An empirical MLR for estimating surface layer DIC and a comparative assessment to other gap-filling techniques for ocean carbon time series

[摘要] Regularized time series of ocean carbon data are necessary forassessing seasonal dynamics, annual budgets, and interannual and climaticvariability. There are, however, no standardized methods for filling datagaps and limited evaluation of the impacts on uncertainty in thereconstructed time series when using various imputation methods. Here wepresent an empirical multivariate linear regression (MLR) model to estimatethe concentration of dissolved inorganic carbon (DIC) in the surface ocean,that can utilize remotely sensed and modeled data to fill data gaps. ThisMLR was evaluated against seven other imputation models using data fromseven long-term monitoring sites in a comparative assessment of gap-fillingperformance and resulting impacts on variability in the reconstructed timeseries. Methods evaluated included three empirical models – MLR, meanimputation, and multiple imputation by chained equation (MICE) – and fivestatistical models – linear, spline, and Stineman interpolation; exponentialweighted moving average; and Kalman filtering with a state space model. Crossvalidation was used to determine model error and bias, while a bootstrappingapproach was employed to determine sensitivity to varying data gap lengths.A series of synthetic gap filters, including 3-month seasonal gaps (spring,summer, autumn winter), 6-month gaps (centered on summer and winter), and bimonthly (every 2 months) and seasonal (four samples per year) sampling regimes, were appliedto each time series to evaluate the impacts of timing and duration of datagaps on seasonal structure, annual means, interannual variability, andlong-term trends. All models were fit to time series of monthly mean DIC,with MLR and MICE models also applied to both measured and modeledtemperature and salinity with remotely sensed chlorophyll. Our MLR estimatedDIC with a mean error of 8.8  µ mol kg −1 among five oceanic sites and20.0  µ mol kg −1 for two coastal sites. The MLR performance indicatedreanalysis data, such as GLORYS, can be utilized in the absence of fieldmeasurements without increasing error in DIC estimates. Of the methodsevaluated in this study, empirical models did better than statistical modelsin retaining observed seasonal structure but led to greater bias in annualmeans, interannual variability, and trends compared to statistical models.Our MLR proved to be a robust option for imputing data gaps over varieddurations and may be trained with either in situ or modeled data dependingon application. This study indicates that the number and distribution ofdata gaps are important factors in selecting a model that optimizesuncertainty while minimizing bias and subsequently enables robust strategiesfor observational sampling.

[学科分类] 大气科学