GPP from two decades of MSG data for terrestrial ecosystem monitoring over Europe and Africa

One of the main carbon fluxes characterizing terrestrial ecosystems and biodiversity is gross primary production (GPP), defined as the amount of carbon fixed by vegetation through photosynthesis, per unit area and unit time. GPP represents the potential carbon uptake of an ecosystem to produce food, wood, and fiber. Therefore, understanding its spatiotemporal variability under future climate change scenarios is essential for environmental management and global sustainable development. The temporal variability can be characterized by analyzing GPP time series, which exhibit non-stationary behavior driven by short-term, seasonal, and long-term variations.

In the last decade, significant advancements have been achieved in the development and production of operational long-term GPP series using Earth Observation (EO)-based data at regional and global scale. This is the case of the 10-day GPP product at 3.1 km (MGPP LSA-411) from geostationary SEVIRI/MSG data within the LSA SAF (Land Surface Analysis SAF) as part of the SAF (Satellite Application Facility) network of EUMETSAT. This product is freely available in the LSA SAF platform since 2018 for addressing near-real-time users’ climate and environmental applications. Currently, the possibility of improving this product using a new version of fAPAR, now under development, is being analyzed. This work aims to provide a 20-year assessment (2004–2023) of terrestrial ecosystem status based on the spatiotemporal analysis of 10-day GPP time series derived from MSG data, following the methodology of the operational MGPP product (LSA-411) but using a novel fAPAR as input based on a deep-learning approach.

In a first stage, the GPP time series is derived by computing daily GPP based on Monteith’s radiation use efficiency concept, which accounts for water stress effects to downregulate the maximum light use efficiency (optimal conditions). A suite of MSG products is used, including the daily downwelling shortwave radiation flux (DIDSSF, LSA-203), daily actual evapotranspiration (LSA-351 and LSA-312.3), and reference evapotranspiration (DMETREF, LSA-303). An evaluation of the derived 10-day GPP time series is performed at local scale using ground-based GPP estimates at 8 eddy covariance (EC) towers from the FLUXNET database. The assessment also includes the comparison with other operational EO-based products, such as the 8-day MODIS, 20-day GDMP and daily SMAP at the same EC towers. The results show high correlations (r > 0.70), between the MGPP and EC estimates, which are very similar to those obtained using MODIS, GDMP and SMAP products.

In a second stage, ecosystem monitoring is performed using the multi-resolution analysis (MRA) based on the wavelet transform (WT). MRA-WT provides a temporal decomposition of the original time series, allowing different signal component to be derived by removing the contribution of specific temporal scales. This approach has been extensively used over the past few decades across several applications. The results show a general greening in the central and eastern Sahel region, eastern Africa (Horn of Africa), eastern Spain and Turkey, which is associated with an increase in precipitation along the period. In contrast, localized negative changes are observed in the Senegal region and southern parts of Africa, mainly attributed to precipitation variability during the same period.