Predictive Modeling of pH in an Aquaponics System Using Bayesian and Non-Bayesian Linear Regression to Inform System Maintenance

Aquaponicsthe farming of
fish and plants in a soilless
systemis a growing niche industry ideal for urban environments
and water-stressed regions. These systems can be a good source of
fresh food but must maintain a delicate balance between the water
quality requirements of the fish, plants, and nitrifying bacteria
or risk decreased production, disease, and death. One of the most
important water quality parameters is pH, which should be maintained
between 6.4 and 7.4 to meet the needs of all three organisms. However,
pH is often unstable and must be continually monitored and adjusted.
The general method for pH maintenance is a triage approach of measurement,
diagnosis of a range violation, and application of either an acidic
or basic compound to lower or raise the pH. A consequence of this
approach can be overcompensation when adding corrective chemicals
or system shock from too sudden of a shift in pH. This could have
negative effects on the health of all of the organisms in the system,
so a better alternative would be to predict pH values in advance and
slowly add the necessary balancing compounds over a longer period.
To predict pH in an aquaponics system, we conducted both traditional
linear regression and Bayesian linear regression using aquaponics
water quality data to develop a predictive model. It was found that
the pH values measured 1 and 2 days prior to the target date could
predict the pH in an aquaponics system with a Nash Sutcliffe Efficiency
score of 0.8 and a root-mean-square error of 0.181. The sensitivity
and specificity of the predictions for range violations were 0.78
and 0.99, respectively. A web application was developed to host this
model as well as to provide options for basic data analysis and visualization.