Vertical Curve Levels: Two Worked Examples from the Field
Parabolic vertical curve computation step by step: grades from station and elevation data, equivalent radius, mid-ordinate, and the reduced level on the curve at any chainage.
A vertical curve is easier than a horizontal one. There is no bearing to carry, no coordinate geometry, and the curve is a parabola rather than an arc. What catches people is the bookkeeping: which tangent you measure the offset from, and whether you add or subtract it.
Both examples below are worked from a route-surveying field diary of October 2003 and have been recomputed from the given data. Chainages are in station notation, so 0+863 is 863 m along the line.
The relations you need
For a parabolic vertical curve of length Lvc between two grades G1 and G2, both as percentages and signed:
Grade G = (Elev₂ − Elev₁)/(Sta₂ − Sta₁) × 100
Equivalent R R = 100 · Lvc / (G1 − G2)
Mid-ordinate M.O. = (Lvc/2)² / 2R
Offset at x E = (x / (Lvc/2))² × M.O.
Curve level crest: tangent level − E
sag: tangent level + E
x is measured from the beginning of the curve, and the tangent level is the level on the straight grade produced through that same chainage. The sign convention is the only real trap: on a crest the curve sits below the tangent, on a sag it sits above.
The equivalent radius is a convenience rather than a true radius. A parabola has no constant radius, and R here is the radius of the circle that would produce the same mid-ordinate. It is useful because it lets you compare two curves at a glance and because sight-distance tables are often tabulated against it.
Example 1: a crest curve
Three points define a change of grade at the middle one. A 200 m parabolic curve is centred on it. Required: the grades, the equivalent radius, the mid-ordinate, and the reduced level on the curve at 0+850.
| Point | Station | Elevation (m) |
|---|---|---|
| PI-5 | 0+600 | 143.451 |
| PI-6 | 0+863 | 144.855 |
| PI-7 | 1+200 | 143.035 |
Grades.
G1 = (144.855 − 143.451)/(863 − 600) × 100 = +0.5338 %
G2 = (143.035 − 144.855)/(1200 − 863) × 100 = −0.5401 %
Rising then falling, so this is a crest and the offsets will be subtracted.
Radius and mid-ordinate.
R = 100 × 200 / (0.5338 − (−0.5401)) = 18 623.71 m
M.O. = 100² / (2 × 18 623.71) = 0.268 m
An equivalent radius of 18.6 km for a 200 m curve sounds absurd until you notice the grades are both about half a per cent. The grade change is only 1.07 per cent, so the curve is very nearly flat and the whole vertical correction across it is 268 mm.
Curve limits.
BC = 0+863 − 100 = 0+763
EC = 0+863 + 100 = 0+963
Level at 0+850. The tangent level comes from PI-5 along G1, a run of 250 m:
Tangent level = 143.451 + 0.5338% × 250 = 144.786 m
x from BC = 850 − 763 = 87 m
E = (87/100)² × 0.268 = 0.203 m
Curve level = 144.786 − 0.203 = 144.583 m
Every figure in this example reproduces from the given data.
Note that 0+850 is before the apex at 0+863, so you are still on the rising grade and the tangent level is still climbing. The 203 mm offset is what the curve has taken off it.
Example 2: a sharp grade break on a short curve
The second example is less comfortable, and more realistic. A 55 m vertical curve sits on a grade break from half a per cent down to over three per cent down.
| Point | Station | Elevation (m) |
|---|---|---|
| PVI-11 | 4+498 | 984.60 |
| PVI-12 | 4+738 | 983.41 |
| PVI-13 | 5+100 | 971.64 |
Grades.
G1 = (983.41 − 984.60)/(4738 − 4498) × 100 = −0.4958 %
G2 = (971.64 − 983.41)/(5100 − 4738) × 100 = −3.2514 %
Both negative, so the road falls throughout and then falls harder. This is still a crest in the geometric sense, because the grade becomes more negative, and the curve lies below the tangent. Calling it a crest when the whole thing is downhill feels wrong the first time, which is why it is better to think in terms of the sign of G1 − G2 than in terms of hills.
Radius and mid-ordinate.
R = 100 × 55 / (−0.4958 − (−3.2514)) = 1995.97 m
M.O. = 27.5² / (2 × 1995.97) = 0.189 m
A grade change of 2.76 per cent over 55 m gives an equivalent radius of 2 km, two orders of magnitude tighter than the first example. The mid-ordinate is similar only by coincidence: a much sharper break over a much shorter curve.
Curve limits and level at 4+725.
BC = 4+738 − 27.5 = 4+710.5
EC = 4+738 + 27.5 = 4+765.5
Tangent level = 984.60 − 0.4958% × 227 = 983.474 m
x from BC = 4725 − 4710.5 = 14.5 m
E = (14.5/27.5)² × 0.189 = 0.053 m
Curve level = 983.474 − 0.053 = 983.421 m
A small note on the last figure. The 2003 notebook recorded 983.422 m. The later compilation “corrected” it to 983.421 m. Recomputing at full precision gives 983.4218, which rounds to 983.422. The original hand calculation was right and the correction introduced the millimetre, not the other way round.
That is a trivial amount and changes nothing on site. It is worth mentioning only because it cuts against the assumption that the recomputed figure always wins. A hand calculation carried carefully can be better than a tidy-up done later, and the only way to know which is which is to redo it from the given data.
A check worth doing every time
Both examples end with a single reduced level, and a single number is hard to sanity-check. Two quick tests catch most errors:
Does the offset vanish at the curve ends and peak in the middle? At x = 0 the offset must be zero, and at x = Lvc/2 it must equal the mid-ordinate. In example 2, at x = 27.5 the offset is (27.5/27.5)² × 0.189 = 0.189, which is the mid-ordinate. Correct.
Is the curve level between the two tangents? On a crest the curve must sit below the back tangent produced and below the forward tangent produced back. If your answer is above either, you have added where you should have subtracted.
That second test takes ten seconds and catches the sign error, which is far and away the most common mistake on vertical curves and the one that puts a formation level out by twice the offset rather than by nothing.
Where the grades come from matters
In both examples the grades were derived from station and elevation pairs rather than given directly. That is normal when you are working from an alignment schedule or a levelling run, and it introduces a dependency worth being explicit about: the grade is only as good as the two elevations and the chainage difference behind it.
In example 1 the grades were computed over 263 m and 337 m. In example 2, over 240 m and 362 m. Long runs like these average out a single bad staff reading. If you derive a grade from two points 20 m apart, a 5 mm level error becomes a 0.025 per cent grade error, and the equivalent radius you compute from it can be out by a large fraction. Derive grades over the longest run the alignment gives you.
