Graph-Guided Regularized Regression of Pacific Ocean Climate Variables to Increase Predictive Skill of Southwestern U.S. Winter Precipitation

Abstract

Understanding the physical drivers of seasonal hydroclimatic variability and improving predictive skill remains a challenge with important socioeconomic and environmental implications for many regions around the world. Physics-based deterministic models show limited ability to predict precipitation as the lead time increases, due to imperfect representation of physical processes and incomplete knowledge of initial conditions. Similarly, statistical methods drawing upon established climate teleconnections have low prediction skill due to the complex nature of the climate system. Recently, promising data-driven approaches have been proposed, but they often suffer from over-parameterization and overfitting due to the short observational record, and they often do not account for spatiotemporal dependencies among covariates (i.e., predictors such as sea surface temperatures). This study addresses these challenges via a predictive model based on a graph-guided regularizer that simultaneously promotes similarity of predictive weights for highly correlated covariates and enforces sparsity in the covariate domain. This approach both decreases the effective dimensionality of the problem and identifies the most predictive features without specifying them a-priori. We use large ensemble simulations from a climate model to construct this regularizer, reducing the structural uncertainty in the estimation. We apply the learned model to predict winter precipitation in the southwestern US using sea surface temperatures over the entire Pacific basin, and demonstrate its superiority compared to other regularization approaches and statistical models informed by known teleconnections. Our results highlight the potential to combine optimally the space-time structure of predictor variables learned from climate models with new graph-based regularizers to improve seasonal prediction.

Hotspots of Predictability: Identifying Regions of High Precipitation Predictability at Seasonal Timescales From Limited Time Series Observations

“…The identified hotspots as well as sources of predictability align reasonably well with previous reports. Specifically, many past studies have identified statistically significant skill in prediction of precipitation totals for the Southwest (Gibson et al., 2021 ; Liu et al., 2018 ; Madadgar et al., 2016 ; Mamalakis et al., 2018 ; McCabe & Dettinger, 1999 ; Pan et al., 2019 ; Schonher & Nicholson, 1989 ; Stevens et al., 2021 ; Zhang et al., 2018 among many others), for Gulf Coast and Southeast regions (Becker et al., 2014 ; Devineni & Sankarasubramanian, 2010a , 2010b ; Kirtman et al., 2014 ), and for precipitation extremes in the Northwest (Gershunov & Cayan, 2003 ; Zarekarizi et al., 2018 ). Moreover, previous studies have highlighted the Pacific and North Atlantic Ocean basins as the main sources of precipitation predictability over North America (see e.g., Dai, 2013 ; Enfield et al., 2001 ; McCabe et al., 2004 ; Newman et al., 2016 ).…”

“…(d) What are the sources of predictability, that is, the large‐scale climate modes that contribute to predictability? Although prediction of precipitation across different CONUS regions has been explored in the past (Gibson et al., 2021 ; Liu et al., 2018 ; Madadgar et al., 2016 ; Mamalakis et al., 2018 ; McCabe & Dettinger, 1999 ; Pan et al., 2019 ; Schonher & Nicholson, 1989 ; Stevens et al., 2021 ; Zarekarizi et al., 2018 ; Zhang et al., 2018 ; Zimmerman et al., 2016 ), fewer studies have approached the problem focusing on the entire CONUS (Devineni & Sankarasubramanian, 2010a , 2010b ; Gershunov & Cayan, 2003 ; Schubert et al., 2016 ), and to the best of our knowledge, our study is the first to explore hotspots of predictability and to systematically address the four questions described above. Results from our analysis may improve water resources management at a federal level, by integrating the knowledge of how predictability is distributed spatially across the US for both dry and wet precipitation extremes.…”

“…In our approach, we assess seasonal predictability by recognizing that one of the key problems in this setting is the limited sample size of high‐quality time series observations. The small sample size cautions against overinterpreting predictability when prediction uncertainty has not been quantitatively assessed and calls for also addressing the high risk of overfitting (DelSole & Banerjee, 2017 ; Ham et al., 2019 ; Stevens et al., 2021 ). So far, no generalized framework of how to assess predictability under this setting exists in the literature.…”

Investigating Predictability of the TRHR Seasonal Precipitation at Long Lead Times Using a Generalized Regression Model with Regularization

“…Yet, it should be noted that the ridge regression does not directly provoke sparsity of the regression model while the LASSO regression tends to assign non-zero value to only one of many correlated predictors which can make the model difficult to interpret. The lack of interpretability of the lasso model is also pointed out in a recent paper from Stevens et al (2021) where a graph-guided variation is used as an extra regularization to improve robustness of the regression model in predicting Southwestern United States winter precipitation. Here, we propose to use the elastic net regularization (Zou and Hastie, 2005) which linearly combines the LASSO and ridge regression regularizations.…”

Learning and Dynamical Models for Sub-seasonal Climate Forecasting: Comparison and Collaboration

“…Such applications aim to advance weather forecast skill using deep learning methods [11,18,27,39]. Despite early studies that show dynamical models outperform statistical models for ENSO seasonal forecasts [2], recent advances in machine learning, especially the development of deep learning, are making the performance of ML models more competitive with dynamical models for both weather [11,17,41] and seasonal [44] prediction.…”