Nutrient removal prediction using hyperspectral reflectance indices and modeling for a pilot constructed channel treatment wetland

The rapid assessment of treatment wetland performance is an important goal to help optimize nutrient removal and pollution mitigation in the design and implementation of treatment channel wetlands technologies. Four pilot-scale surface-flow channel wetlands were constructed to evaluate the potential for treatment of simulated agricultural runoff in the South Texas environment; two channel wetlands were vegetated and the other two were unvegetated and used as a control. The system was operated at four days of hydraulic retention time to evaluate the effect of three different water depths (3, 6 and 9 inches). The results demonstrated a high removal for Nitrogen species in the vegetated channel wetlands, about 85% for TKN, 58% for Nitrates and 97% for ammonia, while the removal of TP removal was about 33% and Orthophosphate removal about 45%. In addition, field spectral measurements using a portable spectroradiometer which acquires continuous spectra from 300-1100 nm were completed. Specific PRI, NDVI and CI vegetation spectral indices were computed from acquired spectral data and compared to nutrient concentrations and changes in concentration levels in the water phase. Predictions of nutrient concentrations in the water column based on linear regressions between nutrient concentrations in the water column and the vegetation indices (PRI, NDVI, CI) yielded reasonable R2 values ranging from 0.26 to 0.52 for different indices. When the rates of change of PRI, NDVI and CI indices (Δ) between different sampling locations were compared to rates of change of nutrient levels in the wetland, the R2 values were also reasonable ranging from 0.51 to 0.64. These results suggest that combinations of the PRI, NDVI and CI index data may be a potentially useful predictor of nitrogen and phosphorous concentrations and uptake rate performance in constructed wetland water columns. The most optimal predictor of nutrient values appeared to be the NDVI index followed by the CI and PRI. Additionally, to demonstrate the potential applicability of this new approach for treatment wetland performance evaluation, a nitrogen dynamics model was developed using uptake rates determined from the field sample testing and radiometric measurements. However, long term data collection will still be necessary to more completely evaluate field scale channel constructed wetland performance removing nutrients through the different seasonal periods of operation. This work suggests that the use of hyperspectral reflectance technology in predicting the nutrient concentrations in a treatment constructed wetlands, and its scale up to field studies and modeling applications is a promising tool for design optimization and refinement.