Skip to main content
ARS Home » Northeast Area » Beltsville, Maryland (BARC) » Beltsville Agricultural Research Center » Environmental Microbial & Food Safety Laboratory » Research » Publications at this Location » Publication #321628
Title: Stressor-response modeling using the 2D water quality model and regression trees to predict chlorophyll-a in a reservoir system

Author
item PARK, YOUNEUN - Orise Fellow
item Pachepsky, Yakov
item CHO, KYUNG HWA - Gwangju Institute Of Science And Technology
item JEON, JIN - Gwangju Institute Of Science And Technology
item KIM, JOON HA - Gwangju Institute Of Science And Technology

Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/4/2016
Publication Date: 3/13/2016
Citation: Park, Y., Pachepsky, Y.A., Cho, K., Jeon, J., Kim, J. 2015. Stressor-response modeling using the 2D water quality model and regression trees to predict chlorophyll-a in a reservoir system. Journal of Hydrology. 529:805-815.

Interpretive Summary: The quality of irrigation, recreational, aquacultural, and many other types of water is affected by algal blooms, as many microbes are harbored and nurtured in these algal communities. There is a need to estimate and forecast algal blooms which are controlled by environmental stressors such as nutrient concentrations in water, temperature, water flow, etc. Existing mechanistic models have the capability to forecast algal blooms based on biogeochemical relations in aqueous ecosystems. However, their output is very complex and in practice cannot be used to facilitate management decisions. We developed a method to compress results of these simulations into easily readable diagrams and rules that relate the monthly bloom algal levels to the levels of individual stressors. Results of this work are important for water quality assessment and management in that the mechanistic modeling results are presented via simple graphics that can be used to assess the status of aqueous ecosystems and and the impact of management decisions on water quality.

Technical Abstract: In order to control algal blooms, stressor-response relationships between water quality metrics, environmental variables, and algal growth should be understood and modeled. Machine-learning methods were suggested to express stressor-response relationships found by application of mechanistic water quality models. The objective of this work was to evaluate the efficiency of the regression trees in development of a stressor-response model for chlorophyll-a concentrations (chl-a) using results of site-specific mechanistic water quality modeling. The CE-QUAL-W2 model was applied to simulate water quality using four-year observations data and additional scenarios of air temperature increases for the Yeongsan Reservoir in Korea. The regression tree modeling was applied to the results of simulations. Given the well expressed seasonality in simulated chl-a dynamics, separate regression trees were developed for each month from May to September. Regression tress provided a reasonably accurate representation of the stressor-response dependence generated by CE-QUAL-W2 model. Different stressors were selected as split variables for different months. Splits by the same stressor variable in most cases showed the same sign of the correlation between this stressor variable and chl-a values. Nutrient content variables appeared to be better predictors of the chl-a responses as compared with abiotic variables. The highest chl-a temperature sensitivity was found in May or June. Regression tree splits by ammonium concentrations resulted in a consistent trend of sensitivity being larger in the group of samples with the larger ammonium concentrations. Regression tree models provided a transparent visual representation of stressor-response relationships for chl-a and its sensitivity. Overall, the representation of stressor-response relationships with the classification and regression tools can be a useful approach to assessment of aquatic ecosystem state and effective determination of significant stressor variables.