Re-defining the Functional Model of the GRACE Total Water Storage Time Series

Klos A., Güneş Ö., Lenczuk A., Aydın C., Bogusz J.

AGU Fall Meeting, Illinois, United States Of America, 12 - 16 December 2022, pp.1

  • Publication Type: Conference Paper / Summary Text
  • City: Illinois
  • Country: United States Of America
  • Page Numbers: pp.1
  • Yıldız Technical University Affiliated: Yes


The Gravity Recovery and Climate Experiment (GRACE) mission, along with its successor GRACE Follow-On mission, have been providing continuous monitoring of hydrological changes over time since 2002. Components of the terrestrial hydrosphere, such as groundwater, surface water, or water contained in snow and glaciers, sum up to the Total Water Storage (TWS) values. The time series of monthly TWS changes recorded from 2002 until now are conventionally modeled by a harmonic regression model, which includes a combination of linear trend plus annual and semi-annual periodic components. This model has long been suitable for many regions of the world. With a changing climate, more frequent and longer periods of drought, and unexpected rainfall, the changes in TWS observed by GRACE are far from linear. Consequently, linear trend contained in the conventional deterministic model is no longer able to reliably represent what is physically present in the series. Therefore, we propose to re-define the TWS time series model. For this purpose, we use the TWS time series released by the GSFC (Goddard Space Flight Center, NASA) center in the form of global RL06 mascon solution. We show that the standard deviation of the linear trend from the actual nonlinear changes present in the time series is regionally dependent and exceeds 10 mm for areas of South America, North America, Greenland, Antarctica, the southern part of Africa, the Tigris-Euphrates basin, the Caspian Sea area, and the Sumatra-Andaman and Tohoku-Oki earthquake-affected areas. In order to reliably represent the nonlinearities, we use polynomial functions. The significance tests show that polynomials of the third or fourth degrees provide the best fit to the TWS time series. Therefore, we propose to add polynomial function to the conventional time series model. The change in the definition of the time series functional model leads to a reduction in the standard deviation of TWS time series residuals and provides a more reliable determination of the uncertainty of the functional model parameters. We compare the polynomial function to climate indices and show that it perfectly captures long-term changes caused by droughts or floods being a consequence of El Nino and La Nina effects, among others.