Utilizing Machine Learning for Constructing Probabilistic Subseasonal Precipitation Forecasts in California

Amid prolonged droughts in California from 2012 to 2016 that strained the state's water resources, severe dry and warm conditions resurfaced in 2019. The years 2020 and 2021 stood out as the fifth and second driest in a century. Given the challenges posed by these extreme climate events, precise Sub-Seasonal (S2S) precipitation prediction gains significance for disaster readiness, mitigation, drought management, and water use policies. While S2S predictions show limited skill beyond two weeks, the last decade has witnessed substantial research to enhance predictive abilities. Advanced statistical and Artificial Intelligence/Machine Learning (AI/ML) methods, either post-processing dynamical model output or improving it, have been a focus.

However, the sheer volume of S2S data renders traditional ML approaches computationally demanding, often inaccessible without institutional computing resources. To tackle this challenge, an alternative emerges in the form of Extreme Learning Machine (ELM), a quicker substitute for neural network-based ML forecasting. ELM, utilizing random initialization and the generalized Moore-Penrose inverse for the output layer, provides a rapid solution, unlike traditional backpropagation weight adjustments.

Nevertheless, ELM's deterministic results necessitate adjustment to suit probabilistic situations. This brings us to Probabilistic Output Extreme Learning Machine (PO-ELM), an enhanced version of ELM that employs sigmoid additive neurons and distinct linear programming for probabilistic predictions. To refine PO-ELM for probabilistic sub-seasonal forecasting, we introduce Extended Probabilistic Output Extreme Learning Machine (EPO-ELM), which generates comprehensive probability distributions. Implementing this, we leverage XCast, a Python library developed by the authors, to utilize EPO-ELM for probabilistic precipitation forecasts based on ECMWF's S2S predictions for December-January-February in California.