Modeling of sea surface temperature using linear models with autocorrelation Indian Ocean

2nd International Conference on Marine Science: Better Insight for the Healthy Ocean

Miftahuddin, M.^1 ; Ilhamsyah, Y.^1,2

Faculty of Mathematics and Science, Syiah Kuala University, Jln. Teuku Nyak Arief Darussalam, Banda Aceh, Aceh

23111, Indonesia^1

Applied Climatology, Graduate School, Bogor Agriculture University, Bogor, Indonesia^2

关键词: Akaike information criterion;    Autoregressive moving average;    Climate parameters;    Climate system;    Eastern equatorial Indian ocean;    Linear modeling;    Sea surface temperature (SST);    Temporal variability;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/176/1/012038/pdf
DOI  :  10.1088/1755-1315/176/1/012038

来源: IOP

【 摘 要 】

Sea surface temperature (SST) is one of several indicators of climate system of the Earth. We used model SST to observe SST dataset from buoy arrays in the eastern equatorial Indian Ocean. Relationship between SST and other climate parameters can be represented in linearity approach. This approach shows that temporal variability of the SST as a dominant effect. Linear model fitting (LMF) has been examined with four treatments, with and without: covariate transformation, interaction, centering, and addition time covariate in the model. The LMF chosen as basic construction in the model with covariate interaction combination and transformation, which increases magnitude of multiple-R2 (56.62%) and adjusted-R2 (56.13%), i.e. 0.31% and 0.43% respectively. This shows that time covariates have a strong significance effect in the model, compared to continuous covariates. However, the model has autocorrelation, which has large Akaike Information Criterion (AIC) value then this deletion of effects can be done through the autoregressive moving average. Moreover we obtained that LMF which suitable to SST is model with AIC value 403.2987 by using three climate features include two time covariates. Furthermore, we observed that using GAM model fitting showed an increase in explained deviance to 65.90%, a significant decrease in AIC from 678.24 to 634.99 and significant increase in adjusted-R2 from 51.20% to 64.40% by using sixteen climate features include two times covariates without interaction and transformation.