IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (Jan 2022)

Toward the Removal of Model Dependency in Soil Moisture Climate Data Records by Using an <inline-formula><tex-math notation="LaTeX">$L$</tex-math></inline-formula>-Band Scaling Reference

  • Remi Madelon,
  • Nemesio J. Rodriguez-Fernandez,
  • Robin van der Schalie,
  • Tracy Scanlon,
  • Ahmad Al Bitar,
  • Yann H. Kerr,
  • Richard de Jeu,
  • Wouter Dorigo

DOI
https://doi.org/10.1109/JSTARS.2021.3137008
Journal volume & issue
Vol. 15
pp. 831 – 848

Abstract

Read online

Building climate data records of soil moisture (SM) requires computing long time series by merging retrievals from sensors on-board different satellites, which implies to perform a bias correction or rescaling on the original time series. Due to their long time span and high temporal frequency, model data could be used as a common reference for the rescaling. However, avoiding model dependence in observational climate data records is needed for some applications. In this article, the possibility of using as reference remote sensing data from one of the $L$-band sensors specifically designed to measure SM is discussed. Advanced Microwave Scanning Radiometer 2 SM time series were rescaled by matching their cumulative distribution functions (CDFs) to those of Soil Moisture and Ocean Salinity (SMOS), Soil Moisture Active Passive (SMAP), and Global Land Data Assimilation System (GLDAS) NOAH model time series. The CDF computation was investigated as a function of the time series length, finding significant differences from four to nine years. Replacing temporal by spatial variance does not allow us to compute better CDFs from short time series. The rescaled time series show a high correlation ($R>0.8$) to the original ones and a low bias with respect to the reference ($< $0.03 m $^{3}\cdot$ m$^{-3}$). The time series rescaled using several SMOS or SMAP datasets were also evaluated against in situ measurements and show performances similar to or slightly better than those rescaled using the model GLDAS. The impact of random errors and gaps of the observational data into the rescaling was evaluated. These results show that it is actually possible to use $L$-band data as reference to rescale time series from other sensors to build long time series of SM.

Keywords