Staff backs year-long machine-learning assessment if Poland Spring volunteers study aquifer

Rick outlined a proposed voluntary periodic assessment Poland Spring has offered that would use multivariable regression and machine-learning techniques on the entire dataset, including background wells, to identify signals of natural precipitation versus local pumping response. "And the software will actually learn from the dataset," Rick said, arguing the method could discern subtle trends and produce more definitive conclusions than the simple well-to-well comparisons used previously.

Rick said the work would not require new monitoring points, would draw on existing data plus one year of additional data collection, and could be submitted in 2027 and reported in early 2028. He framed the assessment as a potential condition of a permit if the parties agree: Poland Spring would provide a proposal for the study, the board would approve a schedule and then the company would execute it as a permit condition. The board asked staff to identify what specific variables and background wells the assessment should include and to return with a concise list for the applicant.