Using ICA to Extract the Water Storage Variations Signals of the Okavango Delta

  Objectives  Studying regional land water storage changes can better understand the characteristics of water storage changes in an area, and provide better help for the study of extreme natural disasters such as drought and flood.

  Methods  To verify the signal decomposition ability of independent component analysis(ICA), the water storage variations in Africa's Okavango delta region from January 2003 to December 2014 was calculated using gravity recovery and climate experiment (GRACE) time-varying earth gravity field model, and the mass change was extracted by principal component analysis and ICA respectively, which was compared with the Global Land Data Assimilation System (GLDAS) hydrological model.

  Results  The results show periodic changes of the water reserves in the northeast of Okavango river, and the correlation coefficient of the time series corresponding to GRACE-IC1 and GLDAS-IC1 between the two datasets of spatial feature distribution in the same position reaches 0.85. The water variations in the Okavango delta area increase from January 2003 to October 2011, the correlation coefficient of the time series corresponding to GRACE-IC2 and GLDAS-IC3 between the two datasets of spatial feature distribution in the same position reaches 0.81. It indicates that GRACE agrees with the GLDAS hydrological model very well in the research area. In addition, Global Precipitation Climatology Center precipitation data and WaterGAP Global Hydrology Model data were introduced to analyze the variation of terrestrial water reserves in the study area.

  Conclusions  Compared to the traditional polynomial fitting method, the ICA can directly extract the spatial-temporal characteristics of the quality change in a specific location in a large area. By comparing the third component of the analysis results of the two GRACE methods, it can be seen that the ICA has stronger decomposition ability than the principal component analysis.