Mediaboss Marketing

What is the problem?

Accurately predicting the intensity of tropical cyclones, defined as the maximum sustained windspeed over a period of time, is a critical yet challenging task. Rapid intensification (RI) events are still a daunting problem for operational intensity forecasting.

Better forecasts and simulation of tropical cyclone (TC) intensities and their track can significantly improve the quality of Moodys RMS tropical cyclone modeling suite. RMS has helped clients manage their risk during TC events in the North Atlantic for almost 20 years. Real time TCs can significantly impact a companys financial, operational, and overall solvency state. Moodys RMS Hwind product helps (re)insurers, brokers, and capital markets understand the range of potential losses across multiple forecast scenarios, capturing the uncertainty in of how track and intensity will evolve.

With the advances in Numerical Weather Prediction (NWP) and new meteorological observations, forecasts of TC movement have progressively improved in global and regional models. However, the model accuracy in forecasting the intensities of TCs remains challenging for operational weather forecasting and consequential assessment of weather impacts such as high winds, storm surges, and heavy rainfall.

Since the current spatial resolution of the NWP model is insufficient for resolving convective scale processes and inner core dynamics of the cyclone, forecast intensities of TCs from operational models are mostly underestimated or low biased. Yet, accurate TC intensity guidance is crucial not only for assessing the impact of the TC, but also for generating realistic projections of storms and their associated hazards. This is essential for effective risk evaluation. Conventional TC intensity forecasting mainly relies on three approaches: statistical, dynamical, and statistical-dynamical methods.

Dynamical models, also known as numerical models, are the most complex and use high performance computing (HPC) to solve the physical equations of motion governing the atmosphere. While statistical models do not explicitly consider the physics of the atmosphere, they are based on historical relationships between storm behavior and storm-specific details such as location and intensity.

The rise of Machine Learning (ML) and Deep Learning (DL) has led to attempts to create breakthroughs in climate modeling and weather forecasting. Recent advances in computational capabilities and the availability of extensive reanalysis of observational or numerical datasets have reignited interest in developing various ML methods for predicting and understanding the dynamics of complex systems.

One of our key objectives is to build a quantum reservoir computing-based model, capable of processing climate model outputs and storm environment parameters, to provide more accurate forecasting, will improve short-term and real-time TC risk analysis.

Official modeling centers use consensus or ensemble-based dynamical models and represent the state of the art in tropical cyclone forecasting. However, these physics-based models may be subject to bias derived from high wind shear, low sea surface temperatures, or the storms location in the basin. By learning from past forecasting errors, we may be able to identify and correct past model biases, thereby greatly enhancing the quality of future forecasting and risk modeling products. The long-term aim is to integrate ML-based elements into coarse global climate models to improve their resolution and include natural dynamical processes currently absent in these models.

Reservoir Computing (RC) is a novel machine-learning algorithm particularly suited to quantum computers and has shown promising results in early non-linear time series prediction tests. In a classical setting, RC is stable and computationally simple. It works by mapping input time series signals into a higher dimensional computational space through the dynamics of a fixed, non-linear system known as a reservoir. This method is efficient, trainable, and has a low computational cost, making it a valuable tool for large-scale climate modeling.

While quantum machine learning has been considered a promising application for near-term quantum computers, current quantum machine learning methods require large quantum resources and suffer from gradient vanishing issues. Quantum Reservoir Computing (QRC) has the potential to combine the efficient machine learning of classical RC with the computing power of complex and high-dimensional quantum dynamics. QRC takes RC a step further by leveraging the unique capabilities of quantum processing units (QPUs) and their exponentially large state space, resulting in rich dynamics that cannot be simulated on a conventional computer. In particular, the flexible atom arrangements and tunability of optical controls within QuEras neutral atom QPU enable the realization of a rich class of Hamiltonians acting as the reservoir.

Recent studies on quantum computing simulators and hardware suggest that certain quantum model architectures used for learning on classical data can achieve results similar to that of classical machine learning models while using significantly fewer parameters. Overall, QRC offers a promising approach to resource-efficient, noise-resilient, and scalable quantum machine learning.

In this project, we are collaborating with QuEra Computing, the leading provider of quantum computers based on neutral-atoms , to explore the benefits of using quantum reservoir computing in climate science and to investigate the potential advantages that the quantum layer from QuEra can bring. QuEra's neutral atom QPU and the types of quantum simulations it can perform give rise to different quantum reservoirs. This unique capability can potentially enhance the modeling of tropical cyclone intensity forecasts and data.

This collaboration involves multiple stakeholders and partners, including QuEra Computing Inc., Moodys RMS technical team, and Moodys Quantum Taskforce. The work is supported by a DARPA grant award, underscoring its significance and potential impact in tropical cyclone modeling and forecasting.

In summary, combining quantum machine learning methods, reservoir computing, and the quantum capabilities of QuEra's technology offers a promising approach to addressing the challenges in predicting tropical cyclone intensity. This collaboration aims to enhance the quality and efficiency of tropical cyclone modeling, ultimately aiding in better risk assessment and decision making in the face of these natural disasters.

See the original post:
Exploring the potential of quantum reservoir computing in forecasting the intensity of tropical cyclones - Moody's Analytics