Applied Field Engineering

Grade is the trade's way of saying slope: rise over run, multiplied by a hundred so it reads as a percentage. The Romans understood the idea long before the notation existed — the aqueducts feeding the capital fell at grades near 0.02 %, and the Pont du Gard section drops about 2.5 cm per kilometre, a precision that still impresses. The trap is the denominator. Grade uses the horizontal run, not the sloped distance you would tape along the ground, and not the hypotenuse. On flat work the difference is invisible; on a 30 % haul road the sloped length is 4.4 % longer than the run, and quantities computed off the wrong one come back to bite at pay time.

A worked example: a parking lot drains 8 in over a 100 ft run. Convert first — 8 in is 0.667 ft — so G = 100 × 0.667 / 100 = 0.67 %, comfortably above the 0.5 % minimum most agencies allow for asphalt and below the 5 % that makes accessible routes non-compliant. Read backwards, the same formula sizes a ramp: to hold 5 % over a 3 ft rise you need 100 × 3 / 5 = 60 ft of run.

This is the single most-used line of arithmetic on a construction site: take a known elevation, walk a measured horizontal distance, and add the grade times that distance. Sign discipline is everything — enter a fall as a negative grade and the formula keeps track for you. It is the same relation that sets pipe inverts: an 8 in sanitary lateral run at −2.0 % from an invert of 253.40 m over 42 m arrives at 253.40 − 0.84 = 252.56 m, and if the downstream structure was built at 252.70 the pipe will not drain and someone is coming back with a saw.

Solved for G it becomes the as-built check — measure both ends, divide the difference by the run — and solved for L it answers the layout question, "how far out do I stake the daylight point?" A worked example: a 1.5 % crowned road starting at 431.20 ft crown elevation reaches 431.20 + 1.5 × 260/100 = 435.10 ft at station 2+60. The trap is using the taped slope distance for L instead of the horizontal projection; on flat grades the error hides, on steep ones it does not.

Percent grade and slope angle describe the same hill in two languages, and they are only interchangeable through the tangent — never by simple proportion. That is the single most common mistake on the topic: a 100 % grade is not vertical, it is 45°, because it rises one unit for every one it runs. Switzerland's Pilatus railway, the steepest cog line in the world, climbs at 48 %, which is a mere 25.6° — steep enough that the carriages are built as stepped terraces, but nowhere near the cliff that "48 percent" suggests to the untrained ear.

The two scales agree closely at small angles, which is why the confusion survives: at 5 % the angle is 2.86°, an error of under 0.1° if you naively read percent as degrees divided by nothing. By 20 % the grade angle is 11.31° and the shortcut is useless. A worked example: a haul road cut at 8 % sits at θ = arctan(0.08) = 4.57°; run it the other way and a 30° talus slope corresponds to G = 100 tan 30° = 57.7 %.

Earthwork drawings almost never label a cut face in percent. They call it 3:1 or 2:1, and the convention in North American civil practice is horizontal first — three metres out for every one up. Read it backwards and a 4:1 lawn slope becomes a 400 % cliff, which is how the argument usually starts on site. The conversion is a plain reciprocal: G = 100/n, so 2:1 is 50 %, 3:1 is 33.3 %, and 4:1 is 25 %. These are not arbitrary numbers — 3:1 is the flattest slope a ride-on mower handles safely, 2:1 is about the steepest that will hold topsoil and vegetation without armouring, and anything steeper generally wants riprap, geogrid or a wall.

A worked example: a detention pond is specified with 4:1 side slopes and the pond is 1.8 m deep, so each bank eats 4 × 1.8 = 7.2 m of horizontal room and the equivalent grade is 100/4 = 25 %. Note that geotechnical reports sometimes quote the same slope V:H instead — always check which number carries the "1".

The average end area method is the trapezoidal rule wearing a hard hat. Plot the cut or fill area of each cross section, assume the area varies linearly between them, and the solid between two stations is the mean of the end areas times the distance. It became standard because highway alignments were already being sectioned every 100 ft — one station — so the field data arrived in exactly the shape the method wants. The constant that trips people is 27: a cubic yard is 3 ft × 3 ft × 3 ft, so an answer of 15,000 ft³ is 555.6 yd³, and every North American earthwork bid is in cubic yards.

A worked example: station 12+00 sections at 120 ft² of cut, station 13+00 at 180 ft², one station apart. V = 100 × (120 + 180)/2 = 15,000 ft³ = 555.6 yd³ — about 28 truckloads at 20 yd³ each. The known bias: when the section shape changes a lot between stations the method overstates the volume, because a linear interpolation of area is a poor stand-in for the real prismoid. Agencies live with it on ordinary terrain and switch to the prismoidal formula where the money is large or the ground is a saddle.

The prismoidal formula is exact for any solid whose cross-sectional area varies as a quadratic along its length — which covers the wedges, prisms, pyramids and transition sections that make up almost all real earthwork. It is Simpson's rule with a surveyor's vocabulary, and the 1-4-1 weighting is what makes it exact where the trapezoidal average end area is merely close. Note the subtlety that catches everyone: A_m is the area of the section physically measured at the midpoint, not the average of A₁ and A₂. Substitute the average and the whole thing algebraically collapses back into the end area method.

A worked example: end areas of 120 ft² and 180 ft² one station apart, with the section actually taken at 0+50 measuring 140 ft². V = 100 × (120 + 4 × 140 + 180)/6 = 14,333 ft³ = 530.9 yd³, some 4.4 % less than the 555.6 yd³ the end area method returns. That difference — the prismoidal correction — is why large-quantity contracts and borrow pit pay quantities specify this method: on a million-yard job, four percent is the profit.

Soil has three volumes and they are never equal: bank (in place, undisturbed), loose (after the bucket has broken it up) and compacted (after the roller). Excavating breaks the interlock between particles and introduces voids, so loose volume always exceeds bank volume. Typical swells are 12 % for damp sand, 25 % for common earth, 40–45 % for clay and 55–65 % for blasted rock. The Panama Canal's Culebra Cut is the monument to this arithmetic: the engineers moved something like 76 million cubic metres of bank material, and every haul, spoil bank and rail schedule had to be sized in loose measure, not bank.

The trap is mixing measures inside one estimate. Excavation is paid by bank cubic yard, trucking is priced by loose cubic yard, and the two differ by a quarter or more. A worked example: a basement of 1,000 BCY in common earth with 25 % swell yields 1,000 × 1.25 = 1,250 LCY to haul; at 12 LCY per truck that is 105 loads, not 84. Running the formula backwards turns a measured truck count back into the bank quantity the owner actually pays for.

Shrinkage is swell's mirror image and it is the one that costs money. Material compacted into an engineered embankment at 95 % of standard Proctor density ends up denser than it was in the borrow pit, so a cubic yard of bank material fills less than a cubic yard of fill. Typical shrinkage runs 10–15 % for common earth and up to 25 % for organic or loosely bedded soils; rock is the exception, since blasted rock fills more than it occupied and is quoted as a negative shrinkage.

Every estimator's classic mistake is balancing cut against fill in bank measure and then discovering the job is short. A worked example: the plans call for 5,400 CCY of embankment in a soil with 10 % shrinkage. The borrow must supply 5,400 / 0.90 = 6,000 BCY — six hundred yards of imported material that a naive one-for-one balance would have missed. Chain the two factors together and you get the full life of a shovel of dirt: 1,000 BCY becomes 1,250 LCY on the trucks and 900 CCY in the finished embankment.

Ready-mix is sold by the cubic yard and delivered in trucks that hold eight to eleven of them, and the driver will not wait while you find a fourth of a yard. Hence the waste allowance: 5 % on a formed slab poured on polyethylene, up to 10 % on footings dug in soft ground where the trench over-excavates itself, and more on a rough-graded subgrade. The 27 in the conversion is the whole game — a cubic yard is 27 ft³ — so a 6 in slab is 0.5 ft thick and 54 ft² of it makes exactly one yard.

A worked example: a 40 ft × 30 ft driveway at 6 in thick is 40 × 30 × 0.5 = 600 ft³ = 22.2 yd³ neat; with 5 % waste, order 23.3 yd³, which in practice means calling for 23.5 yd³ since most suppliers sell in quarter-yard steps. The classic error is entering thickness in inches alongside feet — formula.expert converts for you, but on a napkin a 6 becomes a 12-fold overpour.

Hot-mix asphalt is bought by weight and placed by volume, and this formula is the bridge. Dense-graded mixes compact to roughly 145 lb/ft³ (about 2,320 kg/m³), which is where the paving contractor's rule of thumb comes from: one ton covers about 110 ft² at 1 in thick, or roughly 55 ft² at the more usual 2 in surface lift. Density varies with aggregate — a basalt mix runs heavier than a limestone one — so a mix design sheet beats a rule of thumb whenever the quantity is large.

The trap is compacted versus loose thickness. Mix laid 2.5 in behind the screed rolls down to about 2 in, and ordering against the loose depth buys 25 % too much material that nobody will take back. A worked example: 10,000 ft² of parking lot at a 3 in compacted lift, 145 lb/ft³. Volume = 10,000 × 0.25 = 2,500 ft³, mass = 362,500 lb = 181.25 tons — call it seven truckloads of 25 tons plus change, and add a few percent for the ramp-up and the joint at the end of the day.

A stockpile that has been driven over, or a borrow pit dug with a flat floor, is a frustum: a pyramid or cone with the tip cut off. Its volume needs the geometric mean term √(A₁A₂), and leaving it out — just averaging the two areas — always overstates the pile. This is one of the oldest results in mathematics: problem 14 of the Moscow Mathematical Papyrus, written around 1850 BCE, works the volume of a truncated square pyramid with a 4-cubit base, 2-cubit top and 6-cubit height and gets 56 cubic cubits, exactly what this formula gives. Whoever wrote it had no algebra and no proof we know of, and was still right.

A worked example: a sand pile 20 ft square at the base, 10 ft square on top, 9 ft high. A₁ = 400 ft², A₂ = 100 ft², √(A₁A₂) = 200 ft², so V = (9/3)(400 + 100 + 200) = 2,100 ft³ = 77.8 yd³. The naive average of the end areas would have said 2,250 ft³, a 7 % over-count and, at aggregate prices, a real invoice difference. Set A₂ to zero and the formula degenerates correctly to the full pyramid, V = A₁h/3.

Degree of curve is a nineteenth-century railroad invention that survives because it is easier to lay out in the field than a radius. Under the arc definition, D is the central angle that subtends exactly one 100 ft station measured along the arc, so R·D(in radians) = 100 ft and R = 100 × 180/π ÷ D = 5729.578/D feet. The crews building the transcontinental railroad carried printed tables of one-degree, two-degree and six-degree curves rather than radii, because a chainman could step off stations and turn deflection angles without ever knowing R. Railroads generally use the older chord definition, where the 100 ft is a chord rather than an arc; highways use the arc definition given here, and below about four degrees the two agree to within a foot.

The constant is unit-bound: 5729.578 is feet per degree, so the printed form works only in feet. This solver converts internally, so entering a radius in metres returns the same degree of curve. A worked example: a 4° curve has R = 5729.578/4 = 1,432.39 ft (436.6 m); a tight 12° urban curve is only 477 ft. Highway designers now specify R directly, but D still shows up on every railroad plan sheet and on older as-builts.

Two straight alignments meet at the point of intersection, and the curve that softens the corner has to start back down each tangent by the distance T. Because the two tangent lines are symmetric about the bisector of the deflection angle, T falls straight out of the right triangle formed by the centre, the PC and the PI: the half-angle Δ/2 sits at the centre, R is the adjacent side, and T is opposite. That symmetry is also the field check — the tangent distance is identical on both sides, so a crew that stakes PC and PT and finds unequal distances back to the PI has an error somewhere.

A worked example: a 500 ft radius curve turning through Δ = 60° needs T = 500 tan 30° = 288.68 ft, so the PC lies 288.68 ft before the PI and the PT the same distance beyond it. The trap is the half: Δ/2, not Δ. Using the full angle on that 60° curve returns 866 ft, three times too long, and the mistake announces itself only when the curve refuses to close. Note also that T grows without bound as Δ approaches 180° — a near-reversal of direction needs an enormous tangent run, which is why tight radii are used instead.

If one degree of central angle buys 100 ft of arc, then Δ degrees buys 100Δ/D feet — the neatest consequence of the arc definition and the reason highway stationing runs in 100 ft increments in the first place. Chainage was originally measured with Gunter's chain of 66 ft, but American railroad practice standardised on a 100 ft steel tape, and the whole vocabulary of "station 12+50" (1,250 ft from the origin) follows from it. The 100 here is feet, so the printed formula is unit-bound even though this solver converts whatever units you enter.

A worked example: a 4° curve deflecting 60° runs L = 100 × 60/4 = 1,500 ft of arc, or fifteen full stations. If the PC is at station 20+00, the PT lands at 35+00 — and note that the stationing runs along the curve, not along the tangents, which is why the PT station is not simply the PI station plus the tangent length. Chainage discrepancies at the end of a curve are almost always this mistake. Equivalently, L = RΔ with Δ in radians, which is what the brain evaluates internally.

The long chord is what a total station actually measures when it shoots from the PC to the PT: a straight line through the air, not the arc the pavement follows. It is always shorter than the curve length, and the gap grows quickly with deflection — for Δ = 60° the chord is 4.7 % shorter than the arc, for Δ = 120° it is 17 % shorter. Confusing the two is a classic quantity error, because pavement, guardrail and curb are all paid along the arc while the chord is what a tape across the opening gives.

A worked example: a 500 ft radius curve deflecting 60° has C = 2 × 500 × sin 30° = 500 ft exactly, since a 60° chord equals the radius — the same fact that makes a regular hexagon inscribe in a circle with side length R. The arc, by contrast, is RΔ = 500 × 1.0472 = 523.6 ft. Solving for Δ uses arcsine, which returns the minor arc; a curve that wraps more than halfway round its circle needs the supplement.

The external distance answers the right-of-way question: how far does the pavement stay from the corner the two tangents would have made? Measured from the PI along the bisector to the midpoint of the arc, E is the gap that has to clear whatever sits at the corner — a rock face, a building, a wetland boundary. Because the centre, the PI and the curve midpoint are collinear, R + E is the hypotenuse of the same half-angle triangle that produced the tangent length, giving the secant form.

A worked example: a 500 ft curve deflecting 60° has E = 500(sec 30° − 1) = 500(1.1547 − 1) = 77.35 ft. If the survey shows a building face 60 ft inside the PI on the bisector, that curve will not fit and either the radius has to shrink or the alignment has to shift. Watch the behaviour at large Δ: as the deflection approaches 180° the secant blows up and E runs away to infinity, which is the geometry's way of saying a hairpin needs a small radius, not a big one.

The middle ordinate is the sagitta — the "arrow" of the arc — and it is the quantity that decides whether a driver can see around the inside of a curve. Set M equal to the horizontal clearance from the centreline to a sight obstruction (a cut slope, a noise wall, a row of trees) and the same relation tells you the length of curve over which the sightline is blocked. Highway design manuals print it as a nomograph of R against M for exactly this purpose.

A worked example: a 500 ft curve deflecting 60° has M = 500(1 − cos 30°) = 66.99 ft — so the arc bulges 67 ft off its own long chord. Field crews use the same idea in reverse to measure a curve that was never staked: stretch a chord across a rail or a curb, measure the offset at midspan, and R follows from R ≈ C²/(8M), the small-angle version of this formula. That approximation, incidentally, is how track inspectors measure curvature with a string and a ruler.

K is the length of vertical curve required per one percent of grade change, and it is the single number that carries a design speed into a profile. AASHTO's Green Book tabulates it directly: a crest curve at 50 mph needs K = 84, at 70 mph K = 247, because sight distance grows roughly with the square of speed while curve length grows linearly with K. Sag curves get their own, smaller table driven by headlight throw and rider comfort rather than sight distance. Since A is in percent and L in length, K carries the awkward units of length per percent — feet per percent in the imperial tables, metres per percent in the metric ones — which is why the tables are never interchangeable.

A worked example: a road drops at −2 % and rises at +3 %, so A = 5 %. At a design speed calling for K = 50 ft/%, the crest needs L = 50 × 5 = 250 ft of vertical curve. Halve the grade change and you halve the curve. The trap is the sign: A is the algebraic difference g₂ − g₁, so a −2 % to +3 % sag is a 5 % change, not 1 %, and getting that wrong shortens the curve by a factor of five.

Vertical curves are parabolas, not circles, and for one good reason: a parabola changes grade at a constant rate, so a vehicle travelling at constant speed feels a constant vertical acceleration. The elevation at any point is the tangent grade line plus a correction term that grows with the square of the distance from the BVC. Everything is measured from the beginning of vertical curve, and the standard American profile uses equal tangents, so the BVC sits L/2 before the PVI and the EVC sits L/2 after it.

A worked example: a road entering at g₁ = −3 % meets a +2 % grade over a 400 ft sag, so A = +5 %, and the BVC elevation is 100.00 ft. At x = 100 ft past the BVC, E = 100.00 − 3.00 + 5 × 100²/(200 × 400) = 100.00 − 3.00 + 0.625 = 97.625 ft. The trap is the offset term's denominator: 200L, not 2L — the 100 that converts percent to a decimal hides inside it, and dropping it inflates the curve correction a hundredfold. Solving for x is a quadratic, since a crest crosses any given elevation on both sides of its high point.

On a parabola the grade changes linearly from g₁ at the BVC to g₂ at the EVC, so it passes through zero at the fraction g₁/A of the way along — hence x = −g₁L/A. That point is the crest of a hill or the bottom of a sag, and it is where the drainage engineer needs to be standing: sag low points are where water collects and inlets go, crest high points are where the sightline breaks. If the result comes out negative or larger than L, the profile has no turning point inside the curve at all, meaning both grades run the same way and the road simply steepens or flattens.

A worked example: a road climbing at g₁ = +3 % meets a −2 % descent over a 500 ft crest, so A = −5 %. The high point sits x = −3 × 500/(−5) = 300 ft past the BVC — not at midcurve, which is the intuition to unlearn. The turning point is at midcurve only when the two grades are equal and opposite. Combine this with the elevation formula and you get the crest elevation itself, which is what determines whether a stop sign will be visible over the hill.

Stopping sight distance is two problems glued together: the driver notices nothing for the first two and a half seconds, then brakes. The reaction term is linear in speed, the braking term quadratic, which is why doubling the speed roughly triples the stopping distance rather than doubling it. AASHTO settled on a 2.5 s perception-reaction time — generous compared with the 1.0 to 1.5 s of an alert test driver, deliberately so — and a deceleration of 3.4 m/s² (11.2 ft/s²), a rate that most drivers can achieve on wet pavement without losing steering control.

A worked example at the American design values: 60 mph is 88 ft/s, so the reaction distance is 88 × 2.5 = 220 ft and the braking distance is 88²/(2 × 11.2) = 345.7 ft, totalling 566 ft — which is exactly the figure the Green Book tabulates for 60 mph. The formula as written assumes level ground; on a grade, the effective deceleration becomes a ± g·G/100, so a 6 % downgrade stretches that 566 ft to well over 600 ft. Enter the reduced deceleration directly if you need the graded case.

Over a crest, the pavement itself blocks the view, and the geometry of a sightline grazing a parabola gives this result. The two heights are policy decisions, not measurements: AASHTO uses a driver eye height of 3.5 ft and an object height of 2.0 ft (1.08 m and 0.60 m in metric), and those numbers have drifted over the decades — eye height fell from 3.75 ft as cars got lower, and the object height was once 6 in, on the theory that a driver should be able to see a small obstacle, before it was raised to represent a vehicle's tail lights. Substitute the American values and 200(√3.5 + √2)² collapses to 2,158, the constant printed in every US design manual; the metric pair gives 658.

A worked example: a 4 % grade change with 600 ft of required sight distance needs L = 4 × 600²/2158 = 667 ft of crest curve. The formula assumes S is shorter than L; when the required sight distance overruns the curve a different expression applies, and design manuals give both. Always check the assumption after computing — if S came out longer than L, you used the wrong case.

A vehicle rounding a curve needs a sideways force, and it gets it from two places: the component of its own weight down a banked surface, and friction between tyre and road. The design equation simply says the two together must supply v²/(gR). Highway agencies cap e at 4 % where snow and ice are common (a stalled truck must not slide sideways off a banked curve) and up to 12 % on high-speed rural roads; the friction factor f is taken not from the tyre's real capability but from what a driver will tolerate before feeling uncomfortable, sliding from 0.17 at 30 mph down to about 0.08 at 80 mph.

The familiar shortcuts are this same equation with gravity folded into the units. With V in mph and R in feet, v²/(gR) becomes V²/(15R); with V in km/h and R in metres it becomes V²/(127R), since 3.6² × 9.807 = 127. A worked example: 60 mph on a 1,500 ft radius with f = 0.12 needs e = 100(3600/(15 × 1500) − 0.12) = 4.0 %. The trap is entering the speed in the wrong unit — the term is quadratic, so a 10 % speed error becomes a 21 % error in the required bank.

An azimuth is a direction measured clockwise from north, running 0° to 360°, and the back azimuth is the same line looked at from the other end. The rule is add 180° if the azimuth is under 180°, subtract 180° if it is over — which is one rule, not two, once you take the result modulo 360°. Surveyors need it constantly because a traverse leg measured from A to B has to be carried forward from B, and because the check on a closed traverse is that the back azimuth of the last leg reproduces the first bearing you started from.

Bearings are the older notation and behave differently: a bearing of N 30° E has a back bearing of S 30° W, the same numeric angle with both letters flipped. Azimuths avoid all that quadrant bookkeeping, which is why modern instruments and coordinate geometry work in them. A worked example: a property line runs at azimuth 45°, so its back azimuth is 225°; a line at azimuth 300° reverses to 120°. Note that a magnetic back azimuth taken in the field will not agree with a computed one to better than the local declination and the compass's own reversal error, which is why the check is done on the numbers, not the needle.

A traverse is a chain of measured lines, and before any of it can be plotted each line is resolved into a north-south component (the latitude) and an east-west one (the departure). With azimuths measured clockwise from north, the north component is L cos α, which puts the cosine on the northing — the opposite of the schoolroom habit of putting cosine on x. The sign takes care of itself: azimuths between 90° and 270° return a negative latitude, meaning the course runs south.

The Great Trigonometrical Survey of India, begun in 1802, is the monument to this arithmetic. Every one of the Great Arc's stations was reduced through latitudes and departures computed by hand, by human "computers" working in ledgers, with theodolites weighing half a tonne carried up purpose-built towers; the survey ran for most of a century and its accumulated closure over 2,400 km was measured in metres. A worked example: a 250 ft course at azimuth 60° has a latitude of 250 cos 60° = 125 ft north. The trap is feeding a bearing where an azimuth belongs — N 60° E and azimuth 60° agree, but S 60° E is azimuth 120°.

Departure is the east-west partner of latitude, and with azimuths measured from north it takes the sine. Together the pair converts a field book of distances and directions into coordinate differences that can be added up, adjusted and plotted. Positive departures run east, negative run west, and the sign falls out of the sine automatically for azimuths past 180°.

The reason both components matter is that either one alone is ambiguous. A course with a departure of 125 ft on a 250 ft leg could be heading at azimuth 30° or at 150° — north-east or south-east — and only the latitude settles it. That is precisely why the arcsine brain here is flagged: it returns the northern branch and leaves the choice to you. A worked example: a 250 ft course at azimuth 30° has a departure of 250 sin 30° = 125 ft east and a latitude of 250 cos 30° = 216.5 ft north, and the two squared and added return the 250 ft you started with, which is the first arithmetic check a party chief runs.

Walk a closed traverse and you finish where you started, so in a perfect world the latitudes sum to zero and so do the departures. They never do. Whatever is left over is the misclosure, and because the two residuals are perpendicular the total error is their hypotenuse. The direction of that closing line is informative too — a misclosure that lies consistently along one bearing usually means a systematic problem such as an uncalibrated tape or a mis-set instrument height, while a randomly oriented one is ordinary measurement noise.

A worked example: a five-sided boundary traverse closes with ΣLat = +0.24 ft and ΣDep = −0.32 ft. The linear misclosure is √(0.24² + 0.32²) = 0.40 ft — a tidy 3-4-5 triangle. Whether 0.40 ft is acceptable depends entirely on the perimeter, which is why the closure error is almost always quoted as a precision ratio rather than a raw length. Before adjusting anything by the compass rule or least squares, check the angular closure first: an angle blunder shows up as a misclosure that no distance adjustment can absorb.

An error of a third of a foot means nothing until you know how far you walked to accumulate it. The precision ratio fixes that by dividing the perimeter by the misclosure and quoting the result as 1:N — one part in six thousand, say. Typical standards run 1:5,000 for ordinary boundary work, 1:10,000 for urban surveys and control, and 1:100,000 or better for geodetic networks; rough topographic traverses may be accepted at 1:3,000. The convention long predates electronic distance measurement: with Gunter's chain, which came in 66 ft lengths of a hundred links, a good chainman was expected to hold about 1:5,000 over open ground.

A worked example: a traverse with a perimeter of 2,400 ft closes with a 0.40 ft error, so N = 2,400/0.40 = 6,000 and the survey is reported as 1:6,000 — acceptable for a rural boundary, marginal for a subdivision. Two traps: the ratio always reduces to a numerator of one, so 2:12,000 is not a thing, and a suspiciously good ratio is worth investigating rather than celebrating, since it often means an angle was computed rather than measured.

Differential levelling is the oldest reliable way to move an elevation across a site, and it is nothing but addition and subtraction done in the right order. Sight back to a rod held on the benchmark: that backsight, added to the known elevation, gives the height of instrument — the elevation of the telescope's line of sight, and the number a field book records in its own column. Sight forward to a rod on the new point: that foresight, subtracted from the HI, gives the new elevation. Backsights are plus sights, foresights are minus sights, and the whole discipline of the field book is keeping those two columns straight.

A worked example: benchmark at 100.00 m, backsight 1.52 m, so HI = 101.52 m; foresight to the turning point reads 2.31 m, so the new elevation is 99.21 m. The check is arithmetic and unforgiving: over a closed level loop, the sum of all backsights minus the sum of all foresights must equal the difference between the starting and ending elevations, and it must come back to zero on a loop that returns to the benchmark. Keeping backsight and foresight distances roughly equal at each setup is the other half of good practice — it cancels both collimation error and earth curvature.

Before electronic distance measurement, a level or transit measured distance optically: two extra cross hairs cut a known angle, and the length of rod they span grows in exact proportion to the distance. James Watt is generally credited with the idea around 1771, and instrument makers standardised the geometry so that the stadia interval factor K equals 100 — one foot of rod intercept means a hundred feet of distance, an arrangement so convenient it outlived the technology. The additive constant C accounted for the gap between the instrument's centre and the front of an external-focusing telescope, typically about 1 ft; every modern internal-focusing instrument has C = 0, which is why the constant has almost disappeared from the textbooks.

A worked example: the upper hair reads 6.42 ft and the lower reads 3.00 ft, so s = 3.42 ft and D = 100 × 3.42 + 0 = 342 ft. Stadia is good to roughly 1:300 to 1:500 — fine for topographic detail, useless for boundary work — and this form assumes a horizontal sight. On an inclined sight the horizontal distance picks up a cos²θ factor and the vertical difference a sin θ cos θ term, which is where the old tacheometric tables came in.

Once a traverse has been reduced to coordinates, its area comes for free from the shoelace formula — cross-multiply the corners in order, alternate the signs, halve the total. It is exact, it needs no angles or offsets, and it is what every coordinate geometry package runs under the hood. The sign carries information: list the corners counter-clockwise and the area is positive, list them clockwise and the same magnitude comes back negative. If your answer is negative, you went round the wrong way, and the magnitude is still correct.

The unit to watch is the acre, which is pure surveying archaeology: Gunter's chain of 1620 was 66 ft long precisely so that ten square chains make one acre (10 × 66² = 43,560 ft²), letting a chainman compute acreage without ever leaving the decimal system. A worked example: corners at (0, 0), (300, 0) and (0, 400) ft give A = ½[0(0 − 400) + 300(400 − 0) + 0(0 − 0)] = 60,000 ft², which is 60,000/43,560 = 1.377 acres. For parcels with more than three corners, split them into triangles and add — or run the same shoelace pattern with more terms.

A lapse rate is one subtraction and one division, and the only trick in it is the sign convention. The rate is written as the temperature BELOW minus the temperature ABOVE, so that ordinary air, which cools as you climb, comes out positive. Two thermometers reading 20 °C at the surface and 13.5 °C on a mast 1000 m higher give $\Gamma = (20 - 13.5)/1000 = 0.0065$ °C per metre, which the whole meteorological literature writes as 6.5 °C/km. That number is not an accident of the example. ICAO fixed 6.5 °C/km as the troposphere of the standard atmosphere in 1952, and every pressure altimeter in the world is calibrated against it. Note that only the DIFFERENCE of the two temperatures enters, so a Celsius interval and a kelvin interval are the same thing and the conversion cancels itself.

The rate becomes useful only when it is compared against the rate a parcel of air would cool at if it were lifted with no heat exchange at all. That one is not measured, it is derived. A rising parcel expands and does work against the surrounding pressure, and the energy comes out of its own heat content, so $\Gamma_d = g/c_p = 9.80665/1005 = 0.00976$ K/m, or 9.8 °C/km. Below that value the atmosphere is stable: a lifted parcel cools faster than its surroundings, becomes denser, and sinks back where it came from. Above it the parcel stays warmer than the air around it and keeps climbing, which is instability. Saturated air releases latent heat as it rises and so cools more slowly, roughly 5.4 °C/km in warm air and nearer 8 in cold air, which is why the band between 5.4 and 9.8 is called conditionally unstable.

Everything a plume does follows from that comparison. In unstable air the plume loops, dragged up and down by convective eddies, and a loop that touches down produces a brief ground-level concentration far above anything an hourly average would suggest. In neutral air it cones, spreading symmetrically, which is the case the Gaussian model describes best because it is the case the Gaussian model was fitted to. Under an inversion it fans, flattening into a thin ribbon that can travel many kilometres almost undiluted while the ground underneath sees essentially nothing. The dangerous part is what happens next. When morning sun heats the surface and erodes the inversion from below, the entire night's ribbon is mixed down at once. That is fumigation, and it produces the highest short-term ground-level concentrations most sources ever cause.

Two mistakes are worth naming. The first is measuring the gradient across too thin a layer near the ground, where on a sunny afternoon the bottom few metres can be superadiabatic by tens of degrees per kilometre while the air a hundred metres up is perfectly neutral. A lapse rate is only meaningful over the layer the plume actually occupies. The second is treating the lapse rate as the whole of stability. The Pasquill class that supplies the dispersion coefficients is set by wind speed and solar radiation as well, and a strong wind mixes mechanically no matter what the temperature profile says. Use the lapse rate to know which regime you are in, then take the class from the standard insolation and wind table rather than from the gradient alone.

A part per million by volume is a mole fraction, and a milligram per cubic metre is a mass per volume, so converting between them needs two pieces of information: the molar mass of the gas and the volume one mole occupies. The second comes from the ideal gas law, $V_m = RT/P$, which at 25 °C and 101.325 kPa gives 24.45 L/mol. From there the conversion is just bookkeeping: $C = ppm \times M/24.45$ with M in g/mol and C in mg/m³. One ppm of sulphur dioxide, molar mass 64.06, is $64.06/24.45 = 2.62$ mg/m³; one ppm of carbon monoxide is 1.15 mg/m³; one ppm of nitrogen dioxide is 1.88 mg/m³. Those three numbers appear on wall charts everywhere and this is where they come from.

The reference state is the thing to be careful about, because 24.45 is a choice rather than a constant. At 25 °C the molar volume is 24.45 L/mol, at 20 °C it is 24.06, and at 0 °C it is 22.414 — the number from first-year chemistry. US and industrial-hygiene practice standardises on 25 °C; European emission limits are stated at 0 °C, and the same physical gas therefore converts to a mg/m³ figure 9.1 % higher under the European convention than the American one. That is larger than most compliance margins. A mg/m³ value with no stated reference temperature is an incomplete number, and the correct response to receiving one is to ask rather than to assume.

The other trap is which molar mass to use, and nitrogen oxides are the standing example. Flue gas NOx is mostly nitric oxide, NO, with a molar mass of 30.01, but essentially every regulation requires it to be REPORTED as nitrogen dioxide, molar mass 46.01, on the reasoning that the NO oxidises to NO₂ in the atmosphere within hours. So a NOx concentration in ppm converts with 46.01 regardless of what the analyser is actually seeing, and using 30.01 under-states the mass by a third. Sulphur oxides carry a similar convention, reported as SO₂, and total hydrocarbons are usually reported as methane or as propane depending on the rule, which changes the molar mass by a factor of nearly three.

Two smaller points, both of which cause arguments. Parts per million by volume and parts per million by mass are different quantities for a gas and coincide only when the molar mass of the contaminant equals that of air, 28.96 g/mol; for a gas analyser, ppm always means by volume unless it explicitly says otherwise, while for a liquid or a solid ppm almost always means by mass. And this conversion assumes ideal-gas behaviour, which for trace contaminants in flue gas at atmospheric pressure is accurate to well under a percent — a rare case where the assumption everyone worries about is the one that does not matter.

Stack a column of air on itself and each layer has to carry the weight of everything above it. Write that as a hydrostatic balance, substitute the ideal gas law for the density, hold the temperature constant, and the integration gives an exponential: $P = P_0 e^{-Mgz/RT}$. The group $RT/Mg$ has units of length and is called the scale height, the climb over which pressure falls by a factor of e. At 15 °C it works out to $8.31446 \times 288.15 / (0.0289644 \times 9.80665) = 2395.8/0.28404 = 8435$ m. So a climb of 1000 m from sea level is 0.1186 scale heights, the exponential factor is $e^{-0.1186} = 0.8882$, and the pressure falls from 101.325 kPa to 89.997 kPa.

The molar mass in that expression, 0.0289644 kg/mol, is the value the US Standard Atmosphere assigns to dry air, and it is a weighted average over nitrogen, oxygen, argon and carbon dioxide rather than a property of any single substance. Humid air is lighter than dry air, because a water molecule at 18 g/mol displaces a nitrogen molecule at 28, so a saturated tropical atmosphere has a slightly larger scale height than this constant admits. The effect is under one percent and is usually swamped by the temperature assumption, which is the real weakness here.

That assumption is worth being blunt about. The atmosphere is not isothermal, and this equation knows nothing about the lapse rate. Compare it against the standard atmosphere, which integrates a 6.5 °C/km gradient properly and gives 89.875 kPa at 1000 m: the isothermal answer of 89.997 kPa is high by 0.12 kPa, or about a tenth of a percent, which nobody cares about. Go to 5000 m and the isothermal form using the sea-level temperature returns 56.0 kPa against the standard atmosphere's 54.05 kPa, an error near four percent. The fix is free and is the whole reason the variable here is called the LAYER temperature: feed it the mean temperature of the layer rather than the temperature at the bottom. Using the average of 15 °C and the 17.5 °C below zero found at 5000 m gives 54.06 kPa, which is right to a few hundredths of a percent.

In dispersion work this equation earns its place three ways. It supplies the ambient pressure that Holland's plume-rise equation needs. It sets the air density that converts a stack's volumetric flow to a mass flow, and a plant at 1500 m elevation is moving air about fifteen percent less dense than a plant at sea level, which changes both the exit velocity and the buoyancy flux. And it is the correction that puts a measured emission rate onto the standard conditions a permit is written against. The recurring errors are the obvious two: temperatures in Celsius rather than kelvin, which makes the exponent nonsense, and using the equation above the tropopause, where the real atmosphere switches to an isothermal and then a warming regime and a single-layer model has nothing left to say.

Anemometers sit at 10 m because that is where the World Meteorological Organization put them, and stacks are anywhere but 10 m, so almost every dispersion calculation starts with an extrapolation. The power law is the cheapest one available: $u_2 = u_1 (z_2/z_1)^p$. A 5 m/s reading at 10 m, taken up to a stack top at 80 m over open country with $p = 0.15$, gives $5 \times 8^{0.15} = 5 \times 1.366 = 6.83$ m/s. Turn it around and the same equation fits an exponent from two measured levels: 4 m/s at 10 m and 8 m/s at 100 m give $p = \ln 2/\ln 10 = 0.301$, which is a stable night over rough ground.

The exponent is not a property of the terrain alone, it is terrain and stability together, and the values EPA publishes are worth carrying. Rural: 0.07 for classes A and B, 0.10 for C, 0.15 for D, 0.35 for E and 0.55 for F. Urban: 0.15 for A and B, 0.20 for C, 0.25 for D, 0.30 for E and F. Notice how far the stable rural exponent sits from the rest. On a clear calm night over farmland the wind at 100 m can be more than twice the wind at 10 m, and a calculation that assumes a neutral 0.15 will under-predict the stack-height wind by nearly half. Notice too that the urban exponents are flatter at the stable end, because a city keeps mixing mechanically and thermally long after the countryside has gone still.

The power law is a convenience fit to something better founded. Surface-layer theory gives the logarithmic profile $u(z) = (u_*/k)\ln((z-d)/z_0)$, where $z_0$ is a roughness length running from a millimetre over water to a metre or more over forest, d is the zero-plane displacement, and k is von Karman's constant near 0.40. The log law is what the physics supports, with a Monin-Obukhov correction for stability. The power law survives because it needs one number instead of three and because it is what the regulatory guidance is written in. Both are surface-layer relations, which means both stop being trustworthy above roughly a tenth of the boundary-layer depth: 100 m or so on a stable night, perhaps 150 m in neutral conditions. Extrapolating a 10 m anemometer to a 200 m stack is an act of faith, and on a stable night the nocturnal low-level jet can put a maximum in the profile that no monotonic formula can reproduce.

The consequence for dispersion is that a wind error is charged twice. Plume rise goes as $1/u$, so a wind that is too high shortens the plume, and concentration goes as $1/u$ again in the Gaussian equation, so the same error moves the answer roughly as $u^2$ once the height term is squared. Get the extrapolation wrong by 30 percent and the ground-level concentration can be wrong by a factor of two, which is the whole error budget of the model spent on one input. Two habits keep this honest: record the anemometer height alongside every wind speed, and respect the regulatory floor of about 1 m/s, below which the power law and the Gaussian model both stop meaning anything.

This is nothing more than continuity, $Q = vA$, with the area of a circle written out, and it earns its own page because everything downstream of a stack depends on the answer. Twelve cubic metres a second through a 1.2 m bore gives $4 \times 12/(\pi \times 1.44) = 10.61$ m/s. The one thing to be careful of is which twelve. Fan curves, permits and CEM reports quote flow on at least three different bases: actual cubic metres per second at stack conditions, standard cubic metres at 20 or 25 °C, and dry standard cubic metres with the moisture stripped out. Exit velocity is a physical speed and takes the ACTUAL flow. A boiler stack at 200 °C moves about 1.6 times the volume its standard flow suggests, so using the standard number here under-states the velocity by that same factor and quietly under-states plume rise along with it.

The number to compare the answer against is the wind, not a table. Stack-tip downwash occurs when the exit velocity is less than about 1.5 times the wind speed at the stack top: the plume fails to escape the low-pressure wake immediately behind the stack, gets pulled down into it, and effectively loses part of the height that was paid for. Briggs handled this by subtracting a downwash allowance from the physical stack height, and the practical consequence is that a tall stack with a lazy exit can perform worse than a shorter one with a brisk exit. Industrial designs generally land between 15 and 20 m/s at full load, which clears the wind on all but the worst days.

Part-load is where this bites in the field. A stack sized for 18 m/s at full fire runs at 7 m/s when the boiler modulates down to 40 %, and the downwash condition it comfortably passed on the design sheet fails on an ordinary Tuesday in April. The cheap remedy is a tapered exit cone, which raises the velocity at the tip without touching the fan or the ductwork; it costs a little static pressure and it is very much cheaper than adding steel. The expensive remedy is more height. Doing nothing is common and is why so many buildings have a fume complaint that only appears at part load.

Two more habits worth keeping. Velocity pressure rises with the square of this number and stack noise with roughly its sixth power, so the temptation to solve every dispersion problem by cranking up exit velocity is paid for continuously in fan power and sometimes in a noise complaint. And when the number is going into a Method 5 sampling plan rather than a dispersion model, remember this equation gives the AVERAGE velocity: the local velocity at a traverse point near the wall can be 20 % below it, which is exactly why the sampling rate has to be reset at every point rather than computed once from the mean.

A chimney works because two columns of fluid of different densities sit side by side. Outside there is a column of cold, dense air of height $h$; inside there is a column of hot, light flue gas of the same height. The hydrostatic pressure at the base of each is $\rho g h$, and the difference between them is the draft: $\Delta p = h g (\rho_a - \rho_g)$. A 30 m stack with ambient air at 1.2 kg/m³ and flue gas at 0.7 kg/m³ develops $30 \times 9.80665 \times 0.5 = 147$ Pa, which is 0.59 inches of water column on a manometer. There is no fan in that sentence anywhere — the pressure is manufactured by gravity acting on a density difference, which is why a natural-draft appliance works during a power failure.

The densities are usually not measured but computed, and the ideal gas law is all you need: $\rho = PM/RT$, which for air at ordinary pressure collapses to the useful shortcut $\rho \approx 353/T$ with T in kelvin. That gives 1.20 kg/m³ at 20 °C and 0.70 kg/m³ at 232 °C, which is where the example numbers came from. Flue gas is not air, but its molar mass sits within a couple of percent of air's for ordinary combustion products, so the shortcut is good enough for draft work. The temperature to use is the MEAN gas temperature over the height of the stack, not the temperature at the breeching, and on an uninsulated exterior chimney those two can differ by a hundred degrees. Using the hot inlet temperature over-states the draft, which is the classic way a chimney gets sized too small.

What the equation gives is THEORETICAL draft, and a real chimney never delivers it. Friction along the flue, the entry loss at the breeching, the exit loss at the tip and every elbow in between all consume part of it, and the gas has to be accelerated to exit velocity out of what is left. A common working figure is that available draft is 70 to 80 percent of theoretical, and a tall narrow flue can be worse. That is also why lengthening a chimney has diminishing returns: the driving pressure grows with the first power of height while the friction loss grows with it too, and beyond a point the extra height buys almost nothing.

The same equation, with no changes at all, explains the stack effect in a building. Warm indoor air is the light column, winter outdoor air is the heavy one, and a 60 m tower on a −20 °C day develops something like 200 Pa across its full height. That is what makes ground-floor lobby doors hard to open, drives infiltration in at the bottom and out at the top, and puts a neutral pressure plane somewhere around mid-height where the difference reverses. It is also the mechanism behind cold-flue backdraft: at start-up the gas inside is the same density as the air outside, the draft is zero, and the chimney will not pull until it has warmed through. Any appliance that has to light against a cold exterior masonry chimney is fighting exactly this, and it is why induced-draft fans exist.

Air flowing over a building does not close neatly behind it. It separates at the upwind edge and leaves a cavity of recirculating, highly turbulent air downwind, and anything released into that cavity is brought to the ground almost immediately rather than dispersing. Wind-tunnel work through the 1970s put the vertical extent of the disturbed zone at roughly 1.5 times the smaller of the building's height and its projected width, which is where the formula comes from: $H_{GEP} = h_b + 1.5L$. A 30 m building 40 m wide takes $L = 30$ m, the lesser of the two, giving a GEP height of 75 m. Reading L as the greater dimension is the standard mistake and it over-states the answer every time.

The regulatory history matters for understanding what this number is FOR. Through the 1960s and 70s the cheapest response to a ground-level concentration problem was a taller stack, and the tallest reached over 380 m — dispersion used as a substitute for control, and acid deposition exported hundreds of kilometres downwind. The 1977 Clean Air Act amendments closed that door by capping the stack height a source may take CREDIT for in its dispersion modelling. A source may build any height it likes; it simply may not model above GEP. So this equation sets a ceiling on credit, not a required height, and it is the only equation on this site whose purpose is to limit what you are allowed to claim.

Two provisions go with it. There is a floor of 65 m: any source may claim 65 m of credit regardless of what the buildings around it look like, so the operative GEP height is the greater of 65 m and the formula height. And the formula only applies to structures close enough to matter — the influencing building must lie within five times L of the stack, measured along the wind direction under consideration. Since L and the projected width both change with wind direction, a serious downwash analysis works through the full compass, and the governing structure is often not the obvious one. Modern regulatory practice runs EPA's BPIP pre-processor to sort this out and feeds the result into AERMOD's PRIME downwash algorithm rather than applying the formula by hand.

When the stack cannot be raised to GEP, and often it cannot, the alternatives are all about momentum and geometry. Raising the exit velocity keeps the plume out of the cavity for the same physical height. Removing the rain cap is worth more than most people expect: a flat cap deflects the plume sideways into the wake and is the single most common cause of an entrainment complaint, which is why a drain and a rain guard beat a cap. Grouping several small stacks into one taller flue raises both the height and the momentum. And where the receptor is a fresh-air intake on the same roof rather than the fenceline, ASHRAE's dilution method for intake separation is the right tool — GEP is a regulatory dispersion rule and was never meant to site an intake.

Buoyancy flux is the single number that says how hard a plume wants to rise. It is not a measured quantity, it is a constructed one, and its units of m⁴/s³ are the giveaway: this is a volume flow multiplied by a reduced gravity. Written out, $F = g \, v_s r^2 (T_s - T_a)/T_s$ with the stack RADIUS, which becomes the familiar $g v_s d^2 (T_s-T_a)/(4T_s)$ once you substitute $r = d/2$. A stack 2 m across, discharging at 20 m/s and 400 K into air at 280 K, has $F = 9.80665 \times 20 \times 4 \times 120/(4 \times 400) = 94144/1600 = 58.8$ m⁴/s³, which is a mid-sized industrial boiler.

The temperature ratio is doing something specific and it is worth seeing why. At constant pressure the ideal gas law makes density inversely proportional to absolute temperature, so the fractional density deficit of the hot gas relative to ambient is $(\rho_a - \rho_s)/\rho_a = (T_s - T_a)/T_s$. That fraction multiplied by g is the reduced gravity acting on the plume, which is why the temperature difference has to sit over an absolute temperature and never over a Celsius one. Which absolute temperature belongs in the denominator is a genuine small ambiguity in the literature: Briggs and the EPA formulations use $T_s$, several textbooks use $T_a$, and for a stack at 400 K against air at 280 K the two differ by 43 percent in the denominator but only about 12 percent in the final plume rise, since F enters the rise as a cube root. Pick one, say which one, and do not mix sources.

The value of F sorts a stack into a regime. Briggs' own break point is 55 m⁴/s³, below which the distance to final rise is $x^* = 14F^{5/8}$ and above which it is $34F^{2/5}$. Small F also means the plume may be momentum-dominated rather than buoyancy-dominated, in which case the rise comes from a different equation entirely and this parameter is the wrong tool. And F can be negative. A wet scrubber leaves its exhaust saturated and close to ambient temperature, sometimes below it, and a negative buoyancy flux means a plume that sinks out of the stack rather than rising from it. That is precisely why scrubbed stacks so often carry a reheat burner: the fuel is buying stack height that steel would cost far more to provide.

The errors here are mechanical rather than conceptual. Using the stack cross-sectional AREA where the equation wants the diameter squared costs a factor of $\pi/4$. Using Celsius anywhere in the expression is fatal. And the exit velocity has to be the ACTUAL velocity at stack conditions, not one back-calculated from a flow rate quoted at standard conditions: a flue gas at 400 K occupies about 1.4 times the volume it does at 293 K, so using the standard-condition figure understates both the velocity and the flux by that much.

Briggs' two-thirds law is the result of asking how far a buoyant jet climbs while the atmosphere entrains air into it at a rate proportional to its own rise velocity. The answer is that the rise grows as the two-thirds power of downwind distance and falls inversely with wind speed: $\Delta h = 1.6 F^{1/3} x^{2/3}/u$. Take a plume with $F = 64$ m⁴/s³, at 1000 m downwind in a 5 m/s wind. The cube root of 64 is 4, the two-thirds power of 1000 is 100, so $\Delta h = 1.6 \times 4 \times 100/5 = 128$ m. The 1.6 is not derived, it is fitted, and it comes from Briggs' own 1975 review of plume photographs and lidar traverses at power stations.

Read the exponents and the design advice falls out. Rise depends on the cube root of the buoyancy flux, so quadrupling the heat release buys only 59 percent more rise, which is why nobody solves a dispersion problem by burning more fuel. Rise depends inversely on wind speed, so a plume that climbs 128 m at 5 m/s climbs only 64 m at 10 m/s. That inverse dependence fights with the direct dilution the same wind provides, and the fight is what produces a critical wind speed at which ground-level concentrations are worst, usually somewhere in the 3 to 8 m/s range for a tall buoyant source.

The honesty this equation needs is about its range. It describes TRANSITIONAL rise, the climb the plume is still doing, and it has no idea that the plume eventually levels off. Briggs puts final rise at a downwind distance of $3.5x^*$, with $x^* = 14F^{5/8}$ below 55 m⁴/s³ and $34F^{2/5}$ above it. For the example above, $F = 64$ gives $x^* = 34 \times 64^{0.4} = 34 \times 5.28 = 179$ m, so final rise happens at about 628 m, and the true rise is $1.6 \times 4 \times 628^{2/3}/5 = 1.6 \times 4 \times 73.3/5 = 94$ m. Feeding 1000 m into the equation returned 128 m, an over-prediction of 36 percent, and the error grows without limit the further out you go. Compute $3.5x^*$ first, and if your distance of interest is beyond it, use the rise at $3.5x^*$ and stop.

Two more boundaries. This form is for neutral and unstable air only. In stable air the plume runs out of buoyancy against the ambient stratification and levels off at $\Delta h = 2.6(F/us)^{1/3}$, where s is the stability parameter $g(\partial \theta/\partial z)/T_a$, and in stable calm it is $5F^{1/4}/s^{3/8}$. The equation also divides by wind speed, so it blows up at calm, which is not a prediction that plumes rise infinitely high on still nights but a statement that the model has left its domain. Finally, use the wind speed at STACK TOP, not the anemometer reading at 10 m, or the rise will be over-predicted by whatever the wind profile happens to be worth.

Holland published this in 1953 from observations at Oak Ridge, and it does in one step what Briggs does in several: it adds a momentum term and a buoyancy term and returns a single final rise with no downwind distance in it at all. The structure is easier to see when the bracket is read as two pieces. A stack 2 m across discharging at 15 m/s into a 5 m/s wind, at 1000 mb, with gas at 400 K against air at 280 K, gives $(T_s-T_a)/T_s = 0.3$, a buoyancy term of $2.68\times10^{-3} \times 1000 \times 2 \times 0.3 = 1.608$, a bracket of $1.5 + 1.608 = 3.108$, and a rise of $(15 \times 2/5) \times 3.108 = 6 \times 3.108 = 18.6$ m.

The 1.5 is the momentum coefficient, and $v_s d/u$ is the momentum length scale, so a cold plume with no buoyancy at all still rises $1.5 v_s d/u$ simply because it was thrown upward. The second term carries the buoyancy, and its constant of $2.68\times10^{-3}$ is dimensional: it is only correct with pressure in MILLIBARS, which is hectopascals, because that is the unit Holland's data was tabulated in. This is the single most common way to get a wrong answer out of this equation. Enter 101325 pascals where 1013 millibars belongs and the buoyancy term is inflated a hundredfold. The solver here converts internally from whatever unit you choose, so the constant always sees millibars.