Locally refined spatial predictions of marine sediment carbon stocks from legacy data

The recent “30 by 30” global initiative to protect 30% of the world’s land and ocean by 2030 has increased the need for marine spatial planning decisions to be grounded in locally relevant empirical evidence. Continental shelves play a key role in the cycling of carbon, where marine sediments can act as an important sink of organic carbon (OC). As a result, marine sediments storing carbon have attracted recent scientific attention to elucidate the amount of OC stored and mechanisms influencing its sequestration. Spatial models of marine sediment OC stocks have previously been developed and provide preliminary estimates of standing stocks over wide geographical scales. However, these broad-scale predictions are derived from models of broad scale environmental regimes, which makes them unlikely to capture local spatial variations in environmental conditions and subsequently local variations in OC, reducing their utility for local scale marine spatial planning decisions. This study aims to determine whether legacy data could be used to produce local scale spatial predictions of OC relevant for policy makers. To achieve this aim, local scale predictors relevant for OC were produced/sourced in order to predict local-scale marine sediment OC in the Irish Sea. Legacy data of bottom water temperature (BWT) and bottom water salinity (BWS) measurements were used to bias correct and downscale global models of BWT and BWS. Recently developed high resolution sediment properties (% mud, % sand and % gravel) and locally developed Sediment Mobility and Sediment Disturbance Indices (SMI and SDI, respectively) were also sourced as potential predictors. Public-good, environmental consultancy and government agency repositories were also searched for OC-content data. A Random Forest model was trained to predict OC-content on localised predictors as well as previously identified important predictors of marine sediment OC. The outputs from the localised model were compared to one that was trained on broad-scale predictors to determine the change in model performance and utility for making local scale predictions. This study highlights the value of legacy data in contributing to locally refined spatial predictions of OC-content relevant for marine spatial planning decisions.