Net carbon balance on croplands depends on numerous factors (e.g., crop type, soil, climate) and their interactions. Agroecosystem models are generally used to assess cropland carbon fluxes because of their ability to capture the complex interactive effects of factors influencing carbon balance. For regional carbon flux simulations, generally gridded climate data sets are used because they offer data for each grid cell of the region of interest. However, studies consistently report uncertainties in climate datasets, which affect the accuracy of carbon flux simulations.
The objectives were to 1) determine the uncertainties in daily weather variables of commonly used high resolution gridded climate datasets in the U.S (NARR, NLDAS, Prism and Daymet); 2) estimate their impact on the accuracy of simulated Net Ecosystem Exchange (NEE) under irrigated and non-irrigated corn and soybeans using the Environmental Policy Integrated Climate (EPIC) agroecosystem model, and 3) understand the relative sensitivity of the NEE to various climate variables.
The observational data at four flux tower cropland sites in the U.S Midwest region were used to quantify the uncertainties in the gridded weather datasets, and EPIC simulations were performed at each flux tower site using each gridded climate dataset. Further, sensitivity analysis using Extended Fourier Amplitude Sensitivity Test (EFAST) was conducted.
Results suggest that daily weather variables in all gridded climate datasets display some degree of bias, leading to considerable uncertainty in simulated NEE. The gridded climate datasets produced based on interpolation techniques (i.e. Daymet and Prism) were shown to have less uncertainties, and resulted in NEE estimates with relatively higher accuracy, likely due to their higher spatial resolution and higher dependency on meteorological station observations. The Mean Absolute Percentage Errors (MAPE) values of average growing season NEE estimates for Dayment, Prism, NLDAS and NARR include 22.53%, 23.45%, 62.52% and 66.18%, respectively. The NEE under irrigation (MAPE = 53.15%) tends to be more sensitive to uncertainties compared to the fluxes under non-irrigation (MAPE = 34.19%). Further, this study highlights that NEE responds differently to the individual climate variables and management. Under irrigation management, NEE are more sensitive to temperature. Conversely, under non-irrigation, precipitation is the most dominant factor influencing NEE uncertainty.
These findings demonstrate that careful consideration is necessary when selecting climate data to mitigate uncertainties in simulated NEE. Further, alternative approaches such as integration of remote sensing data products may help reduce the models' dependency on climate datasets and improve the accuracy in the simulated CO2 fluxes.