Temperate lakes may contain both coolwater fish species such as walleye (Sander vitreus) and warmwater fish species such as largemouth bass (Micropterus salmoides). Recent declining walleye and increasing largemouth bass populations have raised questions regarding the future trajectories and management actions for these species. We developed a thermodynamic model of water temperatures driven by downscaled climate data and lake-specific characteristics to estimate daily water temperature profiles for 2148 lakes in Wisconsin, US, under contemporary (1989US, under contemporary ( -2014 and future (2040-2064 and 2065-2089) conditions. We correlated contemporary walleye recruitment and largemouth bass relative abundance to modeled water temperature, lake morphometry, and lake productivity, and projected lake-specific changes in each species under future climate conditions. Walleye recruitment success was negatively related and largemouth bass abundance was positively related to water temperature degree days. Both species exhibited a threshold response at the same degree day value, albeit in opposite directions. Degree days were predicted to increase in the future, although the magnitude of increase varied among lakes, time periods, and global circulation models (GCMs). Under future conditions, we predicted a loss of walleye recruitment in 33-75% of lakes where recruitment is currently supported and a 27-60% increase in the number of lakes suitable for high largemouth bass abundance. The percentage of lakes capable of supporting abundant largemouth bass but failed walleye recruitment was predicted to increase from 58% in contemporary conditions to 86% by mid-century and to 91% of lakes by late century, based on median projections across GCMs. Conversely, the percentage of lakes with successful walleye recruitment and low largemouth bass abundance was predicted to decline from 9% of lakes in contemporary conditions to only 1% of lakes in both future periods. Importantly, we identify up to 85 resilient lakes predicted to continue to support natural walleye recruitment. Management resources could target preserving these resilient walleye populations.

The concentration of dissolved oxygen in aquatic systems helps regulate biodiveristy 1, 2 , nutrient biogeochemistry 3 , greenhouse gas emissions 4 , and drinking water quality 5 . The long-term declines in dissolved oxygen concentrations in coastal and ocean waters have been linked to climate warming and human activity 6, 7 , but little is known about changes in dissolved oxygen concentrations in lakes. While dissolved oxygen solubility decreases with increasing water temperatures, long-term lake trajectories are not necessarily predictable. Oxygen losses in warming lakes may be amplified by enhanced decomposition and stronger thermal stratification 8, 9 or they may increase as a result of enhanced primary production 10 . Here we analyse 45,148 dissolved oxygen and temperature profiles from 393 temperate lakes spanning 1941-2017. We find that a decline in dissolved oxygen is widespread in surface and deep-water habitats. The decline in surface waters is primarily associated with reduced solubility under warmer water temperatures, although surface dissolved oxygen increased in a subset of highly-productive warming lakes, likely due to increasing phytoplankton production. In contrast, the decline in deep waters is associated with stronger thermal stratification and water clarity losses, but not with changes in gas solubility. Our results suggest that climate change and declining water clarity have altered the physical and chemical environment of lakes. Freshwater dissolved oxygen losses are 2.5-10 times greater than observed in the world's oceans 6, 7 and could threaten essential lake ecosystem services 2,3,5,11 .

The rapid growth of data in water resources has created new opportunities to accelerate knowledge discovery with the use of advanced deep learning tools. Hybrid models that integrate theory with state‐of‐the art empirical techniques have the potential to improve predictions while remaining true to physical laws. This paper evaluates the Process‐Guided Deep Learning (PGDL) hybrid modeling framework with a use‐case of predicting depth‐specific lake water temperatures. The PGDL model has three primary components: a deep learning model with temporal awareness (long short‐term memory recurrence), theory‐based feedback (model penalties for violating conversation of energy), and model pretraining to initialize the network with synthetic data (water temperature predictions from a process‐based model). In situ water temperatures were used to train the PGDL model, a deep learning (DL) model, and a process‐based (PB) model. Model performance was evaluated in various conditions, including when training data were sparse and when predictions were made outside of the range in the training data set. The PGDL model performance (as measured by root‐mean‐square error (RMSE)) was superior to DL and PB for two detailed study lakes, but only when pretraining data included greater variability than the training period. The PGDL model also performed well when extended to 68 lakes, with a median RMSE of 1.65 °C during the test period (DL: 1.78 °C, PB: 2.03 °C; in a small number of lakes PB or DL models were more accurate). This case‐study demonstrates that integrating scientific knowledge into deep learning tools shows promise for improving predictions of many important environmental variables.