The Oak Openings Region of Northwest Ohio is well known for rare plant and animal species. It contains the few remaining wet prairie ecosystems in the area. Wet prairie ecosystems are highly sensitive to precipitation and evapotranspiration patterns as they are highly dependent on shallow groundwater levels (water table). A predicted rise in temperature and change in frequency and intensity of storm events may impact the hydrology of wet prairies.
What is the relationship in a wet prairie between water table and precipitation and evapotranspiration? How would the hydrology of a wet prairie change according to the expected climate change in the Midwest?
Hourly groundwater level data were collected using data loggers installed at six different piezometer locations in a wet prairie in the Oak Openings Region from May 2015 to January 2017. Hourly precipitation and air temperature data was obtained from the Toledo Express Airport weather station (TOL), located about 4 km from the study area. Artificial Neural Networks (ANN), a form of machine learning, was used to determine the correlation between temperature, evapotranspiration (estimated from air temperature) and the water table. ANN was used due to the large amount of data available and the complexity of modeling a wet prairie ecosystem that is not well understood physically, and has been intensely managed with ditches.
Results/Conclusions
A time series analysis of the water table and precipitation and evapotranspiration was performed. Pearson correlation tests showed a negative correlation with evapotranspiration (r = -0.46), and a positive, lagged, cross-correlation between the water table level and precipitation time series data. The estimated maximum values of "r" were 0.24, 0.81, and 0.31 for the whole precipitation dataset, for a specific rain event, and for the dry period between two rain events, respectively. These results varied with different seasons, thereby showing a seasonality effect on the correlation between precipitation, evapotranspiration, and water levels. Based on the time series analysis and using a multilayer feed-forward neural network (ANN), a good prediction was obtained between precipitation, evapotranspiration and water table levels for long-term time periods. This ANN model will be used to predict probable changes to the wet prairie hydrology with expected climate change in the Midwest.