There is a moment that happens to most engineers the first time they work in karst terrain. You've done everything right. The survey is accurate. The curve numbers are appropriate. The time of concentration is reasonable. The model runs cleanly. You present your design with confidence.
And then someone who has actually watched that watershed during a storm tells you that roughly a third of the runoff you so carefully routed through your storm system disappears into a hole in the ground about 200 meters upstream, resurfaces as a spring two kilometers away, and has absolutely nothing to do with the drainage infrastructure you just spent three weeks designing.
Welcome to karst. Population: your humbled assumptions.
Karst terrain forms where soluble bedrock, primarily limestone, but also dolomite and gypsum, has been dissolved over geological time by mildly acidic groundwater. The result is a landscape riddled with sinkholes, caves, solution conduits, disappearing streams, and springs. The surface looks more or less normal. Underneath, it is a three-dimensional plumbing system of extraordinary complexity that was built over millions of years without any input from the engineering community.
In Puerto Rico, karst terrain covers approximately 27% of the island's total area, concentrated along the northern coastal plain and extending inland across portions of the island's interior. This is not a marginal geological curiosity. It is a dominant hydrologic feature of one of the most densely populated and actively developed regions of the island.
The northern karst belt sits between the central mountain range and the Atlantic coast. It receives intense tropical rainfall, sits above a critical freshwater aquifer, and contains some of the most ecologically sensitive terrain on the island, mogotes, sinkholes, blind valleys, and the cave systems that have formed over the past several million years. It is also, inconveniently, where a significant amount of development pressure exists.
The foundation of most stormwater engineering is the assumption that water moves overland or through the soil profile in ways that can be reasonably approximated by the methods in our standard toolbox: TR-55, the Rational Method, HEC-HMS, SWMM. These tools were developed primarily for surface-dominated drainage systems. They perform well in environments where infiltration is limited, flow paths are relatively predictable, and the relationship between rainfall and runoff follows the patterns embedded in their calibration datasets.
In karst, the plumbing is underground, the flow paths are three-dimensional, the travel times are highly variable, and a significant fraction of the water budget can move through preferential conduit flow at velocities that bear no resemblance to Darcy's Law. The result is that standard methods can be simultaneously technically correct and practically wrong.
Specifically:
Curve numbers are problematic. TR-55 curve numbers assume that infiltration occurs relatively uniformly across a watershed, governed primarily by soil type and land cover. In karst, point recharge through sinkholes and solution features can locally concentrate infiltration in ways that no CN-based method can adequately capture. A watershed with a CN of 65 on paper can behave like a CN of 40 or lower if a significant portion of the contributing area drains to a sinkhole that acts as a direct connection to the aquifer.
Travel times are unreliable. Sheet flow and channel flow calculations assume that water moves downgradient across the surface or through defined channels. In karst, water can enter a subsurface conduit, travel at velocities of meters per second through a cave system, and emerge at a spring whose geographic location has no obvious relationship to the topographic drainage area. Time of concentration calculations for karst watersheds can be wildly inaccurate in either direction.
Watershed boundaries are approximate at best. In non-karst terrain, topographic divides are reliable drainage area boundaries. In karst, subsurface flow can cross topographic divides entirely. The effective contributing area to a given discharge point may be dramatically larger or smaller than the topographic watershed suggests.
One of the most consequential practical implications of karst hydrology for stormwater design is the behavior of detention facilities. A detention basin placed in karst terrain can drain catastrophically into the subsurface, either through existing solution features that weren't detected during site investigation or through new sinkholes that form after construction as the added hydraulic load accelerates dissolution.
This is not a theoretical concern. Karst sinkhole formation in response to detention pond construction has been documented in multiple jurisdictions. The failure mode is particularly problematic because it tends to happen suddenly, often during or after the first major storm event that fills the basin to design capacity, and because it can create direct hydrologic connections between surface stormwater, which may carry pollutants, sediment, and nutrients, and the underlying aquifer.
In Puerto Rico, the northern karst aquifer is a primary freshwater resource for a significant portion of the island's population. A detention pond that inadvertently creates a conduit to that aquifer is not just a stormwater management failure. It is a water quality threat of a completely different order of magnitude.
Designing for karst requires accepting that the subsurface system is, in most practical respects, unmappable at the resolution needed for conventional stormwater calculations. You will not drill enough borings. You will not run enough tracer tests. The system is too complex, too spatially variable, and too expensive to characterize at engineering precision. This is not a failure of investigation effort. It is a fundamental property of the terrain.
What this means in practice:
First, work with the recharge, not against it. In non-karst terrain, stormwater engineering is largely about preventing infiltration and managing surface flow. In karst, the terrain wants to infiltrate. Designs that work with that tendency, directing runoff to stable recharge zones, using infiltration-based practices in appropriate locations, avoiding impervious cover over high-recharge areas, tend to perform better than designs that try to collect and convey everything on the surface.
Second, protect the aquifer first. Any design that introduces surface water to subsurface karst features needs to account for the fact that treatment and filtration in conduit-dominated karst systems is minimal. Water quality at the source matters enormously because there may be no natural treatment between the surface and the spring.
Third, be conservative about contributing areas. When in doubt, assume your watershed is larger than the topography suggests. The consequences of undersizing infrastructure because you underestimated the contributing area in karst are worse than the consequences of mild oversizing.
Fourth, involve a karst geologist. This is not a standard stormwater problem, and it should not be treated as one. The subsurface characterization work that informs a karst stormwater design requires expertise that goes well beyond what most civil engineers carry in their toolkit.
Karst terrain covers approximately 15% of the Earth's ice-free land surface and provides drinking water to roughly 25% of the global population. In the United States alone, karst underlies significant portions of Florida, Kentucky, Tennessee, Missouri, Texas, and the Appalachian Valley and Ridge province, among others.
The engineering community's general tendency to treat karst as a footnote rather than a governing design condition has produced a long record of infrastructure failures, sinkhole collapses, aquifer contamination events, and stormwater systems that perform exactly as designed on paper and completely differently in the field.
The terrain is not the problem. The mismatch between how we model it and how it actually behaves is the problem. And the first step toward closing that gap is acknowledging, without embarrassment, that the standard toolbox has limits, and that some of those limits are geological rather than computational.
The water will find the cave. It always does. The question is whether we account for that in advance or discover it after the fact.
