The association between internal and external measures of training load in elite rowers during ergometer training

Abstract

While traditional rowing metrics such as stroke rate and split times offer basic performance feedback, they do not fully reflect individual physiological responses, and the relationship between external and internal training loads remains inadequately characterized in rowing ergometer training. This study systematically investigated the correlations between two key internal load measures—training impulse (TRIMP) and session rating of perceived exertion training load (sRPE_TL)—and three biomechanical external load indicators in elite rowers, with the aim of establishing a more comprehensive training monitoring framework. Thirty elite male rowers (age 22.1 ± 3.6 years, VO₂max 64.5 ± 6.1 mL·min⁻¹·kg⁻¹) were monitored throughout a 48-week preparatory training period, with 1196 sessions analyzed. Participants who had missed more than three months of training were excluded. External training load metrics—including duration, stroke rate, stroke count (SC), and 500-m split time—and internal training load measures (TRIMP and sRPE_TL) were collected using standardized procedures. The percentage of boat speed (S%) was calculated by comparing the average boat speed during training to an individually determined maximum speed. Most external load measures, except duration, split time, and stroke rate, showed significant correlations with sRPE_TL (p < .001, FDR-adjusted). Individualized metrics strengthened these relationships, with SC and S% demonstrating moderate correlations with sRPE_TL (r = .41 and r = .40, respectively; p < .001, FDR-adjusted p = .002). TRIMP correlated with all external load metrics except stroke rate and count, most strongly with S% (r = .39, p < .001, FDR-adjusted p = .002). Multivariate regression identified SC, S%, and TRIMP as the strongest predictors of sRPE_TL (marginal R² = .50; model: Y = − 477.8 + 3.477 × S% + 17.262 × SC + 2.156 × TRIMP). In summary, S%, SC, and TRIMP were identified as optimal parameters for monitoring training load, collectively accounting for 50% of the variance in sRPE_TL—outperforming conventional external load metrics. These results support the integration of individualized speed measures with physiological and perceptual indicators to enhance the precision of training prescription in elite rowing programs.

Introduction

Training Load (TL) quantification represents a fundamental challenge in sports science, serving as the critical link between training stimuli and physiological adaptation¹. The dose–response relationship between TL and training adaptation is mediated through two key mechanisms: (1) cumulative metabolic stress that drives cellular adaptation, and (2) sport-specific neuromuscular patterning². This dual-pathway framework explains why precise TL quantification must account for both external work output and individual physiological responses—a challenge particularly salient in endurance sports like rowing.

Traditional monitoring predominantly relies on external load metrics (e.g., power output, distance), which provide objective measures of work performed³. However, these external training load (ETL) metrics exhibit significant limitations: they cannot detect inter-athlete differences in metabolic efficiency⁴, nor reflect the psychological stress components that substantially influence adaptation⁵. This discrepancy between external work and physiological response constitutes the “external-internal load decoupling” phenomenon—particularly pronounced in endurance sports where cardiovascular drift and prolonged neuromuscular fatigue alter the load-response relationship over time⁶.

Rowing training presents unique monitoring challenges due to its combination of cyclic movement patterns, sustained aerobic demand, and high intramuscular coordination requirements. The stationary ergometer, while eliminating environmental variables present in on-water training, introduces specific physiological demands including constrained ventilatory mechanics and continuous upper-body loading⁷. These characteristics necessitate specialized load monitoring approaches that account for: (1) the time-dependent relationship between stroke mechanics and metabolic cost, and (2) the cumulative effect of repetitive motion patterns on perceptual strain.

Current ergometer-based monitoring utilizes stroke rate, total stroke count, and split times as primary ETL measures⁸. However, these measures alone provide an incomplete picture, as rowing physiology demonstrates that internal load arises from the interaction of volume (total strokes), intensity (stroke power), and technique efficiency (force application profile)^9,10. This complexity is exemplified by studies showing that identical stroke counts can produce markedly different physiological strains depending on stroke dynamics and rest intervals.

The relationship between TL metrics warrants particular attention in rowing given the sport’s physiological characteristics. While heart rate (HR) monitoring has been widely adopted, its utility may be limited by the arm-dominated exercise’s unique cardiovascular demands¹¹. Recent evidence suggests session rating of perceived exertion(sRPE) may better capture the combined cardiorespiratory and local muscle fatigue in rowing than traditional physiological measures; however, its relationship with specific ETL measures remains unestablished¹². Prior investigations across different sports have demonstrated moderate correlations between sRPE and running-based ETL measures in rugby players¹¹, while studies in soccer (another team sport) and in sprinting (an individual sport) revealed the mediating role of psychological state¹² and velocity-dependent perceptual variations¹³. However, no studies have systematically examined these relationships in rowing ergometer training.

Therefore, this study aims to establish the quantitative relationships between three key ETL measures (blood lactate concentration, rowing speed, and total stroke count) and two internal load measures (HR-derived training impulse and sRPE) in elite rowers. The findings will provide evidence for developing rowing-specific load monitoring frameworks that more effectively account for the sport’s unique physiological demands than generic models.

Methods

Participants

Thirty elite male rowers (age: 22.1 ± 3.6 years, height: 1.94 ± 0.06 m, body mass: 91.8 ± 5.7 kg) participated in this study. At their most recent assessment, the rowers’ absolute maximal oxygen consumption (VO₂max) was 6110.6 ± 459.3 mL·min⁻¹, with a relative value of 64.5 ± 6.1 mL·min⁻¹·kg⁻¹. Inclusion criteria were: (a) senior team membership for ≥ 2 years, (b) ongoing preparation for the National Games, and (c) ≥ 8 years of systematic training. Exclusion criteria included: (a) a training absence of > 3 consecutive months due to injury or illness in the preceding year, or (b) inability to complete the testing protocols. Prior to all testing sessions, subjects were instructed to refrain from intense exercise for at least 48 h, avoid caffeine and alcohol for 24 h, and maintain their normal hydration and nutritional status. At the time of data collection (the preparatory period, March 2024 to March 2025), the rowers followed a periodized training program comprising 13–15 sessions per week. The rowers were classified as national-level athletes with a mean of 7.2 ± 2.1 years of systematic training experience. Their primary competitive events were 2000-m races in both sculling and sweep rowing. This typically included eight rowing-specific sessions (on-water or ergometer), two strength training sessions, and two functional training sessions per week. The functional training sessions were primarily focused on injury prevention and core stability, incorporating exercises such as plyometrics, dynamic stretching, and movements targeting the shoulder, lower back, and knee joints. All participants provided written informed consent after being fully informed of the study procedures, which were approved by the Ethics Committee of *** (No. 2025A015).

Experimental design

Both internal and external training loads of elite rowers were monitored over a 48-week preparatory period (March 2024–March 2025) during routine training sessions at the National Aquatic Sports Center (Beijing, China). A total of 1248 individual sessions were initially recorded. Data from one rower, who was excluded due to a lower-back injury that resulted in a training absence exceeding three months, were removed from the final analysis.

This study was conducted under non-interventional, observational conditions; the research team did not influence training prescriptions, which were designed and implemented exclusively by the national team coaching staff. The weekly training structure (Table 1) was periodized and included a mix of ergometer and on-water sessions. We controlled for variability in training modality by focusing our analyses on a standardized rowing session. This session was an 18-km, continuous, moderate-intensity exercise performed every Monday morning on Concept2 Model D ergometers.These sessions commenced at 09:00 AM following a standardized 15-min warm-up (10 min of light rowing and 5 min of dynamic stretching) supervised by the coaching staff. Sundays were rest days, minimizing cumulative fatigue carryover. All data were visually inspected for completeness; incomplete entries were excluded, resulting in a final dataset of 1,196 sessions with complete paired external and internal load measures.

Table 1.

Detailed example of training week1.

Time	Monday	Tuesday	Wednesday	Thursday	Friday	Saturday	Sunday
6:00 a.m	Hematological test	On-water 16 km	On-water 20 km	Ergometer 6 km × 3	On-water 14 km	Ergometer 6 km × 3	OFF
8:30 a.m	Ergometer 18 km				Core
3:00 p.m	RT 120 min	Ergometer 10 min × 3	OFF	RT 120 min	On-water 16 km	OFF	OFF
7:00 p.m	FT 30 min	OFF	FT 30 min	OFF	OFF

Abbreviations: RT, resistance training; FT, functional training.

Data collection

Training impulse (TRIMP)

HR was collected simultaneously from each player using the Polar Team Pro 2 system (Polar Electro Oy, Kempele, Finland), which sampled at 1 s intervals throughout each training session. The system consisted of a Pro Shirt with an integrated sensor and a Team Pro smartphone app for real-time monitoring. All devices were inspected and fully charged before each session.

Additionally, TRIMP method¹⁴ for quantifying internal-TL was determined as:

1

△HR_ratio = (HR_exercise-HR_rest) /(HR_max-HR_rest), y is a multiplying factor of 0.64e^1.92x.

whereby e≈2.718 and x = △HR.

The maximum heart rate (HRmax) for each individual was determined based on the highest heart rate achieved during their VO_2max test performed on an ergometer. Resting heart rate (HRrest) was measured after 10 min of supine rest in a quiet, temperature-controlled room (20–22 °C) on the morning of a rest day¹⁵.

Blood lactate

Within 30 s after each monitored session, earlobe capillary blood samples (10 μL) were collected by one of three trained researchers (all holding master’s degrees in sports science with over two years of phlebotomy experience). All researchers completed a standardized training program in capillary blood sampling techniques prior to the study. To ensure measurement consistency, a preliminary intra- and inter-rater reliability test was conducted prior to the formal study. Ten capillary blood samples were analyzed in duplicate by each researcher using the same Lactate Scout + analyzer. The intra-class correlation coefficient (ICC) for intra-rater reliability was > 0.95, and for inter-rater reliability was > 0.90, indicating excellent consistency. Blood lactate concentration was analyzed immediately using a portable Lactate Scout + analyzer (EKF Diagnostics, Magdeburg, Germany), which was calibrated daily using manufacturer-provided standards. Sampling was conducted indoors in a temperature-controlled environment (20–22 °C) with humidity maintained at 40–60%. Participants were instructed to maintain their normal nutritional and hydration practices and to avoid caffeine for 3 h prior to training¹⁶. They were also asked to refrain from using anti-inflammatory medications or nutritional supplements known to affect lactate metabolism for 24 h before testing. Given the focus on practical application and the rapid kinetics of lactate after moderate-intensity exercise, values were not adjusted for plasma volume shifts.

sRPE

Data collection for each training session was conducted with the consent of the athlete’s department and the head coach. Within 30 min post-exercise¹⁷, in a quiet room adjacent to the training facility, rowers provided their sRPE. They were previously familiarized with the CR-10 scale through both written and verbal instructions during two pilot sessions before the study began. The familiarization protocol included watching standardized exercise videos and practicing rating perceived exertion using the scale¹⁸. The reliability of the sRPE scale administration was assessed during these pilot sessions; rowers provided sRPE ratings for standardized exercise bouts, demonstrating good test–retest reliability (ICC = 0.87). A researcher with a PhD in exercise psychology and three years of experience in perceptual scale administration was present to answer questions, but rowers recorded their ratings independently. The sRPE training load (sRPE_TL) was calculated as sRPE × session duration (in minutes).

Percentage of boat speed (S%)

Boat speed was individualized relative to each rower’s maximal capacity. Following a standard warm-up consisting of 10 min of light rowing (50% of perceived maximum), dynamic stretching, and three 30-s build-up strokes, each rower performed a maximal 2000-m time trial on a Concept2 Model D ergometer (Concept2, Morrisville, VT, USA) during the first week of the preparatory phase. The average speed (in m/s) from this test defined their individual maximal speed (Smax). The percentage of boat speed (S%) for any training session was calculated as: S% = (Session Average Speed/Smax) × 100.

Statistical analysis

To account for the repeated-measures design (multiple sessions per athlete), the relationships between internal and external training load (TL) measures were primarily analyzed using linear mixed-effects models with athlete ID included as a random intercept. Pearson’s correlations are reported for comparative purposes, with the strength classified as trivial (0–0.1), small (0.1–0.3), moderate (0.3–0.5), large (0.5–0.7), very large (0.7–0.9), or almost perfect (0.9–1). To build a predictive model for sRPETL, we employed a hierarchical (block-wise) multiple regression approach within the mixed-model framework, where external load metrics were entered in a theoretically informed order. The false discovery rate (FDR) procedure was used to adjust p values for the multiple correlation tests performed. All results are reported with exact p values, effect sizes, and 95% confidence intervals (CIs). Analyses were conducted using SPSS (version 28, IBM Corporation) with the significance level set at α = 0.05. A post hoc power analysis conducted in G*Power (version 3.1) for the key correlation (r = 0.40, α = 0.05) indicated that our sample size (n = 1196 observations) provided a statistical power > 0.99, well above the conventional threshold of 0.80.

Results

Overview of training load

The study collected TL values from 48 training sessions, with complete TL data available for 46 sessions. These sessions yielded 1196 sets of training load data across different rowers, including key metrics such as HR_mean, Bla, TRIMP, sRPE_TL, Total Stroke Count, and Speed.

Table 2 presents the mean values and standard deviations of all measured parameters during the testing period, primarily reflecting the training load data of rowers completing an 18 km ergometer test.

Table 2.

Mean and standard deviation (SD) of the training load metrics (n = 46 training observations).

	Parameter	M ± SD
Internal training load	HRmean (beats/min)	163.78 ± 13.34
	Bla (mmol/L)	4.89 ± 1.55
	TRIMP (AU)	156.72 ± 16.64
	sRPETL (AU)	514.99 ± 117.92
	sRPE (AU)	7.51 ± 1.73
External training load	Duration (min)	68.52 ± 1.74
	Stroke rate (spm)	20.16 ± 0.41
	Stroke count (strokes)	1381.45 ± 40.84
	ST500 (s)	109.21 ± 2.91
	Boat speed (m/s)	4.58 ± 0.12
	Percentage of speed (%)	84.13 ± 2.46

Abbreviations: Spm, strokes per minute; ST500, 500 m Split Time.

Relationship between external and internal loads in ergometer training

After adjusting for repeated measures using a linear mixed model and applying FDR correction for multiple comparisons, most ETL measures, except for DUR, ST500, and SR, showed significant correlations with sRPE_TL. Individualized metrics strengthened these associations. SC and S% exhibited moderate correlations with sRPE_TL (r = 0.41, 95% CI [0.33, 0.48], p < 0.001, FDR-adjusted p = 0.002 and r = 0.40, 95% CI [0.32, 0.47], p < 0.001, FDR-adjusted p = 0.002, respectively). Regarding TRIMP, all ETL indicators except SR and SC demonstrated significant correlations, with S% showing the strongest association (r = 0.39, 95% CI [0.31, 0.46], p < 0.001, FDR-adjusted p = 0.002) (Fig. 1).

Fig. 1.

Correlation matrix between session rating of perceived exertion (sRPE), sRPE-based external training load (sRPE_TL), training impulse (TRIMP), blood lactate concentration ([La]), and mean heart rate (HR_mean) with six performance/load parameters. The color bar represents the range of Pearson correlation coefficients. Abbreviations for parameters: DUR (duration); SR (stroke rate); SC (stroke count); ST500 (500 m split time); S% (percentage of speed). Statistical significance is denoted by asterisks: *p < 0.05, **p < 0.01.

Mixed-effects regression analysis for sRPE_TL

A linear mixed-effects model, with athlete as a random intercept, was fitted to predict sRPE_TL. The overall significance and variance explained by the model are summarized in Table 3. The model was statistically significant (F(3, 59) = 28.15, p < 0.001). The conditional R² (variance explained by both fixed and random effects) was 0.54, and the marginal R² (variance explained by fixed effects alone) was 0.50, indicating that the predictors (S%, SC, and TRIMP) collectively explained half of the variance in sRPE_TL after accounting for individual differences. The Durbin-Watson statistic of 1.72 indicated no autocorrelation. As shown in Table 3, the regression component accounted for the majority of the variance in the model, with a highly significant F-statistic confirming that the set of predictors reliably explains variations in sRPE_TL.

Table 3.

Analysis of Variance (ANOVA) for the linear mixed-effects model predicting sRPE_TL.

	Sum of squares	df	Mean square	F	p
Regression	504,512.71	3	168,170.90	28.15	< .001
Residual	357,578.50	59	6060.65
Total	862,091.21	62

The model includes athlete as a random intercept. Degrees of freedom were calculated using the Satterthwaite approximation.

The detailed parameter estimates for each predictor are presented in Table 4. As shown, all predictors were statistically significant (exact p values reported). Collinearity was not a concern (all VIFs < 2). The final regression equation was: Y = − 477.8 + 3.477 × S% + 17.262 × SC + 2.156 × TRIMP. The results in Table 4 indicate that all three training load metrics contribute uniquely to the prediction of perceived exertion. The positive coefficients suggest that increases in boat speed percentage, stroke count, and TRIMP are all associated with higher sRPE-based training loads, with boat speed percentage appearing to be the most influential factor per unit change.

Table 4.

Parameter estimates from the linear mixed-effects model predicting sRPE_TL..

Variable	Unstandard estimate			t	P	95% C I		VIF
	B	Standard error	df			Lower	Upper
Intercept (β0)	− 477.878	133.580	25	− 3.577	< .001	− 739.99	− 215.77
S%	3.477	.929	59	3.742	< .001	1.65	5.30	1.569
SC	17.262	7.708	59	2.239	.029	2.12	32.40	1.456
TRIMP	2.156	.749	59	2.876	.006	0.69	3.63	1.591

The model included athlete as a random intercept. Degrees of freedom (df) were calculated using the Satterthwaite approximation. CI = Confidence Interval.VIF = Variance Inflation Factor; S% = Speed; SC = Stroke Count; TRIMP = Training Impulse.

Discussion

This study aimed to investigate the workload and internal load of rowers during ergometer training. Additionally, it explored the potential relationship between external and internal load metrics. Furthermore, significant correlations were observed between most ETL measures (i.e., performance indicators) and internal load measures. Our findings indicate that ETL measures such as SC, Speed, and S% exhibit the strongest correlations with sRPE_TL and TRIMP. These results align with those reported by Dellavalle¹⁹, who observed associations between sRPE and peak watts resistance, training intensity rating, as well as blood lactate levels. Similarly, another study demonstrated correlations between average heart rate, maximum heart rate, and RPE²⁰.

Boat speed plays a critical role in rowing competitions and training²¹. Higher velocities demand greater force and power output per stroke²². Given that rowers develop distinct technical styles and achieve varying maximum speeds, S% serves as a more individualized metric. A rower’s physical conditioning, stroke rate, pull amplitude, and training status collectively determine the ratio between average and maximum speed. Furthermore, during ergometer training—unlike on-water sessions where boat speed fluctuates significantly—this parameter demonstrates better consistency with internal load than it does in the on-water environment. Consistent with prior research, increased running speed has been linked to elevated internal load²³, resulting in moderate-to-large correlation coefficients (r ≥ 0.43) between external and internal load metrics.

The increased propulsive demand elicits greater muscle recruitment and oxygen consumption²⁴, which in turn amplifies cardiovascular strain and perceptual effort. The consistency of these relationships under controlled power meter conditions reinforces the validity of ETL measures as surrogate markers of internal load. Unlike absolute velocity, S% effectively normalizes for inter-individual variations in technical proficiency and physiological limits. Notably, rowers maintaining near-maximal velocity demonstrate a disproportionate surge in internal load—a phenomenon driven by the quadratic increase in hydrodynamic resistance with velocity. As water resistance scales approximately with the square of velocity, higher handle speeds during the drive phase demand exponentially greater force production²⁵. This nonlinear relationship may underlie the superior predictive power of S% over absolute velocity in correlation analyses.

Other studies have identified stronger correlations between ETL and sRPE compared to TRIMP methodology, concluding that the former represents a more robust indicator²⁶. Our results confirm this phenomenon also occurs in rowing training. This may relate to work-to-rest ratio²⁷—higher ratios could yield diminished perceptual responses due to extended recovery periods. In contrast, continuous ergometer training without intervals leads to cumulative cardiopulmonary stress that becomes comprehensively reflected in sRPE. As perceived exertion integrates both psychological and physiological stressors, it encompasses more contributing factors than TRIMP, making sRPE_TL more closely associated with completed ETL. Therefore, although conventional practice prioritizes rowers’ heart rate²⁸ or blood samples²⁹ as fundamental monitoring metrics, coaches should incorporate subjective perceived exertion when designing and adjusting training loads.

Consistent with the recent findings by Murphy, et al.³⁰, our multiple stepwise regression analysis demonstrated that a combination of various ETL measures could explain more variance in sRPE_TL than any single load metric alone. Notably, S% exhibited the highest coefficient value, suggesting that the ratio between boat speed and individual 2 km time-trial speed serves as a particularly effective indicator for monitoring training load. This observation aligns with other studies emphasizing the importance of incorporating multiple ETL measures to better predict perceived exertion and internal training load than using a single metric^31–33.

The sRPE method has been established as a comprehensive metric for monitoring training load³⁴. Our findings reveal that common ETL measures can only explain a limited proportion of the variance in sRPE_TL, suggesting that a substantial portion remains unaccounted for. Such an approach enables more accurate and holistic training load monitoring, empowering practitioners to fine-tune training loads and facilitate active recovery, thereby optimizing training programs. Collectively, these measures can mitigate the risk of overtraining and help rowers achieve peak performance³⁵. However, it is important to note that excessive data collection may impose financial burdens and disrupt rowers’ training routines. Thus, judicious dimensionality reduction of metrics can alleviate practitioners’ cognitive load, improve rowers’ compliance with monitoring protocols, and—most critically—streamline decision-making in training load adjustments. Our results demonstrate that for rowing ergometer training monitoring, incorporating metrics such as perceived exertion, boat speed, and TRIMP can sufficiently capture rowers’ training load with reasonable accuracy.

Our hypothesis—that both TRIMP and sRPE_TL would correlate with most ETL measures, with the latter demonstrating stronger associations—was substantiated by the findings. When monitoring external load via the S% metric, greater consistency with internal load responses is achieved, enhancing monitoring efficacy. This suggests that such individualized metrics better reflect rowers’ perceptual responses to training stimuli than non-individualized external metrics (e.g., duration or stroke rate alone). By minimizing redundant data collection, these approaches reduce interference with rowers’ practice time.

The limited variance in sRPE_TL explained by conventional ETL measures indicates that rowers’ internal load is influenced by factors beyond these basic measures. This finding underscores the need to move beyond a reliance on isolated objective data (e.g., heart rate) and adopt a multidisciplinary framework for effective training monitoring. In practice, coaches and practitioners should leverage available resources to integrate data streams as comprehensively as possible. An integrated approach that combines internal load metrics (such as sRPE and TRIMP), ergometer-recorded ETL (particularly individualized metrics like S%), and rowing-specific skill assessments is likely to significantly enhance monitoring efficacy. Such integration enables more precise training load modulation and facilitates proactive recovery strategies. Collectively, these refinements may mitigate overtraining risks and support rowers in attaining peak performance.

This study has several limitations that should be acknowledged. First, the findings are based on a homogeneous sample of elite male rowers from a single team, which may limit their generalizability to other populations. Second, the controlled ergometer setting, while methodologically advantageous for standardizing measurements, does not fully capture the biomechanical complexity of on-water rowing. Third, the cross-sectional design limits insights into the longitudinal dynamics of training load across seasons.

Additionally, the use of Monday training sessions, which were intentionally selected to minimize carryover effects from prior training due to Sunday’s designated rest day, may introduce bias in representing overall weekly load distribution. While this approach enhances internal validity by controlling for residual fatigue, it may not fully reflect the cumulative load patterns typically observed mid-week. A principal limitation remains the focus on validating monitoring metrics without establishing a direct link to competitive performance outcomes. Finally, although key variables were controlled, unmeasured factors such as individual recovery capacity may have influenced the results. Future research should therefore validate these findings in more diverse populations, in on-water environments, and through longitudinal designs that directly correlate the proposed metrics (e.g., S%, SC, TRIMP) with enhancements in actual race performance, while also incorporating load data from multiple time points across the training week. Building on the practical utility of S%, SC, and TRIMP identified in this study, subsequent investigations should focus on developing integrated monitoring systems that seamlessly incorporate these key metrics. Future work could also establish sport-specific thresholds for these parameters to guide training periodization and help identify early signs of maladaptation in elite rowers.

Conclusions

The results demonstrate significant correlations between TRIMP and sRPE_TL with various ETL measures including SC, ST500, S, and S%, with the strongest association observed when ETL was expressed as S%. Notably, SC, speed, and TRIMP collectively accounted for approximately 50% of the variance in sRPE_TL, suggesting that conventional ETL measures have limited capacity to explain the full spectrum of internal load variation. Collectively, these findings indicate that S%, SC, and TRIMP represent the most appropriate training load parameters for coaches and sports scientists to monitor elite rowers’ training load. This study provides coaches with practical insights for implementing load monitoring in web-based rowing training programs, highlighting the utility of these specific metrics for training optimization.

Author contributions

D.X.conducted experimental testing, data processing, and analysis. Y.J.conceptualized and designed the study and drafted the manuscript. B.X. conceptualized the study, liaise with athletes and coaches and edited the final draft. All authors read and approved the final manuscript.

Funding

General Administration of Sport of China,(23QN015)

Data availability

The data that support the findings of this study are openly available upon request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.