1 Introduction

Shifts in the provision of nutrients can seriously impact ocean health and productivity (Bristow et al., 2017). On one hand, elevated anthropogenically derived nutrients can lead to eutrophication and ecosystem degradation (Anderson et al., 2002). On the other hand, climate warming and stratification are predicted to reduce the supply flux and surface nutrient availability (J. K. Moore et al., 2018). Changes in both directions can pose major threats to marine ecosystems. Coastal regions near large population centers experience substantial increases in nutrient inputs (Andersen et al., 2017). In contrast, there is a long-term deepening of the phosphocline but a stable nitracline in oligotrophic regions (Gerace et al., 2025b). Yet, despite suggestions for contrasting regional trends—rising nutrient inputs in coastal zones and declining availability in the open ocean—a comprehensive global empirical assessment of nutrient dynamics remains lacking.

Biogeochemical theory predicts a biome-dependent link between recent climate change and shifts in upper ocean nutrients (Bopp et al., 2001). Biological nitrogen fixation is common in oligotrophic regions and appears to buffer stratification-induced changes in nitrogen (N) but not phosphorus (P) supply (Gerace et al., 2025b). However, N fixation can be suppressed by a low iron supply (C. M. Moore et al., 2009), and N fixation is lower or possibly absent in mesotrophic regions (Sohm et al., 2011). Hence, N fixation may not buffer nitrate levels everywhere, and we predict a stratification-induced decline in both N and P in regions with excess nitrate. Furthermore, surface nutrient are already depleted by biological uptake in low latitude regions meaning that any physical changes may not impact already vanishingly low concentration levels. Concurrent with climate warming, coastal regions also experience increasing run-off from land (Anderson et al., 2002). This elevated horizontal nutrient supply could either buffer or exceed impacts from offshore stratification. Finally, the ocean is a dynamic fluid environment, and advection may mask any spatially unique shifts in nutrient levels. Thus, while there are strong theoretical mechanisms for regional trends in upper ocean nutrients, empirical observations are needed to elucidate the resulting changes.

Climate predictions mainly focus on the depletion of mixed-layer nutrients. On a contemporary time-scale, P can be regarded as conserved meaning that a surface depletion leads to accumulation elsewhere. For N, it is more complex. Here, N flux changes can be buffered by biological fixation, leading to a stable surface but rising subsurface concentration from remineralization of this additional “source” (Somes et al., 2016). Furthermore, denitrification leads to a depletion of N in subsurface low oxygen zones. Hence, the strength of biological feedbacks are important for long-term trends and vertical reorganizations in nutrient availability (Hutchins & Capone, 2022). The deep compared to surface ocean inventory is much larger, so any water-column reorganizations may be undetectable deeper in the water column. However, a model study showed that warming and elevated phytoplankton productivity led to the expected nutrient depletion in the surface layer of the Southern Ocean (J. K. Moore et al., 2018). Concurrently, reduced upwelling and sinking organic matter remineralization led to a deep ocean “nutrient trapping.” This accumulation deeper in the water column resulted in a reduced advective nutrient flux out of the Southern Ocean and suppressed productivity in other regions (Primeau et al., 2013). Thus, climate warming and associated biogeochemical processes will likely affect the entire water column nutrient profile.

The goal of this study is to detect centennial nutrient trends across biomes and the water column. Nutrient concentrations from the World Ocean Database consist of 9,103,674 phosphate and 4,758,805 nitrate measurements from the top 800 m from 1925 to 2025 (Figure 1). However, the observations are heterogeneously sampled across time, space, and depth. This heterogeneity poses significantly analysis challenges and can introduce key biases that needs to be addressed to increase confidence in observed trends. First, like most oceanographic factors, nutrient concentrations exhibit seasonal variation that can mask or bias interannual trends (Gerace et al., 2025b). The World Ocean Atlas is the companion climatological objectively analyzed mean from the same observations. Hence, we use nutrient concentration anomalies by subtracting the monthly climatological mean to remove seasonal biases in the data. Second, the spatial distribution of observations is regionally biased (Figure 1). On the one hand, trends in well-sampled regions are more robust, whereas there is high uncertainty at places only sampled a few times. On the other hand, well-sampled regions can dominate the analysis (Figure 1e) but not necessarily represent a global trend. There is no correct “answer” for dealing with this spatial bias. Instead, we conduct a series of analyses with varying adjustment weights for sampling coverage. Third, there is also a temporal bias in measurement coverage where certain years are more intensely sampled, overlapped with a shift in technique, and earlier samples likely have higher uncertainty (Figure 1c). To assess this set of temporal biases, we removed individual decades and examined the resulting trends. Fourth, changes in nutrients occur across the full water column driven mechanistically by both physical and biological processes leading to complex but potentially distinct and connected vertical profile shifts. Furthermore, there is a strong vertical autocorrelation between observations across the water column. Traditional statistically techniques will struggle with this problem by either treating each depth independently or assuming a shape for the profile shift. To detect a non-linear yet-to-be determined trend across the water column, we designed a machine-learning approach using an autoencoder to quantify and compare such inter-connected shifts in vertical nutrient profiles to our biogeochemical expectations while allowing structural flexibility in both the vertical and time-evolving dimensions. Fifth, we contrasted the observed trends to historical and future predictions made by ocean biogeochemical models. This comparison enabled us to test if current biogeochemical theory composed of known processes captures the observed trends across biomes as well as the impact of future warming. Combined, this study provides the first globally comprehensive evaluation of the 3-D changes to key ocean nutrient fields.

Global coverage of nutrient measurements. The analysis included 9,103,674 and 4,758,805 phosphate (a) and nitrate (b) measurement covering 21096 unique 1° × 1° spatial grid cells of the surface ocean (Mishonov et al., 2024). Despite the extensive coverage, there are also clear biases in the temporal (c), vertical (d), and regional (e) data coverage.

2 Materials and Methods

2.1 Nutrient Data

The analysis is based on a combination of the World Ocean Database (WOD) (Mishonov et al., 2024) and World Ocean Atlas (WOA) (Garcia et al., 2024). WOD includes full water column in situ measurements from 1925 to 2024 (Figure 1). This includes 4,758,805 nitrate and 9,103,674 phosphate measurements (unit: μM). The measurements cover all ocean regions but with a significant bias toward the northern hemisphere and in particular regions adjacent to East Asia, North America, and Europe (Figures 1a and 1b). We restricted the analysis to the top 800 m due to the lower data coverage deeper in the water column (Figure 1d). Coastal sites are defined as having a bottom depth shallower than 500 m and vice-versa for pelagic sites. Oligotrophic regions are defined with surface layer nitrate less than 1 μM and phosphate less than 0.2 μM (Martiny et al., 2019). The World Ocean Atlas represents the objectively analyzed mean of the WOD data set gridded at 1° resolution. All in situ observations are binned at 1° resolution and specific depth intervals matching the WOA 3-D grid, resulting in 51% of ocean 1° grid cells covered. To correct for seasonal variation, the monthly mean value at specific grids from WOA is then subtracted, resulting in nutrient concentration anomalies. In total, 15 regions are defined based on a combination of surface nutrient concentrations (oligo- or mesotrophic), latitude, longitude, and bottom depth (Table S1 in Supporting Information S1).

2.2 Long-Term Anomaly

The long-term anomaly is estimated using three approaches. We use different approaches to adjust the weight (and hence importance) of sampling effort to balance biases in space and time. Without weighting, intensely sampled regions and periods will dominate the long-term signal. The advantage with a no-weight approach is that the robustness of long-term trends is likely proportional to effort. Alternatively, we can introduce weights to balance the influence of sampling intensity. The advantage here is that regions and periods more evenly contribute to the estimated trend. We use a “robust” regression approach with a bisquare weight function to limit the influence of outliers. Approach “binned” is designed to fully control for temporal biases in sample collection and hence each period is equally weighted. Here, observations are binned in 10 years intervals from 1925 to 1965 and 2 years intervals from 1965 to 2025. The varying bin-size applied is picked due to the low number of observations prior to 1965. Figure S1 in Supporting Information S1 also displays the anomaly change only using samples collected after 1965. For approach “all,” the trend is estimated using all observations. Here, all samples are equally weighted. Approach “All weighted” represents a compromise between “binned” and “all” approaches. Before the long-term regression, we apply a weight to all samples calculated as 1/log(1 + N_grid,i + N_time,j). N_grid,i represents the number of observations in an 1° × 1° grid cell, and N_time,j represents the number of observations for year j. The error bars represent the 95-percentile uncertainty for the mean nutrient anomalies. For Figure 2, we only used observations in the top 30 m whereas Figure 3 includes observations in the top 800 m for pelagic and top 200 m for coastal regions.

Long-term trend in upper ocean nutrient concentrations. Nutrient concentration anomalies in the upper ocean (≤30 m depth) for nitrate in coastal (a), mesotrophic (+1 μM, (b), and oligotrophic (c) regions and for phosphate in coastal (d), mesotrophic (+0.2 μM, (e) and oligotrophic regions (f). Concentration anomalies are binned by 10-year intervals until 1960 and then 2 years afterward. Nutrient concentration anomalies are observed concentrations corrected for monthly variations (by subtracting monthly climatological means). The F statistic rejects a constant model within machine precision (p < 1E−200) for all biomes and nutrients. Coastal locations have bottom depth less than 500 m. Pelagic locations are divided into mesotrophic and oligotrophic based on a threshold concentration (nitrate = 1 μM and phosphate = 0.2 μM).

Water-column nutrient trends across biomes. (a) Long-term trend in nitrate (μM/yr). (b) Long-term trend in phosphate (nM/yr). Each dot represents the “All weighted” trend for each depth interval. The line represents a “robust LOWESS” moving average of 8 depths.The coastal trend is capped at 200 m due to low data coverage deeper in the water column. Note difference in unit for nitrate (μM) and phosphate (nM).

2.3 Site and Depth Specific Trends

We also binned all observations at 1° resolution and depth intervals matching the WOA grid, and the annual mean anomaly (i.e., seasonally corrected) is calculated. This approach is used to eliminate any spatial sampling bias. We introduced a minimum sampling threshold for spatial bins to be included to ensure statistical robustness in these trends (6 for nitrate and 12 for phosphate observations). We also evaluated using other thresholds for the minimum number of observations in a spatial grid cell required for estimating trends (Figure S12 in Supporting Information S1). The thresholds reflect the higher sampling intensity for phosphate versus nitrate. Next the long-term (yearly) bin-specific slope is quantied using a linear least-square regression (Nutrient anomaly ∼ sampling year). The resulting trends by depth and biome are displayed.

2.4 Significance Test Using Temporal Randomization and an Autoencoder

We want to evaluate the significance of long-term trends across depths across the water column (i.e., top 800 m for pelagic and top 200 m for coastal). Conventional statistical approaches are not well-suited for this problem. One approach is comparing a null model for each individual depth trend. However, due to interdependence of shifts across depths, we here want to test if the water column nutrient profiles are changing. A parametric model for the whole water column requires an a priori defined functional form that we do not have a this stage. Instead, we want to evaluate the significance of a water column non-linear change in nutrient concentrations but with structural flexibility.

To generate a null distribution, we first randomize the sampling year of all observations. This is repeated 1,000 times. For each randomization, the long-term trend in nutrient anomalies for each depth and biome are estimated, and the vertical trend profiles are calculated. Next, we synthesize the overall profile for each randomization using an unsupervised machine-learning approach. An autoencoder is a type of neural network that estimates a simplified representation of noisy data set. Here, we train the autoencoder on the 1,000 randomized nutrient anomaly trend profiles using Matlab (AutoEnc: “hiddensize” = 10, “MaxEpochs” = 100, “L2WeightRegularization” = 0.001, “SparsityRegularization” = 4, “SparsityProportion” = 0.05, “ScaleData” = false). Then, we calculate the Euclidian distance between each random trend profile and the autoencoder prediction. This provides an estimate of the random prediction error for the full water column trend. Last, the distance between the observed trend and autoencoder prediction is calculated. If the observed profile is truly distinct from randomized profiles, then the distance should be significantly larger. The significance is calculated as the fraction of randomizations with a Euclidian distance less than observed.

2.5 Earth System Model Analysis

We compared the observed vertical nutrient trends to Earth system model simulations (Table S2 in Supporting Information S1). This includes both historical and future changes under both intermediate (SSP2-4.5) and severe (SSP5-8.5) climate change scenarios. All model simulations are part of the Coupled Model Intercomparison Project, Phase 6 (CMIP6) (Séférian et al., 2020). Due to the coarse spatial grid, we do not include any coastal comparisons. We again divide locations into oligotrophic and mesotrophic sites using the same surface ocean nutrient thresholds. We then calculate the trend for each model cell (unique latitude, longitude, and depth layer) using least square regression. Next, we estimate the vertical profile of nutrient trends for oligotrophic and mesotrophic regions. Finally, we compare the observed and predicted trends in the surface layer (Figure 4).

Comparison between observed and modeled surface nutrient trends. Surface modeled and observed trends for nitrate in mesotrophic (a) and oligotrophic (b) regions and for phosphate in mesotrophic (c) and oligotrophic (d) regions. The three modeled scenarios include historical, moderate warming (SSP2-4.5) and severe warming (SSP5-8.5) for the surface ocean (0–30 m). For the box plots, the red line is the median, the “box” spans the 25 and 75 percentiles, the whiskers cover 99.3% of the observations, and red crosses are outlier beyond this range. See Table S1 in Supporting Information S1 for the specific Earth system models included in this analysis. Observed “All weighted” trends are for the top 30 m layer shown in Figure 3.

3 Results

3.1 Long-Term Surface Nutrient Anomaly

We observe significant centennial trends in the surface ocean nutrient concentrations (Figure 2). In coastal biomes, nutrient concentrations have significantly risen leading to an integrated addition of 0.3–1.6 μM nitrate and 70–90 nM phosphate. The large range for nitrate reflect regional uncertainty in the trend (Figure S2 in Supporting Information S1). For phosphate, Atlantic coastal regions share the overall trend, whereas the directionality is less certain for Pacific coastal regions. Nitrate in mesotrophic biomes display a weakly significant increase. This increase is sensitive to the exact period analyzed (Figure S1 in Supporting Information S1). There is also clear divergence in regional mesotrophic trends with equatorial waters showing declining nitrate level, whereas the subpolar North Pacific and Indian Ocean display increases. Hence, there is less confidence in a uniform long-term surface nitrate trend in mesotrophic regions. In contrast to nitrate, phosphate significantly decline in mesotrophic biomes. This decline is robust to the exact period sampled, and no individual mesotrophic region shows a significantly opposite phosphate trend (Figure S1 in Supporting Information S1). Oligotrophic biomes display significant declines in both nitrate and phosphate. Albeit significant, the nitrate trend is an order of magnitude smaller than coastal and mesotrophic regions. In contrast, the phosphate decline in oligotrophic regions is strong and seen in all regions and mirror the recently observed deepening of the phosphacline (Gerace et al., 2025b). With the exception of changes to nitrate in mesotrophic regions, we do not detect strong temporal or regional biases. Thus, there is confidence in biome-dependent long-term trends in nutrient inventories but weak confidence in regionally unique shifts.

3.2 Vertical Re-Organization of Nutrients

We detect substantial water-column reorganizations of nutrients (Figure 3). In mesotrophic regions, nitrate is mostly stable right at the surface, but below the mixed layer, there is a large accumulation of nitrate with a peak around 100 m. Individual mesotrophic regions share this nitrate trend profile although equatorial regions are experiencing a decline in the surface. (Figure S3 in Supporting Information S1). For mesotrophic phosphate, there is generally a decline near the surface, but no clear accumulation. Regionally, phosphate is also declining in the surface whereas a few regions show signs of a surface accumulation (Figure S4 in Supporting Information S1). Deeper in the water column, the North Pacific show signs of accumulation and the North Atlantic a depletion. In oligotrophic regions, nitrate is stable or slightly declining at the surface (Figure 3). Oligotrophic biomes experience a nitrate accumulation deeper in the water column. This N accumulation is shared across ocean regions (Figure S3 in Supporting Information S1). Across oligotrophic regions, phosphate is again declining in the upper part and stable or possibly slightly increasing deeper in the water column (Figure S4 in Supporting Information S1). In coastal regions, nitrate accumulates across the water column although this is mainly seen in the northern hemisphere. Coastal phosphate also increases in the surface layer but stable deeper in the water column (Figure 3 and Figure S5 in Supporting Information S1). We also applied a different approach to estimate trends by first binning all samples in 1° × 1° × unique depth and then estimate the median trend in each bin (Figure S6 in Supporting Information S1). This approach weighs regions evenly independent of temporal sampling intensity. However, the resulting water column trends are similar to Figure 3. Overall, the vertical patterns underscore the distinct nutrient dynamics across ocean biomes, highlighting both shared and region-specific biogeochemical processes that shape the vertical distribution of nitrate and phosphate.

There is considerable uncertainty for each individual depth trend. Hence, trends at only a few depths are significantly different from zero. At the same time, the distinct profiles and closely matched vertically adjacent shifts strongly indicate a non-random biogeochemical reorganization of nutrients (Figure 3). We designed a machine-learning approach using an autoencoder to quantify and compare such inter-connected shifts in water column profiles to our biogeochemical expectations while allowing structural flexibility in the vertical dimension. First, we artificially scramble the sampling years 1000 times to estimate the chance of a spurious long-term trend. Second, we train an unsupervised neural network model designed to “learn” the overall shape of each random water column trend (i.e., with the autoencoder). Third, we use the trained model to evaluate if the observed full water column trends are significantly distinct from spurious shifts. We find that spurious trends display overall flat but noisy vertical profiles (Figure S7 in Supporting Information S1). In contrast, observed mesotrophic and oligotrophic full water column nutrient trends are significantly distinct from such random profiles (Figure S8 in Supporting Information S1). For example, the distinct sub-surface accumulation of nitrate or the upper-ocean depletion of phosphate are not seen in the random trends (Figure S7 in Supporting Information S1). The coastal water column changes are also significantly distinct from random profiles. Thus, we detect significant vertical reorganizations of nutrient concentrations in pelagic and possibly also in coastal biomes.

3.3 Earth System Models Underestimate Observed Nutrient Trends

Ocean biogeochemical models substantially underestimate current trends in ocean nutrients (Figure 4). We next compared the observed and model-simulated shifts in nitrate and phosphate. This is done for both mesotrophic and oligotrophic pelagic regions, whereas the models are too spatially coarse to accurately simulate coastal dynamics. A diverse set of biogeochemical models predict historical trends that are approximately an order of magnitude less than diagnosed from observations (Figure 4). The key exception is the observed nitrate increase in mesotrophic regions, whereas models suggest a weak decline. Models agree on a climate-driven acceleration of both nitrate and phosphate surface depletion. Declines are most pronounced mesotrophic regions under the high radiative forcing (and high warming) scenario (SSP5-8.5). Models also predict an accumulation of nitrate at depth (Figures S9–S11 in Supporting Information S1). Accumulation is most intense under SSP5-8.5 and peaks deeper in the water column (Figure S9 in Supporting Information S1). For phosphate, the current decline matches the intermediate climate change scenario (SSP2-4.5) in mesotrophic regions, whereas oligotrophic regions experience stronger surface declines than seen in any model. However, there is disagreement among the models at which depth horizon the strongest phosphate declines occur (Figure S10 and S11 in Supporting Information S1). In some models, the decline occurs near the surface, whereas others suggest strong subsurface depletion. Thus, the comparison between observed and simulated changes suggests that (a) warming and stratification have strong impacts on surface ocean depletion, (b) models partially support a sub-surface nitrate accumulation, and (c) ocean biogeochemical models may have adopted overly conservative biological parameterizations, leading to underestimating current nutrient changes.

4 Discussion

4.1 Biome Dependence of Long-Term Nutrient Shifts

The observations support the hypothesis of long-term nutrient shifts being biome dependent. Consistent with past studies, there is a general decline in surface phosphate in pelagic regions (Gerace et al., 2025b; Yasunaka et al., 2016). Reduced physical transport and increased phytoplankton demand can both lead to this change. Given evidence of declining primary production (Ryan-Keogh et al., 2025; Silsbe et al., 2025), a physical mechanism is the more probable driver. Subsurface phosphate shows weak or absent accumulation, possibly reflecting the reorganization of nutrients but a large deep-water inventory. In contrast, surface nitrate trends vary. The nearly stable nitrate levels in oligotrophic waters are likely controlled by exhaustive phytoplankton assimilation (Garside, 1985) and limited potential for further decline. Diverging trends in deeper nitrate and phosphate point to a biological influence with a shift in the balance between nitrogen supplied by fixation versus physical transport, leading to a subsurface N surplus (Yoshikawa et al., 2013). Accumulation of N but not P in oligotrophic regions aligns with high N fixations rates being observed in this biome (Sohm et al., 2011). However, nitrate accumulation in mesotrophic regions could reflect intensified N fixation in traditionally low-fixation zones (Zehr & Capone, 2020) but also lateral advection from oligotrophic areas. There are many recent reports of N fixation in regions with excess nitrate (Fonseca-Batista et al., 2019; Shiozaki et al., 2017), supporting that subsurface N accumulation in mesotrophic regions can have a biological origin. Coastal accumulations, especially in the North Atlantic, likely stem from anthropogenic runoff, whereas declines in the Southern Hemisphere may reflect less anthropogenic impacts combined with an offshore nutrient decline. Considering both nutrients jointly, the long-term nutrient trends suggest a generally rising N:P in the surface due to higher N fixation balancing nitrate but a phosphate decline (Gerace et al., 2025b). This, in turn, should result in higher biomass N:P and when remineralized in the mesopelagic contribute to nitrate accumulation at depth. Overall, the synthesis reveals strong contempory nutrient trends that are faster—but mostly consistent in direction - with biogeochemical expectations.

4.2 Biogeochemical Models Underestime Warming Impacts on Nutrients

Despite a general alignment with observations, ocean models substantially underestimate observed long-term changes. This mismatch likely stems from complex and model-specific limitations (Séférian et al., 2020). Physically, models may not fully capture changes in upper water column stratification (Sallée et al., 2021; Treguier et al., 2025). Biologically, a key feedback appears to be the regulation of N fixation, whose spatial distribution and controls remain uncertain, especially in nitrate-rich areas (Zehr & Capone, 2020). Our results highlight a potentially significant role of N fixation in mesotrophic nutrient trends. Ocean models often impose a temperature control on N fixation (J. K. Moore et al., 2001), restricting this process to low-latitude regions. Thus, models may underestimate the strength of biological feedbacks to buffer the nitrogen supply in some surface ocean ecosystems. These model-restraints on N fixation could result in overestimating future surface nitrate depletion under severe climate change. Additionally, N fixation is sensitive to subtle variations in N:P:Fe stoichiometry, which many models struggle to capture (Martiny et al., 2019; Wang et al., 2019). Uncertain model parameterizations may also dampen sensitivity to contemporary shifts. Observations thus offer new constraints for improving simulations of phytoplankton growth and cycling of ocean nutrients.

4.3 Potential Biases and Uncertainties in Long-Term Trends

The heterogenous nature of nutrient observations, including varying measurement techniques and spatio-temporal coverage presents a challenge for this analysis. Analysis design choices balancing spatial or temporal coverage versus sampling intensity do affect the resulting trends (Figures S1 and S12 in Supporting Information S1). At one end of the spectrum, we evenly weight each location independently of sampling intensity and thus reduce spatial biases. On the other hand, trends from intensely sampled locations or periods are likely more robust, and one can reasonably argue should carry more weight in the analysis. Trends including surface phosphate depletion, nitrate depletion in oligotrophic regions, and subsurface nitrate accumulations in both mesotrophic and oligotrophic regions are consistent across this robustness versus coverage spectrum (Figure S11 in Supporting Information S1). Model simulations of historical changes support mechanistically such changes, providing further confidence in the trend direction. The trend in surface mesotrophic nitrate is less certain, and it was sensitive to temporal sampling biases, regionally varying, and models predict an opposite trend. Coastal surface phosphate accumulation is also sensitive to analysis design choice and hence less certain. The study underscores that despite the large number of existing observations, there is a great need for continued monitoring across ocean regions to reduce uncertainties of nutrient trends.

5 Conclusions

The results reveal that anthropogenic impacts are reshaping the foundation of ocean ecosystems. Declines in surface nutrients can limit phytoplankton growth—the base of the marine food web—potentially reducing oceanic carbon uptake and destabilizing trophic dynamics (Chust et al., 2014). Meanwhile, intensified nitrogen fixation and nitrate accumulation at depth are shifting nutrient stoichiometry and the vertical profile. These shifts are enhancing N:P nutrient ratios and phosphorus limitation, likely to impact phytoplankton diversity and function (Kwiatkowski et al., 2018). Such trends both confirm and expand prior findings pointing toward a systemic transition toward more P-stressed conditions (Gerace et al., 2025b). The subsurface nitrate accumulation could facilitate large episodic nutrient inputs during physical stratification breakdowns and promote irregular growth including blooms and carbon sequestration (Gupta et al., 2022). Although coastal runoff drives nutrient enrichment in some regions, particularly near population centers (Anderson et al., 2002), such increases may soon be overshadowed by offshore basin-scale nutrient depletions in many places. Crucially, models agree that surface depletion will accelerate with continued climate change, posing a serious and growing threat to ocean productivity and ecosystem health.

Conflict of Interest

The author declares no conflicts of interest relevant to this study.

Data Availability Statement

All nitrate and phosphate observations were retrieved from the World Ocean Database 2023 on 16 February 2025 (Mishonov et al., 2024). We also used objectively analyzed means from the World Ocean Atlas 2023 (Garcia et al., 2024) and CMIP6 data files previously presented (Gerace et al., 2025a). Formatted datafiles and scripts are available here (Martiny, 2026).