Field-measured drag area is a key correlate of level cycling time trial performance

Introduction

A cyclist’s ability to produce and sustain mechanical power output is highly dependent upon physiological characteristics, particularly $\dot{V}{O}_2$ max, lactate threshold (LT), and economy (Coyle et al., 1988; Coyle, 1995). In the laboratory, where resistive forces are controlled or minimized, these physiological factors have been successfully used to predict simulated time trial performance (Coyle et al., 1988; Loftin & Warren, 1994; Coyle, 1995; Bishop et al., 2000; Bentley et al., 2001; Lamberts et al., 2012). However, in the field when cycling on level terrain at constant velocities (>40 km hr⁻¹), more than 90% of the total resistance (R_TOT) impeding the forward motion of a bicycle-rider system is determined by aerodynamic resistance (R_a) (Pugh, 1974; Di Prampero et al., 1979; Martin et al., 1998; Olds & Olive, 1999). Thus, measures of a cyclist’s ability to supply mechanical power do not always predict performance time in time trial racing (Hoogeveen & Schep, 1997; Balmer, Davison & Bird, 2000). For example, Balmer, Davison & Bird (2000) demonstrated that while peak mechanical power output assessed during a graded exercise stress test does correlate highly (r = 0.99, p < 0.001) with average mechanical power output during a 16.1 km field time trial, neither the laboratory peak mechanical power output nor the average mechanical power output during the time trial correlated well with performance time (r = 0.46, p > 0.05). One possible explanation for these results is that the resistance impeding the forward motion faced by competitive cyclists is variable enough that mechanical power output alone may not predict performance (Jeukendrup & Martin, 2001).

Measuring the aerodynamic resistance force (R_a) during cycling can be complex. Direct measures include wind-tunnel tests (Davies, 1980; Kyle, 1991; Martin et al., 1998), motorized towing (Di Prampero et al., 1979; Capelli et al., 1993), and coasting deceleration methods (De Groot, Sargeant & Geysel, 1995; Candau et al., 1999). Even though these direct measurements are accurate, they are impractical for most researchers and practitioners. For example, wind-tunnel testing facilities are inaccessible and costly. In addition, during wind tunnel testing, the bikes are not physically moving through the air and the pedaling motion introduces noise in the force measurement system during each pedal stroke. Motorized towing and coasting deceleration methods are also elaborate in their setup and execution meaning they are not an ideal substitute to a wind tunnel.

Because measuring the aerodynamic resistance force can be complex, it is sometimes assumed to be directly proportional to measures or estimates of the projected frontal area (A_P) of the bicycle and rider (Di Prampero et al., 1979; De Groot, Sargeant & Geysel, 1995; Olds & Olive, 1999; Heil, 2001; Anton et al., 2007). This assumes, though, that between individuals, aerodynamic resistance changes predictably with changes in A_P. However, evidence exists to the contrary. Previous research shows a lack of proportionality between an individual’s measured A_P and aerodynamic resistance (Kyle, 1991; De Groot, Sargeant & Geysel, 1995). This discrepancy must be due to variability in the coefficient of drag (C_d, see equation (2)), which is influenced by the shape of the bicycle and rider, varies greatly between individuals, and does not change proportionally with changes in projected frontal area (Debraux et al., 2011).

Recently, it has been demonstrated that measuring mechanical power output and speed in the field with cycle mounted power meters is a viable and accessible technique for determining an individual’s aerodynamic and rolling resistances (Martin et al., 2006; Debraux et al., 2011; Lim et al., 2011). This technique, in combination with standard physiological profiling, could improve our ability to predict field performance since it has been argued that physical factors resisting forward motion play a larger role in performance outcome than physiological variables (Jeukendrup & Martin, 2001). Anton et al. (2007) have previously shown that projected frontal area alone is correlated to performance time during a level time trial (r = − 0.73). However, their estimated frontal area did not improve the prediction of level time trial performance compared to just maximal mechanical power output measured in the lab. As described above, A_P is not directly proportional to aerodynamic resistance. Thus, the aerodynamic characteristic of a cyclist (drag area (A_d)), which includes both A_P and C_d, may provide a better correlation for time trial performance.

Accordingly, the purpose of this study was to quantify physiological determinants of endurance performance in conjunction with physical factors that contribute to resistance during cycling. Correlations between the various determinants/factors and level time trial performance were then compared. We hypothesized a cyclist’s physiological capacity or average power output would not predict level time trial performance time unless normalized to some representation of aerodynamic resistance. Additionally, we hypothesized that estimates of aerodynamic resistance, such as direct and indirect assessments of projected frontal area, would not correlate to performance as highly as the actual aerodynamic characteristic of the cyclist.

Methods

Results

The primary components of resistance during cycling are presented in Table 2. Because the A_P was calculated with the rider only while the A_d was calculated for the entire bicycle and rider system, it would be technically incorrect to calculate a C_d for the bicycle and rider system or rider alone. Still, dividing A_d by the A_P would give a estimated C_d of 1.03 ± 0.12. A correlation of 0.735 was found between A_P and A_d. No relationship, however, was found between A_P and the C_d (r = − .170, p = 0.486).

The results of the time trial are presented in Table 3. Of the physiological variables measured, $\dot{V}{O}_2$ peak (l min⁻¹) (r = 0.83, p < 0.001), power at $\dot{V}{O}_2$ peak (r = 0.67, p < 0.001), and power at LT (r = 0.69, p < 0.001) were significantly correlated to field-measured power output during the time trial. These physiological measures, however, were not strongly or significantly related to time trial performance time ($\dot{V}{O}_2$ peak r = − 0.42, p = 0.08; power at $\dot{V}{O}_2$ peak r = − 0.43, p = 0.07; power at LT r = − 0.45, p = 0.06). In addition, although significant, the field-measured power output during the time trial was correlated to performance time with an r-value of only −0.59. Thus, as hypothesized, non-normalized physiological measures were not strongly related to time trial performance time.

The correlation of physiological measures and performance time greatly increased when normalized to aerodynamics. When normalized to A_d, power at $\dot{V}{O}_2$ peak (watts m⁻²) (r = − 0.92, p < 0.001) and power at LT (r = − 0.85, p < 0.001) were better correlated to performance time (Fig. 1). Other normalized laboratory physiological performance measurements were either not significant or if significant were only moderately correlated. The correlation to performance time was best when mean field-measured power output was normalized to either k (r = − 0.92, p < 0.001) or A_d (r = − 0.92, p < 0.001). Figure 2 displays the relationship between performance time relative to the mean field-measured power output as well as mean field-measured power output normalized to body mass, A_P, and A_d. Rolling resistance alone or when used to normalize power or physiological measures was not related to performance time.

Both k (r = − 0.32, p = 0.18) and A_d (r = − 0.32, p = 0.17) were not correlated to field-measured power output during the time trial. However, both k (r = 0.85, p < 0.001) and A_d (r = 0.85, p < 0.001) alone were significantly related to performance time. Although significant (p = 0.04), measured A_P alone had a lower correlation (r = 0.47) and no methods for determining Est A_P correlated with performance time (r = 0.010 to 0.099). Furthermore, field-measured power output normalized to A_P was only modestly correlated to performance time (r = − 0.75).

Assuming that all of the resistance faced by the cyclists during the time trial was due to aerodynamic and rolling resistance, estimates of power for each component are given. Aerodynamic power was calculated from the k measured for each subject’s bicycle position while rolling resistance was assumed to be the remainder of the power. During the time trial, aerodynamic and rolling resistance accounted for 89.77% and 10.23% of the total power at 272 ± 23 and 31 ± 12 watts, respectively.

Except for the Olds et al. (1995) and Heil (2001) methods (P > 0.999 and 0.322, respectively), the mean value for A_P (0.338 ± 0.049 m²) was significantly over predicted by all other estimations of A_P as a percent of BSA (p < 0.001; Pugh (34), P = 0.011). The range for Est A_P as a percent of BSA was 0.306 ± 0.020 m² to 0.495 ± 0.032 m². For predictions of A_P based on anthropometric algorithms, except for the Heil (2002) method (P = 0.983), all other methods significantly (p < 0.001) over predicted our digitized A_P. These Est A_P ranged from 0.399 ± 0.025 m² to 0.550 ± 0.028 m². No correlation was found between total mass and A_P (r = 0.302, p = 0.208).

Stepwise regression was also performed to predict time trial performance time and power. In this multivariate analysis, the best correlates of level time were the mean field-measured power to A_d (r = − 0.92, p < 0.001, y = − 0.53 × time trial power to drag area + 2358.6). The correlation coefficient for this relationship was not significantly different from that for power to k (r = − 0.92, p < 0.001), which indicates that variations in ambient air density between the different time trial days was unlikely to have had much impact on performance time in our group of subjects. If power at $\dot{V}{O}_2$ peak and LT are normalized to parameters that affect resistance, the best correlate of performance time was power at $\dot{V}{O}_2$ peak normalized to A_d (r = − 0.92, p < 0.001, y = − 0.48 × power at $\dot{V}{O}_2$ peak to A_d + 2394.8). This correlation was not significantly different from that obtained using the mean field-measured power normalized to A_d or k. For power output during the time trial, the best correlate from all the variables measured in the laboratory is $\dot{V}{O}_2$ peak (l min⁻¹) (r = 0.83, p < 0.001, $y=54.67\times \dot{V}{O}_2$ peak + 48.8).

Discussion

As hypothesized, a cyclist’s mean field-measured power output normalized to our field-determined aerodynamic resistance (i.e., k or A_d) was the best correlate of level time trial performance time (r = − 0.92, p < 0.001). These results demonstrate and confirm that aerodynamic resistance is the primary resistance that is faced when cycling on level terrain. Accordingly, lab-measured physiological indices of performance (e.g., power at $\dot{V}{O}_2$ peak), though significantly correlated to field-measured power output (r = 0.64), were not significantly related to performance time (r = − 0.43), unless normalized to aerodynamic resistance (r = − 0.92). Of note, our field-determined drag area alone was the single best correlation with performance time on the level (r = 0.85, p < 0.001), confirming that aerodynamic resistance in our population played the most significant role in level time trial performance. Further, our field-determined drag area improved correlations to performance time better than either A_P or Est A_P. In addition, rolling resistance was not a distinguishing predictor of performance reflecting its smaller contribution to the resistance to forward motion during cycling.

In agreement with Balmer, Davison & Bird (2000), we found the majority of laboratory measures were not significantly correlated to level time trial performance time. In contrast to our findings though, a number of investigators have shown stronger correlations between laboratory measures of performance and level time trial performance in the field (Hawley & Noakes, 1992; Nichols, Phares & Buono, 1997; Smith, Dangelmaier & Hill, 1999; Anton et al., 2007). It is likely that these conflicting findings are due to differences in the subject populations recruited. For example, in the study by Anton et al. (2007), the subject population had a relatively homogenous frontal surface area (0.35 ± 0.02 m²) and a relatively heterogeneous maximal mechanical power output (490 ± 56 Watts) compared to the subjects from our study (0.34 ± 0.05 and 362 ± 30, respectively). This homogeneity in frontal surface area likely resulted in a greater correlation between laboratory measures and level time trial performance. Because level time trial performance is essentially a balance between resistive forces (primarily aerodynamic resistance) and propulsive forces (mechanical power output), homogeneity in one factor results in the other heterogeneous factor having a greater correlation. Furthermore, it is our belief that correlations in these previous studies would be stronger if the laboratory performance measures were normalized to the subject’s actual aerodynamic drag area. For instance, in our study, power at LT had a correlation of −0.45 with performance time but when this same laboratory measure was normalized to A_d, the correlation became −0.85.

Our finding that level time trial performance time is best predicted by normalizing field-measured power output to either k or A_d (r = − 0.92, p < 0.001), while perhaps self-evident, underscores the significant absence of actual aerodynamic measures in studies attempting to predict cycling time trial performance in the field. Martin et al. (2006) found measured aerodynamic drag can be used to predict sprinting speed in cyclists. Our study furthers these findings, and is the first to highlight the importance of measured aerodynamics during longer duration events at slower speeds compared to those seen during sprinting. In our study, the mean field-measured power was significantly correlated to level time (r = − 0.59, p < 0.01) but only explained 35% of the variability in level time. However, when field-measured power was normalized to aerodynamics, 85% of the variability was explained by highlighting that differences in aerodynamics are a critical performance determinant.

We found aerodynamic drag alone represented as either k or A_d, was significantly correlated to performance time (r = 0.85, p < 0.001), explaining 72% of the variability in performance time. The heterogeneous equipment and positions subjects used in the time trial may have aided this finding. Out of our nineteen subjects, nine used time trial bicycles, four used standard road bicycles with aerodynamic handlebars, and the remaining six used a standard road bicycle with no specific aerodynamic equipment. However, while each group had significantly different mean values for A_d, within the nine subjects using time trial bicycles, our results were similar with mean field-measured power normalized to aerodynamics the single best predictor of performance time (r = − 0.88, p < 0.001), aerodynamics the next best predictor (r = 0.72, p < 0.001), and field-measured power not predicting performance time (r = − 0.33, p > 0.05). The range in A_d between subjects riding a time trial bicycle (0.258 to 0.355 m²) may have also aided in this finding. This range in A_d though, also highlights the variability that exists between individuals with similar equipment and the importance of determining drag area to predict level time trial performance.

Because aerodynamic drag has been difficult to measure in the past, many studies have attempted to use estimates of the projected frontal area (Est A_P) or actual measures of projected frontal area (A_P) to represent aerodynamic drag (A_d) (Swain, 1994; Anton et al., 2007). Estimating A_d in this manner assumes a constant coefficient of drag (C_d), which along with A_P makes up A_d. However, in the present study, A_P only accounted for 54% of the variability in A_d. This discrepancy is largely due to individual differences in C_d. Assuming a constant C_d requires that riders must be riding in the same environmental conditions while using the same bike, geometry, clothing, and equipment. However, the range of C_d values between our subjects (0.823 to 1.262) highlights the individual variability in this measure. Furthermore, in agreement with previous work (Kyle, 1991; De Groot, Sargeant & Geysel, 1995; Debraux et al., 2011), we found no relationship between A_P and C_d (r = − 0.17, p = 0.49) demonstrating their independence and individual importance in determining A_d. Accordingly, when normalizing field measured power output during the time trial, we found that the correlation to performance time was significantly lower using A_P (r = − 0.75) or Est A_P (−0.71) compared to normalizing using our field-determined A_d (r = − 0.92). Thus, actual projected frontal area or an estimate of projected frontal area does not explain all of the variability in A_d and when used to normalize power is not necessarily better in the prediction of level time trial performance than A_d alone (r = − 0.85).

The results from the present study suggest that field-testing with a power meter may be a better alternative to traditional laboratory testing (i.e., $\dot{V}{O}_2$ peak, LT, economy) for athletes and coaches wanting to predict time trial performance. The field-measured drag area in combination with field-measured power output provided the best prediction of time trial performance compared to any combination of laboratory testing. Furthermore, we demonstrate that determining drag area does not need to be difficult. Our field-measured determination of drag area was the single best predictor of time trial performance and requires only a power meter, speedometer, and flat road making it far more economical than other tools (i.e., a wind tunnel). Thus, athletes can easily quantify and improve performance in events like level time trials by using our field-measured drag area technique to optimize the relationship between aerodynamic resistance and power output.

In conclusion, to predict cycling performance in the field, a cyclist’s ability to produce power must be considered relative to the forces that resist forward motion. Accordingly, level time trial performance is best predicted by normalizing the mean field-measured power output to aerodynamic resistance per velocity squared (k) or the drag area (A_d). Despite the fact that physiological variables are commonly thought of as the most important determinant of performance, we found that alone, measures of aerodynamic resistance were better related to level performance than $\dot{V}{O}_2$ peak, LT, or economy. Moreover, the use of projected frontal area (A_P) or estimates of projected frontal area (Est A_P) are not as accurate as directly measured A_d.

Supplemental Information