Few studies have utilized machine learning techniques to predict or understand the Madden-Julian oscillation (MJO), a key source of subseasonal variability and predictability. Here we present a simple framework for real-time MJO prediction using shallow artificial neural networks (ANNs). We construct two ANN architectures, one deterministic and one probabilistic, that predict a real-time MJO index using maps of tropical variables. These ANNs make skillful MJO predictions out to ~17 days in October-March and ~10 days in April-September, outperforming conventional linear models and efficiently capturing aspects of MJO predictability found in more complex, dynamical models. The flexibility and explainability of simple ANN frameworks is highlighted through varying model input and applying ANN explainability techniques that reveal sources and regions important for ANN prediction skill. The accessibility, performance, and efficiency of this simple machine learning framework is more broadly applicable to predict and understand other Earth system phenomena.

Recent work using CMIP5 models under RCP8.5 suggests that individual multimodel-mean changes in precipitation and wind variability associated with the Madden-Julian oscillation (MJO) are not detectable until the end of the 21st century. However, a decrease in the ratio of MJO circulation to precipitation anomaly amplitude is detectable as early as 2021-2040, consistent with an increase in dry static stability as predicted by weak-temperature-gradient balance. Here, we examine MJO activity in multiple reanalyses (ERA5,MERRA-2, and ERA-20C) and find that MJO wind and precipitation anomaly amplitudes have a complicated time evolution over the record. However, a decrease in the ratio of MJO circulation to precipitation anomaly amplitude is detected over the observational period, consistent with the change in dry static stability. These results suggest that weak-temperature-gradient theory may be able to help explain changes in MJO activity in recent decades.

Tropical islands are highly dependent on rainfall to provide resources for drinking and agriculture. Hence, understanding changes to precipitation under a changing climate is critical for societal planning. Decadal variability in the climate system causes the strength of SST gradients to vary across the tropical Pacific that can cause precipitation patterns to transition from one decade to the next, even in the presence of longer term climate trends. To study 21st Century changes to tropical rainfall patterns in the presence of decadal variability, we use the CESM1 Large Ensemble (Kay et al. 2015) forced under RCP8.5. Each ensemble member uses different initial conditions that can be used to examine climate projections on short term (e.g. weeks, years) through long term (e.g. century) time scales. Since climate models contain climate variability such as El Niño Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO), each ensemble member can have diverse projection outcomes in a given decade. While all ensemble members eventually show an El Niño-like warming pattern by 2100 relative to 1985-2005, before the mid 21st Century, preferential SST warming and precipitation intensity in the tropics in any given 20-year period can be weighted toward the west, central, or east Pacific. Further, while Niño3.4 SST generally goes up relative to 1985-2005, the tropical Pacific east-west temperature gradient change does not show as consistent an upward trend. Implications of SST and precipitation change patterns for Guam, Samoa, Hawaii, and Puerto Rico are examined. Spearman’s correlation is used to examine the relationship between station island precipitation and the east-west Pacific SSTs gradient change. A strong negative correlation relative to gradient change is found for Guam, in contrast to Samoa having a high positive correlation. This study highlights the importance of decadal climate variability for understanding changes in water resources in island nations in a changing climate. This study was conducted as part of the Earth System Modeling and Education Institute summer REU program at Colorado State University.

Boreal-wintertime hindcasts in the Unified Forecast System with the tropics nudged toward reanalysis improve United States (U.S.) West Coast precipitation forecasts at Weeks 3-4 lead times when compared to those without nudging. To diagnose the origin of these improvements, a multivariate k-means clustering method is used to group hindcasts into subsets by their initial conditions. One cluster characterized by an initially strong Aleutian Low demonstrates larger improvements at Weeks 3-4 with nudging compared to others. The greater improvements with nudging for this cluster are related to the model error in simulating the interaction between the Aleutian Low and the teleconnection patterns associated with the Madden-Julian oscillation (MJO) and El Niño-Southern Oscillation (ENSO). Improving forecasts of tropical intraseasonal precipitation, especially during early MJO phases under non-cold ENSO, may be important for producing better Weeks 3-4 precipitation forecasts for the U.S. West Coast.

Changes to the eastern North Pacific tropical intraseasonal oscillation (ISO) at the end of the 21st Century and implications for tropical cyclone (TC) genesis are examined in the Shared Socioeconomic Pathways (SSP585) scenario of the Coupled Model Intercomparison Project phase 6 (CMIP6) data set. Multimodel mean composite low-level wind and precipitation anomalies associated with the leading intraseasonal mode indicate that precipitation amplitude increases while wind amplitude weakens under global warming, consistent with previous studies for the Indo-Pacific warm pool. The eastern North Pacific intraseasonal precipitation/wind pattern also tends to shift southwestward in a warmer climate, associated with weaker positive precipitation anomalies near the coast of Mexico and Central America during the enhanced convection/westerly wind phase. Implications for the modulation of TC genesis by the leading intraseasonal mode are then explored using an empirical genesis potential index (GPI). In the historical simulation, GPI shows positive anomalies in the eastern North Pacific in the convectively enhanced phase of the ISO. The ISO’s modulation of GPI weakens near the coast of Mexico and Central America with warming, associated with a southward shift of GPI anomalies. Further examination of the contribution from individual environmental variables that enter the GPI shows that relative humidity and vorticity changes during ISO events weaken positive GPI anomalies near the Mexican coast with warming and make genesis more favorable to the southwest. The impact of vertical shear anomaly changes is also to favor genesis away from the coast. These results suggest a weaker modulation of TCs near the Mexican Coast by the ISO in a warmer climate.

Due to the coupled nature of the earth system, precipitation forecast errors at S2S lead times are caused by a combination of errors/biases from the atmosphere, ocean, ice and land across a range of spatial and temporal scales. We show that UFS precipitation errors over the U.S. at Weeks 3-4 can be directly related to biases in simulating tropical dynamics. In particular, the inability of the UFS to realistically simulate the Madden-Julian oscillation (MJO) leads to biases in the teleconnection to North America that produces these errors. When the tropics are nudged to produce an accurate representation of the MJO and other tropical disturbances, U.S. West Coast precipitation biases are substantially reduced. A clustering analysis is used to show that the greatest forecast improvements with nudging occur during warm ENSO events when MJO convection is in the Indian Ocean and about to move into the Maritime Continent. Physical mechanisms that explain the improvement in tropical-extratropical teleconnections during certain MJO and ENSO states will be discussed. We will also present future plans to combine state-of-the-art developments in machine learning with process-based diagnostics of the tropical moisture and moist static energy (MSE) budgets to understand and correct precipitation biases in coupled UFS hindcasts. In particular, we will discuss how model biases and errors in tropical variability (e.g. MJO) and associated teleconnections to midlatitudes lead to errors in U.S. precipitation on S2S timescales, and present methods to reduce these errors via post-processing on a forecast-by-forecast basis.

The subseasonal Madden-Julian oscillation (MJO) is among the most important modes of tropical variability on the planet and is a dominant driver of subseasonal-to-seasonal (S2S) prediction globally. The past decade has seen substantial advances in MJO prediction using dynamical forecast models, which now routinely outperform traditional statistical MJO forecasts (e.g. multiple linear regression models). At the same time, an increasing body of literature has demonstrated that machine-learning methods represent a new frontier in Earth science, opening the door to more advanced statistical forecast models of the MJO. In this study, we explore whether state-of-the-art machine learning methods can be used to make real-time MJO forecasts that outperform traditional statistical models and do comparably well to dynamical models. In particular, we utilize neural networks trained on observational tropical fields to attempt to make skillful forecasts of MJO convection out to several weeks lead time. Through contrasting the machine-learning models’ behavior with simpler statistical models and dynamical forecast models, we explore the advantages and disadvantages of statistical versus dynamical forecasts. A novel aspect of our analysis is the use of cutting-edge techniques to allow us to visualize how our neural network models makes their predictions. These techniques, such as layer-wise relevance propagation, can lead to new insights into regions of MJO predictability, allowing us to better interpret sources of MJO prediction skill within the machine-learning model. We further diagnose whether our machine-learning models contain well-known aspects of MJO prediction found in dynamical models, such as an increase in prediction skill during boreal winter or during certain phases of the stratospheric quasi-biennial oscillation.