Gap filling strategies for annual estimates of soil respiration

Background/Question/Methods

Soil respiration (R_soil) is one of the largest CO₂ fluxes of in the global carbon cycle. Quantifying the contribution of R_soil to the global carbon cycle requires annual estimates integrated across large spatial scales. R_soil records generally contain gaps. Filling data gaps is therefore requisite to accurately predict R_soil. However, the reliability of various strategies for filling gaps in R_soil records and scaling survey respiration measurements to an annual time scale has not yet been assessed. Here, we: 1) conducted a literature survey for gap filling strategies used to estimate annual R_soil, and 2) evaluated the performance of different gap filling methods by analyzing the errors introduced when filling artificial gaps into annual R_soil datasets for various ecosystem types. Gap filling methods evaluated included linear and quadratic interpolation, monthly average, and exponential temperature-dependence models assuming a) a single temperature sensitivity (E) and reference R_soil (R_ref) over the entire year, b) constant E and varying R_ref, and c) varying E and R_ref. Artificial gaps were introduced to the datasets at 11 gap fractions (0-95% of existing data) and in a pattern replicating bi-monthly survey measurements ( >99% “gap”) and filled using each method.

Results/Conclusions

Our literature survey analysis showed that a wide variety of gap filling methods have been used in R_soil records. However, the linear interpolation method along with the temperature-dependence R_soil model assuming a single E and R_ref over the entire year were the gap filling methods more widely used. Our results on gap filling methods performance showed that uncertainty in annual R_soil estimates increased along with gap fraction length. For short and medium gap fraction lengths (5-to-85% of the existing data), all gap filling strategies performed reasonably well. However, the relative differences between the different gap filling methods to predict the annual R_soil sums were larger for longer gap fraction lengths (95 and 99% of the existing data). Taken together our results emphasize the need to standardize gap filling methods for R_soil. We use our results to provide guidance as to which gap filling methods should be used to provide defensible annual R_soil estimates.