Introduction

In high-performance sports, the analysis of athletes´ activity data has become crucial for gaining insights into the physical and physiological demands during competition. This information is essential for optimizing performance and improving overall athletic abilities [1]. Activity data quantifies the athletes´ work during competition and can be categorized into external and internal load data. While the external load describes the physical work of the athletes like distances covered or velocities reached, the internal load describes the individual physiological responses to the given external load such as heart rate increases or lactate accumulation [10]. Nowadays, wearable technologies such as global positioning system (GPS) sensors and accelerometers allow easy, automatic, and precise recording of external load data to describe athlete activity profiles [23]. Based on this objective quantification of the physical work of the athlete on the field, benchmarks can be developed to individualize the intensity and duration of exercise during exercise sessions [5]. These benchmarks allow for a tailored adaptation of training and aim to optimize the preparation of the athlete for the anticipated physical stress during competition [5].

In team sports such as field hockey (FH), the intermittent and unpredictable nature of sports challenges the elaboration of benchmarks, since external loads are highly variable and dynamic. Therefore, a FH player is not only required to sustain physical performance over time (4 × 15 min), but also to follow the unpredictable increments and reductions in match intensity through a FH match. The overall demands of a player can be quantified using the accumulation of recorded activity data. Throughout four quarters of 15 min each, FH players cover 6 to 8 km per match at an average speed of 7.5–9 km/h [21]. From this total distance, ~ 10% is covered at intense running speeds > 16 km/h, while > 50% of the distance is covered by jogging and walking, so there is high variability in the intensity of activity [15]. One limitation of accumulative analysis of activity data is that it barely reflects the intermittent and variable character of activity profiles in FH. To overcome this issue, more recent data analysis approaches investigated activity data within shorter time windows to identify the most intense periods (MIPs) within the intermittent activity profiles [3]. These MIPs are further stated as worst-case scenarios since they display the upper limit of physiological demands that an athlete can experience during a match [25]. The MIP approach is less sensitive to variations in competition intensities and avoids activity peaks are washed out by FH-specific phases of inactivity like penalties or time-outs [26]. Therefore, the MIP approach provides complementary information on the on-field demands of FH players by reflecting the potential extremes of on-field demands.

The first investigations in FH extracted MIPs from competitive FH matches and observed considerable differences between average match intensities (about 120 m/min) and peak match intensities based on the MIP approach (> 200 m/min) [6, 16]. Consequently, training intensities adapted according to accumulated values would be likely to underestimate the peaks of intensity during competitive matches [25]. In particular, a comparison of the training load during 5-min peak intervals in matches and 5 min of small-sided games (SSGs) during training indicated that the latter only partially reached the intensities of game demands [7]. While all studied SSG variations (2 vs. 2, 3 vs. 3, 4 vs. 4) demonstrated higher overall acceleration values compared to the competition, substantially lower values were observed for distance-based metrics (Duthie [7]). A retrospective analysis in soccer came to similar conclusions after demonstrating that training sessions failed to reproduce MIPs from competitions in various distance-based MIP metrics such as total distance covered or high-intensity running distance [17]. Aiming to prepare athletes for competitive demands, it seems likely that training sessions in team sports do not reach the peak intensities players experience during matches. Therefore, the extraction of MIPs from competitions and training can serve as valuable input to re-calibrate training practices and intensities.

Therefore, the present study aimed to compare the MIPs extracted from training and competitive sessions in a high-level male FH team. To identify the potential discrepancies between peak intensities during training and competition, an 8-week competitive in-season cycle was investigated retrospectively. Based on previous research [7, 17], it was hypothesized that MIPs from competition exceed those extracted from training sessions. In order to take the intermittent character of FH into account, this study further not only investigated distance-related metrics, but also explored acceleration-related outcomes as an important cornerstone of FH performance with regards to match-competition discrepancies [7]. Upon identification of discrepancies in MIPs between competition and training, this study may provide important implications for the design of training sessions according to actual competition-derived peak intensities.

Methods

Sample

In total, data sets from 20 elite FH players (21.2 ± 2.4 years) were collected throughout the 8-week in-season cycle. All players were part of a german field hockey team competing in the first german division. The players were amateurs, trained three times a week, and competed in one to two matches on the weekend. Further, all players performed individual strength and conditioning sessions besides their regular training and profession/ education. The players were informed about the analysis of the data collected throughout the training and competition since the data was embedded into the training monitoring process of the team. Through written informed consent, every player agreed to analyze his data for scientific purposes. Further, the ethics committee of the affiliated university approved the analysis of the training data.

Training Cycle

The 8-week period analyzed in the present study resembles the first competitive part of the season. During this period, the team trained three times a week (excluding individual strength and conditioning sessions) and played one to two competitive matches on weekends. In total, 30 sessions were recorded throughout the cycle (11 matches), resulting in a total number of n  = 372 individual raw data sets. The weekly training cycle was organized into an intense training session on Tuesday (T1), a technical training session on Wednesdays with a focus on team tactics and corner shooting (T2), and a game-orientated training session on Thursday (T3). On the weekends, competitive matches (C) took place. In three of the eight weeks, two competitive matcheshad to be played on one weekend.

Extraction of MIPs

Running and acceleration data was recorded using the Polar Team Pro sensor (Polar Electronics, Kempele, Finland). The sensor was connected to a chest belt at the beginning of each session and the recording session was started at the beginning of the warm-up period. The sensor records positional data at 10 Hz due to a GPS sensor and further records acceleration data at 200 Hz due to an inertial measurement unit (IMU). The system has been used previously to analyze activity data in outdoor team sports [19]. For the determination of peak match demands, raw data including sample points, accumulative distance (in m), acute speed (in km/h) and acceleration (in m/s2) was exported from the Polar Team Pro cloud at a sampling rate of 10 Hz. To extract the MIPs for each session throughout the training cycle, the raw data for each session was imported into MATLAB (R2020b, The MathWorks) and analyzed based on original scripts. The MIPs were calculated by applying the rolling window approach since this approach outperformed the fixed period method to identify peak intensities [3]. Following Delves et al. [6], we chose to roll window lengths of 1, 2, 3, 4 and 5 min to calculate MIPs. Since the raw data displayed the accumulated distance at 10 Hz, the algorithm first calculated sample-to-sample changes and then computed the maximum moving sum over the above-mentioned window lengths (1–5 min) as session MIP. For high-intensity running (HIR, > 16 km/h), a similar approach was chosen, but only sample-to-sample changes resulting from measured acute speeds higher than 16 km/h were considered. For sprinting (SPRINT; > 20 km/h), only sample-to-sample changes resulting from measured acute speeds > 20 km/h were considered. Then, sample-to-sample changes were computed before the computation of moving sums over the above-specified window lengths. For the acceleration metrics, both negative and positive acceleration values were turned into absolute values, changing negative into non-negative values, and moving averages over the above-specified window length were calculated over the whole time series [4].

For training sessions, the whole session was analyzed starting with the warm-up and ending after the cool-down activity. For matches, the recorded session was cut into two sections (Sect. 1: quarter 1 + quarter 2, Sect. 2: quarter 3 + quarter 4) and each section was analyzed as a single session for MIP detection. In the final step, the higher MIP of both sections was identified as the competition MIP. MIPs served as dependent variables and were calculated in m/min for total distance (TD), HIT distance (m/min > 16 km/h < 20 km/h), and SPRINT distance (m/min > 20 km/h). The MIPs for the acceleration load (ACCEL) were calculated in m/s2/min. The session type (C vs. T1 vs. T2 vs. T3]) served as the independent variable.

Statistics

All statistical analyses were performed using SPSS (Version 28, IBM, New York). First, general differences in MIPs between training and competition sessions were explored. Therefore, data for all different types of training sessions (T1, T2, and T3) was pooled and compared to the data obtained from C sessions. The data were checked for normal distribution by applying the Kolmogorov–Smirnov test and since the majority of outcomes under the independent variable “training” displayed a non-parametric distribution, Mann–Whitney-U-Tests were applied to analyze statistical differences in MIPs (P < 0.05). An unpaired statistical test was chosen since the group of players participating at the single sessions throughout the eight week cycle slightly changed from session-to-session. To interpret training-competition discrepancies between different MIP parameters and window lengths, effect sizes for Mann–Whitney-U-Tests were calculated via r = z/√n, where z was extracted as the Mann–Whitney-U z-value [13]. For interpretation, r-values < 0.03 were interpreted as small effects, values > 0.03 and < 0.05 as medium effects and values > 0.05 as large effects.

For a more particular insight into the training cycle, differences between different types of training sessions and matches were analysed through Kruskall-Wallis H analysis of variances (ANOVAs) with the factor TYPE (T1,T2,T3,C). Again, non-parametric statistical tests were chosen since normal-distribution was not accepted, particularly for parameters in the T2 group. According to the Kruskall Wallis H output, main-effects of TYPE on MIPs and post-hoc differences in MIPs between the different session types were extracted. To account for multiple comparisons, Bonferroni-correction was applied and the level of significance was set to P < 0.05.

Results

Descriptive Absolute Data

Across the 8-week in-season cycle, 30 sessions (11 matches) were recorded and resulted in 372 individual activity data recordings (C = 144, T1 = 56, T2 = 84, T3 = 88). An overview of average values from T1, T2, T3, and C sessions is displayed in Table 1.

Table 1 Overview of average and SD values of different external load metrics extracted from training and competitive sessions throughout an 8-week in-season period in high-level male FH players
Full size table

MIPs

To assess differences in MIPs between competitive matches and training sessions, Mann–Whitney-U-Tests were performed. The analysis revealed significant differences between MIPs extracted from competition and training sessions for the metrics TD, ACCEL, HIR, and SPRINT (P < 0.001). An overview of P-statistics and corresponding effect sizes r is presented in Table 1. Large effect sizes were observed for TD (r = 56–0.64), HIR (r = 0.60–0.62), and SPRINT (r = 0.52–0.63). For ACCEL, moderate sizes were observed (r = 0.38–0.47).

Outcome-dependent trends have been observed for the associations between effect size and MIP window length across parameters. While effect sizes for ACCEL increased with longer MIP window length, effect sizes tend to decrease with longer window length for SPRINT MIPs. For TD and HIR, no specific interactions between window length and effect size have been observed. Figures 1 and 2 further display histogram plots of the MIPs extracted from pooled training sessions and competition.

Fig. 1
figure 1

Raincloud plots of most-intense-periods (MIPs) extracted through moving sum calculations of activity data based on different window lengths (1 to 5 min). Clouds and raindrops display the histogram distribution of individual MIPs across the 8-week in-season cycle. The y-axis displays the calculated kernel density of individual values for the parameters total distance covered (TD, cyan = training, black = competition) in m/min and acceleration load accomplished (ACCEL, pink = training, black = competition) m/s2

Full size image
Fig. 2
figure 2

Raincloud plots of most-intense-periods (MIPs) extracted through moving sum calculations of activity data based on different window lengths (1 to 5 min). Clouds and raindrops display the histogram distribution of individual MIPs across the 8-week in-season cycle. The y-axis displays the calculated kernel density of individual values for the parameters high-intensity-running [HIR (> 16 km/h)] distance covered (yellow = training, black = competition) and sprinting [SPRINT (> 20 km/h)] distance covered (green = training, black = competition) in m/min

Full size image

Session-Specific Comparisons

Kruskall-Wallis H ANOVAs revealed main-effects for TYPE on TD- [H (3) = 137.54–198.56, P < 0.001], HIR- [H(3) = 167.21–183.29, P < 0.001], SPRINT- [H(3) = 106.08–156.27, P < 0.001], and ACCEL-related MIPs [H(3) = 75.42–96.68, P < 0.001]. Mean outcomes of session-specific MIPs are provided in Table 2.

Table 2 Overview of external load variables and most intense periods (MIPs) for different session types (T1 = Tuesday, T2 = Wednesday, T3 = Thursday, C = Matchday) across in-season cycle
Full size table

Post-hoc analysis revealed specific differences between the different session types. Comparing T1 and C sessions, C sessions demonstrated higher MIPs for TD, HIR, SPRINT and ACCE across all window lengths. For the comparison of T2 and C sessions, higher values for TD, HIR, SPRINT and ACCEL were observed for C across different window lengths either. Also for the comparison of T3 and C, MIPs from C exceeded those from T3 significantly Regardless of window length.

Among the different training session types, T1 and T3 exceeded T2 sessions for MIPs related to TD, HIR and ACCEL, but not for SPRINT. No differences in MIPs were observed comparing T1 and T3 sessions. A particular visualization of the post-hoc results is provided in Fig. 3. The Kruskall-Wallis H test-statistics can be found in Table 3.

Fig. 3
figure 3

Overview of session-specific (T1 = Tuesday, T2 = Wednesday, T3 = Thursday, C = Matchday) means and standard deviations of most-intense-periods (MIPs). MIPs were extracted through moving sum calculations of activity data based on different window lengths (1 to 5 min). Section a displays peak total distances covered (TD). Section b displays peak distances covered through high intensity running (HIR, > 16 km/h). Section c displays peak distances covered through sprinting (SPRINT, > 20 km/h). Section d displays peak acceleration loads (ACCEL). Statistically significant post-hoc differences (P < 0.05): # = signficantly lower than C, § = significantly lower than T1, ß = significantly lower than T3

Full size image
Table 3 Non-parametric Kruskall-Wallis H analysis of variances (ANOVAs) comparing the most intense periods (MIPs) throughout an 8-week in-season cycle across different session types (T1 = Tuesday, T2 = Wednesday, T3 = Thursday, C = Matchday)
Full size table

Discussion

The present study aimed to retrospectively compare MIPs extracted from training and competitive sessions throughout an 8-week in-season cycle in an elite male FH team. The analysis of more than 370 individual recordings revealed that peak intensities during training neither reached competition intensities for distance- nor acceleration-based metrics, independent of the session type. A comparison of effect sizes between the different inspected outcomes (TD, HIR, SPRINT, and ACCEL) indicated higher discrepancies between training and competition for distance-based metrics. No systematic differences were observed concerning the length of the window chosen for MIP extraction. The observed discrepancies between training and competition intensities provide important implications for the design of training sessions in FH and team sports.

As a key finding of this study, the analysis revealed that distance-based MIPs during training significantly failed to reach the magnitude of peak intensities observed during competitive matches. As shown in the histogram plots in Figs. 1 and 2, TD-, HIR- and SPRINT-based MIPs extracted form training sessions did not reach the intensities the players experienced during competitive matches on the weekend. In comparison to a previous study extracting MIPs from matches in high-level female FH players by McGuinness et al. [16] and Delves [6], the MIPs extracted in the present study were slightly lower, particularly for high-intensity MIPs (HIR and SPRINT). However, the MIPs experienced during the competition were still ~ 40%–60% higher than the peak intensities observed during the training sessions. Therefore, it might be stated that the intensity of training sessions was not high enough to prepare the players for competition intensities.

Discrepancies in intensities between training and competitive session intensity have been observed and reported in previous analyses of team sports. For instance, Oliva-Lozano et al. [17] analyzed activity data from professional soccer players over a consecutive in-season period and reported that match days induced a higher load on the players compared to training sessions. Similar observations have been made in professional handball players [8] and youth soccer players [22] through retrospective analyses. On the one hand, a tapering of training load between matches seems reasonable to reduce the metabolic and mechanical load and boost the performance capacities of a player [24]. However, exercise sessions that are not representative of competitive characteristics are likely to fail to induce the physiological overload needed for physiological adaptations relevant to performance [5]. The present study identified a discrepancy between training and matches also for the peak intensities experienced during exercise sessions. One issue of not preparing players for competitive MIPs during training is that the latter is associated with fatigue-initiated reductions in physical performance. Schimpchen et al. (2021) observed reduced high-intensity running activity for five consecutive minutes after individual MIPs. Therefore, MIPs seem to display severe physiological stressors athletes are not familiarized with during regular exercise sessions.

A possible reason for this discrepancy between peak intensities could be found in training routines in team sports, such as the focus on technical-tactical content during the season and the utilization of SSGs to simulate game situations [25]. For instance, Duthie et al. [7] already reported in a sample of FH players that typical training contents, e.g. SSGs, do not reach the distance-related intensity of competitive matches, but do so for acceleration-based metrics. Lacome et al. [12] also described this phenomenon concerning MIPs in soccer players comparing SSGs and competitive matches [12]. Among other reasons, the reduction of court dimensions and potential running distances was discussed as a possible reason for lower distance-based intensities. Findings from Martin-Garcia [14] on soccer players support the observation that particularly for accumulated high-speed activities, SSGs seem insufficient to stimulate competition-like intensities. Further, the ecological context of training drills seems to affect the peak intensities a player can experience. For instance, Sansone et al. (2023) reported that the focus on skill-related exercises exposes players to lower intensities than game-based exercises [20]. Therefore, the present study supports previous studies claiming insufficient intensities in (FH?) team sports training, with a specific focus on MIPs as short, but strenuous periods of activity. A possible solution could be increasing the court dimensions in the SSG to maintain the spatial characteristics of the team sports and avoid reductions in speed and running demands [12]. Taking into account the effects of the modification of SSG dimensions on heart rate data [11], it might also be beneficial to further consider internal load responses for the re-calibration of training intensity in monitoring routines.

In addition to distance-based metrics, acceleration-based MIPs revealed discrepancies between training and competition intensities.Based on effect size estimation and the observation of MIP distribution in Fig. 1, it can be stated that training-competition discrepancies for acceleration-based MIPs are less drastic than for distance-based metrics (TD, HIR, and SPRINT). Furthermore, the effect sizes indicate higher differences between competition and training for a larger MIP window length. Considering that several studies suggested that SSGs better match acceleration- than distance-based competition intensities and partly extend the latter [7, 12, 17], the reduction of court dimensions seems to be effective in preserving game-like acceleration demands. However, the significant differences in acceleration data observed in the present study are likely to result from the pooling of all recorded exercise sessions within the analyzed period of the season. Considering the variability of training intensities in terms of micro-cycle periodization, it seems legit that not all exercise sessions reach competition intensity [17, 18]. Furthermore, the duration of SSGs is frequently restricted to bouts of a few minutes with intermittent breaks [7], so an increased intensity over several minutes cannot be reached in such types of game-based exercise.

As a final observation of this study, it is worth noting that the discrepancy between training and competition was in parts moderated by the window length chosen for MIP extraction. Several studies indicated that the length of the analyzed window modulates the severity of differences between the average and peak intensity and highlighted the importance of the choice of window length for the description of peak intensities [6, 16]. For comparison of training and competition data, the length of the chosen window had a substantial impact on ACCEL-based MIPs (Table 1) and amplified the discrepancy between training and competition. Further, the different metabolic demands associated with high-intensity exercise such as SSGs of varying window length (e.g. 1 min vs. 5 min) should be considered when using objective MIP values as benchmarks for training design [9]. To prepare an athlete for the intermittent and variable demands of the competition, it could be important to implement SSG formats of varying durations for simulating worst-case scenarios experienced during competition across different time scales. As an alternative approach, variables which were observed to be underload during training sessions (e.g. HIR MIPs) could be targeted by specific exercise drills. For instance, repeated sprints and transition runs from HIR to SPRINT could induce the needed stimuli which optimize athletes´ preparation for competitive matches.

Limitations

Despite the important implications of this study for exercise planning, several potential limitations must be considered. First, only one team was analyzed, so team-specific routines may have contributed to the discrepancy between training and competition MIPs. Therefore, coaching philosophy could have played an important role in the specific composition of the training cycle and the desired intensities during the trainings sessions. Although such discrepancies have been reported previously [12], it is highly recommended to objectively compare MIPs from training and competition for each team specifically before re-calibrating training intensities. Further, the present approach is restricted to the physical demands of team sports. While training routines such as SSGs seem to not match the physical requirement of the competition yet, an intensification of non-physical aspects such as tactics and techniques compared to the match demands might be still given [2]. Therefore, coaches and scientists are requested to treat the observation of the present study as a piece of complementary information for the optimization of training.

Conclusions

The present study objectively identified discrepancies in peak intensities between training and competitive sessions in a male first-division field hockey team. The discrepancies were evident for both distance- as well as acceleration-based metrics and are likely to be caused by training routines related to SSGs. Therefore, coaches and scientists should consider an evaluation of actual match peak intensities as benchmarks for the adjustment of training routines. Next to peak intensities, aspects such as volume should be further considered to optimally prepare athletes for competing demands. Future studies should aim to replicate similar findings in other intermittent sports and further try to elaborate training routines that avoid an underestimation of competition demands in training sessions.