Fig.3 Comparison of (a) spatial PAR and site PAR, (b) tower GPP and LUE-EF GPP (uncorrected spatial data), (c) tower GPP and LUE-EF GPP (corrected spatial data), (d) LUE-EF GPP with uncorrected and LUE-EF GPP corrected spatial data, (e)spatial EVI (500m resolution) and spatial EVI (10km resolution), (f) tower GPP and LUE-NDWI GPP (uncorrected spatial data), (g) tower GPP and LUE-NDWI GPP (corrected spatial data), and (h) LUE-NDWI GPP with uncorrected and LUE-NDWI GPP corrected spatial data. All comparisons are based on site scale.
At the global scale, the biases in spatial data inputs had a great impact on the simulated GPP even for the less sensitive model LUE-NDWI. Fig.4a shows the global average annual GPP distribution from 2000 to 2018, simulated by the LUE-NDWI model using corrected input data layers. The spatial pattern of GPP agrees well with previous studies. However, the impact of data biases on the spatial pattern of simulated GPP was obvious and not uniform across space (Fig.4b). The area overestimated is much larger than underestimated area when the data biases were not attended, and the area fractions with GPP biases at (-50%)-(-30%), (-30%)-(-10%),10%-30%, and 30%-50%, were 8%, 19%, 27%, 31% and 15%, respectively. After data correction, area of GPP serious reduction occurs in the mountain systems of the Tibetan plateau in Asia, northern Africa and South America region. Area of serious growth were observed in Australia, northwest North America and Siberia. The global annual average GPP estimated by the LUE-NDWI, after input data correction, was about 125.6Pg C yr-1. Without data correction, the LUE-NDWI model would overestimate global GPP by 18% (Fig 4c and 4d). The corresponding global growth rate of GPP decreased from 0.34 to 0.17 Pg C yr-1 after input data correction (Fig. 4e).
The comparison of global GPP products simulated by 22 models is shown in Fig 4c. The GPP products were from TRENDY and other studies. Large differences can be seen from these models with long-term GPP averages varying from 92.7 to 178.9 Pg C yr-1 with more GPP estimates concentrated in the 120-130 Pg C yr-1 (Fig 4d). The interannual variabilities simulated by these models were also quite different. In general, the inter-annual variabilities of LUE and biophysical models were greater than those of the machine learning models. It is worth noting that the GPP simulations from the Revised-ECLUE, PR, LPJ-GUESS, and data LUE-NDWI (after correcting input data biases) models were all within the interquartile range (IQR) of the 22 GPP products throughout all years. In addition, the GPP from the LUE-NDWI model, after data correction, was the closest to the median GPP value of the 22 global models. The trends of GPP simulated by the 22 models also varied greatly from -0.25 to 0.84 Pg C yr-1 (Fig 4e). The trends of the machine learning models were smaller than those of other models. Most model showed positive trends, only the revised-ECLUE and CLASS-CTEM models showed downward trends, and some models demonstrated no significant trends (i.e., MODIS, FLUXCOM_ANN, FLUXCOM_MARS, and FLUXCOM_RF).
<Fig 4 roughly here>