English
 
Privacy Policy Disclaimer
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Journal Article

Multi-decadal temporal reconstruction of Sentinel-3 OLCI-based vegetation products with multi-output Gaussian process regression

Authors

D.Kovács,  Dávid
External Organizations;

Reyes-Muñoz,  Pablo
External Organizations;

/persons/resource/kberger

Berger [Richter],  Katja
1.2 Global Geomonitoring and Gravity Field, 1.0 Geodesy, Departments, GFZ Publication Database, Deutsches GeoForschungsZentrum;

Mészáros,  Viktor Ixion
External Organizations;

Caballero,  Gabriel
External Organizations;

Verrelst,  Jochem
External Organizations;

External Ressource
No external resources are shared
Fulltext (public)

5028074.pdf
(Publisher version), 28MB

Supplementary Material (public)
There is no public supplementary material available
Citation

D.Kovács, D., Reyes-Muñoz, P., Berger [Richter], K., Mészáros, V. I., Caballero, G., Verrelst, J. (2024): Multi-decadal temporal reconstruction of Sentinel-3 OLCI-based vegetation products with multi-output Gaussian process regression. - Ecological Informatics, 83, 102816.
https://doi.org/10.1016/j.ecoinf.2024.102816


Cite as: https://gfzpublic.gfz-potsdam.de/pubman/item/item_5028074
Abstract
Operational Earth observation missions, like the Sentinel-3 (S3) satellites, aim to provide imagery for long-term environmental assessment to monitor and analyze vegetation changes and dynamics. However, the S3 archive is limited in temporal availability to the year 2016. Although S3 provides continuity of previous missions, key vegetation products (VPs) including leaf area index (LAI), fraction of photosynthetically active radiation (FAPAR), fractional vegetation cover (FVC), and leaf chlorophyll content (LCC), can be reliably produced from Ocean and Land Colour Instrument (OLCI) data only since the sensors' launch. To overcome this limitation, our study proposes a reconstruction workflow that extends the data record beyond its data acquisition. By using multi-output Gaussian process regression (MOGPR) fusion, we explored guiding predictor VPs from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor for the reconstruction of multi-decadal (spanning two decades, 2002–2022) temporal profiles of four OLCI-derived VPs (S3-MOGPR), moving past S3's launch. We first evaluated three MODIS-derived inputs as predictor variables: LAI, FAPAR, and the Normalised Difference Vegetation Index (NDVI) over nine sites with distinct land covers from the Ground-Based Observations for Validation (GBOV) service. Each predictor produced a distinct time series for the four reconstructed S3 VPs. To determine which predictor variable most accurately reconstructs data streams of the targeted variable, all S3-MOGPR VPs were compared to satellite-based products from the Copernicus Global Land Service (CGLS). MOGPR models were trained for 2019 and compared to reference data. Since MODIS LAI demonstrated the best reconstruction performance of all predictors, S3-MOGPR VPs were fully reconstructed from 2022 back to 2002 using guiding MODIS LAI and evaluated with in-situ data. The most consistent reconstructed product was FVC (, NRMSE = 0.17) over mixed forests compared to CGLS estimates. FVC also yielded the highest validation statistics (, , NRMSE = 0.14) over croplands. The highest correlation coefficients were achieved by the predictor variable LAI reconstructing FVC with mean , and NRMSE = 0.11 among all sites of 0.91 and 0.88, respectively. In the absence of both satellite and ground-based LCC reference measurements, the reconstructed LCC profiles were compared to the OLCI and MERIS Terrestrial Chlorophyll Index (OTCI, MTCI). The correlation metrics provided strong evidence of the reconstructed LCC product's integrity, with the highest correlation over deciduous broadleaf, mixed forests and croplands (). The lowest correlations for all reconstructed variables appeared over evergreen broadleaf forests, driven by the absence of seasonal patterns. Altogether, by leveraging the flexibility of the MOGPR algorithm with guiding historical data, contemporary EO data can be extrapolated into the past.