Constraining estimates of global soil respiration by quantifying sources of variability

Quantifying global soil respiration (R_SG ) and its response to temperature change are critical for predicting the turnover of terrestrial carbon stocks and their feedbacks to climate change. Currently, estimates of R_SG range from 68 to 98 Pg C year^-1 , causing considerable uncertainty in the global carbon budget. We argue the source of this variability lies in the upscaling assumptions regarding the model format, data timescales, and precipitation component. To quantify the variability and constrain R_SG , we developed R_SG models using Random Forest and exponential models, and used different timescales (daily, monthly, and annual) of soil respiration (R_S ) and climate data to predict R_SG . From the resulting R_SG estimates (range = 66.62-100.72 Pg), we calculated variability associated with each assumption. Among model formats, using monthly R_S data rather than annual data decreased R_SG by 7.43-9.46 Pg; however, R_SG calculated from daily R_S data was only 1.83 Pg lower than the R_SG from monthly data. Using mean annual precipitation and temperature data instead of monthly data caused +4.84 and -4.36 Pg C differences, respectively. If the timescale of R_S data is constant, R_SG estimated by the first-order exponential (93.2 Pg) was greater than the Random Forest (78.76 Pg) or second-order exponential (76.18 Pg) estimates. These results highlight the importance of variation at subannual timescales for upscaling to R_SG. The results indicated R_SG is lower than in recent papers and the current benchmark for land models (98 Pg C year^-1 ), and thus may change the predicted rates of terrestrial carbon turnover and the carbon to climate feedback as global temperatures rise.