Probabilistic Forecasting of El Niño Using Neural Network Models

Petersik, Paul; Dijkstra, Henk A.

doi:10.1029/2019gl086423

Cited by 38 publications

(28 citation statements)

References 43 publications

(57 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Other similar works include a standard neural network applied by Lubkov et al [42] to forecast El Niño and La Niña based on a set of global climatic indices of atmosphere-ocean system oscillations, with a lead time of 3 months. Petersik and Dijkstra [43] applied Gaussian density neural network and quantile regression neural network to forecast ONI, for lead times of up to 21 months. Their input features include ONI previous values, volume of water above the 20°C isotherm in the tropical Pacific, Dipole mode index, zonal wind stress anomaly, sea surface height anomaly, and season.…”

Section: Related Workmentioning

confidence: 99%

Forecasting El Niño and La Niña Using Spatially and Temporally Structured Predictors and a Convolutional Neural Network

Hashemi

2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

El Niño-Southern Oscillation (ENSO) is characterized by large-scale fluctuations of sea surface temperature in the central and eastern tropical Pacific accompanied by changes in the atmospheric circulation. ENSO events are of two main types: El Niño and La Niña. Oceanic Niño index (ONI) determines the five consecutive 3-month running mean of sea surface temperature (SST) anomalies, in the Niño 3.4 region (5°S-5°N, 170°W-120°W). El Niño is a phenomenon in the equatorial Pacific Ocean characterized by a value of greater than 0.5°C for ONI. La Niña is a phenomenon in the equatorial Pacific Ocean characterized by a value of less than-0.5°C for ONI. The lingering of either of these two phenomena could induce severe droughts, while either of them following the other could cause massive floods. In both cases, deaths and substantial pecuniary loss is unavoidable, making their forecast of great significance. This study takes over the challenge of forecasting these two phenomena with one year lead time which has proven difficult in the literature. This research's contribution is restructuring ENSO events' predictors, including SST, sea level pressure, surface wind speed, and wind stress in a spatially and temporally meaningful way and designing a convolutional neural network (CNN) that takes advantage of this structure to forecast ENSO events of different types (i.e. Central Pacific and Eastern Pacific) in the next year. Not only a high precision in forecasts was achieved, but also it was shown that the proposed model has the potential to achieve higher recalls if a larger number of samples from the positive class would become available.

show abstract

Section: Related Workmentioning

confidence: 99%

Forecasting El Niño and La Niña Using Spatially and Temporally Structured Predictors and a Convolutional Neural Network

Hashemi

2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…One of the key features of statistical physics-based approaches reviewed here is the ability to better and reliably forecast the complex Earth phenomena, such as El Niño events, extreme rainfall in the Eastern Central Andes, Indian summer monsoon, the collapse of the Atlantic multidecadal oscillation and the occurrence of earthquakes. Moreover, we observed that artificial intelligence and deep learning techniques [413] have achieved great success during recent years in many fields, such as phase transitions in statistical physics [414] , [415] , [416] , data-driven Earth system science [417] , ENSO forecasts [294] , [295] , Indian monsoon rainfall [418] , as well as the forecasting aftershock patterns [395] and detecting earthquakes [419] in seismic systems. In fact, we are confident that the combination and complement of network-based and artificial intelligence-based skills will boost each other.…”

Section: Discussionmentioning

confidence: 99%

“…In Ref. [294] , Petersik and Dijkstra applied Gaussian density neural network and quantile regression neural network ensembles to predict ENSO. Classical machine learning models for prediction usually only predict a point value, e.g., the ONI.…”

Section: Applicationsmentioning

confidence: 99%

Statistical physics approaches to the complex Earth system

et al. 2021

View full text Add to dashboard Cite

Global warming, extreme climate events, earthquakes and their accompanying socioeconomic disasters pose significant risks to humanity. Yet due to the nonlinear feedbacks, multiple interactions and complex structures of the Earth system, the understanding and, in particular, the prediction of such disruptive events represent formidable challenges to both scientific and policy communities. During the past years, the emergence and evolution of Earth system science has attracted much attention and produced new concepts and frameworks. Especially, novel statistical physics and complex networks-based techniques have been developed and implemented to substantially advance our knowledge of the Earth system, including climate extreme events, earthquakes and geological relief features, leading to substantially improved predictive performances. We present here a comprehensive review on the recent scientific progress in the development and application of how combined statistical physics and complex systems science approaches such as critical phenomena, network theory, percolation, tipping points analysis, and entropy can be applied to complex Earth systems. Notably, these integrating tools and approaches provide new insights and perspectives for understanding the dynamics of the Earth systems. The overall aim of this review is to offer readers the knowledge on how statistical physics concepts and theories can be useful in the field of Earth system science.

show abstract

“…This should facilitate a transfer learning process and build a general model, able to interpret and predict radar measures, that can then be specialized onto specific instruments and/or different meteo/climatic regions. We also intend to study the potential for a probabilistic nowcast, estimating the spatial and temporal uncertainty that can be obtained from neural network model, for instance through a quantile regression and the use of a pinball loss function [35].…”

Section: Discussionmentioning

confidence: 99%

Performance Comparison between Deep Learning and Optical Flow-Based Techniques for Nowcast Precipitation from Radar Images

Marrocu

Massidda

2020

Forecasting

View full text Add to dashboard Cite

In this article, a nowcasting technique for meteorological radar images based on a generative neural network is presented. This technique’s performance is compared with state-of-the-art optical flow procedures. Both methods have been validated using a public domain data set of radar images, covering an area of about 104 km2 over Japan, and a period of five years with a sampling frequency of five minutes. The performance of the neural network, trained with three of the five years of data, forecasts with a time horizon of up to one hour, evaluated over one year of the data, proved to be significantly better than those obtained with the techniques currently in use.

show abstract

Probabilistic Forecasting of El Niño Using Neural Network Models

Cited by 38 publications

References 43 publications

Forecasting El Niño and La Niña Using Spatially and Temporally Structured Predictors and a Convolutional Neural Network

Forecasting El Niño and La Niña Using Spatially and Temporally Structured Predictors and a Convolutional Neural Network

Statistical physics approaches to the complex Earth system

Performance Comparison between Deep Learning and Optical Flow-Based Techniques for Nowcast Precipitation from Radar Images

Contact Info

Product

Resources

About