Almost all existing remote sensing models (e.g., regression models, parametric models, process models and machine learning models) rely on plant functional type (PFT)-based parameter settings, multiple data sources (e.g., meteorological data) and key indices (e.g., leaf area index, LAI), limiting their estimation accuracy and spatial generalization capability.

Therefore, we developed an End-To-End Satellite-based model (ETES) to improve GPP estimation. ETES only utilizes input variables from original satellite observations and Global Land Surface Satellite (GLASS) downward shortwave radiation data. It replaces the traditional vegetation types data with a set of numeric variables (named as Seasonal Characteristics of Vegetation Types and Growth, SCVTG) derived from the curve of vegetation index time series within each growing cycle.

We provide an example dataset containing processed site records paired with 5-km satellite data. The remote sensing data in that example is the mean of 5-km pixels' reflectance. We construct the ETES model for GPP estimation in multi-layer perceptron method (MLP), and it contains a dataset file, model file and example code.

Tips: It would have a better 5-km result if applying the satellite data with higher spatial resolution and integrating estimation results to 5 km scale, because of the non-linear reponse of ETES model. In some way, the non-linear relationship is the main process existing within the real world.

• We developed an end-to-end satellite-based GPP estimation model (ETES), lessening the reliance on multiple data sources.
• We designed a new plant continuous variable set, named as Seasonal Characteristics of Vegetation Types and Growth (SCVTG).
• SCVTG can reflect the continuous spatiotemporal differences in plant functional types and phenology.
• ETES is free from plant functional types, meteorological data and traditional parameters.
• ETES can improve the GPP estimation accuracy, with an average 27.89% reduction of RMSE (monthly scale) compared to similar GPP products.

This a conceptual example showing the SCVTG (Seasonal characteristics of vegetation types and growth) variable set. The SCVTG could characterize the differences in vegetation types and phenology

Comparison of each combination scheme during the screening experiment. (a) The R2 comparison of daily GPP estimation; (b) the RMSE comparison of daily GPP estimation; and (c) the comparison of relative feature importance among schemes. Each scheme contains different input variables. The red boxes (FLUX-SW) represent schemes using shortwave radiation data from flux observations, and the blue boxes (GLASS-SW) represent schemes using shortwave radiation data from remote sensing products. The symbol *, **, *** shows the significance level of t-test between two schemes (e.g., Plan A' vs. Plan B') at the 0.05, 0.01 and 0.001 level, respectively. The relative feature importance (showed with the area) is calculated with each input variable’s mean |SHAP|. Note the differences between SCVTG-based schemes (Plan A, B, C, D, E) and PFT-based schemes (Plan A', B', C', D', E'), and the differences between the FLUX-SW schemes and the GLASS-SW schemes.

Comparison of the overall GPP accuracies at different temporal scales. 8 daily comparison: (a) MOD17, (b) GOSIF, (c) Fluxcom, and (d) ETES. Monthly comparison: (e) MOD17, (f) GOSIF, (g) GPP-NIRv, (h) Fluxcom, and (i) ETES. Yearly comparison: (j) MOD17, (k) GOSIF, (l) GPP-NIRv, (m) Fluxcom, and (n) ETES. The blue–purple color indicates a lower scatter density, while the yellow–green color indicates a higher scatter density.



Comparison of GPP time series in different models for 10 typical sites with different vegetation types. Each point represent an 8 daily GPP estimation (red color, models’ result) or an 8 daily GPP observation (blue color, FLUXNET). (a) MOD17 vs. FLUXNET, (b) GOSIF vs. FLUXNET, (c) Fluxcom vs. FLUXNET, (d) ETES vs. FLUXNET. CRO: cropland, DBF: deciduous broad-leaf forest, EBF: evergreen broad-leaf forest, ENF: evergreen needle-leaf forest, MF: mixed forest, GRA: grassland, OSH: open shrub, SAV: savanna, WSA: Woody Savanna, WET: wetland.