Curve Number Selection for Mixed Land Use Watersheds

The curve number method gives you a runoff depth, but not a peak flow. It is nonlinear: runoff depth Q changes disproportionately as the maximum retention term S changes, so a small error in CN does not stay small. It grows, especially at rainfall depths near the threshold where P equals Ia, defined as 0.2S.

CN itself is a rescaled version of S, mapped onto a 0 to 100 scale through S = (1000 / CN) − 10. It compresses hard at the top, so a shift of a few points near CN 90 or above changes computed runoff far more than the same shift would near CN 60. This matters enormously once a watershed stops being a single, uniform surface, because a site with parking, rooftops, lawn, woodland, and farmland all draining to one point contains CNs from 98 down to 30 or lower for good woodland on well-drained soils. Picking one CN to represent that whole mix, by eyeballing it as "mostly residential" or "mostly open," is a structural error that misrepresents how the land actually behaves during a storm. Research evaluating how CN gets implemented in urbanized watersheds, comparing a single basin-wide value against disaggregated sub-areas, has found that the implementation choice itself changes peak flow predictions materially. Which CN value gets picked matters, but so does the method used to build it, and that is the argument this piece works through from here.

The three inputs that determine a single-area CN

Before any composite can be built, each individual sub-area needs a CN that is actually defensible, and that requires three separate judgment calls: hydrologic soil group, cover type and treatment, and hydrologic condition. Get any one wrong across a sub-area and the error doesn't stay contained. It spreads through every acre that shares that soil, that cover, or that condition rating.

Hydrologic soil group sits at the center of most of the trouble. The data source matters too. If you pull HSG data from an old county soil map or a legacy GIS layer without checking whether it has since been reclassified, you can quietly bias an entire pervious sub-area, so start with the NRCS Web Soil Survey, and use it alongside the NRCS Field Office Technical Guide and published soil survey databases.

Cover type is the second input, and it comes with its own built-in assumptions. Land cover also moves over time, and research tracking land use and land cover change has shown that watershed-scale CN values shift as development, clearing, or revegetation proceeds. Cover classification needs to reflect the site as it stands now, not as an old aerial photo recorded it.

Hydrologic condition is the third input, and it's the one most likely to get skipped. Condition ratings for pasture, open space, and woodland run from Good, with more than 75 percent ground cover and light or occasional grazing, down through Fair at 50 to 75 percent, to Poor below 50 percent and compacted. Yet condition is the input engineers skip most often when they lack site-visit data, defaulting every pervious sub-area to "Fair" or "Average" out of convenience. So you end up flattening real variation across the watershed into a single number that reflects nobody's actual site conditions.

One more adjustment sits downstream of these three: antecedent moisture condition, applied after individual CNs are chosen. Standard tables assume AMC II, average moisture. Dry conditions (AMC I) and wet conditions (AMC III) require separate conversion equations, and that conversion is itself nonlinear: the gap between AMC I and AMC III widens as the underlying CN(II) value climbs. That means the AMC choice carries the most weight precisely on the high-CN sub-areas, impervious cover above all, where it can swing computed runoff more than people expect from what looks like a routine moisture adjustment.

Why HSG is not spatially uniform across a real watershed

Soil boundaries and land use boundaries rarely line up. A single parcel zoned residential can sit across two or three distinct soil series, each with a different HSG, so if a sub-area spans more than one soil series, you need to split it before you assign a CN to it. Lumping that variation together, treating a mixed-soil parcel as if it sat entirely on one HSG, is probably the single most common way pervious zones end up with CNs that are too low, understating runoff from land that doesn't drain as well as the simplified map suggests. Research into HSG classification uncertainty backs this up directly: soil assignments near group boundaries already carry a measurable spread in the resulting CN, and that spread gets worse, not better, when a single HSG is forced onto a large area that actually contains more than one soil type.

The fix is to treat the land use polygon and the soil mapping unit as two separate layers that need to be intersected before any lookup happens. Each cover type and soil group combination gets its own row with its own area, and you build the area-weighted composite from the full matrix of these intersections. The tool changes; the requirement to intersect before weighting does not.

Dual-rated soils force a judgment call, and no software can make it for you. Spatially explicit, high-resolution CN datasets covering the contiguous United States, updated annually, have shown that this kind of dynamic soil-land cover interaction gets missed by static, county-level soil surveys. Treat legacy soil data as a starting point to check against current sources, not a final answer, especially on any parcel with documented grading, clearing, or drainage work.

Computing the area-weighted composite CN correctly, with a worked example

With sub-areas properly classified and split along soil boundaries, the composite CN is the area-weighted mean of the individual sub-area values: CN_composite equals the sum of each CN_i times its area A_i, divided by the total area. This weighting scheme is the only defensible way to combine sub-area values, because it preserves the linear relationship between area and maximum retention S before that nonlinear runoff equation ever gets applied.

The process runs in four steps. First, delineate and measure every sub-area, where each unique combination of cover type, soil group, and hydrologic condition gets its own row in the table, with all areas tracked in consistent units throughout, whether acres or square feet. Second, assign a CN to each sub-area using the TR-55 table appropriate to its cover type, drawing on standard values such as 98 for paved parking and roofs across all soil groups, 89 through 95 for a commercial district at roughly 85 percent impervious depending on soil group, 61 through 87 for quarter-acre residential lots at roughly 38 percent impervious, 39 through 80 for open space in good condition, and 30 through 77 for woodland in good condition. Each row should carry a record of the source table, the soil group used, and the condition rating applied, since that documentation is what lets a reviewer or a regulator audit the composite without rebuilding the whole analysis from the ground up.

Third, multiply each sub-area's CN by its area to get a CN times A figure for that row, sum those products across all rows, divide by total contributing area, and round to the nearest whole number for design use. A ten-acre site works as a clean illustration of that arithmetic. One and a half acres of woodland on Group B soil in good condition, at CN 55, contributes 82.5. Those four products sum to 799.5, and dividing by the total ten acres gives a composite CN of approximately 80. That number of 80 then becomes an input, not an answer: it converts to S equal to 1000 divided by 80, minus 10, or 2.5 inches of maximum retention, and an initial abstraction Ia of 0.2 times S, or 0.5 inches, both of which feed into the runoff equation from the first section. State drainage guidance that applies the CN loss model typically requires the composite value to ship alongside its full land use breakdown, because the weighting table itself counts as a required piece of the submittal, not a scratch calculation left on the engineer's desk.

The nonlinear aggregation error: why averaging CN instead of S produces a biased result

Because the relationship between CN and runoff depth bends rather than runs straight, averaging CN values directly across sub-areas with very different values does not produce the same answer as averaging the physical quantity those CNs represent. Averaging CN arithmetically instead of averaging S, the maximum retention term, and only converting back to CN afterward, produces a composite runoff estimate that runs systematically high. The bias doesn't scatter randomly from one study to the next. It pushes toward over-prediction in any watershed that mixes high-CN and low-CN land covers, and that describes most mixed land use sites by definition. Work on the theoretical underpinnings of the CN method has examined how these aggregation assumptions, built into the structure of a watershed model, shape the runoff predictions that come out the other end: the level at which values get aggregated is a modeling decision with real consequences, not a footnote to skip past.

So you need to run the path through S, not through CN directly. Convert each sub-area's CN_i to its corresponding S_i using S_i = (1000/CN_i) − 10. Compute the area-weighted average of those S values across all the sub-areas. Then convert that average back to a composite CN: CN_composite = 1000 / (S_composite + 10). This route produces a lower composite CN, and therefore less computed runoff, than arithmetic CN averaging, consistent with the proportionality the method was built on.

The gap between the two methods widens with the spread of CNs in the mix. Analysis of CN implementation choices in urbanized watersheds has confirmed that different aggregation approaches produce different peak flow predictions for the identical physical watershed, which settles the point that aggregation method is a variable a practitioner has to choose deliberately and document, not an afterthought folded into the spreadsheet.

Temporal and spatial CN variability in real watersheds

A composite CN you calculate correctly today can still become wrong tomorrow, because the land cover it describes changes and because real antecedent moisture conditions drift away from the AMC II assumption baked into standard tables. CN selection is a value that needs to be checked against current conditions every time it gets used in a new study, not a lookup performed once and filed away.

Land cover change drives a large share of that drift. Research forecasting land use and land cover change in Brazilian watersheds has shown that CN parameters move substantially as urbanization, deforestation, or agricultural conversion progresses across a basin. The consequence for practice is straightforward: a study built on historical land use data will understate runoff in a watershed that is actively developing, and it will overstate runoff in a watershed where vegetation has since recovered. Annual dynamic CN datasets covering the contiguous United States from 2008 through 2021 track this kind of year-to-year change, driven by land cover shifts, drought cycles, and agricultural management practices, and the range of variation recorded across those years suggests that treating CN as fixed over a project's design life can introduce more uncertainty than a soil group misclassification would.

Antecedent moisture carries its own version of this problem. The AMC adjustment equations that convert CN(II) to CN(I) and CN(III) compound this, since that conversion is itself nonlinear and the gap between dry and wet conditions widens as CN climbs, so for a heavily impervious mixed-use watershed, the AMC choice alone can move computed runoff more than reclassifying an entire sub-area's soil group would. None of this is an argument for walking away from the method. It is an argument for checking three things before trusting a composite CN: confirm the land cover data is current, write down the AMC assumption used in the calculation explicitly, and flag any known or planned land cover changes that fall within the project's design life.

The disaggregated sub-basin approach

A single composite CN can get the total runoff volume right and still get the hydrograph wrong. So when a mixed land use watershed contains sub-areas with very different CN values, different flow path lengths to the outlet, or separate points of concentration, folding everything into one composite number suppresses the shape and timing of the resulting hydrograph even when the volume underneath it checks out. Volume accuracy doesn't guarantee peak flow accuracy, and for most design purposes, peak flow is the number that actually governs pipe sizing, detention volume, and floodplain mapping.

So if sub-basin routing is called for, you don't abandon the composite CN method. State drainage design guidance generally treats the CN loss model as one piece of a larger hydrograph method, where CN supplies the loss function and the sub-basin structure governs how that loss translates into timing. CN selection and basin delineation are coupled decisions that have to be made together, not a sequence where one gets settled before the other is even considered. A watershed with a wide CN spread, divergent routing paths, or a regulatory requirement for sub-basin analysis calls for disaggregation instead, because at that point the implementation method decides the accuracy of the result as much as the CN value itself ever did.