Improved retrieval of land surface biophysical variables from time series of Sentinel-3 OLCI TOA spectral observations by considering the temporal autocorrelation of surface and atmospheric properties

Estimation of essential vegetation properties from remote sensing is crucial for a quantitative understanding of the Earth system. Ill-posedness of the model inversion problem leads to multiple interpretations of one satellite observation, and using prior information is a promising way to reduce the ill-posedness and increase the accuracy of land surface products. Tobler's first law of geography states that everything is related to everything else, but near things are more related than distant things. Likewise, it is expected that the state of an object at a single moment is related to the state at every other moment, but temporally near attributes are more related than distant ones. This temporal autocorrelation is a vital source of prior information and can be used to improve the retrieval accuracy. In this study, we develop a retrieval framework that makes use of the temporal autocorrelation and dependence of land surface and atmospheric properties. We apply this retrieval algorithm to Sentinel-3 Ocean and Land Colour Instrument (OLCI) satellite data to derive land surface biophysical variables with a focus on leaf area index (LAI) from top-of-atmosphere (TOA) radiance observations. The results from both a synthetic dataset and a real satellite dataset show that the use of the temporal continuity as a priori information improves the accuracy of the estimation of land surface properties, such as leaf chlorophyll content and LAI. Compared with the MODIS LAI products, much less unrealistic short-term fluctuations are found in the LAI retrievals from OLCI with the time-series retrieval approach across different land cover types including cropland, forest and savannah. Field measurements of LAI at two forest sites quantitatively confirm that the estimated LAI from OLCI is reasonably accurate with R-2 > 0.65 and RMSE < 1.00 m(2)m(-2). Overall, the time series retrieval results in more robust and smoother time series than standard retrievals of LAI from individual scenes, more stable retrievals than the MODIS LAI product, and values of LAI that match better with reported measurements in the field. The present retrieval framework can make better use of time series of spectral observations and potentially of multi-sensor observations.

Automated summary

The study demonstrates the use of temporal autocorrelation to improve the accuracy of remote sensing retrieval of vegetation properties, particularly leaf area index. The method shows promising results in estimating land surface biophysical variables and reduces unrealistic short-term fluctuations in the retrieved data, showing potential for application in various land cover types and improving upon existing products such as MODIS LAI.