Spatio-temporal approach to moving window block kriging of satellite data v1.0

J.M. Tadić, X. Qiu, S. Miller and A.M. Michalak


Statistical gap filling techniques make it possible to obtain spatially and temporally contiguous estimates of land surface and atmospheric properties based on limited and noisy satellite observations. In this paper we extend the basic principles of spatial gap filling (based on nonstationary kriging) into the spatiotemporal domain. The paper presents methodological advances for doing so: alterations of the subsampling techniques, modeling of the variability, and kriging. The method yields estimates at target spatial and temporal resolutions, together with quantitative uncertainty estimates. The approach is demonstrated using solar induced fluorescence as well as atmospheric carbon dioxide and methane concentrations as measured by three different satellites.

Figure: (a) IASI XCH4 (0-4 km) retrievals (ppb) for March 2, 2013 (sea only), (b) Contiguous IASI maps for Northern Hemisphere for the same day obtained using Spatio-temporal Moving Window Block Kriging at 2.5 × 2° spatial resolution and (c) associated uncertainties, given as 1-sigma (σ) (ppb).


Numerous existing satellites observe physical or environmental properties of the Earth system. Many of these satellites provide global-scale observations, but these observations are often sparse and noisy. By contrast, contiguous, global maps are often most useful to the scientific community (i.e., level 3 products). We develop a spatiotemporal moving window block kriging method to create contiguous maps from sparse and/or noisy satellite observations. This approach exhibits several advantages over existing methods: 1) it allows for flexibility in setting the spatial resolution of the level 3 map, 2) it is applicable to observations with variable density, 3) it produces a rigorous uncertainty estimate, 4) it exploits both spatial and temporal correlations in the data, and 5) it facilitates estimation in real time. Moreover, this approach only requires the assumption that the observable quantity exhibits spatial and temporal correlations that are inferable from the data. We test this method by creating Level 3 products from satellite observations of CO2 (XCO2) from the Greenhouse Gases Observing Satellite (GOSAT), CH4 (XCH4) from the Infrared Atmospheric Sounding Interferometer (IASI) and solar-induced chlorophyll fluorescence (SIF) from the Global Ozone Monitoring Experiment–2 (GOME-2). We evaluate and analyze the difference in performance of spatio-temporal vs. recently developed spatial kriging methods.

Tadić, J.M., X. Qiu, S. Miller, A.M. Michalak (2017) "Spatio-temporal approach to moving window block kriging of satellite data v1.0", Geoscientific Model Development, 10, 709-720, doi:10.5194/gmd-10-709-2017.