An official website of the United States government
Abstract. Deep learning (DL) rainfall–runoff models outperform conceptual, process-based models in a range of applications. However, it remains unclear whether DL models can produce physically plausible projections of streamflow under climate change. We investigate this question through a sensitivity analysis of modeled responses to increases in temperature and potential evapotranspiration (PET), with other meteorological variables left unchanged. Previous research has shown that temperature-based PET methods overestimate evaporative water loss under warming compared with energy budget-based PET methods. We therefore assume that reliable streamflow responses to warming should exhibit less evaporative water loss when forced with smaller, energy-budget-based PET compared with temperature-based PET. We conduct this assessment using three conceptual, process-based rainfall–runoff models and three DL models, trained and tested across 212 watersheds in the Great Lakes basin. The DL models include a Long Short-Term Memory network (LSTM), a mass-conserving LSTM (MC-LSTM), and a novel variant of the MC-LSTM that also respects the relationship between PET and evaporative water loss (MC-LSTM-PET). After validating models against historical streamflow and actual evapotranspiration, we force all models with scenarios of warming, historical precipitation, and both temperature-based (Hamon) and energy-budget-based (Priestley–Taylor) PET, and compare their responses in long-term mean daily flow, low flows, high flows, and seasonal streamflow timing. We also explore similar responses using a national LSTM fit to 531 watersheds across the United States to assess how the inclusion of a larger and more diverse set of basins influences signals of hydrological response under warming. The main results of this study are as follows: The three Great Lakes DL models substantially outperform all process-based models in streamflow estimation. The MC-LSTM-PET also matches the best process-based models and outperforms the MC-LSTM in estimating actual evapotranspiration. All process-based models show a downward shift in long-term mean daily flows under warming, but median shifts are considerably larger under temperature-based PET (−17 % to −25 %) than energy-budget-based PET (−6 % to −9 %). The MC-LSTM-PET model exhibits similar differences in water loss across the different PET forcings. Conversely, the LSTM exhibits unrealistically large water losses under warming using Priestley–Taylor PET (−20 %), while the MC-LSTM is relatively insensitive to the PET method. DL models exhibit smaller changes in high flows and seasonal timing of flows as compared with the process-based models, while DL estimates of low flows are within the range estimated by the process-based models. Like the Great Lakes LSTM, the national LSTM also shows unrealistically large water losses under warming (−25 %), but it is more stable when many inputs are changed under warming and better aligns with process-based model responses for seasonal timing of flows. Ultimately, the results of this sensitivity analysis suggest that physical considerations regarding model architecture and input variables may be necessary to promote the physical realism of deep-learning-based hydrological projections under climate change.
more »
« less
Modular Compositional Learning Improves 1D Hydrodynamic Lake Model Performance by Merging Process‐Based Modeling With Deep Learning
Abstract Hybrid Knowledge‐Guided Machine Learning (KGML) models, which are deep learning models that utilize scientific theory and process‐based model simulations, have shown improved performance over their process‐based counterparts for the simulation of water temperature and hydrodynamics. We highlight the modular compositional learning (MCL) methodology as a novel design choice for the development of hybrid KGML models in which the model is decomposed into modular sub‐components that can be process‐based models and/or deep learning models. We develop a hybrid MCL model that integrates a deep learning model into a modularized, process‐based model. To achieve this, we first train individual deep learning models with the output of the process‐based models. In a second step, we fine‐tune one deep learning model with observed field data. In this study, we replaced process‐based calculations of vertical diffusive transport with deep learning. Finally, this fine‐tuned deep learning model is integrated into the process‐based model, creating the hybrid MCL model with improved overall projections for water temperature dynamics compared to the original process‐based model. We further compare the performance of the hybrid MCL model with the process‐based model and two alternative deep learning models and highlight how the hybrid MCL model has the best performance for projecting water temperature, Schmidt stability, buoyancy frequency, and depths of different isotherms. Modular compositional learning can be applied to existing modularized, process‐based model structures to make the projections more robust and improve model performance by letting deep learning estimate uncertain process calculations.
more »
« less
- PAR ID:
- 10483828
- Publisher / Repository:
- DOI PREFIX: 10.1029
- Date Published:
- Journal Name:
- Journal of Advances in Modeling Earth Systems
- Volume:
- 16
- Issue:
- 1
- ISSN:
- 1942-2466
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
This study developed a hybrid model for predicting dissolved oxygen (DO) using real-time sensor data for thirteen parameters. This novel hybrid model integrated one-dimensional convolutional neural networks (CNN) and long short-term memory (LSTM) to improve the accuracy of prediction for DO in water. The hybrid CNNLSTM model predicted DO concentration in water using soft sensor data. The primary input parameters to the model were temperature, pH, specific conductivity, salinity, density, chlorophyll, and blue-green algae. The model used 38,681 water quality data for training and testing the hybrid deep learning network. The training procedure for the model was successful. The training and test losses were both nearly zero and within a similar range. With a coefficient of determination (R2) of 0.94 and a mean squared error (MSE) of 0.12, the hybrid model indicated higher performance compared to the classical models. The normal distribution of residual errors confirmed the reliability of the DO predictions by the hybrid CNN-LSTM model. Feature importance analysis indicated pH as the most significant predictor and temperature as the second important predictor. The feature importance scores based on extreme gradient boosting (XGBoost) for the pH and temperature were 0.76 and 0.12, respectively. This study indicated that the hybrid model can outperform the classical machine learning models in the real-time prediction of DO concentration.more » « less
-
Abstract Backpropagation is widely used to train artificial neural networks, but its relationship to synaptic plasticity in the brain is unknown. Some biological models of backpropagation rely on feedback projections that are symmetric with feedforward connections, but experiments do not corroborate the existence of such symmetric backward connectivity. Random feedback alignment offers an alternative model in which errors are propagated backward through fixed, random backward connections. This approach successfully trains shallow models, but learns slowly and does not perform well with deeper models or online learning. In this study, we develop a meta-learning approach to discover interpretable, biologically plausible plasticity rules that improve online learning performance with fixed random feedback connections. The resulting plasticity rules show improved online training of deep models in the low data regime. Our results highlight the potential of meta-learning to discover effective, interpretable learning rules satisfying biological constraints.more » « less
-
Fitch, T.; Lamm, C.; Leder, H.; Teßmar-Raible, K. (Ed.)Is analogical reasoning a task that must be learned to solve from scratch by applying deep learning models to massive numbers of reasoning problems? Or are analogies solved by computing similarities between structured representations of analogs? We address this question by comparing human performance on visual analogies created using images of familiar three-dimensional objects (cars and their subregions) with the performance of alternative computational models. Human reasoners achieved above-chance accuracy for all problem types, but made more errors in several conditions (e.g., when relevant subregions were occluded). We compared human performance to that of two recent deep learning models (Siamese Network and Relation Network) directly trained to solve these analogy problems, as well as to that of a compositional model that assesses relational similarity between part-based representations. The compositional model based on part representations, but not the deep learning models, generated qualitative performance similar to that of human reasoners.more » « less
-
Abstract In many regions globally, snowmelt‐recharged mountainous karst aquifers serve as crucial sources for municipal and agricultural water supplies. In these watersheds, complex interplay of meteorological, topographical, and hydrogeological factors leads to intricate recharge‐discharge pathways. This study introduces a spatially distributed deep learning precipitation‐runoff model that combines Convolutional Long Short‐Term Memory (ConvLSTM) with a spatial attention mechanism. The effectiveness of the deep learning model was evaluated using data from the Logan River watershed and subwatersheds, a characteristically karst‐dominated hydrological system in northern Utah. Compared to the ConvLSTM baseline, the inclusion of a spatial attention mechanism improved performance for simulating discharge at the watershed outlet. Analysis of attention weights in the trained model unveiled distinct areas contributing the most to discharge under snowmelt and recession conditions. Furthermore, fine‐tuning the model at subwatershed scales provided insights into cross‐subwatershed subsurface connectivity. These findings align with results obtained from detailed hydrogeochemical tracer studies. Results highlight the potential of the proposed deep learning approach to unravel the complexities of karst aquifer systems, offering valuable insights for water resource management under future climate conditions. Furthermore, results suggest that the proposed explainable, spatially distributed, deep learning approach to hydrologic modeling holds promise for non‐karstic watersheds.more » « less
