# Tws Forecasting Essay

To achieve the objective of TWS forecasting in this thesis, the regions selected for extreme events analysis in section 3.2.6 were used. Correlation was done between area averaged TWS of each region and sea surface temperature indices of major global oceans (Pacific Ocean: from different Niño regions such as Niño 1.2, Niño 3, Niño 3.4 and Niño 4; and Indian Ocean SST, DMI). The Pearson product moment sample coefficient of correlation was used, which is defined as: ………………………………………….……...….…3.17

Where is correlation coefficient formally called Pearson product moment sample coefficient of correlation, Covariance between X and Y, and are the standard deviations for the variables X and Y, here TWS and sea surface temperature

*…show more content…*

Then, as many predictors as possible were selected depending on the magnitude of the absolute value correlation coefficient they have with TWS. Stepwise multiple linear regressions were then used to avoid co linearity and over fitting. In this process, many combinations of predictors (SSTs) are regressed on TWS producing many different regression models. In stepwise regression, the models are assessed by the residual sum of squares. The criteria for including a predictor are an improvement significant at the 0.15 levels (Diro et al., 2008). The model with good predicting skill with possible minimum predictors was taken finally from the

*…show more content…*

Monthly maps of equivalent water height anomaly for 151 months from April 2002 to February 2016 have been made (see Appendix 1). Maps of soil moisture storage from GLDAS were also made for 2002 to 2015. In this case, we made maps of absolute magnitude of soil moisture instead of removing the long term mean (See Appendix 2 for more information). Maps of groundwater anomaly were also produced and monthly climatology is shown in Figure 4.3. Spatial climatology of TWS variability from GRACE, soil moisture storage from GLDAS and groundwater variation calculated from TWS variations from GRACE and soil moisture storage variations from GLDAS were produced for all months from January to December averaged over the years from 2002 to 2014 (see Figure 4.1 to Figure 4.3