Synergistic estimates of global 4-day 500 m gross primary production, evapotranspiration, and ecosystem water use efficiency from satellite data

Gross primary production (GPP) and evapotranspiration (ET) are essential components of global carbon and water cycles, respectively, while the ratio of GPP to ET, also known as ecosystem water use efficiency (WUE), reflects the trade-off between carbon gain and water loss in terrestrial ecosystems. Simultaneous estimates of GPP, ET, and WUE from satellite data with high accuracies are highly challenging due to negligence or inadequate representation of co-variation of GPP and ET in current models. This study develops a novel and practical model for Synergistic estimates of global 4-day 500 m gross primary Production, Evapotranspiration, and ecosystem water use Efficiency (SynPEE), by combining the multivariable convolutional neural network (MCNN) and a synthesis of in-situ observations at 314 globally distributed sites, satellite remote sensing datasets, and ERA5-land reanalysis datasets from 2000 to 2020. The newly proposed SynPEE model is prominently superior in (1) explicitly considering the synergistic relationship among the GPP, ET, and WUE; (2) achieving high-accuracy estimations of GPP, ET, and WUE simultaneously; and (3) avoiding the outliers of WUE estimates that are commonly found in the un-synergistic models. Validated against in-situ observations by a spatial 10-fold cross-validation scheme, the SynPEE model was proven to overall outperform the un-synergistic models (CNN_IN) constructed for separate estimates of GPP, ET and WUE. Moreover, the SynPEE model also showed much better performances than four state-of-the-art RS products, i.e., BESSv2, PMLv2, FLUXCOM, and MODIS. Furthermore, the spatio-temporal patterns of the 8-day and yearly GPP, ET and WUE estimates by the SynPEE model were generally consistent with those of the four state-of-the-art products. The SynPEE model has great potential of generating time-series products of high-accuracy global GPP, ET and WUE, which is promising to enhance our understanding of land–atmosphere interactions of carbon and water, thus better serving for terrestrial carbon and water management.