Cloud Computation and Google Earth Visualization of Heat/Cold Waves: A Nonanticipative Long-Range Forecasting Case Study

TL;DR

This paper presents a cloud-based, nonanticipative analog algorithm for long-range heat and cold wave forecasting, utilizing high-performance virtual machines and Google Earth visualization to achieve high accuracy without probability distributions.

Contribution

It introduces a novel cloud computing approach combined with Google Earth visualization for accurate long-range weather wave predictions using nonanticipative analog models.

Findings

Up to 36.4% of heat waves predicted accurately.

Up to 33.3% of cold waves predicted accurately.

Predictions are 100% accurate based on sign comparison.

Abstract

Long-range forecasting of heat/cold waves is a topical issue nowadays. High computational complexity of the design of numerical and statistical models is a bottleneck for the forecast process. In this work, Windows Server 2012 R2 virtual machines are used as a high-performance tool for the speed-up of the computational process. Six D-series and one standard tier A-series virtual machines were hosted in Microsoft Azure public cloud for this purpose. Visualization of the forecasted data is based on the Google Earth Pro virtual globe in ASP.NET web-site against http://gearth.azurewebsites.net (prototype), where KMZ file represents geographic placemarks. The long-range predictions of the heat/cold waves are computed for several specifically located places based on nonanticipative analog algorithm. The arguments of forecast models are datasets from around the world, which reflects the concept of teleconnections. This methodology does not require the probability distribution to design the forecast models and/or calculate the predictions. Heat weaves at Annaba (Algeria) are discussed in detail. Up to 36.4% of heat waves are specifically predicted. Up to 33.3% of cold waves are specifically predicted for other four locations around the world. The proposed approach is 100% accurate if the signs of predicted and actual values are compared according to climatological baseline. These high-accuracy predictions were achieved due to the interdisciplinary approach, but advanced computer science techniques, public cloud computing and Google Earth Pro virtual globe mainly, form the major part of the work.