Marathon Pace and Elevation Adjustments

Marathon pace prediction is dominated by Riegel's 1.06 power-law formula and Daniels' VDOT lookup tables. Both assume a flat course on a non-windy day at moderate temperatures. The first time a runner trains for Boston, NYC, or any course with meaningful elevation, the predictions become a rough first approximation that needs an explicit correction.

This article tests Riegel plus elevation correction against actual race-result data, walks through the Minetti metabolic-cost curve that anchors the correction, applies the model to Boston's Newton hills, and ends with a pacing protocol that holds for any hilly marathon.

The metabolic-cost-of-grade curve

Margaria 1965 made the first serious measurement of the metabolic cost of running on inclines.^[4] Minetti and colleagues 2002 produced the modern reference curve over a wider range of grades.^[5] The metabolic cost C of running at a given grade i (positive for uphill, negative for downhill) is well-fit by:

C(i) = 155.4×i^5 - 30.4×i^4 - 43.3×i^3 + 46.3×i^2 + 19.5×i + 3.6   J/kg/m

  At grade  0:  C =  3.6 J/kg/m  (level running)
  At grade  +5%: C =  4.6 J/kg/m  (+28%)
  At grade +10%: C =  6.4 J/kg/m  (+78%)
  At grade  -5%: C =  3.0 J/kg/m  (-17%)
  At grade -10%: C =  3.0 J/kg/m  (-17%)  ← the curve is asymmetric

Two important properties. First, uphill costs more than downhill recovers. A 5 percent uphill costs 28 percent more energy; a 5 percent downhill saves only 17 percent. Net-zero elevation courses are slower than flat courses. Second, downhill cost flattens at moderate negative grades. Beyond about -10 percent, eccentric loading and brake forces stop saving energy and start costing it.

Minetti 1994 explored extreme grades.^[7] At -20 percent, the metabolic cost of running starts to rise again because of the muscular effort required to control the descent.

The Strava grade-adjusted-pace heuristic

Strava operationalised the Minetti curve at scale. The platform's grade-adjusted pace (GAP) algorithm corrects observed pace for elevation grade using a smoothed version of the Minetti polynomial.^[6] Validated against millions of GPS-tracked runs, the practical heuristic emerges:

• Uphill: roughly +6 to +8 sec/km per 1 percent grade. At 2 percent grade for 1 km, expect 12 to 16 seconds slower than flat pace at the same effort.
• Downhill: roughly -3 to -4 sec/km per 1 percent grade. At -2 percent grade for 1 km, expect 6 to 8 seconds faster than flat pace at the same effort.
• Net elevation rule: approximately 3 sec/km per 10 m of net climb across an undulating course. A course with 200 m of net gain over 42 km adds roughly 60 sec/km × ... actually translates to a few seconds per kilometre slower than the same fitness on a flat course.

The heuristic holds best for grades between -8 percent and +8 percent, which covers the great majority of road marathon courses. Steep climbs (above 10 percent) and steep descents (below -10 percent) need explicit Minetti-curve calculation.

Pfitzinger's rule for marathon-specific elevation

Pfitzinger and Douglas 2019 give a practitioner rule that aggregates the Minetti curve over a marathon distance:^[2]

• Per metre of climbing: roughly 8 to 10 seconds slower marathon time at the same effort.
• Per metre of descending: roughly 3 to 4 seconds faster marathon time, capped on courses with steep descents because of muscular damage and brake forces.

For a marathon with 200 m of gross climbing and 200 m of gross descending (net zero, but undulating), the cost is approximately 200 × 8 - 200 × 3.5 = 900 seconds = 15 minutes slower than a perfectly flat course at the same fitness. For a marathon with 200 m of net gain (such as Big Sur), the cost is approximately 1,600 seconds = 27 minutes.

Validation: Riegel + elevation against Boston Marathon results

Boston is the canonical test case because its elevation profile is well-documented and its results are public. Course profile:

• Net elevation: -137 m (Boston is a net downhill course).
• Steep early downhill: first 6 km drops 90 m.
• Newton hills: miles 16 to 21 (km 26 to 34) include 88 m of climbing across four hills. Heartbreak Hill is the last and steepest.
• Final descent into Boston: -50 m over the last 8 km.

Naive elevation correction (8 sec/m up, 3 sec/m down):

Gross climbing: ~150 m (Newton hills + smaller climbs)
Gross descending: ~287 m (early descent + Newton recoveries + final descent)

Cost of climbs:    150 × 8 = 1,200 sec slower
Recovery of descents: 287 × 3.5 = 1,005 sec faster
Net:               -195 sec faster than flat (3:15 less)

Boston is a net-fast course IF the runner arrives at mile 17 fresh.

Diaz and colleagues 2019 analysed pacing across the world's fastest marathon courses and reported that elite runners pace Boston 1.5 to 2 percent slower than Berlin or Chicago at matched fitness, despite Boston's net-downhill profile.^[9] The discrepancy is explained by two factors:

• Newton-hills late-race penalty. Climbing 88 m at miles 16 to 21 occurs after 26 km of running; the muscular damage from the early downhill miles is substantial by the time the climbs hit, and the metabolic cost of the climbs is amplified.
• Quadriceps damage from early downhill. The first 6 km drops 90 m, which beats up the quadriceps eccentrically. By the late race, eccentric load tolerance is reduced and downhill running is slower than the flat-pace heuristic would predict.

The honest Boston correction: take the flat-marathon Riegel prediction, subtract 1 to 2 minutes for the net-downhill profile, then add 2 to 4 minutes for the Newton-hills late-race penalty and quadriceps damage. The net is a Boston time that is typically 1 to 2 minutes slower than the flat-marathon Riegel prediction at the same fitness.

Validation against undulating courses (NYC, San Francisco, Big Sur)

NYC has gross climbing of approximately 240 m across the five-borough course; the Verrazano Bridge climb at the start, the 59th Street Bridge at mile 16, and the Central Park rolling hills late account for most of it. The straight Pfitzinger rule predicts ~32 minutes slower than a flat course; the observed elite pace difference between NYC and Berlin is roughly 1 to 2 percent of finish time, or 2 to 5 minutes for a 2:10 men's elite or 2:30 women's elite. The model overstates NYC's slowdown for elites, who manage the climbs efficiently; for amateurs, the model is closer to accurate, with 5 to 8 minute slowdowns vs flat at the same fitness.

Big Sur (net gain ~225 m, gross climbing ~600 m due to extensive undulation) is the canonical hilly marathon. The Pfitzinger rule predicts ~50 minutes slower than a flat course; the observed amateur pace difference is closer to 25 to 45 minutes depending on training preparation. The model overstates Big Sur's slowdown for runners who specifically train hill repeats; for runners who arrive untrained for hills, the model is conservative.

Padulo and colleagues 2012 measured running kinematics at varying velocities and slopes and noted that trained hill runners adopt different stride patterns on inclines that reduce the metabolic cost relative to untrained runners.^[3] The Minetti curve is fit on a population mean; trained hill runners sit below the curve, untrained runners sit above.

The pacing protocol for a hilly marathon

1. Compute flat-marathon Riegel prediction from a recent half-marathon race, derated for volume if necessary.
2. Apply gross-climb correction: +8 sec per metre climbed, -3 sec per metre descended. Cap at -1 percent on net-downhill courses (the recovery flattens).
3. Apply late-race-hills penalty: +1 to 2 percent on the corrected time if hills sit after mile 16.
4. Pace by effort, not by clock, in hilly sections: hold heart rate or RPE steady on climbs even if pace drops 30 to 60 sec/km. Recover the time on the descents only if quad fatigue allows.
5. Train for the course: at least 6 to 8 hill-specific sessions in the 12-week build-up. The training reduces the effective slowdown by 30 to 50 percent.

Worked example: 3:30 marathoner targeting Boston

Inputs
  Recent half-marathon:    1:38 (volume-supported)
  Riegel flat marathon:    3:25 (1.06 exponent, mid-volume runner)

Boston correction
  Gross climbing:           150 m × 8 sec = 1,200 sec slower
  Gross descending:         287 m × 3 sec = 861 sec faster
                            (cap at -2.5% of finish time = -307 sec)

  Net elevation correction: -307 sec (favourable)
  Newton hills penalty:     +90 sec (1% of 3:25)
  Quad damage tax:          +90 sec (1% for early downhill)

Predicted Boston time:      3:25 - 5:07 + 1:30 + 1:30 = 3:22:53

Pacing strip
  Miles 1-6 (downhill start):  hold predicted average pace, RESIST the urge to push
  Miles 7-15 (rolling):         on predicted pace, RPE ~7
  Miles 16-21 (Newton hills):   hold RPE ~7, accept 20-30 sec/km slowdown
  Miles 22-26 (final descent):  recover pace IF quads allow, else hold

The pacing strip is conservative on the early descent because that's where most amateur runners destroy their Boston race. The early downhill feels easy and a 3:30 target lets the runner descend at 3:15-pace in the first 10 km; the cost shows up at mile 18 when the quads can no longer absorb the descents.

Cross-link tools

• Race Time Predictor for the flat-marathon Riegel baseline.
• Running Pace Calculator for kilometre/mile splits including elevation-adjusted target paces.

• Riegel and VDOT predict flat-marathon times; hilly courses need explicit elevation correction.
• The Minetti 2002 metabolic-cost curve quantifies the grade-by-grade cost of running. Strava's GAP heuristic operationalises it at scale.
• Practical rule: +8 sec/m climbed, -3 to -4 sec/m descended, capped on net-downhill courses.
• Boston runs 1 to 2 minutes slower than a flat-marathon Riegel prediction because the Newton hills hit at mile 17 and the early downhill damages quads.
• Big Sur and NYC predictions over-state the slowdown for trained hill runners and under-state it for runners who never see hills in training.
• Pace by effort on climbs; recover on descents only if quad fatigue allows. Train hill-specific sessions in the 12-week build to halve the effective slowdown.