magistrsko delo
Uroš Perkan (Author), Gregor Skok (Mentor), Žiga Zaplotnik (Co-mentor)

Abstract

V okviru magistrske naloge razvijemo model za napovedovanje vremena ConvCastNet, ki temelji na konvolucijskih nevronskih mrežah. Podatke za učenje nevronske mreže pridobimo iz ERA5 reanalize in jih interpoliramo na ekviangularno mrežo z ločljivostjo 3°, pri čemer uporabimo 10 spremenljivk, ki opisujejo stanje ozračja, površine morja in kopnega ter 3 statična polja. Razvijemo nov način obravnave robnih pogojev s konvolucijskimi nevronskimi mrežami, s katerim se znebimo računskih težav, povezanih z matrično predstavitvijo podatkov. Model optimiziramo z analizo delovanja komponent konvolucijskih nevronskih mrež in algoritma za njihovo učenje. Vremensko napoved izvajamo z avtoregresijo dnevno povprečenih spremenljivk. ConvCastNet pri napovedovanju višine geopotenciala na 500 hPa ploskvi doseže 7,8 dni z vrednostjo koeficienta korelacij anomalij večjo od 0,6, s čimer se pri tej metriki približa rezultatom najnovejših modelov strojnega učenja in fizikalnemu determinističnemu visokoresolucijskemu modelu (HRES) Evropskega centra za srednjeročne vremenske napovedi (ECMWF). Pri napovedovanju poti tropskih ciklonov ConvCastNet več dni sledi njihovi dejanski trajektoriji in njihovo intenziteto povečuje ali zmanjšuje sorazmerno ERA5 reanalizi. Kljub temu imajo njegove napovedi omejeno uporabnost, saj je za oceno vpliva takšnih ekstremnih dogodkov na ljudi pomembno napovedovanje njihovih trenutnih in lokalnih karakteristik, kar zaradi dnevnega povprečenja podatkov in 3° horizontalne ločljivosti ni izvedljivo. Analiza napak modelskih napovedi v letu 2022 pokaže, da je njihova magnituda največja na baroklinih območjih v visokih geografskih širinah. Normiranje napake z naravno variabilnostjo pa pokaže, da te glede na tipično hitrost spreminjanja vremena najhitreje naraščajo v tropskih predelih.

Keywords

strojno učenje;napovedovanje vremena;konvolucijske nevronske mreže;tropski cikloni;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FMF - Faculty of Mathematics and Physics
Publisher: [U. Perkan]
UDC: 551.509
COBISS: 164351491 Link will open in a new window
Views: 9
Downloads: 9
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Weather forecasting with convolutional neural networks
Secondary abstract: We develop a weather forecasting model ConvCastNet, which is based on convolutional neural networks. The training data are obtained from ERA5 reanalysis and are interpolated to an equiangular grid with a 3° spatial resolution. The forecast is carried out with autoregression of daily averaged variables, encompassing 10 atmospheric, sea and land variables and 3 static fields. A novel approach to the treatment of boundary conditions in convolutional neural networks is introduced, effectively mitigating challenges associated with the matrix representation of data. We conduct a comprehensive analysis of convolutional neural networks components and their training algorithm’s performance, thus optimising the forecast model. ConvCastNet achieves significant success in predicting geopotential at 500 hPa pressure level, with 7.8 days of useful forecast, defined by the lead time at which the anomaly correlation coefficient is greater than 0.6. This result comes close to the results of the latest machine learning models and the deterministic high-resolution model (HRES) of the European Centre for Medium-Range Weather Forecasts (ECMWF). ConvCastNet accurately predicts tracks of tropical cyclones for several days in advance and adjusts their intensity proportionally to ERA5 reanalysis. Nevertheless, due to the daily averaging of the data and a spatial resolution of 3°, the forecasts have limited utility, as predicting current and local weather characteristics holds greater significance in assessing the impact of such extreme events on people. An analysis of ConvCastNet forecast errors in the year 2022 demonstrates that their magnitude is greatest in highly baroclinic regions in high latitudes. Normalizing the errors with the atmospheric natural variability reveals that the error growth relative to the typical rate of local weather variability is greatest in the tropics.
Secondary keywords: machine learning;weather forecasting;convolutional neural networks;tropical cyclones;
Type (COBISS): Master's thesis/paper
Study programme: 0
Thesis comment: Univ. v Ljubljani, Fak. za matematiko in fiziko, Oddelek za fiziko
Pages: IX, 81 str.
ID: 19937473