Lake health indicators change with shifting environmental conditions

Understanding drivers of water quality is critical for anticipating and managing lake ecosystems under global change. While theoretical relationships between drivers (air temperature) and water quality (chlorophyll-a) are well understood, it remains unclear whether these relationships remain stationary. Further, the temporal scale over which data are analyzed may influence driver-response relationships. To identify changing relationships, we fit univariate autoregressive models between drivers and lake trophic state over 20+ years in Lake Rotoehu, New Zealand using the Trophic Level Index (TLI), a widely used ecological indicator of lake water quality and nutrient status. We tested model performance over three window types: 1) the full time series, 2) three discrete 8-year windows, and 3) 8-year moving windows. Model performance and significance varied over time, indicating non-stationarity between drivers and TLI. Additionally, driver importance to the TLI also varied over window type. Specifically, meteorological and internal loading drivers were top predictors during full and discrete windows, but moving windows revealed the importance of other drivers (e.g., water level) which would otherwise have been unidentified. Examination of model parameters revealed that direction and magnitude of driver-TLI relationships also varied over window type, indicating that parameter values fit over a single window may not capture variability in relationships. We demonstrate the importance of using trophic state indicators dynamically, acknowledging shifting driver-response relationships to support adaptive freshwater management under environmental change.