Locating Wetlands for Better Model Development and Planning

As hydrologic engineers, we would be remiss not to use all information available to us in the course of understanding water flow within a drainage basin. Increasingly, geospatial data, high-resolution terrain models, and machine learning are available to engineers for a variety of purposes, including the development of hydrologic models. The Wetland Intrinsic Potential (WIP) tool, developed by researchers at the University of Washington in Seattle, uses multivariable datasets, lidar-derived terrain information, multi-scale topographic analysis, and a random forest machine-learning model to identify wetlands that may be missed by conventional mapping methods.

As hydrologists recognize, wetlands influence flood storage, stormwater routing, groundwater recharge, habitat protection, permitting and land-use decisions. When existing inventories miss small, seasonal, forested or otherwise difficult-to-detect wetlands, project teams may underestimate site constraints or overlook important hydrologic functions.

WIP was specifically developed to help locate “cryptic wetlands” that can be difficult to detect because they’re seasonally inundated, hidden beneath forest canopy, obscured by terrain or lacking visible standing water when aerial imagery is collected. Instead of relying only on direct visual indicators, WIP uses spatially derived proxies for wetland indicators such as hydrology, hydrophytic vegetation, and hydric soils. This methodology has proven effective. In the Hoh River watershed test case in Washington, the tool identified 17,300 acres (27.0 square miles) of wetlands using a 0.5 probability threshold, compared with 7,500 acres (11.7 square miles) mapped by the National Wetlands Inventory.

Topography, Hidden Hydrology and Wetland Indicators

Wetland professionals have long relied on three key indicators when identifying wetlands in the field: 1) hydrophytic vegetation, 2) hydric soils, and 3) wetland hydrology. WIP follows this same conceptual framework but uses topographic and remotely sensed clues to detect the hydrologic systems that often support wetland formation.

Changes in elevation, slope gradient, land-surface curvature, depth to water, flow accumulation, and other terrain attributes can reveal a lot about surface and subsurface flows, saturated soils, drainage patterns, and locations where hydrophytic vegetation is more likely to occur. These indicators are especially useful where wetlands are difficult to see directly from imagery or where field access is limited.

National Agriculture Imagery Program data, USGS hydrography data, soils data, lidar-derived digital elevation models and other datasets can be used to derive important wetland indicators. These include hydraulic conductivity, soil traits influencing water retention or drainage, wetland-associated vegetation, depth-to-water indices, topographic wetness indices, and other hydrologic indicators evaluated at multiple spatial scales. Considered together, these indicators provide a probability-based view of wetland potential across a given area of interest.

The model works by evaluating the input datasets and generating a probability of wetland presence for each digital elevation model grid cell or pixel. Users can set a probability threshold for classifying areas as wetlands. In the Hoh River watershed example, a threshold of 0.5 was used. A lower threshold would classify more land area as potential wetland, while a higher threshold would reduce the area classified as wetland. This probability-based output provides a more-nuanced picture than a simple binary wetland/non-wetland map, particularly in areas with gradual wet-to-dry transitions.

Beyond identifying additional wetland areas, WIP also reduced the number of wetlands likely missed; this often is called “omission error,” meaning areas that actually are wetlands but aren’t identified as wetlands in the map or model output. In the Hoh River watershed test case, WIP reduced this missed-wetland error by more than 33 percent compared with the National Wetlands Inventory, while also improving the overall accuracy of wetland identification by approximately 8 percent. This means that WIP can be useful as a screening tool in places where small, seasonal, forested or otherwise difficult-to-detect wetlands may be overlooked by existing inventories.

WIP is flexible in order to be easily adaptable to different site conditions. It is available as an ArcGIS toolbox and can be modified to integrate region-specific landscape clues or wetland indices. As a result, WIP can be used in many ways: hydrologic model development, but also for planning and related resource protection efforts like wellhead protection zoning and shoreline erosion control.

The developers of WIP stress that it is not a replacement for established wetland identification methodologies, but rather a complementary screening and mapping tool. Field investigation and formal wetland delineation procedures remain necessary where precise boundaries or permitting decisions are required. But WIP can help hydrologists and planners identify areas that warrant closer review, especially where conventional mapping may miss less-visible, but hydrologically key, wetlands.

The growth of methodologies that combine geospatial data, terrain analysis, remote sensing, and machine learning charts a path towards more integrated and fact-based decision-making. Tools that help us locate important hydrologic systems like wetlands can only help in the development of models that accurately reflect surface and subsurface flow dynamics.