Nonlinear chaos in temperature time series: Part I: Case studies

TL;DR

This paper demonstrates that nonlinear modeling of temperature time series using Lorenz-type models can achieve stable 100-day forecasts, outperforming existing methods and avoiding complex fluid dynamics calculations.

Contribution

Introduces a nonlinear Lorenz-type model for local temperature forecasting that remains stable over 100 days, surpassing traditional methods in accuracy and computational efficiency.

Findings

Achieves stable 100-day temperature forecasts.

Outperforms existing forecast methods for periods above 11 days.

Provides a computationally efficient alternative to Navier-Stokes based models.

Abstract

In this work we present 3 case studies of local temperature time series obtained from stations in Europe and Israel. The nonlinear nature of the series is presented along with model based forecasting. Data is nonlinearly filtered using high dimensional projection and analysis is performed on the filtered data. A lorenz type model of 3 first order ODEs is then fitted. Forecasts are shown for periods of 100 days ahead, outperforming any existing forecast method known today. While other models fail at forecasting periods above 11 days, ours shows remarkable stability 100 days ahead. Thus finally a local dynamical system if found for local temperature forecasting not requiring solution of Navier-Stokes equations. Thus saving computational costs.