Search for a command to run...
PY-SOM-FFN-4D Temperature and Carbon observation-based products These files represent an early beta of the heat and carbon observation-based products determined using the PY-SOM-FFN-4D method for the period 2004-01-01 to 2023-12-31. The files present are: dissolved-inorganic-carbon-v6_4d_mm_2004-2024.nc temperature-v3_4d_mm_2004-2024.nc This product is produced as part of the European AI4PEX research project to contribute towards the improvement of high-resolution Earth System Modelling of extreme events. The temperature product provided is generated using the Feed-Forward Network method as the Self-Organising Map method was not found to provide significant improvement in the output. The product is derived using primarily Earth observation data in combination with an observation-based product of ocean mixed-layer depth. Spatiotemporal input features are also included: spatial information is encoded using n-vectors (Gade, 2010, The Journal of Navigation) and time is encoded using a monotonic index for trend and a circular encoding for seasonality seasonality (Gregor, Kok & Monteiro, 2017, Biogeosciences). The model hyperparameters are tuning through four-fold temperal-block cross-validation. The optimised hyperparameters are kept constant for all model levels. The first level is predicted by training the model on the listed inputs. Each following subsequent layer is initialised with the model weights of the previous layer before being trained with the same input features. The EN 4.2.2.g10 temperature observation data is used as target data. The following predictor data were used: Sea Surface Temperature: resampled to 1x1 degree from the NOAA Geo-Polar Blended Global Sea Surface Temperature Analysis (Level 4) data product (Night) Sea Surface Height: resampled to 1x1 degree from the Copernicus Global Ocean Gridded L 4 Sea Surface Heights and Derived Variables Reprocessed 1993 Ongoing data product Cloud Cover Fraction: resampled to 1x1 degree from EUMETSAT CM SAF Cloud Product (Low + High) Mixed-Layer Depth: (Buongiorno Nardelli, 2020, Earth System Science Data The following data were used for training and validation: Ocean Temperature: Met Office Hadley Centre observation dataset EN 4.2.2 quality controlled subsurface ocean temperature and salinity profiles and objective analyses (Good, Martin & Rayner, 2013, Journal of Geophysical Research: Oceans) The data product provided contains an ensemble mean and standard deviation of 20 four-dimensional estimates. The product is still under-development and being prepared for publication. The dissolved inorganic carbon product provided is generated using the Feed-Forward Network method as the Self-Organising Map method was not found to provide significant improvement in the output. The product is derived using secondary observation-based data products. Time is encoded as an input feature using a monotonic index for trend. The model hyperparameters are tuning through four-fold temperal-block cross-validation. The optimised hyperparameters are kept constant for all model levels. The first level is predicted by training the model on the listed inputs. Each following subsequent layer is initialised with the model weights of the previous layer before being trained with the same input features. The GLODAPv2 dissolved inorganic carbon observation data is used as target data. The following predictor data were used: Ocean Temperature: the PY-SOM-FFN-4D temperature observation-based product Atmospheric CO2 Mixing Ratio: Dlugokencky et al. (2021) NOAA Global Monitoring Laboratory Ocean Salinity: modified from analysis product of Good, Martin & Rayner (2013, Journal of Geophysical Research: Oceans) EN.4.2.2.g10 data were obtained from: https://www.metoffice.gov.uk/hadobs/en4/ and are © British Crow Copyright, Met Office, provided under a Non-Commercial Government Licence http://www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/.) Dissolved Oxygen: resampled to 1x1 degree from GOBAI-O2 data product (Sharp et al., 2022, NOAA National Centers for Environmental Information) The following data were used for training and validation: Dissolved Inorganic Carbon: resampled to 1x1 degree from the GLODAPv2.2023 data product of Olsen et al. (2016) and Key et al. (2015) The data product provided contains an ensemble mean and standard deviation of 20 four-dimensional estimates. The product is still under development and being prepared for publication.