certified A320 simulator found that higher SDNN and RMSSD and lower LF/HF ratio were associated with better manoeuvre performance, with an interquartile range increase in SDNN associated with a 37% increase in the odds of passing a manoeuvre.²⁹

Cardiovascular function indicators oscillate in different rhythms, each carrying distinct physiological significance.³⁰ Spectral analysis enables the differentiation of these oscillations into several frequency bands.³¹ Among them, the most relevant are the low-frequency (LF) band (0.04–0.15 Hz) and the high-frequency (HF) band (0.15–0.40 Hz). The LF band reflects both sympathetic nervous system activity and vagal influence, with their proportion depending on situational factors such as body position.^32,33 Meanwhile, the HF band primarily represents vagal activity associated with physiological respiratory arrhythmia.³³ The balance between sympathetic and parasympathetic activity is often expressed through the LF/HF ratio, with higher values generally indicating increased mental workload.³⁴

However, while an increased LF/HF ratio is commonly interpreted as a sign of elevated workload, some studies challenge this notion.³⁵ For instance, Kim et al. found that the LF/HF ratio rises under stress in individuals who adopt passive coping strategies, whereas it decreases in those favoring active coping mechanisms.³⁶ Other researchers question the simplification of the sympathetic/parasympathetic relationship into a strictly linear dependence as represented by the LF/HF ratio.^37,38

Another key physiological indicator for assessing mental workload is the R-R interval, which represents the time between consecutive R waves in an electrocardiogram (ECG).³⁴ While R-R interval analysis provides valuable insight, it is prone to measurement noise. However, its ease of sampling due to the distinct prominence of R waves makes it a practical tool.^34,39 A decrease in a mean R-R interval signifies an acceleration of heart rate (HR), often indicating heightened workload or stress.⁴⁰

One of the most widely used HRV parameters in occupational health studies is the standard deviation of NN intervals (SDNN), which reflects overall HRV.^41,42 Borchini et al. identified a link between occupational stress and SDNN during the workday.⁴³ In aviation contexts, SDNN has been confirmed to decrease with sympathetic activation during high cognitive load phases such as take-off and landing, and is significantly correlated with airspeed error rate.⁴⁴ In a simulated flight multitasking study, SDNN was negatively correlated with subjective mental workload under low task-load conditions.⁴⁵ A systematic review confirmed that for ECG-based assessment in real-world field settings, SDNN is among the most reliable and consistently workload-sensitive metrics available.¹⁰ Time-domain HRV analysis, as one of the simplest measurement methods, evaluates both heart rate and successive interbeat intervals.⁴ The four key recommended time-domain parameters are SDNN, the HRV triangular index, SDANN, and RMSSD (including pNN50 and NN50).²⁸

Methodology

The need to obtain the ability to determine the capacity of individual airports resulted in the following solution. Case study measurements were carried out on an air traffic control tower (TWR) simulator within the Czech Air Navigation Institute (CANI). This simulator consists of several screens that realistically display the environment of the airport and its immediate surroundings, thus providing a faithful simulation of air traffic control operating conditions. The simulator is designed to replicate real air traffic control conditions as accurately as possible, including visual and communication elements, which ensure a high level of credibility of the tested environment.

Three selected air traffic controllers (designated ATC1, ATC2, and ATC3) were tested in a 50-min simulation exercise. The experiment was conducted during the morning. All test subjects reported being in good health, well-fed, and adequately hydrated, ensuring readiness for high-demand operational conditions. All controllers hold TWR and APP licenses. The exercise was designed so that the air traffic controllers were under heavy load according to Eurocontrol methodology.⁴⁶ Table 1 shows the recorded working time of one hour for an ATC and the associated workload. If an air traffic controller spends more than 42 min per hour communicating, either on the frequency with pilots or coordinating with other controllers, it is considered an overload. This excessive communication can lead to a loss of situational awareness, meaning the controller may struggle to keep track of aircraft positions, flight plans, and potential conflicts in the airspace. For the case of 50-min heavy workload exercise, the ATC must speak 54–69% of the total time, i.e., more than 27 min and less than 34.5 min.

Table 1.

Workload and recorded working times of ATCs.⁴⁶

Threshold	Interpretation	Recorded working time during 1 h
70% or above	Overload	42 min +
54–69%	Heavy load	32–41 min
30–53%	Medium load	18–31 min
18–29%	Light load	11–17 min
0–17%	Very light load	0–10 min

Throughout the experiment, the heart activity of the test subjects was monitored using the FlexiGuard system developed for the purpose of monitoring heart and movement activity in the Integrated Rescue System units at the Department of Information and Communication Technologies in Medicine of the Faculty of Biomedical Engineering, Czech Technical University in Prague.⁴⁷ The system is used as a support tool for training in demanding conditions and on ATC and Air Aviation simulation equipment. Structurally, it is a robust version of a single-lead ECG monitor with a sampling frequency of 1000 Hz, supplemented by a 3D accelerator for actigraphy, a contact thermometer on the body surface, and a thermometer with a hygrometer to monitor the environment beneath clothing. In addition, the unit is equipped with a GPS unit, which allows for the localization of the subject in the field. The recorded data are stored inside the unit and the information obtained from them is wirelessly transmitted to the surveillance system. Simultaneously, photoplethysmographic (PPG) recording was performed in synchronization with a second single-lead ECG for control purposes. For this experiment, ECG data were used for subsequent analysis of HRV in a standardized environment of Kubios (version 2.1) and further processed in R.⁴⁷

Physiological indicators were monitored with the aim of demonstrating that the workload of air traffic controllers varies significantly between individuals, even under identical working conditions and operational situations. Specifically, the physiological indicators workload LF/HF, SDNN, and mean RR were monitored at 5-min intervals. Based on practical experience from regular operations, the following hypotheses were established:

$\bar{H^{1}}$ : There are statistically significant differences in LF/HF values between individual ATCs .

$\bar{H^{2}}$ : There are statistically significant differences in SDNN values between individual ATCs.

$\bar{H^{3}}$ : There are statistically significant differences in mean RR values between individual ATCs .

Results

Table 2 shows the values of the studied parameters “LF/HF”, “SDNN”, and “mean RR”, which were measured for individual research participants ATC1, ATC2, and ATC3. The aim of this chapter is to show that individual variations in workload between air traffic controllers are significant in each of the physiological parameters studied.

Table 2.

Values of the parameters “LF/HF”, “SDNN”, and “mean RR”.

Time (5 min)	LF/HF $(A^{1})$			SDNN $(A^{2})$			Mean RR ( $A^{3})$
Time (5 min)	ATC1 ( $A_{1}^{1}$ )	ATC2 ( $A_{2}^{1}$ )	ATC3 $(A_{3}^{1}$ )	ATC1 $(A_{1}^{2}$ )	ATC2 $(A_{2}^{2})$	ATC3 $(A_{3}^{2})$	ATC1 $(A_{1}^{3})$	ATC2 $(A_{2}^{3})$	ATC3 $(A_{3}^{3})$
1–5	6.333	5.491	3.417	71.8	33.1	53.3	618.1	636	672
6–10	4.073	7.048	4.329	50.4	26.2	51.3	632.2	621.9	666.7
11–15	3.908	8.532	4.074	53.7	24.3	43.9	639.1	603	654.6
16–20	7.731	9.313	4.264	43.4	22.8	45.6	628.2	612.2	637.4
21–25	6.109	7.285	3.213	51.6	19.8	46.8	654.6	631.6	680.8
26–30	4.316	5.884	2.85	43.7	35	42.4	647.5	630.1	683.9
31–35	4.875	7.247	4.692	44.9	26	48	646.8	630.5	683
36–40	4.132	6.613	2.476	52.1	28.6	41.3	674	659.7	684.6
41–45	5.46	10.776	5.6	47.9	27.5	43.3	652.5	629.9	659.7
46–50	4.922	8.331	4.689	43.4	32	43.2	659.6	658.1	656.2
$m e d_{i}^{k}$	4.898	7.266	4.169	49.15	26.85	44.75	647.15	630.3	669.35
${\bar{x}}_{i}^{k}$	5.186	7.652	3.960	50.29	27.53	45.91	645.2	631.3	667.8

We will denote the three studied workload parameters “LF/HF”, “SDNN”, and “mean RR” successively with the symbols $A^{1}, A^{2}, A^{3}$ . Furthermore, for the parameter $A^{k}$ , $k \in {1, 2, 3}$ we will mark the measured data from participants ATC1, ATC2, and ATC3 successively with the symbols $A_{1}^{k}, A_{2}^{k}, A_{3}^{k}$ . This notation is also given in the columns of measured values in Table 2.

We will further assume that the data in each column of Table 2 is the realization of some random variable. Our goal is to compare, for the selected parameter $A^{k}$ , $k \in {1, 2, 3}$ , whether the data of individual participants $A_{1}^{k}, A_{2}^{k}, A_{3}^{k}$ differ significantly from each other or not. More precisely, whether the data from one participant has lower values or, conversely, higher values than from another. As a first step in this analysis, it is necessary to realize that each participant responded to the same event in a given time period. This means that the data $A_{1}^{k}, A_{2}^{k}, A_{3}^{k}$ are dependent on each other.

Statistical data processing

At the beginning of the analysis, a Kolmogorov–Smirnov test of normality was performed on the data columns in Table 2. For each column, the hypothesis that the data originated from a random sample with a normal distribution was tested. This hypothesis was not rejected, as p > 0.51.

However, the sample size was small, with only 10 values available. Such a small number of measured values raises concerns about the possibility of a Type II error. Now, depending on whether we assume that the data in the columns come from a random sample with a normal probability distribution or not, we will perform two types of tests, the Friedman and Two-way ANOVA test.

Statistical data processing – The friedman test

In this subsection, we will assume that the distribution from which the data comes is unknown. If the data are dependent and, moreover, come from an unknown distribution, the Friedman test is often chosen to compare the equality of their typical values. It is used to assess the agreement of the medians of random variables, the realizations of which are the data $A_{1}^{k}, A_{2}^{k}, A_{3}^{k}$ . The median characterizes the position of the “center” of the density of the unknown probability distribution and is therefore also a good characteristic with which to determine whether the measured values for one of the participants are lower or higher than the values measured for the other participants.⁴⁸

Using the Friedman test, we will determine whether the medians of the distributions from which the data $A_{1}^{k}, A_{2}^{k}, A_{3}^{k}$ originate are the same or not. Let us denote the median of the parameter distribution by the symbol $m e d_{i}^{k} A_{i}^{k}$ , where $i, k \in {1, 2, 3},$ we choose the standard significance level $α = 0.05$ and test the hypothesis:

$H^{k} : m e d_{1}^{k} = m e d_{2}^{k} = m e d_{3}^{k}$ compared to

$\bar{H^{k}} :$ there are $i, j \in {1, 2, 3}$ such that $m e d_{i}^{k} \neq m e d_{j}^{k}$

The Friedman test results^33,34 are presented in Table 3.

Table 3.

Friedman test results.

Parameter $A^{k} :$	$A^{1}$	$A^{2}$	$A^{3}$
p-value	0.0012	0.0000	0.0017
$H^{k} : m e d_{1}^{k} = m e d_{2}^{k} = m e d_{3}^{k}$	$H^{1} :$ rejected	$H^{2} :$ rejected	$H^{3} :$ rejected

We reject the hypothesis that the medians of individual columns are the same for all parameters (Table 3). Therefore, we will perform a post-hoc analysis for each parameter. Its goal is to identify columns whose medians differ. Therefore, for each parameter $A^{k}, k \in {1, 2, 3}$ at the significance level $α = 0.05$ there will be three tests performed:

$H_{i j}^{k} : m e d_{i}^{k} = m e d_{j}^{k}$ compared t o

$\bar{H_{i j}^{k}} : m e d_{i}^{k} = m e d_{j}^{k}$ , where $i, j \in {1, 2, 3}, i \neq j$ .

The results of these tests are shown in Table 4.

Table 4.

Post-hoc analysis.

Parameter:	$A^{1}$
column medians comparison $A_{i}^{1}, A_{j}^{1}$ , $i, j \in {1, 2, 3}, i \neq j$	$A_{1}^{1}, A_{2}^{1}$	$A_{1}^{1}, A_{3}^{1}$	$A_{2}^{1}, A_{3}^{1}$
p-value	0.0370	0.5029	0.0010
$H_{i j}^{1} : m e d_{i}^{1} = m e d_{j}^{1}$	$H_{12}^{1} :$ rejected	$H_{13}^{1} :$ accepted	$H_{23}^{1} :$ rejected
parameter:	$A^{2}$
column medians comparison $A_{i}^{2}, A_{j}^{2}$ , $i, j \in {1, 2, 3}, i \neq j$	$A_{1}^{2}, A_{2}^{2}$	$A_{1}^{2}, A_{3}^{2}$	$A_{2}^{2}, A_{3}^{2}$
p-value	0.0004	0.6438	0.0102
$H_{i j}^{2} : m e d_{i}^{2} = m e d_{j}^{2}$	$H_{12}^{2} :$ rejected	$H_{13}^{2} :$ accepted	$H_{23}^{2} :$ rejected
parameter:	$A^{3}$
column medians comparison $A_{i}^{3}, A_{j}^{3}$ , $i, j \in {1, 2, 3}, i \neq j$	$A_{1}^{3}, A_{2}^{3}$	$A_{1}^{3}, A_{3}^{3}$	$A_{2}^{3}, A_{3}^{3}$
p-value	0.1733	0.1733	0.001
$H_{i j}^{3} : m e d_{i}^{3} = m e d_{j}^{3}$	$H_{12}^{3} :$ accepted	$H_{13}^{3} :$ accepted	$H_{23}^{3} :$ rejected

Based on the results shown in Table 4, we can conclude that:

for the parameter $A^{1}$ there are statistically significant differences in the medians in columns $A_{1}^{1}$ , $A_{2}^{1}$ , and in columns $A_{2}^{1}$ , $A_{3}^{1};$

for the parameter $A^{2}$ there are statistically significant differences in the medians in columns $A_{1}^{2}$ , $A_{2}^{2}$ , and in columns $A_{2}^{2}$ , $A_{3}^{2}$ ;

for the parameter $A^{3}$ there are statistically significant differences in the medians in the columns $A_{2}^{3}$ , $A_{3.}^{3}$

For clarity, Figure 1 shows the characteristics of the data sets from each section using a box plot. In each graph, symbols are used on the horizontal axis $A_{i}^{k}, i, k \in {1, 2, 3}$ to indicate the data “LF/HF”, “SDNN”, and “mean RR” from the given respondent. The vertical axis shows the characteristics of these data (median, 25% quantile, and 75% quantile).

Figure 1.

Representation of data set characteristics using a box plot.

Statistical data processing – The ANOVA test

Now we will assume that the data in the columns come from a random sample with a normal probability density distribution. (This is true for each column with p -value $> 0.51$ ). In this case, we can analyze the data from Table 2 using the so-called Two-way ANOVA method. For a fixed parameter $A^{k}$ , $k \in {1, 2, 3}$ we will assume for each observation in Table 1 that

x_{i j}^{k} = μ^{k} + ξ_{i}^{k} + η_{j}^{k} + ε_{i j}^{k}

where k is the index of the selected section,

μ

is the overall mean value in section k,

ξ_{i}^{k}

is the mean value of the event factor (in the i -th row of parameter

A^{k}

in Table 2),

η_{j}^{k}

is the mean value of the participant factor (in the j -th column of the parameter

A^{k}

in Table 2), and

ε_{i j}^{k}

is a random error. Now we will test two hypotheses at the significance level

α = 0.05

$H_{ξ}^{k} : ξ_{1}^{k} = ξ_{2}^{k} = \dots = ξ_{10}^{k} = 0$ versus the alternative hypothesis $\bar{H_{ξ}^{k}}$

$H_{η}^{k} : η_{1}^{k} = η_{2}^{k} = η_{3}^{k} = 0$ versus the alternative hypothesis $\bar{H_{η}^{k}}$

The evaluation of the Two-way ANOVA test is shown in Table 5.

Table 5.

Two-way ANOVA test.

Section $A^{k} :$	$A^{1}$	$A^{2}$	$A^{3}$
$H_{η}^{k} : η_{1}^{k} = η_{2}^{k} = η_{3}^{k} = 0$ , $p$ -value	0.0000	0.0000	0.0000
$H_{η}^{k} : η_{1}^{k} = η_{2}^{k} = η_{3}^{k} = 0$	$H_{η}^{1} :$ rejected	$H_{η}^{2} :$ rejected	$H_{η}^{3} :$ rejected
$H_{ξ}^{k} : ξ_{1}^{k} = ξ_{2}^{k} = \dots = ξ_{10}^{k} = 0$ , $p$ -value	0.0335	0.1139	0.0057
$H_{ξ}^{k} : ξ_{1}^{k} = ξ_{2}^{k} = \dots = ξ_{10}^{k} = 0$	$H_{ξ}^{1} :$ rejected	$H_{ξ}^{2} :$ accepted	$H_{ξ}^{3} :$ rejected

From the first three rows of Table 5, it follows that for the parameters $A^{k}$ , $k \in {1, 2, 3}$ we reject the hypothesis of the agreement of the mean values of the individual columns. This result corresponds well with the outcome of the Friedman test, according to which (see Table 3), for the parameters $A^{k}$ , $k \in {1, 2, 3}$ . We reject the hypothesis of equality of the medians of the individual columns. Both tests therefore lead, at a significance level of α = 0.05, to the conclusion that for each of the observed parameters $A^{k}$ , $k \in {1, 2, 3}$ at least one participant has values that differ from the others.

From the last two rows of Table 5, we also find that for parameters A¹ and A³, we reject the hypothesis of equality of the mean values of the individual rows. This means that the type of event to which the participant must respond also affects the values of parameters A¹ and A³. However, this effect is not statistically significant for parameter A³.

Discussion

The measured values show significant individual differences in workload among air traffic controllers, even though the simulation conditions were identical for all participants. The validity of hypotheses $\bar{H^{1}}$ , $\bar{H^{2}}$ , and $\bar{H^{3}}$ was demonstrated. Individual factors and their influence on air traffic controller workload have been discussed several times in literature,^{3,4,11,16,21,49,50} but the specific differences in workload that these factors may cause have not yet been systematically investigated. Based on the results of the case study, it can be concluded that a detailed study and systematic consideration of these differences is a key factor. This approach allows obtaining relevant information for optimizing airport capacity planning, effective management of operational areas, adaptation of work shifts and rest cycles. Since each air traffic controller is a unique individual with varying levels of stress resilience, it is recommended to introduce regular measurement of objective physiological parameters. This would allow continuous monitoring of workload and detection of excessive stress. These measures would have a direct impact on strengthening safety standards and improving the efficiency of air traffic.

The LF/HF ratio, indicating the degree of activation of the sympathetic nervous system, showed the highest values in participant ATC2 (median 7.266), which indicates higher stress activation and higher workload. In contrast, the values of the LF/HF parameter in the other two participants ATC1 and ATC3 (medians 4.898 and 4.169) were, at the significance level $α = 0.05$ , statistically indistinguishable. This indicates a similar level of workload for them. The workload for ATC2 is 48.34% higher than for ATC1 and 74.28% higher than for ATC3.

SDNN, which reflects the overall heart rate variability and the ability to autonomically regulate stress, was measured lowest in participant ATC2 (median 26.85), indicating lower heart rate variability and thus higher workload compared to the other two participants, the difference in the SDNN value is therefore 45.37% compared to ATC1 and 39.99% compared to ATC3. Participants ATC1 and ATC3 had, at the significance level $α = 0.05$ , the median values of this parameter identical (median 49.15 and 44.75). These higher values indicate better physiological adaptability to workload than participant ATC2.

Mean RR, which is inversely proportional to heart rate and can indicate stress level, was, at the significance level $α = 0.05$ , only different between ATC2 (median 630.3) and ATC3 (median 669.35), the difference is therefore 5.83%. A high Mean RR value corresponds to a lower heart rate and lower perceived workload in ATC3. Participant ATC1 (median 647.15) showed, at the significance level $α = 0.05$ , Mean RR parameter values identical to the values of the other two participants ATC2 and ATC3.

In practice, it is useful to use the HR coefficient instead of RR values. This means that the obtained RR is normalized by the resting value, which reflects the subject's condition prior to the experiment. This eliminates the effects of health status, physical fitness, age, and similar factors. In our case, values from the resting phase before the experiment were used. However, since these were not morning values but resting values measured directly before the experiment, the resting value was further reduced by 15%. This approach is implemented in practice by CASRI and CTU FBMI KIT.

The resulting adjusted values cannot be used for HRV analysis; however, the coefficient allows for a better comparison of different subjects in similar situations regardless of each participant's resting HR (see Figure 2).

Figure 2.

Representation of HR coefficient differences using a box plot.

The observed pattern, in which ATC2 consistently exhibited the highest sympathetic activation across all three parameters while ATC1 and ATC3 remained statistically indistinguishable on two of three measures, is unlikely to reflect differences in task exposure, as all three controllers faced identical traffic scenarios. The most plausible explanations lie in individual differences in autonomic reactivity, coping strategy, and baseline physiological resilience. As Kim et al.³⁶ demonstrated, the LF/HF ratio responds differently to identical stressors depending on whether an individual adopts an active or passive coping style, which may account for the markedly elevated sympathetic dominance observed in ATC2 despite equivalent task demands. The relatively preserved SDNN values of ATC1 and ATC3, both approaching normative ranges reported by Nunan et al.⁵¹ for healthy resting adults, indicate that these two controllers maintained adequate parasympathetic reserve throughout the exercise, suggesting either greater stress resilience, greater cognitive reserve,⁵² or a lower individual sensitivity of the autonomic nervous system to this particular type of operational demand.

These results clearly confirm that workload is individual and cannot be universally attributed to objective situational parameters alone. Although all controllers underwent the same simulation scenario, their physiological responses differed significantly. This suggests that subjective perception of workload and individual stress reactivity play a key role in air traffic controllers’ job performance.

These findings are consistent with several lines of evidence in the literature. Zaytsev⁵⁰ demonstrated that while mean ATC reaction times remain stable across different times of day, individual-level responses diverge significantly, revealing that group averages systematically obscure the variability that matters most in safety-critical settings. Triyanti et al.¹¹ confirmed that ATC mental workload is not only high but shaped by individual factors including experience, personality, and stress management capacity. The present study strengthens both findings by providing objective physiological evidence: even when individual factors and personality traits are held constant in the sense that all three controllers faced the same traffic scenario, the autonomic nervous system responses diverged by 74% in LF/HF and 45% in SDNN. Importantly, a seminal population study of 205 ATCs by Zeier⁵³ found that while mean psychophysiological stress indicators fell within normal ranges, approximately 10 to 15% of controllers showed elevated stress symptom values to an extent indicating serious stress problems at work or in private life,⁵³ confirming that a subgroup of controllers is consistently more vulnerable, a pattern the present findings replicate at the physiological level in real time.

The safety implications are direct and quantifiable. A 10% degradation in operator performance would already be considered operationally significant in any safety-critical system. The present study documents differences exceeding 50% across all three HRV parameters between the highest- and lowest-loaded controller working the same scenario at the same time. This magnitude is not a marginal effect. As workload increases, the ATC tends to employ less time-consuming procedures, progressively reduce flight information, and relax self-imposed qualitative criteria, a pattern that can lead to a very risky situation when the decision-making capacity is stretched to its maximum.⁵⁴ If two controllers handle the same traffic volume but one is operating at physiologically extreme load while the other retains substantial reserve capacity, the system is not in a stable state. The risk of error, separation loss, or situational awareness degradation is asymmetrically distributed in a way that fixed traffic-based capacity thresholds cannot detect.

These findings have three concrete operational consequences. First, physiological monitoring during simulations and potentially during real operations makes it possible to identify controllers whose autonomic stress profiles consistently place them at elevated risk, enabling targeted resilience training and psychological support before deficits translate into operational errors. Second, personalised shift scheduling informed by individual physiological baselines can distribute cognitive demands more equitably, reducing the accumulated fatigue that is among the most consistently documented contributors to ATC performance degradation.⁵⁵ Third, and most fundamentally, the results challenge the prevailing model of airport capacity as a fixed threshold expressed in aircraft movements per hour. If the effective operational capacity of a given sector is bounded not only by traffic complexity but by the physiological state of the individual controller on duty, then capacity declarations must incorporate a real-time human factors dimension.^14,15 A controller with a median LF/HF of 7.3 and SDNN of 26.85 ms, as observed for ATC2 in this study, is operating in a physiological regime that the Eurocontrol heavy-load classification^21,46 does not distinguish from a controller with a median LF/HF of 4.2 and SDNN of 44.75 ms. Yet from an autonomic and cognitive reserve perspective, these two individuals are not managing the same workload, even when every external parameter of their operational environment is identical.

The results of this study are directly comparable with and contextually positioned against the broader body of HRV-based workload research in aviation. Table 6 summarises key previous studies on HRV.

Table 6.

Comparison of key previous studies on physiological workload assessment in aviation.

Study	N	Task / operational context	Interval	LF/HF ratio	SDNN	Mean RR
Cao et al. (2019)²⁹	30 commercial pilots	A320 sim: manoeuvres of varying difficulty; 3 CO₂ conditions	30 s	5.0–7.0 (range across all active manoeuvres)	34.1 ± 12.7 ms Lowest on hardest manoeuvres	Not separately reported
Li et al. (2022)⁴⁵	26 participants	Flight multitasking – 3 load levels (varying subtask count)	Not specified	LF/HF measured – not significant across task load levels	SDNN negatively correlated with subjective workload	Not separately reported
Yang et al. (2023)⁵⁶	68 cadet pilots	Simulated flight: plane tracking, anti-gravity pedalling, reaction tasks	Task-based	LF/HF significantly higher under high CTL vs. low CTL	SDNN significant (p < 0.05)	Not separately reported
Yuan et al. (2025)⁴⁴	Flight cadets	A320 sim: takeoff, climb, cruise, descent, landing	30 s windows (40% overlap)	LF/HF increase during takeoff and landing vs. cruise	SDNN decrease during takeoff	Not separately reported
Wang et al. (2025)⁵⁷	40 pilots	Flight sim: foggy vs. clear weather	Per-minute	LF/HF significant across workload levels	SDNN decrease in flight vs. rest	Not separately reported
Wang et al. (2024)¹⁸	29 studies (review)	Multiple flight tasks and phases across 29 studies	Various	LF/HF: inconsistent – increase in some studies, n.s. in others	SDNN: most frequently investigated, generally decreasing	Mean RR: generally decreasing with higher MWL

Several important observations emerge from this comparison. First, the present study is the only one in this table that explicitly holds the operational scenario constant across participants while treating inter-individual physiological variation as the primary dependent variable. All other studies that measured physiological responses manipulated task demands across conditions or compared group-level responses. This makes the present contribution complementary to rather than redundant with existing work. Second, and equally significant, is the near-complete absence of ATC-specific studies reporting explicit numerical HRV values comparable to those presented in Table 6. The search of the available literature identified only two ATC-focused HRV studies: Socha et al.,³⁰ who monitored heart activity of ATCOs under varying traffic loads and airspace complexity using LF/HF-based frequency analysis, and Fürstenau et al.,⁵⁸ who examined cardiovascular biomarkers including lg(LF/HF) during ATC simulation with varying traffic volumes. Both studies confirmed the expected directional relationship between traffic load and HRV, but neither reported interval-level median values of LF/HF, SDNN, or mean RR with the temporal resolution applied in the present study

It must be acknowledged directly that several existing methods (other than HRV) substantially outperform the present approach in terms of workload classification accuracy. EEG-based ML classifiers have achieved accuracies as high as 98.18% in controlled laboratory settings,⁵⁹ and multimodal approaches combining EEG with fNIRS have reached approximately 90% in binary workload classification.^60,61 The present study does not report a classification accuracy because it does not attempt workload-level classification. Its objective is fundamentally different.

For HRV indicators, the literature broadly shows that LF/HF increases and SDNN decreases under elevated mental workload,¹⁸ though results across studies are inconsistent. In simulated flight multitasking studies, there were no significant differences in HRV (SDNN, LF/HF) among different task load levels, but mean HR increased with workload, highlighting that not all HRV parameters respond consistently across all aviation contexts.⁴⁵ The present study is therefore notable in showing clear and statistically significant differences in all three measured parameters, though not within individuals across task conditions, but between individuals under the same condition. This between-subject sensitivity is precisely what the majority of prior HRV aviation studies, which average responses across participants, systematically obscure.

A science mapping analysis of ATC workload research confirms that many different techniques and methods have been used, with mixed results across studies, and that the field is still developing.⁶² In this context, the present study's contribution is not to propose the most accurate workload classifier, but to demonstrate with appropriate statistical rigour that a operationally deployable, wearable ECG system can reliably detect between-controller physiological differences.

It must be honestly acknowledged that the ECG-HRV approach used here does not achieve the classification accuracy of ML-based EEG or multimodal methods reported in laboratory studies.^25,59 This is not an inherent failure of the method but a consequence of the deliberate prioritisation of operational deployability over classification precision. A 74% difference in median LF/HF between two controllers is not a subtle statistical signal requiring sophisticated ML extraction, it is a substantial and operationally consequential difference detectable by straightforward non-parametric comparison. The statistical methodology applied yielding p-values below 0.002 for LF/HF and SDNN provides robust convergent evidence that does not depend on model assumptions or classification thresholds. The real comparative advantage of the proposed approach is therefore its combination of three properties that no existing higher-accuracy alternative simultaneously possesses: (1) full operational compatibility with ATC equipment and environment; (2) continuous, real-time, non-intrusive physiological monitoring throughout an entire shift or exercise; and (3) statistical sensitivity sufficient to detect the large-magnitude inter-individual differences that, as this study demonstrates, genuinely exist and cannot be assumed away by treating controllers as physiologically interchangeable.

The most significant limitation is the small sample size (N = 3). While a case study design is appropriate for an exploratory proof-of-concept, the findings cannot be generalised to the broader ATC population, and the observed differences may not be representative of controllers varying in age, experience, gender, or chronotype. It should be mentioned that in the technical field there are studies related to statistical data processing techniques that may offer a different approach, see^63,64 and it would certainly be interesting in the future to try their use on our dependent small-sample physiological data.

No standardised morning baseline ECG was recorded, the pre-exercise rest values used for HR coefficient normalisation were measured immediately before the exercise and required an empirical correction of 15%, reducing the precision of individual normalisation for SDNN and mean RR. Future studies should incorporate a standardised 5-min supine resting baseline at a fixed time of day.

The study relies exclusively on ECG-derived HRV, capturing only the autonomic cardiovascular dimension of workload. It does not reflect cortical processing load, attentional resources, or fatigue accumulation. A multimodal approach incorporating respiratory rate, skin temperature, fNIRS, or EEG would provide a more complete picture. Additionally, the LF/HF ratio remains methodologically contested,^37,38 as respiratory rate and individual coping style can shift spectral power between bands independently of cognitive load.³⁶ Future studies should complement LF/HF with RMSSD.

Finally, all measurements were conducted in a simulator. While the FlexiGuard system has been validated in operational contexts and the CANI simulator provides high-fidelity TWR conditions, it remains an open question whether physiological responses in simulation faithfully replicate those during real operations, as the absence of consequential risk may attenuate sympathetic arousal in some individuals.

Conclusion

This study provided evidence that workload in air traffic control is highly individual and cannot be simply inferred from external factors such as traffic density or situation complexity. The data obtained directly confirms individual differences in physiological responses to workload. Individual differences in physiological responses indicate that workload assessment should be based not only on operational parameters but also on objective biometric data.

The originality of this study lies in its explicit focus on between-controller variability rather than the within-individual workload-versus-traffic relationship that dominates the ATC literature.^14,23 Most prior studies manipulate traffic volume and measure how a given controller's physiology responds.^4,13 The present study held the task scenario constant across all three participants, isolating individual physiological reactivity as the primary variable. This design reveals that the same heavy-load exercise generated LF/HF differences of up to 74.28%, SDNN differences of up to 45.37%, and mean RR differences of 5.83% between controllers, magnitudes not previously quantified in a controlled TWR simulator context.

A further distinguishing feature is the FlexiGuard platform, which imposes no constraints on headgear, head movement, or voice communication, making it genuinely compatible with real ATC shift deployment in a way that laboratory EEG or fNIRS setups cannot achieve.

The study also proposes a conceptual shift in how airport capacity is defined. Rather than treating sector capacity as a fixed number of aircraft movements per hour as prescribed by Eurocontrol methodology,^21,46 the results support understanding effective capacity as a dynamic interval conditioned on the real-time physiological state of the individual controller on duty.

Future work should replicate these findings with a larger, stratified sample and adopt a longitudinal design to assess whether individual reactivity profiles are stable traits. Future studies should also explore whether brief, pre-shift HRV assessments measuring baseline autonomic tone can serve as practical screening tools to predict which controllers are entering a shift in a physiologically vulnerable state.

In summary, this study provides original empirical evidence that ATC workload is not uniformly determined by traffic complexity but is profoundly shaped by individual physiological characteristics, calling for a reconsideration of how workload is monitored, how capacity is defined, and how individual controllers are supported in safety-critical environments.

Footnotes

Acknowledgements

The conducted experiments were carried out in cooperation with the Ministry of Defence workplace, CASRI, p.o., which performed the recording of physiological parameters and their subsequent processing. The sensing technology is the result of long-term collaboration between the CASRI, p.o. team and CTU KIT.

ORCID iDs

Tomáš Hoika

Jiří Millek

Šárka Hošková-Mayerová

Jiří Jánský

Zbyšek Korecki

Ethical considerations

The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board (or Ethics Committee) of the University of Defence (protocol code 11/2025, date of approval 6/10/2025).

Consent to participate

All participants provided written informed consent prior to participating.

Consent for publication

Not applicable. This study presents only fully anonymized data with no possibility of identifying individual participants.

Author contributions

Conceptualization, TH and ZK; methodology, TH, JM and SH-M; validation and formal analysis, SH-M, JJ and JM; investigation and data curation, JM and JJ; writing—original draft preparation and review and editing, TH, SH-M, JJ, and JM; visualization, JM and JJ; supervision SH-M; project administration and funding acquisition, SH-M and ZK. All authors have read and agreed to the published version of the manuscript.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Ministry of Defence of the Czech Republic, grant numbers AIROPS (22572/2022-HA) and VAROPS (23602/2023-HA).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability

Data used to support the findings are included within the article.

References

Marchitto

. Cognitive workload and complexity in air traffic control: simulation of operational tasks in eye tracking studies. Universidad de Granada, 2015.

Athenes

Averty

Puechmorel

, et al. ATC complexity and controller workload: Trying to bridge the gap. In: Proceedings of the International Conference on HCI in Aeronautics, 2002.

Majumdar

Ochieng

. Factors affecting air traffic controller workload: multivariate analysis based on simulation modeling of controller workload. Transp Res Rec 2002; 1788: 58–69.

Wiediartini

Ciptomulyono

Dewi

. Physiological and biochemical measures of mental workload of air traffic controllers: a systematic literature review. In: Proceedings of the 2nd Stifler Annual Conference, 2021. DOI: 10.46254/SA02.20210677

Cummings

Tsonis

. Deconstructing complexity in air traffic control. Int J Aviat Psychol 2006; 16: 297–316.

NASA. NASA Task Load Index (TLX), https://www.nasa.gov/human-systems-integration-division/nasa-task-load-index-tlx/ (accessed 31 March 2025).

Bolton

Biltekoff

Humphrey

. The mathematical meaninglessness of the NASA task load Index: a level of measurement analysis. IEEE Trans Hum-Mach Syst 2023; 53: 590–599.

Hart

. NASA-Task Load Index (NASA-TLX); 20 years later. Proc Hum Factors Ergon Soc Annu Meet 2006; 50: 904–908.

Orn

. Everything you need to know about the NASA-TLX. Research Collective, https://research-collective.com/everything-you-need-to-know-about-the-nasa-tlx/ (2023, accessed 31 March 2026).

10.

Diarra

Theurel

Paty

. Systematic review of neurophysiological assessment techniques and metrics for mental workload evaluation in real-world settings. Front Neuroergonomics 2025; 6. DOI: 10.3389/fnrgo.2025.1584736

11.

Triyanti

Azis

Prasetyawan

, et al. Individual factors related to mental workload in air traffic controller. In: Gutierrez

AMJ

Goonetilleke

Robielos

RAC

(eds) Convergence of ergonomics and design. Springer, 2021, pp.271–278. DOI: 10.1007/978-3-030-63335-6_28

12.

Zamarreño Suárez

Arnaldo Valdés

Pérez Moreno

, et al. Understanding the research on air traffic controller workload and its implications for safety: a science mapping-based analysis. Saf Sci 2024; 176: 106545.

13.

Mogford

Guttman

Morrow

, et al. The complexity construct in air traffic control: a review and synthesis of the literature. Storming Media, 1995.

14.

Pagnotta

Jacobs

de Frutos

, et al. Task difficulty and physiological measures of mental workload in air traffic control: a scoping review. Ergonomics 2022; 65: 1095–1118.

15.

Young

Brookhuis

Wickens

, et al. State of science: mental workload in ergonomics. Ergonomics 2015; 58: 1–17.

16.

Tao

Tan

Wang

, et al. A systematic review of physiological measures of mental workload. Int J Environ Res Pub Health 2019; 16: 2716.

17.

W-C

Zhang

Kearney

, et al. Psychophysical coherence training regulating air traffic controller’s heart rate variability and resilience to fatigue. In: Harris

W-C

(eds) Engineering psychology and cognitive ergonomics. Springer, 2021, pp.142–150. DOI: 10.1007/978-3-030-77932-0_12

18.

Wang

Houghton

Majumdar

. Detecting and predicting pilot mental workload using heart rate variability: a systematic review. Sensors 2024; 24: 3723.

19.

Di Mascio

Carrara

Frasacco

, et al. How the tower air traffic controller workload influences the capacity in a complex three-runway airport. Int J Environ Res Pub Health 2021; 18: 2807.

20.

Dismukes

Berman

Loukopoulos

. The limits of expertise: rethinking pilot error and the causes of airline accidents. Ashgate, 2007.

21.

Eurocontrol. Airport capacity methodology assessment. https://www.eurocontrol.int/sites/default/files/publication/files/nom-apt-acap-acamman-v1-1.pdf (accessed 6 July 2025).

22.

Żyluk

Zieja

Tomaszewska

, et al. Service life prediction for rotating electrical machines on aircraft in terms of temperature loads. Energies 2023; 16: 18.

23.

Radüntz

Fürstenau

Mühlhausen

, et al. Indexing mental workload during simulated air traffic control tasks by means of dual frequency head maps. Front Physiol 2020; 11: 300.

24.

Luzzani

Pogliano

Buraioli

, et al. ECG, respiration, fNIRS, and eye tracking for stress and mental workload monitoring in human-machine interaction. IEEE Access 2025; 13: 1–1.

25.

Mark

Curtin

Kraft

, et al. Mental workload assessment by monitoring brain, heart, and eye with six biomedical modalities during six cognitive tasks. Front Neuroergonomics 2024; 5. DOI: 10.3389/fnrgo.2024.1345507

26.

Dybvik

Erichsen

Snider

, et al. fNIRS, EEG, ECG, and GSR reveal an effect of complex, dynamically changing environments on cognitive load, affective state, and performance, but not physiological stress. Front Hum Neurosci 2025; 19: 1459653.

27.

KKH

SCM

, et al. Recognising situation awareness associated with different workloads using EEG and eye-tracking features in air traffic control tasks. Knowledge-Based Syst 2023; 260: 110179.

28.

Malik

Bigger

Camm

, et al. Heart rate variability: standards of measurement, physiological interpretation, and clinical use. Euro Heart J 1996; 17: 354–381.

29.

Cao

MacNaughton

Cadet

, et al. Heart rate variability and performance of commercial airline pilots during flight simulations. Int J Environ Res Pub Health 2019; 16: 37.

30.

Socha

Hanáková

Valenta

, et al. Workload assessment of air traffic controllers. Transp Res Procedia 2020; 51: 243–251.

31.

Shaffer

Ginsberg

. An overview of heart rate variability metrics and norms. Front Pub Health 2017; 5: 58.

32.

Houle

Billman

. Low-frequency component of the heart rate variability spectrum: a poor marker of sympathetic activity. Am J Physiol – Heart Circ Physiol 1999; 276: H215–H223.

33.

Aysin

Colombo

Aysin

. Comparison of HRV analysis methods during orthostatic challenge: HRV with respiration or without? In: Proceedings of the 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2007, pp.5047–5050. DOI: 10.1109/IEMBS.2007.4353474

34.

Kitamura

Murai

Wakida

. Evaluation of mental workload for a newly-appointed pilot using salivary NO3− concentration and LF/HF values, compared with port-coordinators. In: Proceedings of the 2017 Joint 17th World Congress of IFSA and 9th International Conference on SCIS (IFSA-SCIS), 2017, pp.1–5. DOI: 10.1109/IFSA-SCIS.2017.8023280

35.

Pilmannová

Hanáková

Socha

, et al. Student air traffic controllers’ performance under conditions of increased workload. In: Human Factors in Transportation. AHFE Open Access, 2023. DOI: 10.54941/ahfe1003839

36.

Kim

Koo

Lee

, et al. Application and limitation of frequency domain, LF/HF component in heart rate variability as an acute stress index. In: Proceedings of the International Conference on Biomedical Engineering and Systems (ICBES), 2014. https://www.avestia.com/ICBES2014_Proceedings/papers/128.pdf

37.

Billman

. The LF/HF ratio does not accurately measure cardiac sympatho-vagal balance. Front Physiol 2013; 4: 26.

38.

von Rosenberg

Chanwimalueang

Adjei

, et al. Resolving ambiguities in the LF/HF ratio: lF-HF scatter plots for the categorization of mental and physical stress from HRV. Front Physiol 2017; 8: 60.

39.

Delliaux

Delaforge

Deharo

J-C

, et al. Mental workload alters heart rate variability, lowering non-linear dynamics. Front Physiol 2019; 10: 565.

40.

Guo

Wang

Qin

, et al. Assessment of flight fatigue using heart rate variability and machine learning approaches. Front Neurosci 2025; 19. DOI: 10.3389/fnins.2025.1621638

41.

Järvelin-Pasanen

Sinikallio

Tarvainen

. Heart rate variability and occupational stress: systematic review. Industr Health 2018; 56: 500–511.

42.

Solhjoo

Haigney

McBee

, et al. Heart rate and heart rate variability correlate with clinical reasoning performance and self-reported measures of cognitive load. Sci Rep 2019; 9: 14668.

43.

Borchini

Veronesi

Bonzini

, et al. Heart rate variability frequency domain alterations among healthy nurses exposed to prolonged work stress. Int J Environ Res Pub Health 2018; 15: 13.

44.

Yuan

Jia

Zhang

, et al. Pilot mental workload analysis in the A320 traffic pattern based on HRV features. Front Neuroergonomics 2025; 6. DOI: 10.3389/fnrgo.2025.1672492

45.

Xie

, et al. Evaluating mental workload during multitasking in simulated flight. Brain Behav 2022; 12: e2489.

46.

Eurocontrol. Pessimistic sector capacity estimation. Eurocontrol, 2008.

47.

Kliment

Smrčka

Hána

, et al. Wearable modular telemetry system for the integrated rescue system operational use. J Sensors 2017; 2017: 9034253.

48.

Hollander

Wolfe

Chicken

. Nonparametric statistical methods. 3rd ed. Wiley, 2013. DOI: 10.1002/9781119196037

49.

Wickens

Mavor

McGee

. Flight to the future: human factors in air traffic control. National Academy Press, 1997.

50.

Zaytsev

. Impact of day time on performance of air traffic controllers. Czech Technical University, 2023. http://hdl.handle.net/10467/111294

51.

Nunan

Sandercock

GRH

Brodie

. A quantitative systematic review of normal values for short-term heart rate variability in healthy adults. Pacing Clin Electrophysiol 2010; 33: 1407–1417.

52.

Stern

. What is cognitive reserve? Theory and research application of the reserve concept. J Int Neuropsychol Soc 2002; 8: 448–460.

53.

Zeier

. Workload and psychophysiological stress reactions in air traffic controllers. Ergonomics 1994; 37: 525–539.

54.

Costa

. Occupational stress and stress prevention in air traffic control. International Labour Office, 1995. https://skybrary.aero/sites/default/files/bookshelf/1643.pdf

55.

Peukert

Claus

Meyer

. Subjective and objective fatigue dynamics in air traffic control. Industr Health 2025; 63: 431–442.

56.

Yang

Tang

Cong

, et al. Development and validation of an assessment index for quantifying cognitive task load in pilots under simulated flight conditions using heart rate variability and principal component analysis. Ergonomics 2024; 67: 515–525.

57.

Wang

Guo

, et al. Machine learning evaluation model of pilot workload in a low-visibility environment. Sci Rep 2025; 15: 20518.

58.

Radüntz

Mühlhausen

Freyer

, et al. Cardiovascular biomarkers’ inherent timescales in mental workload assessment during simulated air traffic control tasks. Appl Psychophysiol Biofeedback 2021; 46: 43–59.

59.

Pušica

Mijović

Leva

, et al. Machine learning performance in EEG-based mental workload classification across task types: a systematic review. Front Neuroergonomics 2025; 59. DOI: 10.3389/fnrgo.2025.1621309

60.

Zhang

Luo

Cao

, et al. Evaluating pilot mental workload using fNIRS-based functional connectivity features with a deep residual shrinkage network under emergency flight scenarios. Int J Hum–Comput Interact 2025; 41: 10443–10458.

61.

Cao

Garro

Zhao

. EEG/fNIRS based workload classification using functional brain connectivity and machine learning. Sensors 2022; 22: 7623.

62.

Suarez

Lopez de Frutos

Puntero

, et al. Quantifying air traffic controller mental workload. In: Proceedings of the SESAR Innovation Days (SIDs 2014), 2014.

63.

Karasu

. A novel hybrid model using demand concentration curves, chaotic AFDB-SFS algorithm and Bi-LSTM networks for heating oil price prediction. Electronics 2025; 14: 4814.

64.

Karasu

Altan

. Crude oil time series prediction model based on LSTM network with chaotic Henry gas solubility optimization. Energy 2022; 242: 122964.