Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions
Abstract
Abstract Recently, recurrent deep networks have shown promise to harness newly available satellite‐sensed data for long‐term soil moisture projections. However, to be useful in forecasting, deep networks must also provide uncertainty estimates. Here we evaluated Monte Carlo dropout with an input‐dependent data noise term (MCD+N), an efficient uncertainty estimation framework originally developed in computer vision, for hydrologic time series predictions. MCD+N simultaneously estimates a heteroscedastic input‐dependent data noise term (a trained error model attributable to observational noise) and a network weight uncertainty term (attributable to insufficiently constrained model parameters). Although MCD+N has appealing features, many heuristic approximations were employed during its derivation, and rigorous evaluations and evidence of its asserted capability to detect dissimilarity were lacking. To address this, we provided an in‐depth evaluation of the scheme's potential and limitations. We showed that for reproducing soil moisture dynamics recorded by the Soil Moisture Active Passive (SMAP) mission, MCD+N indeed gave a good estimate of predictive error, provided that we tuned a hyperparameter and used a representative training data set. The input‐dependent term responded strongly to observational noise, while the model term clearly acted as a detector for physiographic dissimilarity from the training data, behaving as intended. However, when the trainingmore »
- Authors:
-
- Pennsylvania State Univ., University Park, PA (United States); Stanford Univ., CA (United States)
- Pennsylvania State Univ., University Park, PA (United States)
- Publication Date:
- Research Org.:
- Pennsylvania State Univ., University Park, PA (United States)
- Sponsoring Org.:
- USDOE Advanced Research Projects Agency - Energy (ARPA-E); National Science Foundation (NSF)
- OSTI Identifier:
- 1755316
- Alternate Identifier(s):
- OSTI ID: 1786917
- Grant/Contract Number:
- SC0016605; EAR #1832294
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Water Resources Research
- Additional Journal Information:
- Journal Volume: 56; Journal Issue: 12; Journal ID: ISSN 0043-1397
- Publisher:
- American Geophysical Union (AGU)
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 54 ENVIRONMENTAL SCIENCES; Soil moisture; Monte Carlo dropout; LSTM; uncertainty; deep learning; Bayesian inference
Citation Formats
Fang, Kuai, Kifer, Daniel, Lawson, Kathryn, and Shen, Chaopeng. Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions. United States: N. p., 2020.
Web. doi:10.1029/2020wr028095.
Fang, Kuai, Kifer, Daniel, Lawson, Kathryn, & Shen, Chaopeng. Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions. United States. https://doi.org/10.1029/2020wr028095
Fang, Kuai, Kifer, Daniel, Lawson, Kathryn, and Shen, Chaopeng. Mon .
"Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions". United States. https://doi.org/10.1029/2020wr028095. https://www.osti.gov/servlets/purl/1755316.
@article{osti_1755316,
title = {Evaluating the Potential and Challenges of an Uncertainty Quantification Method for Long Short–Term Memory Models for Soil Moisture Predictions},
author = {Fang, Kuai and Kifer, Daniel and Lawson, Kathryn and Shen, Chaopeng},
abstractNote = {Abstract Recently, recurrent deep networks have shown promise to harness newly available satellite‐sensed data for long‐term soil moisture projections. However, to be useful in forecasting, deep networks must also provide uncertainty estimates. Here we evaluated Monte Carlo dropout with an input‐dependent data noise term (MCD+N), an efficient uncertainty estimation framework originally developed in computer vision, for hydrologic time series predictions. MCD+N simultaneously estimates a heteroscedastic input‐dependent data noise term (a trained error model attributable to observational noise) and a network weight uncertainty term (attributable to insufficiently constrained model parameters). Although MCD+N has appealing features, many heuristic approximations were employed during its derivation, and rigorous evaluations and evidence of its asserted capability to detect dissimilarity were lacking. To address this, we provided an in‐depth evaluation of the scheme's potential and limitations. We showed that for reproducing soil moisture dynamics recorded by the Soil Moisture Active Passive (SMAP) mission, MCD+N indeed gave a good estimate of predictive error, provided that we tuned a hyperparameter and used a representative training data set. The input‐dependent term responded strongly to observational noise, while the model term clearly acted as a detector for physiographic dissimilarity from the training data, behaving as intended. However, when the training and test data were characteristically different, the input‐dependent term could be misled, undermining its reliability. Additionally, due to the data‐driven nature of the model, data noise also influences network weight uncertainty, and therefore the two uncertainty terms are correlated. Overall, this approach has promise, but care is needed to interpret the results.},
doi = {10.1029/2020wr028095},
journal = {Water Resources Research},
number = 12,
volume = 56,
place = {United States},
year = {Mon Nov 09 00:00:00 EST 2020},
month = {Mon Nov 09 00:00:00 EST 2020}
}
Works referenced in this record:
Ensemble Kalman filter data assimilation for a process-based catchment scale model of surface and subsurface flow: EnKF FOR A MODEL OF SURFACE AND SUBSURFACE FLOW
journal, October 2009
- Camporese, Matteo; Paniconi, Claudio; Putti, Mario
- Water Resources Research, Vol. 45, Issue 10
Deep Classifiers from Image Tags in the Wild
conference, January 2015
- Izadinia, Hamid; Russell, Bryan C.; Farhadi, Ali
- Proceedings of the 2015 Workshop on Community-Organized Multimodal Mining: Opportunities for Novel Solutions - MMCommons'15
Pitfalls and improvements in the joint inference of heteroscedasticity and autocorrelation in hydrological model calibration: Technical note
journal, July 2013
- Evin, Guillaume; Kavetski, Dmitri; Thyer, Mark
- Water Resources Research, Vol. 49, Issue 7
Changing ideas in hydrology — The case of physically-based models
journal, January 1989
- Beven, Keith
- Journal of Hydrology, Vol. 105, Issue 1-2
Prolongation of SMAP to Spatiotemporally Seamless Coverage of Continental U.S. Using a Deep Learning Neural Network
journal, November 2017
- Fang, Kuai; Shen, Chaopeng; Kifer, Daniel
- Geophysical Research Letters, Vol. 44, Issue 21
Deep learning
journal, May 2015
- LeCun, Yann; Bengio, Yoshua; Hinton, Geoffrey
- Nature, Vol. 521, Issue 7553
Quantifying Uncertainty in Discrete-Continuous and Skewed Data with Bayesian Deep Learning
conference, July 2018
- Vandal, Thomas; Kodra, Evan; Dy, Jennifer
- KDD '18: The 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining
Developing a Spatial Framework of Common Ecological Regions for the Conterminous United States
journal, April 2001
- McMAHON, Gerard; Gregonis, Steven M.; Waltman, Sharon W.
- Environmental Management, Vol. 28, Issue 3
A Low Rank Weighted Graph Convolutional Approach to Weather Prediction
conference, November 2018
- Wilson, Tyler; Tan, Pang-Ning; Luo, Lifeng
- 2018 IEEE International Conference on Data Mining (ICDM)
Treatment of input uncertainty in hydrologic modeling: Doing hydrology backward with Markov chain Monte Carlo simulation: FORCING DATA ERROR USING MCMC SAMPLING
journal, December 2008
- Vrugt, Jasper A.; ter Braak, Cajo J. F.; Clark, Martyn P.
- Water Resources Research, Vol. 44, Issue 12
Robust Climate Policies Under Uncertainty: A Comparison of Robust Decision Making and Info-Gap Methods
journal, April 2012
- Hall, Jim W.; Lempert, Robert J.; Keller, Klaus
- Risk Analysis, Vol. 32, Issue 10
Approximation capabilities of multilayer feedforward networks
journal, January 1991
- Hornik, Kurt
- Neural Networks, Vol. 4, Issue 2
Theory-Guided Data Science: A New Paradigm for Scientific Discovery from Data
journal, October 2017
- Karpatne, Anuj; Atluri, Gowtham; Faghmous, James H.
- IEEE Transactions on Knowledge and Data Engineering, Vol. 29, Issue 10
Speech recognition with deep recurrent neural networks
conference, May 2013
- Graves, Alex; Mohamed, Abdel-rahman; Hinton, Geoffrey
- ICASSP 2013 - 2013 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Structuring and evaluating decision support processes to enhance the robustness of complex human–natural systems
journal, January 2020
- Moallemi, Enayat A.; Zare, Fateme; Reed, Patrick M.
- Environmental Modelling & Software, Vol. 123
Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks
journal, January 2018
- Kratzert, Frederik; Klotz, Daniel; Brenner, Claire
- Hydrology and Earth System Sciences, Vol. 22, Issue 11
HESS Opinions: Incubating deep-learning-powered hydrologic science advances as a community
journal, January 2018
- Shen, Chaopeng; Laloy, Eric; Elshorbagy, Amin
- Hydrology and Earth System Sciences, Vol. 22, Issue 11
Learning Deep Architectures for AI
journal, January 2009
- Bengio, Y.
- Foundations and Trends® in Machine Learning, Vol. 2, Issue 1
Bayesian analysis of input uncertainty in hydrological modeling: 2. Application: INPUT UNCERTAINTY IN HYDROLOGY, 2
journal, March 2006
- Kavetski, Dmitri; Kuczera, George; Franks, Stewart W.
- Water Resources Research, Vol. 42, Issue 3
Deep learning for healthcare: review, opportunities and challenges
journal, May 2017
- Miotto, Riccardo; Wang, Fei; Wang, Shuang
- Briefings in Bioinformatics, Vol. 19, Issue 6
Use long short-term memory to enhance Internet of Things for combined sewer overflow monitoring
journal, January 2018
- Zhang, Duo; Lindholm, Geir; Ratnaweera, Harsha
- Journal of Hydrology, Vol. 556
Detecting Arbitrary Oriented Text in the Wild with a Visual Attention Model
conference, January 2016
- Huang, Wenyi; He, Dafang; Yang, Xiao
- Proceedings of the 2016 ACM on Multimedia Conference - MM '16
Developing a Long Short-Term Memory (LSTM) based model for predicting water table depth in agricultural areas
journal, June 2018
- Zhang, Jianfeng; Zhu, Yan; Zhang, Xiaoping
- Journal of Hydrology, Vol. 561
An initial assessment of SMAP soil moisture retrievals using high-resolution model simulations and in situ observations: SMAP Comparisons
journal, September 2016
- Pan, Ming; Cai, Xitian; Chaney, Nathaniel W.
- Geophysical Research Letters, Vol. 43, Issue 18
Generic error model for calibration and uncertainty estimation of hydrological models: GENERIC ERROR MODEL
journal, November 2008
- Götzinger, Jens; Bárdossy, András
- Water Resources Research, Vol. 44, Issue 12
Gaussian Processes for Machine Learning
book, January 2005
- Rasmussen, Carl Edward; Williams, Christopher K. I.
- The MIT Press
Global Data Sets of Vegetation Leaf Area Index (LAI)3g and Fraction of Photosynthetically Active Radiation (FPAR)3g Derived from Global Inventory Modeling and Mapping Studies (GIMMS) Normalized Difference Vegetation Index (NDVI3g) for the Period 1981 to 2011
journal, February 2013
- Zhu, Zaichun; Bi, Jian; Pan, Yaozhong
- Remote Sensing, Vol. 5, Issue 2
Comparison of NLDAS-2 Simulated and NASMD Observed Daily Soil Moisture. Part I: Comparison and Analysis
journal, October 2015
- Xia, Youlong; Ek, Michael B.; Wu, Yihua
- Journal of Hydrometeorology, Vol. 16, Issue 5
Sustainable water resource management under hydrological uncertainty: WATER RESOURCES AND HYDROLOGICAL UNCERTAINTY
journal, November 2008
- Ajami, Newsha K.; Hornberger, George M.; Sunding, David L.
- Water Resources Research, Vol. 44, Issue 11
Long Short-Term Memory
journal, November 1997
- Hochreiter, Sepp; Schmidhuber, Jürgen
- Neural Computation, Vol. 9, Issue 8
Deep learning in neural networks: An overview
journal, January 2015
- Schmidhuber, Jürgen
- Neural Networks, Vol. 61
Modeling residual hydrologic errors with Bayesian inference
journal, September 2015
- Smith, Tyler; Marshall, Lucy; Sharma, Ashish
- Journal of Hydrology, Vol. 528
Ignorance is bliss: Or seven reasons not to use uncertainty analysis: OPINION
journal, May 2006
- Pappenberger, F.; Beven, K. J.
- Water Resources Research, Vol. 42, Issue 5
An evaluation of the impact of model structure on hydrological modelling uncertainty for streamflow simulation
journal, October 2004
- Butts, Michael B.; Payne, Jeffrey T.; Kristensen, Michael
- Journal of Hydrology, Vol. 298, Issue 1-4
Brain wave classification using long short-term memory network based OPTICAL predictor
journal, June 2019
- Kumar, Shiu; Sharma, Alok; Tsunoda, Tatsuhiko
- Scientific Reports, Vol. 9, Issue 1
Deep Convolutional Encoder‐Decoder Networks for Uncertainty Quantification of Dynamic Multiphase Flow in Heterogeneous Media
journal, January 2019
- Mo, Shaoxing; Zhu, Yinhao; Zabaras, Nicholas
- Water Resources Research, Vol. 55, Issue 1
Enhancing Streamflow Forecast and Extracting Insights Using Long‐Short Term Memory Networks With Data Integration at Continental Scales
journal, September 2020
- Feng, Dapeng; Fang, Kuai; Shen, Chaopeng
- Water Resources Research, Vol. 56, Issue 9