Long-term reorganization of ocean nutrient distribution

Oceans rely on nutrients like nitrate and phosphate to support the growth of phytoplankton and marine productivity. Using nearly a century of global ocean data, this study shows that nutrients are changing in different ways depending on location. This entry contains scripts needed for data analysis presented in this manuscript, as well as the nutrient data files. The data analysis consists of four parts: (i) preprocessing, (ii) estimating long-term surface nutrient trends, (iii) estimating vertical trends and their significance, and (iv) analysis of CMIP6 data. The data consists of the pre-processed World Ocean Database observations.

Methods
All analyses are done using Matlab (v2023a). There are four groups of scripts. Group 1 consists of scripts to preprocess World Ocean Database observations. These are not needed as the shared data files (TwodN.csv and TwodP.csv) have all needed information. Group 2 consists of scripts required to estimate the long-term surface nutrient trends using different approaches. The key script is ‘regression_anomaly_wRandom_function.m’ as it describes the various regression approaches. Group 3 consists of scripts required to estimate the vertical profile trends and associated significance. ‘random_longterm_depth_parallel.m’ randomizes the observation along the temporal axis and estimates the randomized trends. ‘regression_anomaly_depth_top30.m’ estimate the vertical trend profile for Figure 3. ‘AutoEnconder_LongTerm.m’ uses an AutoEncoder to ‘learn’ the unsupervised profile of the randomized profiles and compared to the observed. Group 4 consists of scripts needed for processing CMIP6 model outputs. It is assumed that you have already processed the raw data files into a standardized grid (lat, lon, depth, time). ‘CMIP_trend_analysis.m’ calculates and the trends required for Figure 4 and ‘plot_trends_combined.m’ is used for plotting.