Open AccessDevelopment of a statistical forecast model to improve accuracy based on statistical analysis of weather historical data for the Kalmyk region
Author(s) -
V. M. Goryaev,
Gerenzel Yashkulovna Kazakova,
Д. Б. Бембитов,
E. N. Dzhakhnaeva,
E V Sangadjieva
Publication year - 2019
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/350/1/012058
Subject(s) - smoothing , exponential smoothing , series (stratigraphy) , random walk , autoregressive integrated moving average , time series , moving average , computer science , statistics , term (time) , mathematics , econometrics , paleontology , physics , quantum mechanics , biology
The article presents the main assumption that the averaging and smoothing models for the time series of meteorological data are local-stationary with slowly varying values. Accordingly, to estimate the current average, a local average was chosen, which was used to forecast for the near future and can be considered as a compromise between the average model and the random walk model (without drift). The same strategy can be used to evaluate and extrapolate a local trend. Average monthly values of temperature, humidity and pressure for the period from January 1928 to December 2018 were used as initial data at 4 synoptic stations. Comparative statistics on dependencies and forecasts are given. The selected method — a moving average — is often referred to as a smoothed version of the original series, since short-term averaging in a given window has the effect of smoothing irregularities in the desired series. The objective of the study was to adjust the degree of smoothing for the moving average, as an optimal balance between the performance of averaging models and the simplicity of the random walk model or between quality and cost. The task was solved on the basis of the software implementation of the SARIMA model, which required a lengthy adjustment of the initial data and significant manipulations with time series, however, in the end, a successful model was selected.