| Terrestrial ecosystem absorbs CO2through photosynthesis,providing the ability to partially offset anthropogenic CO2emissions,which effectively slows the rising trend of global warming.Given the central role of terrestrial ecosystem in the global carbon cycle,it is crucial to quantify their carbon sequestration capacity（i.e.,gross primary productivity,GPP）.As an important part of terrestrial ecosystems,cropland GPP has become the largest source of uncertainty in global GPP estimation due to the complexity of planting types and growing seasons.How to use limited observational data to reduce the uncertainty of global cropland GPP simulation has been become a scientific problem.In this study,the Markov Chain Monte Carlo method（MCMC）was used to estimate the posterior distribution of the maximum light use efficiency parameters（ε*）of 26 crops in the EC-LUE model,and based on crop type distribution data,the Monte Carlo method was used to estimate global cropland GPP and its uncertainty range in 2001,2005 and 2010,the impact of model parameters and input datasets on the estimation of global cropland GPP was further explored.Specifically,global cropland GPP based on single-parameter and multi-crop parameter simulations,and global cropland GPP based on pure cropland pixel and mixed pixel simulations were compared,respectively.The main conclusions are as follows:（1）Theε*parameters of the 26 crops in the EC-LUE model had obvious differences,ranging from 1.29 to 5.23 MJ/g C,the highest value was sorghum,and the lowest value was cotton.Overall,C4 crops（sorghum,corn,etc.）parameter values were significantly higher than those of C3 crops（wheat,soybean,etc.）,which indicated the necessity of distinguishing different crop parameters when the model estimated cropland GPP.The uncertainty of the model parameters was affected by the model performance and the number of observational data,and the sorghum with the highest uncertainty of theε*parameter reaches 1.86MJ/g C.The cropland GPP based on multi-crop parameter simulation showed significant improvement at the site scale.Both R and RMSE were significantly better than the MODIS-GPP dataset.The ratio of observed value to simulated value remained at 0.95-1.05 at most sites.Therefore,it is a feasible method to estimate global cropland GPP using multi-crop parameters in the model.（2）Monte Carlo simulations showed that in 2001,2005,and 2010,the global cropland GPP was 11.92±0.73 Pg C,12.24±0.75 Pg C,and 12.62±0.78 Pg C,respectively.Among the 26crops,only three food crops,wheat,maize and rice,contributed more than 40%of the global cropland GPP.High GPP values were mainly concentrated in some maize and wheat growing areas,such as the Midwest of the United States,the Ganges Plain in Asia,and some islands in Southeast Asia,with the highest value exceeding 1200 g C m-2yr-1.There were 8 crops with uncertainty above 0.1 Pg C,including wheat（0.42 Pg C）,maize（0.3 Pg C）,rice（0.22 Pg C）,other perennials（0.19 Pg C）,pasture（0.17 Pg C）,soybean（0.17 Pg C）,sorghum（0.17 Pg C）,barley（0.11 Pg C）,which were mainly affected by the planting area of crops and the uncertainty of model parameters,so the estimation of global cropland GPP can only consider the difference of the model parameters of these crops,which will not produce obvious deviation.Compared with other GPP datasets,the results of this study were close to those of SIF-GPP and NIRv-GPP,and higher than those of MODIS-GPP,FLLUXCON,and Revised-EC-LUE,and the results of this study were also within the range of the previously estimated cropland GPP（10.42-16.03 Pg C）.（3）Global cropland GPP simulated by a single parameter does not change in annual amount.The specific differences were mainly reflected in various crops.The GPP of C4 crops such as sorghum and corn was seriously underestimated,while the C3 crops with low light use efficiency was overrated.In terms of spatial pattern,the simulated GPP of the two cases was more different.In some areas with C4 crops or C3 crops distribution with high light use efficiency,such as the North American plains,the Sahel region,etc.,the cropland GPP simulated based on a single parameter was significantly lower that the cropland GPP based on multi-crop parameter simulation,the opposite situation was shown in C3 crop planting areas such as Eastern Europe;the simulated cropland GPP based on mixed pixels was more than 0.5Pg C yr-1higher than the cropland GPP based on pure cropland pixel simulation,the GPP of different crops also showed a similar situation.The cropland GPP simulated by mixed pixels in the humid/semi-humid area was higher,but the opposite was true in the arid/semi-arid area,which was mainly due to the difference in land use types in different regions.The cropland in the humid/semi-humid area are mostly near forests,shrubs,etc.,the vegetation index would be overestimated during the resampling process,while the cropland in the arid/semi-arid area is near grassland,etc,the vegetation index would be underestimated during the resampling process.This ultimately leads to the difference between the two simulated scenarios. |
