The crux of the problem is spatial heterogeneity. Conventional evapotranspiration models were developed for agricultural fields and large-scale hydrological budgets, where vegetation is comparatively uniform. A bioretention basin is nothing of the sort. Within a few meters, plant height can jump from bare ground to four-meter shrubs, buildings and highway walls cast shifting shadows, and soil moisture swings dramatically between the flooded basin floor and its drier flanks. Because evapotranspiration, or ET, is governed by leaf area, solar radiation, vapor pressure deficit, and soil water availability, this patchiness propagates directly into uncertainty. Reported estimates of how much stormwater ET removes from green infrastructure range from nearly zero to almost 90 percent, a spread so wide it borders on useless for design purposes.

To pin down what is actually happening, the team spent a growing season measuring ET directly at eleven fixed plots within the Philadelphia basin using the closed dynamic chamber method. Transparent chambers were sealed over collars set into the soil, and a laser-based gas analyzer recorded water vapor accumulation once per second, allowing the researchers to calculate fluxes from the rate of buildup. Measurements were taken five to six times per day across seven dates from June through October, deliberately spanning dry spells and post-storm conditions. After discarding about 17 percent of records for quality reasons, 302 usable flux measurements remained, each tied to a specific combination of plant stature, topographic position, and microclimate.

The spatial context came from an unusual pairing of technologies. Terrestrial LiDAR scanning produced a bare-earth digital elevation model of the basin, while drone-based photogrammetry yielded a digital surface model of everything above it; subtracting the two mapped vegetation height across the site. To capture shading from the highway wall, nearby buildings, and a billboard, the researchers used a regional aerial LiDAR dataset and the Area Solar Radiation calculator in ArcGIS to compute shade coefficients for every pixel on every day of the study. These coefficients, typically between 0.75 and 0.95, scaled the measured solar radiation to reflect how much light actually reached each patch of ground. The basin was then segmented into 967 roughly plant-sized polygons using an image-clustering algorithm, and all environmental variables were aggregated to that scale.

With these ingredients, the team built a statistical model of ET that explicitly incorporated plant height, topographic position, shade, vapor pressure deficit, solar radiation, and soil moisture, complete with interaction terms and random effects for plot and date. The model explained most of the observed variation, with fixed effects alone accounting for 65 percent of variance. Its coefficients told a coherent physiological story: ET rose strongly with solar radiation and vapor pressure deficit, was lower on the drier upper flanks of the basin, and increased with plant height, reflecting greater leaf area per unit ground. Interestingly, the effect of radiation weakened in wetter soils, likely because near-saturated conditions restrict both evaporation and plant transpiration, while the effect of moisture on ET differed between the basin floor and its slopes.

Scaling the model across the whole basin revealed daily ET ranging from zero to about 6 millimeters per day, peaking in early and midsummer and declining to roughly 1 millimeter per day by mid-autumn. The clearest spatial pattern was elevated ET beneath the tallest vegetation, particularly a cluster of shrubs hugging the highway wall. Just as telling was what the sensitivity analysis showed. Setting all plant heights to zero reduced season-wide ET by 41 percent, while assigning every segment a uniform 100-centimeter height inflated it by 60 percent; even using the basin’s mean height of 36 centimeters everywhere depressed the total by 13 percent. Shade mattered too: removing all structural shading boosted seasonal ET by 11 percent, while assuming only a quarter of incoming radiation reached the basin cut it nearly in half. Topographic position, by contrast, had little effect on the basin-wide total.

How did the conventional models fare? The team ran six widely used equations, including the two Penman-Monteith variants endorsed by the American Society of Civil Engineers and the FAO, Priestley-Taylor, Hargreaves-Samani, Matt-Shuttleworth, and Granger-Gray. Most overpredicted ET on average. The Penman-Monteith, Priestley-Taylor, and Hargreaves-Samani models consistently overestimated relative to the empirical estimates, while Matt-Shuttleworth and Granger-Gray, which were less responsive to atmospheric conditions, overpredicted at the low end of the range and underpredicted at the high end. Digging into the discrepancies, the researchers found that differences between Penman-Monteith and empirical estimates grew systematically with wind speed, explaining up to 39 percent of the bias for the ASCE variant, whereas accounting for soil-water limitation made little difference; water stress was infrequent during the study period.

The practical payoff came from testing three correction strategies. Applying shade coefficients to the solar radiation inputs improved agreement for the Penman-Monteith and Priestley-Taylor models. Multiplicative landscape coefficients, analogous to crop coefficients in agriculture and calculated here at roughly 0.5 to 1.0, also helped those same models. But the strongest results came from additive corrections: simply subtracting a regression-derived offset from Penman-Monteith and Priestley-Taylor estimates raised concordance correlation coefficients to about 0.88 or 0.89, a dramatic improvement. The caveat is that additive correction can produce negative daily ET values when uncorrected estimates are small, so it must be used with care. None of the corrections improved the already-decent Matt-Shuttleworth and Granger-Gray estimates, and the Hargreaves-Samani model remained the poorest performer regardless of adjustment, a notable finding given that some reviews have recommended it for its minimal data requirements.

The implications ripple outward from one Philadelphia basin. For designers of similar systems, the message is that Penman-Monteith and Priestley-Taylor, once corrected, can credibly represent ET in water balances used for regulatory compliance and volume-reduction credit, rather than being ignored or trusted blindly. The study also suggests a design lever: because plant height and shade dominate basin-scale ET, landscape architects can deliberately position larger plants in unshaded locations to boost atmospheric water removal, provided the coarse, fast-draining soil media can supply enough moisture between storms. Deep-rooted shrubs and trees may be particularly valuable here, drawing water from a broader depth range during dry spells.

The authors are candid about limitations. Their empirical model was calibrated on a single basin, and ET from the tallest plants, which covered only about 10 percent of the area but exerted outsized influence, required extrapolation beyond the height of the flux chambers. Microclimatic variables other than solar radiation were assumed uniform across the basin, and lacking inflow data, the team could not compute ET’s share of total stormwater removal. Still, the basin contains many features common to bioretention design everywhere, and the growing availability of high-resolution LiDAR, drone imagery, and weather data makes locally calibrated corrections increasingly feasible. As cities pour billions into green stormwater infrastructure to combat sewer overflows and climate-driven downpours, this study delivers a refreshingly actionable insight: the atmosphere’s share of the water budget is measurable, modelable, and, with a few well-chosen coefficients, finally predictable.

Subject of Research: Spatial heterogeneity in evapotranspiration estimates for urban bioretention basins

Article Title: Incorporating spatial heterogeneity into evapotranspiration estimates for bioretention basins

Article References: Caplan, J. S., Bouda, M., Salisbury, A. B., Alonzo, M., Nyquist, J. E., Toran, L., & Eisenman, S. W. (2026). Incorporating spatial heterogeneity into evapotranspiration estimates for bioretention basins. Hydrology and Earth System Sciences, 30(18), 6075-6093. https://doi.org/10.5194/hess-30-6075-2026

Image Credits: AI Generated

DOI: 10.5194/hess-30-6075-2026

Keywords: evapotranspiration, bioretention basins, green stormwater infrastructure, Penman-Monteith, spatial heterogeneity, urban hydrology, flux chambers, LiDAR, shade coefficients, landscape coefficients, Philadelphia, stormwater management