A global monthly climatology of oceanic total dissolved inorganic carbon (DIC): a neural network approach (NCEI Accession 0222469)

This dataset contains global monthly climatology of oceanic total dissolved inorganic carbon (DIC). (DIC) monthly climatology was created from a neural network approach (Broullón et al., 2020). The neural network was trained with GLODAPv2.2019 (Olsen et al., 2019) and LDEOv2016 (Takahashi et al., 2017) data, using as predictor variables position (latitude, longitude and depth), year, temperature, salinity, phosphate, nitrate, silicate and dissolved oxygen. pCO2 from LDEOv2016 and AT from Broullón et al. (2019) were used to compute DIC surface values to increase the surface coverage in the training data. The relations extracted between the predictor variables and DIC were used to obtain the climatology passing through the network global monthly climatologies of the predictor variables: temperature and salinity fields of the World Ocean Atlas version 2013 (WOA13), filtered WOA13 oxygen (fifth-order one-dimensional median filter through the depth dimension; see Broullón et al., 2019 for details) and nutrients computed using CANYON-B (Bittig et al., 2018) over the three previous fields. The obtained climatology has a 1°x1° spatial resolution and 102 depth levels between 0 and 5500 m, with a monthly resolution from 0 to 1500 m and an annual resolution from 1550 to 5500m.

本数据集包含全球海洋总溶解无机碳（total dissolved inorganic carbon，DIC）的逐月气候态数据。该DIC逐月气候态基于神经网络方法（Broullón等人，2020）构建而成。训练所用的数据集包含GLODAPv2.2019（Olsen等人，2019）与LDEOv2016（Takahashi等人，2017）的观测数据，预测变量包括位置（纬度、经度与深度）、年份、温度、盐度、磷酸盐、硝酸盐、硅酸盐以及溶解氧。为提升训练数据的表层观测覆盖度，研究人员还引入了LDEOv2016的pCO2数据与Broullón等人（2019）提出的AT数据，用于计算DIC表层数值。研究人员基于预测变量与DIC间的关联关系，将训练好的神经网络应用于以下预测变量的全球逐月气候态数据以生成目标气候态：世界海洋图集2013版（World Ocean Atlas 2013，WOA13）的温度与盐度场、经过滤波处理的WOA13溶解氧数据（沿深度维度执行五阶一维中值滤波，详细信息参见Broullón等人，2019），以及基于上述三场数据通过CANYON-B模型（Bittig等人，2018）计算得到的营养盐数据。本次生成的气候态数据集空间分辨率为1°×1°，在0至5500米的深度范围内共设置102个深度层：其中0至1500米区间采用逐月分辨率，1550至5500米区间则采用年分辨率。