Eighteen competitive female swimmers participated in this study. Two groups with the same number of swimmers were formed according to their performance level (see main characteristics in Table 1). According to the classifications (Gonjo and Olstad, 2021), the 9 swimmers of the elite group had a performance level corresponding to a FINA point level >700 points. These swimmers were part of the French national swimming team and competed at the highest level (Beijing 2008, London 2012 Olympic Games and Roma 2009, Shanghai 2011 World Swimming Championships; whereas the 9 sub-elite female swimmers competed only in French National Championships (FINA point level < 650). Elite swimmers completed a larger weekly volume of swimming training (33,600 ± 8300 m·wk^-1 vs. 22, 300 ± 3400 m·wk^-1) and dry-land training volume (109 ± 118 min·wk^-1 vs. 54 ± 71 min·wk^-1) than sub-elite swimmers. All of the participants practiced the front crawl tumble turn on a regular basis and used it for competition races. Swimmers also used only the dolphin kicking technique for underwater propulsion. Adult participants and the parents of swimmers under 18 were informed in detail about the procedures and gave informed consent prior to the experimentation. The ethics committee of the University approved the investigation.

We reused an extended protocol proposed for elite male swimmers (Puel et al., 2012, 2022) to assess a larger set of performance variables. The participants were asked to perform several front crawl tumble turns as fast as possible until entire satisfaction of all the partners of the study (the swimmer herself, her coach and the researchers). This sequence was placed after the warm-up. A warm-up and a few practice turns were always done beforehand. Finally, never more than three turns were performed consecutively, without a long rest period, so that fatigue would not influence the results (Blanksby et al., 1995). Coaches timed each turn and researchers ensured that both the swimmer’s feet hit the force platform mounted on the turning wall. The Sportlab software comprising five separate software programs programmed in the LABVIEW 6.1 environment automated the detection of the head passing the virtual 3m line, permitting calculation of the 3mRTT parameter, and time related to the performance at the turn.

The derivation of the horizontal position of the head allowed for obtaining the horizontal velocities, and times for the 3 m before the turn wall and 3 m after. The calculated differences in seconds between the automatic and operator detections for the times of end of push-off, end of glide, start of contact, start of rotation and resumption of swimming were 0 sec., 0 sec., -0.6 msec. [-6 : 5, 95%CI], 1.5 msec. [-0.1 : 4, 95%CI], 0.25 msec. [-0.8 : 2, 95%CI] respectively. The intraclass correlation coefficient between the automatic and manual detections was 0.99. After 3 consecutive turns (with 3 min interval between each attempt), the swimmer had to rest to prevent fatigue (Blanksby et al., 1995). Except for a few cases, the maximum number of turns performed by each swimmer did not exceed 3. Only the best-time turn of each swimmer was analyzed.

The swimmers were analyzed when passing through a parallel piped calibrated space with mean dimensions of 4.11 x 1.11 x 1.89 m for the horizontal (main movement direction), vertical (pool depth) and lateral (lane width) directions, placed in contact with the turning wall and the water surface. The calibration structure was composed of 5 to 7 air-filled vertical PVC tubes, all 0.02 m in diameter, and of equal length. Each end was marked in contrasting color to be used as a calibration point. Each tube was attached to a thin non-elastic wire, itself attached to a weight on the pool floor. Four contrasting points marked on the force platform completed the structure. The structure was completed by four contrasting points marked on the corners of the vertical force platform placed just under the water surface and measured 0.6 m high and 0.4 m wide. Thus, the calibration structure included at least 14 and at most 18 calibration points. This type of configuration is in accordance with the methods of Challis and Kerwin (1992), who preferred use of control points distributed around the outside, rather than within the space to be calibrated.

The calibration system was left in place during the passage of swimmers, leaving more time for testing, which was limited by the autonomy of mini-DV cameras and batteries. This configuration also meets the recommendations of Chen et al. (1994), who stated that a homogeneous distribution of calibration points is required, and the recommendations of Wood and Marshall (1986), who stated that the sights be located as close as possible to the site of movement, and the volume should contain the movement. Last, a similar configuration was used by Elipot et al. (2009, 2010) to study the underwater phase following the swim start.

A set of 4 to 8 stationary mini-DV stationary mini-DV video cameras (Sony DCR-HC62E and DCR-HC96E, shutter speed: 1/120 s) were located underwater at different depths in waterproof cases (Sony SPK-HCD). The 2D image coordinates were transformed to 3D object-space coordinates using the direct linear transformation algorithm (DLT) (Abdel-Aziz and Karara, 1971; Elipot et al., 2009) in Sport Lab software. Initially intended to allow 3D reconstruction from two cameras, the software was modified to allow reconstruction from two up to eight cameras. This increase in the number of cameras that can be used simultaneously increases the number of points reconstructed. Multiplication of the shots maximizes the number of reconstructed points, and refines the reconstruction of these points. Thus, during each experiment, all available cameras were used. This arrangement also made it possible to ensure the reconstruction even when various complications arose such as poor sealing (presence of water or destruction of the camera), poor quality of a mini DV tape, loss of images during digitalization, defective camera battery (incomplete recording), poor framing (the camera is not in the right place). The digitized video files initially sampled at 25 frames/second were opened in Virtual Dub software. The bob double video filter was used to oversample the interlaced videos to double the frequency (i.e. 50 frames/second, 50 hertz).

The interlacing consisted of recreating 50 numbered frames/s, alternating even and odd frames, and respecting the chronology of the video (two hundredths of a second separated each consecutive image). For a configuration with 14 test patterns and five cameras (see figure above), the average 3D position deviation calculated was 12.8 mm with a maximum deviation of 28.7 mm. The distance between the real coordinates of the calibration points and the calculated coordinates along the three axes x, y and z were 3.7 mm, 10.7 mm and 3.5 mm. In comparison with the dimensions of the calibrated space, the errors were similar to those presented by Payton et al. (2002) and Coleman and Rankin (2005); i.e., respectively 0.09, 1.11, and 0.46% (7.1, 9.8 and 5.1 mm) of the horizontal, vertical, and lateral calibrated dimensions. The angle between each pair of consecutive cameras ranged from 30 to 55° (Figure 1) with synchronization obtained using an underwater strobe flash. The temporal precision was 0.02 s. A piezoelectric 3D force platform (Kistler 9253B12, 2000 Hz) was also mounted underwater on the turning wall. Kinetic data capture was carried out by an 8-channel charge amplifier and a DAQ system with BioWare software (Kistler 9865E1Y28 and 5691A1). Kinematic and kinetic data were synchronized at the end of the push-off.

One complete turn was analyzed for each swimmer by manual tracking of the center of the head visible during most of the movement (Costill et al., 1992). The center of the head was determined as the middle of the inter-ear segment (geometric center of the skull) (Winter, 2004). The approach began at the same time as the turn, i.e., when the swimmer's head passed the line located 3 m before the turn wall. Turning started when the swimmer's head began to submerge underwater (Blanksby et al., 1996). The software program was configured to detect this moment automatically, but manual observation remained the detection solution chosen by the majority. The end of the rollover corresponded to the beginning of the contac,t and was detected via the dynamometric signals. The approach time and the roll over time were calculated. Sixteen other anatomical points were also tracked (left and right hallux, ankles, knees, hips, shoulders, elbows, wrists, and fingertips).

Given the sagittal symmetry of the swimmer during the contact phase, a seven-segment model (feet, legs, thighs, trunk, head, arms, and forearms-hands; Winter, 2004) was used to assess the body extension indexes. For each of the segments, the distal and proximal points were respectively the tips of the hallux and the ankle; ankle and knee; knee and hip; hip and shoulder; the center of the head; elbow and shoulder; fingertips and elbow. The anatomical landmarks of the proximal points and the positions of the center of gravity of the segments relative to the proximal points are indicated in Winter (2004). The experimenters were trained in detection of the center of gravity of the segments from the videos of swimmers showing several points of view. Kinematic and kinetic data were smoothed by the Savitzky-Golay filtering method (degree of polynomials: r=2; sizes of the moving window: w_k=13, w_d=65; Savitzky and Golay, 1964), used to study underwater (Domenici et al., 2000) and human movements (Sibella et al., 2007). Speeds were computed by the same method (Staggs, 2005).

Kinematic (absolute and relative times, horizontal speeds and distances, depths, body extension indexes, Figure 2) and kinetic variables (forces, impulses, and decompositions of the push-off force vector at the horizontal force peak) were computed for each phase and the global turn performance. The time taken to swim from 3 m in to 3 m out the turning wall (3mRTT), and the approach duration (AT) were also determined. Speeds were the instantaneous horizontal speed of the swimmer’s head 3 m before the turning wall (VIn), the speed 1 m before the beginning of the rotation (V1mR), the maximal horizontal speed (Vmax), the speed at the end of the contact (when glide began) (VG), and the speed 3 m after the wall (VOut). Distances were horizontal distance between the swimmer’s head and the turning wall, the head-to-wall distance when swimmer’s speed reduced to 2.2 m/s (D22), when swimmer’s speed went down to 1.9 m/s (D19), or the head-to-wall distance at the end of the glide (when underwater propulsion began) (UD).

According to the methodology indicated in Puel et al., 2012, 2022, lower limb extension indexes were determined as the ratios between the hips-to-wall distance and the sum of foot, leg, and thigh lengths at chosen key times: the first contact, the end of placement (Lyttle et al., 1999; Prins and Patz, 2006), the force peak, and the end of push-off (respectively, CLLei (Contact Lower Limb extension index), PoLLei (Push-off Lower Limb extension index), PeLLei (Peak Lower Limb extension index), and GLLei (Glide Lower Limb extension index). Similarly, upper body extension indexes CUBei (Contact Upper Body extension index), PoUBei (Push-off Upper Body extension index), PeUBei (Peak Upper Body extension index), GUBei (Glide Upper Body extension index) were computed as the ratio between the fingers-to-hips distance and the sum of trunk, arm, forearm, and hand lengths respectively. The lower limb extension index was derived as the ratio of the horizontal distance from the hips to the wall to the sum of the segmental lengths of the feet, legs and thighs (Puel et al., 2022). The larger the ratio, the more extended the swimmers' lower limbs were. Conversely, the smaller the ratio, the more the feet, legs and thighs were flexed. The upper body extension index was computed as the ratio of the horizontal distance between the fingers and hips to the sum of the segmental lengths of the trunk, arms and forearms and hands (Puel et al., 2022). The greater the ratio, the more extended the swimmers' upper limbs were. Conversely, the smaller the ratio, the more the hands, forearms and arms were bent.

For kinetic variables the maximal value of the horizontal component of the force during the push-off (the force peak) (Pe) was determined and normalized by the swimmer's body weight (nPE). Five impulse variables were computed for the three components of force (lateral, vertical, and horizontal, orientations in accordance to the calibrated space dimensions) and during chosen phases (placement, from the push-off beginning to the force peak, and during the whole push-off). The lateral impulse during the placement (i.e. the integration of the lateral component of the force applied by each swimmer on the turning wall between the first contact and the beginning of the push-off sub-phase, LBI), the lateral impulse during the push-off (LPoL, N.s), the vertical impulse during the push-off (VPoL, N.s), the horizontal impulse during the push-off (HPoL, N.s) and the horizontal impulse to the force peak (HPeIP, N.s) were determined. Finally, the vertical and lateral angles of the force vector at the horizontal peak were also computed (VA°, LA°) (Figure 3).

All variables were compared between the two groups using a Student t test for independent samples. The normality of both samples was verified by the Shapiro-Wilk test. In the case of non-normal distributions, which occurred for only 2 of the 56 variables studied, the non-parametric Mann-Whitney test for independent samples was used. Effect sizes (d) were also computed to complement and classify the magnitude of the differences between samples (Cohen, 1988; Nakagawa and Cuthill, 2007) where≤ |d|≤ 0.5 was deemed small, 0.5 ≤ |d| ≤ 0.8, |moderate, and d| > 0.8 large. (Cohen, 1988). Differences in results were deemed statistically significant when p ≤ at 0.05. Finally, for the group considered as a whole, Pearson correlation coefficients were calculated between technical skills (composed of all kinematic and kinetic variables) and 3mRTT turning performance. The magnitude of the correlations was interpreted with reference to Taylor (1990) with correlations ≤ 0.35 considered low, between 0.36 and 0.67 modest, and between 0.68 and 1 strong.

Swimmers in the elite group swam faster than sub-elite swimmers 3 m before the wall and during the approach which is consistent with recent work by Marinho et al. (2020) who demonstrated in elite swimmers faster approach speeds (5m-in, the time between reaching the 45 m mark and contact with the wall) in 100 m races compared to 200 m races. In our study, the duration of this phase was likely related to a higher swimming speed (the average performance of the elite swimmers in the 200 m front crawl was 01:59.5 sec. versus 02:15.30 sec. for the sub-elites). These outcomes align with the research of Born et al. (2021) who observed for the 5 m-in a difference of 24 hundredths between the 25th and 90th percentiles (3.28 sec. for 01:54.95 sec. vs. 3.52 sec. for 02:03.04 sec.) on elite competitors in the European Short Course Championships. Faster approach times for elite versus sub-elite swimmers have also been identified for other strokes such as breaststroke (Sánchez et al., 2021), backstroke, and butterfly (Born et al., 2021), likely due to higher energy and technical qualities allowing for higher swim speed. Another explanation is better regulation and less decrease in velocity on approach to the turn wall (David et al., 2022; Seifert et al., 2018; Simbana et al., 2018) obtained in particular by taking visual information (Seifert et al., 2018) and adjusting the turning distance (i.e. tuck index) (David et al., 2022).

Elite swimmers moved faster than sub-elite swimmers at the end of the push-off and 3 m after the wall. This outcome is consistent with recent work reporting faster underwater speed (between the contact and head breaking through the water surface) and shorter 5-m out in the fastest races (Marinho et al., 2020) and among the best swimmers in the 200 freestyle (18 hundredths of a difference between the 25th and 90th percentiles, i.e. 2.09 sec. for 01:54.95 sec. vs. 2.27 sec. for 02:03.04 sec) (Born et al., 2021; Marinho et al., 2020).

Just before the end of the push-off, elite swimmers reached their maximal speed that was significantly greater than the sub-elite swimmers. As already observed in other research, these better push-off in elite swimmers is probably related to a higher horizontal push-off impulse (Jones et al., 2018; Pereira et al., 2006) and to lateralization of the push-off (Puel et al., 2012; 2022) which were two variables significantly correlated with fast turn times. More practice by elite swimmers probably helped them maintain a higher horizontal speed further (at the end of the push-off) than sub-elite swimmers. This level of performance was likely related to exerting a lateral impulse allowing a more rectilinear trajectory, and/or adopting a more streamlined position along the push-off. These results are complementary to a recent work (Puel et al., 2022) that emphasizes that longitudinal (twisting) rotations during placement, and the depth of the glide phase after push-off were higher in the female swimmers who performed the best turn times.

Another important difference between the two groups was the propulsion distance underwater. The elite swimmers started the propulsion underwater 0.33 m further from the wall than the sub-elites which did not seem to relate to the difference in (standing) height between the two groups which was only 0.09 m. At this time, the horizontal speed was nearly the same for both group (1.88 m/s for elites and 1.87 m/s for sub-elites) and very close to the lower speed recommended by Lyttle et al. (2000, i.e. 1.90 m/s). This difference in distance could indicate that sub-elite swimmers were not as efficient as elite swimmers on contact and glide phases. The gliding phase has been very well documented by Goya et al. (2003) for a cohort of 30 subjects, male and female swimmers, from trained to elites. The swimmer must be aligned during the push-off, a posture that was more accentuated in expert swimmers.

Other authors have also studied the gliding phase using computational fluid dynamics (Marinho et al., 2013). In this study, the drag coefficient showed the lowest value in the prone position, followed by the lateral position with 45° rotations (0.29%, 0.15%, 0.01% increase in drag for 1.5, 2.0 and 2. 5 m/s, respectively), the lateral position with 90° rotations (1.03%, 0.94%, 0.64% increase), and the supine position (2.21%, 1.42%, 0.96% increase in comparison with the prone position), in which the highest value of drag coefficient was observed. In light of these results, it would seem that an efficient turn strategy would be to adopt a slightly lateral position when placing the feet, to twist by pushing on the wall (Puel et al., 2022) to achieve the glide in prone position.

All contact-calculated body extension indices were higher in trend for elite swimmers than for sub-elite swimmers (Figure 3). For the lower limbs, anterior studies indicate that contacting the wall with the legs more extended (about 100-120°) could shorten the turn time (Takahashi et al., 1983; Blanksby et al., 1996; Cossor et al., 1999; Araujo et al., 2010). On the upper body, no previous studies have analyzed the profile of swimmers during the contact. Moreover, the correlations calculated on the whole group showed a greater extension of the legs at contact, and at the start of the push in the fastest swimmers. The values of the two variables Contact Lower Limb extension index and Contact Upper Body extension index could demonstrate superior ability for elites who were better able to place their arms during the rotation by directing them forward in the main direction of next movement. These results indicate that the expert swimmers exhibited a better profile than the non-expert swimmers throughout the contact. In contrast, the non-experts' hands, forearms and arms were more flexed compared to the experts. At push-off, elite swimmers were more streamlined than sub-elite swimmers, thus maximizing the speed production due to the extension of legs. As lateral and vertical components of the contact force presented very few differences between groups, the horizontal component seemed to differ slightly on some parameters linked to an earlier and higher force peak for elite female swimmers.

This larger horizontal impulse during push-off in elite swimmers is consistent with the results of Miyashita et al. (1992) who had noted increasing powers of legs extension between age group (13 years), college (19 years) and American national level swimmers (18 years). However, in the present study, the horizontal impulse during push-off although superior among expert swimmers was not correlated with 3mRTT for the entire group. This result implies that a strong impulse is not enough to achieve a fast tumble-turn if it is not combined with technical capabilities such as body extension to reduce resistance and push-off lateralization (produce a push-off with an angle that allows to transform a lateral torsional movement into a straight trajectory). The effects of the temporal position of the force peak on the speed curve were introduced by Klauck (2005). Using an underwater push-off simulation, the author emphasized the influence of water and its drag effect on the swimmer speed. Unlike an aerial push-off (e.g., countermovement jump), the temporal distribution of the force applied on the turning wall determines the end-of-contact speed. Klauck (2005) argued that better turns could be performed with a force peak occurring closer to the end of the contact that permits the highest possible end-of-contact speed. However, the elite female swimmers did not perform turns in this sequence (Figure 4) as their force seems to peak before the sub-elites’, and could thus contribute to the highest horizontal push-off impulse.

For the full group of swimmers, the signs of the correlation coefficients (direction of effect) for the variables of lateral impulse during placement, and lateralization, of the push-off indicate that the best turns were characterized by a higher lateral impulse during placement as well as a greater lateralization of the push-off. The high values of lateral impulse during placement could be the consequence of a fast turn and a high approach speed. The fastest swimmers also appeared to be better able to exert lateral force during the push-off, allowing them to move horizontally forward during the glide as proposed by Pereira et al. (2006).

Finally, one could argue that the difference in performance between the two groups was related to the age difference. However, all of the swimmers of both groups were female adults according to the criteria as described by Marshall and Tanner (1969). Further evidence of physical maturity in the youngest participants was that their growth (in stature) was limited over the two years following the experiment (data not shown) in agreement with the conclusions of Silva et al. (2007). In addition, it should be noted that approximately ten years after the data was collected, no swimmer from the sub-elite group reached the elite level. We interpret this outcome that the differences observed were most likely largely related to the level of expertise, and superior training in the elite group and not the age differences per se. Thus, the criteria that differentiate the turn and swimming performances of the two groups of swimmers were their expertise, body height and body mass. These differences corresponded to more years of practice and training for elite swimmers, indicated by higher force and muscle mass and better performance at the 200 m front crawl (L) and at the tumble turn (3m RTT). However, these differences -which should give proportionally greater attribution to factors related to the strength of the swimmers- were not revealed by our results, thus supporting the assertion that mastery of the technical features of the tumble turn could be essential to make the best turns (Puel et al., 2012, 2022). The age and the level of practice of the expert swimmers probably also explain the results obtained. Indeed, the additional years of practice, expert coaching, and in some circumstances, biomechanical support, likely all contributed to improved performance. Coaches and practitioners should structure drills to train swimmers to approach the wall at the highest possible speed, direct the push-off horizontally forward, and then streamline to create the highest speed.