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.
Key Points
Process‐Guided Deep Learning (PGDL) models integrate advanced empirical techniques with process knowledge
We used PGDL to accurately predict lake water temperatures for various conditions
PGDL performance improved significantly when pretraining data included diverse conditions generated by an existing process‐based model
Inputs of dissolved organic carbon (DOC) to lakes derived from the surrounding landscape can be stored, mineralized or passed to downstream ecosystems. The balance among these OC fates depends on a ...suite of physical, chemical, and biological processes within the lake, as well as the degree of recalcintrance of the allochthonous DOC load. The relative importance of these processes has not been well quantified due to the complex nature of lakes, as well as challenges in scaling DOC degradation experiments under controlled conditions to the whole lake scale. We used a coupled hydrodynamic-water quality model to simulate broad ranges in lake area and DOC, two characteristics important to processing allochthonous carbon through their influences on lake temperature, mixing depth and hydrology. We calibrated the model to four lakes from the North Temperate Lakes Long Term Ecological Research site, and simulated an additional 12 'hypothetical' lakes to fill the gradients in lake size and DOC concentration. For each lake, we tested several mineralization rates (range: 0.001 d(-1) to 0.010 d(-1)) representative of the range found in the literature. We found that mineralization rates at the ecosystem scale were roughly half the values from laboratory experiments, due to relatively cool water temperatures and other lake-specific factors that influence water temperature and hydrologic residence time. Results from simulations indicated that the fate of allochthonous DOC was controlled primarily by the mineralization rate and the hydrologic residence time. Lakes with residence times <1 year exported approximately 60% of the DOC, whereas lakes with residence times >6 years mineralized approximately 60% of the DOC. DOC fate in lakes can be determined with a few relatively easily measured factors, such as lake morphometry, residence time, and temperature, assuming we know the recalcitrance of the DOC.
The General Lake Model (GLM) is a one-dimensional open-source code designed to simulate the hydrodynamics of lakes, reservoirs, and wetlands. GLM was developed to support the science needs of the ...Global Lake Ecological Observatory Network (GLEON), a network of researchers using sensors to understand lake functioning and address questions about how lakes around the world respond to climate and land use change. The scale and diversity of lake types, locations, and sizes, and the expanding observational datasets created the need for a robust community model of lake dynamics with sufficient flexibility to accommodate a range of scientific and management questions relevant to the GLEON community. This paper summarizes the scientific basis and numerical implementation of the model algorithms, including details of sub-models that simulate surface heat exchange and ice cover dynamics, vertical mixing, and inflow–outflow dynamics. We demonstrate the suitability of the model for different lake types that vary substantially in their morphology, hydrology, and climatic conditions. GLM supports a dynamic coupling with biogeochemical and ecological modelling libraries for integrated simulations of water quality and ecosystem health, and options for integration with other environmental models are outlined. Finally, we discuss utilities for the analysis of model outputs and uncertainty assessments, model operation within a distributed cloud-computing environment, and as a tool to support the learning of network participants.
Salting our freshwater lakes Dugan, Hilary A.; Bartlett, Sarah L.; Burke, Samantha M. ...
Proceedings of the National Academy of Sciences - PNAS,
04/2017, Letnik:
114, Številka:
17
Journal Article
Recenzirano
Odprti dostop
The highest densities of lakes on Earth are in north temperate ecosystems, where increasing urbanization and associated chloride runoff can salinize freshwaters and threaten lake water quality and ...the many ecosystem services lakes provide. However, the extent to which lake salinity may be changing at broad spatial scales remains unknown, leading us to first identify spatial patterns and then investigate the drivers of these patterns. Significant decadal trends in lake salinization were identified using a dataset of long-term chloride concentrations from 371 North American lakes. Landscape and climate metrics calculated for each site demonstrated that impervious land cover was a strong predictor of chloride trends in Northeast and Midwest North American lakes. As little as 1% impervious land cover surrounding a lake increased the likelihood of long-term salinization. Considering that 27% of large lakes in the United States have >1% impervious land cover around their perimeters, the potential for steady and long-term salinization of these aquatic systems is high. This study predicts that many lakes will exceed the aquatic life threshold criterion for chronic chloride exposure (230 mg L−1), stipulated by the US Environmental Protection Agency (EPA), in the next 50 y if current trends continue.
Ecosystem respiration Solomon, Christopher T.; Bruesewitz, Denise A.; Richardson, David C. ...
Limnology and oceanography,
20/May , Letnik:
58, Številka:
3
Journal Article
Recenzirano
Odprti dostop
We assembled data from a global network of automated lake observatories to test hypotheses regarding the drivers of ecosystem metabolism. We estimated daily rates of respiration and gross primary ...production (GPP) for up to a full year in each lake, via maximum likelihood fits of a free-water metabolism model to continuous high-frequency measurements of dissolved oxygen concentrations. Uncertainties were determined by a bootstrap analysis, allowing lake-days with poorly constrained rate estimates to be down-weighted in subsequent analyses. GPP and respiration varied considerably among lakes and at seasonal and daily timescales. Mean annual GPP and respiration ranged from 0.1 to 5.0 mg O2 L−1 d−1 and were positively related to total phosphorus but not dissolved organic carbon concentration. Within lakes, significant day-to-day differences in respiration were common despite large uncertainties in estimated rates on some lake-days. Daily variation in GPP explained 5% to 85% of the daily variation in respiration after temperature correction. Respiration was tightly coupled to GPP at a daily scale in oligotrophic and dystrophic lakes, and more weakly coupled in mesotrophic and eutrophic lakes. Background respiration ranged from 0.017 to 2.1 mg O₂ L−1 d−1 and was positively related to indicators of recalcitrant allochthonous and autochthonous organic matter loads, but was not clearly related to an indicator of the quality of allochthonous organic matter inputs.
High‐frequency physical observations from 40 temperate lakes were used to examine the relative contributions of wind shear (u*) and convection (w*) to turbulence in the surface mixed layer. Seasonal ...patterns of u* and w* were dissimilar; u* was often highest in the spring, while w*increased throughout the summer to a maximum in early fall. Convection was a larger mixed‐layer turbulence source than wind shear (u*/w* < 0.75) for 18 of the 40 lakes, including all 11 lakes <10 ha. As a consequence, the relative contribution of convection to the gas transfer velocity (k, estimated by the surface renewal model) was greater for small lakes. The average k was 0.54 m day−1 for lakes <10 ha. Because u* and w*differ in temporal pattern and magnitude across lakes, both convection and wind shear should be considered in future formulations of lake‐air gas exchange, especially for small lakes.
Key Points
Convection was consistently a larger turbulence source than wind for small lakes
Convection and wind shear have different diurnal seasonal and geographic pattern
Wind‐only methods for gas exchange are not appropriate for small lakes
Hypolimnetic oxygen depletion during summer stratification in lakes can lead to hypoxic and anoxic conditions. Hypolimnetic anoxia is a water quality issue with many consequences, including reduced ...habitat for cold-water fish species, reduced quality of drinking water, and increased nutrient and organic carbon (OC) release from sediments. Both allochthonous and autochthonous OC loads contribute to oxygen depletion by providing substrate for microbial respiration; however, their relative contributions to oxygen depletion across diverse lake systems remain uncertain. Lake characteristics, such as trophic state, hydrology, and morphometry, are also influential in carbon-cycling processes and may impact oxygen depletion dynamics. To investigate the effects of carbon cycling on hypolimnetic oxygen depletion, we used a two-layer process-based lake model to simulate daily metabolism dynamics for six Wisconsin lakes over 20 years (1995–2014). Physical processes and internal metabolic processes were included in the model and were used to predict dissolved oxygen (DO), particulate OC (POC), and dissolved OC (DOC). In our study of oligotrophic, mesotrophic, and eutrophic lakes, we found autochthony to be far more important than allochthony to hypolimnetic oxygen depletion. Autochthonous POC respiration in the water column contributed the most towards hypolimnetic oxygen depletion in the eutrophic study lakes. POC water column respiration and sediment respiration had similar contributions in the mesotrophic and oligotrophic study lakes. Differences in terms of source of respiration are discussed with consideration of lake productivity and the processing and fates of organic carbon loads.
Based on empirical and synthetic research, lakes make, store, and mineralize organic carbon (OC) at rates that are significant and relevant to regional and global carbon budgets. Although some ...global-scale studies have examined specific processes such as carbon burial and CO₂exchange with the atmosphere, most studies of lake carbon cycling are from single systems, focus only on a specific habitat, and do not account for all of the major terms in OC budgets. Hence, most lake OC budgets are incomplete, leaving some key processes highly uncertain. To advance the analysis of the role of the inland waters in C-cycling, ecosystem science needs a new generation of studies that confront these shortcomings. Here we address research needs and priorities for improving OC budgets. We present ten key research questions and recommend a framework for essential ecosystem-scale studies of lake OC cycling. Answers to these ten questions will not only improve carbon budgets but also provide robust estimates of lake contributions to global and regional carbon cycling. In addition, studies of lake carbon budgets will provide relative autochthonous and allochthonous carbon fluxes, indicate sources and rates of carbon burial, improve quantification of lake-atmosphere carbon exchanges, better integrate lakes with terrestrial and lotic carbon dynamics, promote understanding of how climate and land-use change will impact lakes, and enable tests of ecological theory related to subsidies and food web stability.
•Annual lake phosphorus dynamic explained mostly by recycling and sedimentation•Mass balance unable to fully reproduce lake phosphorus cycling•Process-guided machine learning outperforms neural ...network or process alone•Missing long-term trend in lake phosphorus likely due to changes in loading
Phosphorus (P) loading to lakes is degrading the quality and usability of water globally. Accurate predictions of lake P dynamics are needed to understand whole-ecosystem P budgets, as well as the consequences of changing lake P concentrations for water quality. However, complex biophysical processes within lakes, along with limited observational data, challenge our capacity to reproduce short-term lake dynamics needed for water quality predictions, as well as long-term dynamics needed to understand broad scale controls over lake P. Here we use an emerging paradigm in modeling, process-guided machine learning (PGML), to produce a phosphorus budget for Lake Mendota (Wisconsin, USA) and to accurately predict epilimnetic phosphorus over a time range of days to decades. In our implementation of PGML, which we term a Process-Guided Recurrent Neural Network (PGRNN), we combine a process-based model for lake P with a recurrent neural network, and then constrain the predictions with ecological principles. We test independently the process-based model, the recurrent neural network, and the PGRNN to evaluate the overall approach. The process-based model accounted for most of the observed pattern in lake P; however it missed the long-term trend in lake P and had the worst performance in predicting winter and summer P in surface waters. The root mean square error (RMSE) for the process-based model, the recurrent neural network, and the PGRNN was 33.0 μg P L−1, 22.7 μg P L−1, and 20.7 μg P L−1, respectively. All models performed better during summer, with RMSE values for the three models (same order) equal to 14.3 μg P L−1, 10.9 μg P L−1, and 10.7 μg P L−1. Although the PGRNN had only marginally better RMSE during summer, it had lower bias and reproduced long-term decreases in lake P missed by the other two models. For all seasons and all years, the recurrent neural network had better predictions than process alone, with root mean square error (RMSE) of 23.8 μg P L−1 and 28.0 μg P L−1, respectively. The output of PGRNN indicated that new processes related to water temperature, thermal stratification, and long term changes in external loads are needed to improve the process model. By using ecological knowledge, as well as the information content of complex data, PGML shows promise as a technique for accurate prediction in messy, real-world ecological dynamics, while providing valuable information that can improve our understanding of process.