Publications

Carbon pricing of shipping requires timely national estimates of maritime emissions, yet no study has evaluated how to forecast the new monthly, AIS-based statistics that allocate ship emissions to economies. We compare local statistical models, global machine-learning models, their combinations and five reconciliation methods on 148 hierarchically organised series, using 13 rolling origins and horizons of 1–12 months. Automatic exponential smoothing and an equal-weight combination of exponential smoothing, ARIMA and Theta were the most accurate and best calibrated; global gradient boosting and pooled autoregression were significantly less accurate and their conformal intervals covered only 68–69% of outcomes at a nominal 95%. Reconciliation methods that ignore scale heterogeneity tripled bottom-level errors, whereas top-down proportions and shrinkage-based trace minimisation preserved accuracy. Simple local models remain the safe default for short, heterogeneous panels.

Keywords: Cross-learning; Forecast reconciliation; Exponential smoothing; Conformal prediction; Shipping decarbonisation; Gradient boosting; Rolling-origin evaluation

Logistic and other S-shaped growth curves are widely used to project the diffusion of wind and solar power and to infer saturation levels from early data. This paper tests whether they forecast better than simple trend methods. Using annual generation in 47 national wind and 35 national solar markets, it produces 13,440 rolling-origin forecasts from seven methods at origins 2014–2019 and horizons of one to five years, and it tracks how fitted ceilings change as data accumulate. The logistic curve was no more accurate than exponential extrapolation in mean rank and was significantly less accurate than the Gompertz and Bass curves, a damped-trend model and an equal-weight combination; the damped trend was the most accurate single method for both technologies. All S-curves under-forecast: five years ahead, actual generation exceeded the logistic forecast in 73% (wind) and 77% (solar) of cases. Ceilings were unstable. The median ceiling estimated with data to 2024 was 2.4 times (wind) and 11.1 times (solar) the ceiling estimated with data to 2014, 83% and 67% of the 2014 ceilings had been exceeded by 2024, and up to a third of fits ended on the bounds of the parameter space. The findings indicate that the carrying capacity of renewable electricity diffusion is policy- and cost-dependent rather than a fixed property of national systems. Planners should treat S-curve ceilings as scenarios rather than forecasts, combine growth curves with damped trends and re-estimate ceilings every year.

Keywords: Technology diffusion; Logistic growth; Renewable electricity; Forecast evaluation; Damped trend; Carrying capacity; Energy transition

The emission savings from electrifying transport, heating and industry depend on the carbon intensity of the electricity that replaces fossil fuels. This paper asks whether national power systems are converging towards low carbon intensity or forming persistent clubs. Applying the Phillips–Sul log t test and club clustering algorithm to the carbon intensity of electricity generation in 191 countries over 2000–2024, I reject convergence of the full panel (t = −52.9) and identify three convergence clubs. A low-carbon club of 19 hydro-, nuclear- and geothermal-based systems fell from a mean of 105 to 43 gCO₂/kWh; a transition club of 31 countries, mostly European and Latin American, fell from 290 to 155 gCO₂/kWh; and a high-carbon club of 141 countries, including China, India, the United States, Germany, Japan and Türkiye, declined by only 8% to 531 gCO₂/kWh. Eleven countries that started below 200 gCO₂/kWh joined the high-carbon club by adding fossil capacity. The partition is robust to the filtering parameter, the truncation fraction, the merging rule and the exclusion of small economies. An ordered logit shows that club membership is predicted by initial hydro and nuclear shares and income, but not by the initial coal share once non-fossil shares are controlled for. Replacing a gas boiler with a heat pump cuts CO₂ by a median of 95% and 79% in the low-carbon and transition clubs but only 18% in the high-carbon club, which illustrates how club membership conditions the payoff of electrification policies.

Keywords: Club convergence; Carbon intensity of electricity; Phillips–Sul test; Energy transition; Electrification; Ordered logit

From December 2023, attacks on merchant ships in the Red Sea diverted much of the Asia–Europe trade around the Cape of Good Hope. This paper estimates the effect of the diversions on CO₂ emissions from shipping, using the OECD's monthly estimates that allocate emissions to the economy of residence of the ship operator for 140 economies between January 2019 and June 2026. For each economy and for regional and world aggregates, I construct counterfactual paths with exponential smoothing and a combination of statistical models estimated on pre-crisis data, and I benchmark the crisis-period gap against the gaps at 13 placebo origins before the crisis. In the first twelve months, emissions of the 140 economies exceeded the counterfactual by 50 Mt (5.6%), about ten times the average placebo gap; the excess persisted through mid-2026. The effect was concentrated: operators resident in Europe and Asia accounted for 45.5 of the 48.5 Mt of economy-level excess, whereas operators in the Americas and Oceania showed no increase. Among the 20 largest emitters, the excess was large and significant for economies that host major container lines, such as Switzerland, France, Denmark, Singapore and Germany (11–23%), and small for tanker- and bulk-dominated fleets such as Greece, Norway and Japan. In emissions-weighted regressions, the excess rose with the container share of the national fleet. The results quantify a climate cost of chokepoint disruption that falls on a few carriers' home economies, with implications for the EU ETS and for maritime decarbonisation policy.

Keywords: Red Sea crisis; Shipping emissions; Rerouting; Counterfactual forecasting; Placebo inference; Container shipping; Maritime chokepoints

Adjusted net savings (ANS), or genuine savings, measure whether a country's total wealth, including natural capital, is rising or falling, and persistently negative values signal weak unsustainability. This paper studies how the whole cross-country distribution of ANS evolves, rather than its mean. Using World Bank data for 160 countries from 1990 to 2021, I estimate annual and five-year Markov transition matrices over six ANS classes, compute ergodic distributions, mobility indices and mean first-passage times, test homogeneity over time and across resource dependence, and estimate stochastic kernels. The share of countries with negative ANS fell from 32% in 1990 to 14% in 2019, but negative savings are persistent: a country below −5% of GNI remains there for five years with probability 0.54 and needs about 14 years on average to reach non-negative savings. The ergodic distribution implies that 12–16% of countries would remain negative savers in the long run. The annual chain understates five-year persistence, so the process is not first-order Markov at the annual frequency. Resource dependence changes the dynamics fundamentally (likelihood-ratio p < 10⁻⁶): among countries with natural resource rents of at least 10% of GDP, the ergodic share of negative savers is 25–27%, against 9–13% for other countries, and the distribution is polarised between high-saving and depleting economies. These patterns, robust to rent thresholds, a balanced panel and the exclusion of pandemic years, identify a resource-linked trap in weak sustainability and support Hartwick-type rules for reinvesting resource rents.

Keywords: Adjusted net savings; Genuine savings; Weak sustainability; Distribution dynamics; Markov chains; Natural resource rents; Resource curse

Cross-country studies often find that carbon taxes reduce national CO₂ emissions substantially, but adopters may already be on declining emission paths. This paper uses the 2026 release of the EDGAR database (1980–2025) for 156 countries, 26 of which adopted a carbon tax between 1990 and 2024, and compares ten estimators for staggered adoption under a common falsification test that moves each adoption three years earlier and uses only pre-adoption data. Estimators that rely on parallel trends, including heterogeneity-robust difference-in-differences, imputation, local-projection and synthetic difference-in-differences estimators, imply that adoption reduced emissions by 16–29 log points on average over the first decade, but all of them also find reductions of 6–38 log points in the three years before adoption. Adopters' emissions were already falling relative to those of other countries. Only the two estimators that model country-specific trajectories, cohort-specific linear trends and an interactive fixed effects counterfactual, pass the test; they yield effects of +2.6 (95% CI −4.0 to 11.2) and −2.8 (−17.9 to 15.5) log points. The interactive fixed effects estimator also passes a placebo-outcome test on agricultural methane, which the linear-trend estimator fails. Neither sectoral nor price-group estimates show robust reductions, and country-level effects do not increase with the tax rate. At the rates adopted so far, a median of US$4 per tonne in the first decade, carbon taxes have not detectably bent national trajectories beyond pre-existing trends. The results caution against inferring policy effectiveness from parallel-trends estimates and support pricing at levels and coverage that can shift trajectories.

Keywords: Carbon tax; CO₂ emissions; Staggered difference-in-differences; Interactive fixed effects; Synthetic difference-in-differences; Falsification test; Policy evaluation

Low-wind weeks are a central risk for power systems that rely on offshore wind. This paper studies how offshore wind capacity should be allocated across sites in European seas to limit that risk. Daily capacity factors are simulated for a 15-megawatt reference turbine at 30 existing or planned offshore sites from the North Sea to the Black Sea, using reanalysis winds from January 1981 to October 2026. Capacity shares are chosen to maximise the mean output of the worst 5% of weeks, either for the empirical distribution or for the worst distribution within a Wasserstein ball, solved as an exact linear programme on bounded support. I show that, for small radii, robustification adds a penalty proportional to the largest capacity share and therefore acts as a soft cap on concentration. In a rolling out-of-sample evaluation over 1,347 weeks from 2001 to 2026, the risk-based allocations raise the lower-tail output by 14–19% relative to equal shares, 25–30% relative to minimum variance and 50–57% relative to the single best site, while also increasing mean output relative to the first two. They combine the North Sea and the Baltic with Atlantic sites whose winds peak in summer, such as the Canary Islands, and with Aegean sites driven by the summer Etesian winds. Robustness improves the lower tail slightly when only three to five years of data are available, and cross-validation selects positive radii far less often when twenty years are available. The results quantify the value of cross-basin planning for offshore wind.

Keywords: OR in energy; Distributionally robust optimization; Conditional value-at-risk; Offshore wind; Portfolio selection

This study compares two shocks to European tourism demand: the loss of Russian visitors after February 2022 and extreme summer heat. Using monthly nights spent in 30 countries to August 2026, origin-specific annual data and daily temperatures at 80 destination points, it combines continuous-exposure difference-in-differences with wild cluster bootstrap inference, synthetic difference-in-differences for the most exposed destinations and an accounting decomposition. Russian nights in the EU fell by 85% and did not recover, yet destinations recovered unevenly: other markets replaced the lost demand in Cyprus, Finland and Lithuania by 2025, whereas Latvia and Estonia remained 23–36% below their synthetic counterparts. Ten additional days above 32 °C in a month had no detectable effect on nights spent. Diversion to Türkiye and Serbia shows that the source market was redirected rather than lost. Source-market concentration, not heat, was the binding vulnerability.

Keywords: Tourism demand; Destination resilience; Source markets; Geopolitical shocks; Extreme heat; Synthetic difference-in-differences; Europe

Time series foundation models produce probabilistic forecasts without training and could simplify forecasting in decision support systems, but independent evidence on whether they improve decisions is scarce. This study evaluates four foundation models (Chronos-Bolt in two sizes, Chronos-2 and TimesFM 2.5) against five statistical methods on 470 monthly series: weather at 60 ports (wind, temperature, precipitation and solar irradiance, to September 2026) and fossil CO₂ emissions by country and sector (to August 2026). In a rolling-origin design with twelve-month horizons, 3,302 series–origin pairs are evaluated on point accuracy, quantile loss, interval calibration and the expected cost of newsvendor-type decisions, with rank-based tests, conformal recalibration and a robustness check on targets that postdate all model releases. The foundation models ranked first in both families. Chronos-2 had the lowest scaled quantile loss for port weather, 5% below exponential smoothing and 9% below climatology, and TimesFM 2.5 had the lowest loss for emissions, 9% below the seasonal naive method, which nevertheless gave the most accurate point forecasts. Gains over the best statistical method were small for wind, precipitation and solar irradiance, but the ranking persisted for post-release targets. Raw 80% intervals were miscalibrated for several methods, including Chronos-2 for weather (74% coverage); split conformal recalibration restored coverage to 78–82% for every method at almost no cost in sharpness. The best method depended on the cost asymmetry of the decision. For decision support systems, foundation models are a strong default when combined with conformal recalibration, simple benchmarks as monitors and data-quality screening.

Keywords: Time series foundation models; Probabilistic forecasting; Conformal prediction; Forecast evaluation; Decision costs; Port operations; Carbon emissions

Composite indices of national energy-transition performance inform investors and policymakers, yet their rankings depend on contestable weights. This paper evaluates 84 countries on seven criteria for 2025, the latest year with complete data: carbon dioxide emissions per capita, their change since 2015, emissions per unit of output, the carbon intensity of electricity, the share of wind and solar power, its change since 2015 and the effective carbon tax rate. All analyses use one common weight polytope in which each criterion carries between 5% and 35% of the weight. On it I compute optimistic and pessimistic benefit-of-the-doubt scores by linear programming and the full rank distribution by stochastic multicriteria acceptability analysis with 100,000 hit-and-run weight vectors. Six countries are in the top ten under 95% of admissible weights and six in the bottom ten, but the average country moves by 26 positions across the polytope. I show that the holistic acceptability index with linear meta-weights is an affine function of the expected rank; in the data it reproduces the equal-weight ranking, whereas optimistic scores reorder the middle of the distribution. For Türkiye, ranked between 33rd and 68th, a carbon price of US$70 per tonne would raise the expected rank from 48 to 25, against 38 for ten more percentage points of wind and solar power; with both and a 25% cleaner power mix it would rank in the top quartile under 83% of admissible weights. Rank ranges and acceptabilities, rather than single ranks, give a defensible basis for benchmarking.

Keywords: Multiple criteria analysis; Stochastic multicriteria acceptability analysis; Benefit of the doubt; Composite indicators; Energy transition; Hit-and-run sampling; Robustness

Ports are exposed to wind, heavy rain and heat, but the size and persistence of weather-related losses in port activity are rarely measured at scale. This study combines daily port calls and cargo volumes estimated from vessel tracking data for 434 ports from January 2019 to October 2026 with daily weather at each port. Poisson models with port-by-month, date and port-by-weekday fixed effects identify the effects of weather from day-to-day variation within ports. Port calls fell by 35% on days with maximum winds of at least 14 m/s and by 18% on days with at least 50 mm of precipitation; cargo volumes fell more. Extreme heat did not reduce activity. Activity dropped already on the day before strong winds and recovered only partly afterwards: over the window from two days before to ten days after a strong-wind day, the cumulative effect was as large as the same-day effect, so the lost calls were not made up within ten days. Dry bulk carriers were the most and roll-on/roll-off ships the least sensitive. Ports that rarely experience strong winds lost about twice as much activity per windy day as ports where such days are common, which points to adaptation. Strong winds and heavy rain removed 0.66% and 0.16% of port calls in the sample, with up to 5% at exposed ports in North-Western Europe and East Asia. Weather risk to ports is concentrated, persistent and partly adaptable, which matters for adaptation planning and for the emissions of ships waiting at anchor.

Keywords: Ports; Extreme weather; Climate adaptation; AIS data; Port calls; Poisson regression; Empirical Bayes

The EU Carbon Border Adjustment Mechanism (CBAM) entered its definitive period in January 2026. This paper provides early evidence on its trade effects using monthly extra-EU imports of 231 HS4 products from 171 partners, January 2019 to August 2026. Poisson triple-difference models compare CBAM goods with related goods and liable partners with exempt EFTA partners; event studies and synthetic difference-in-differences trace the dynamics. Imports of CBAM goods from liable partners rose by 36% in the last quarter of 2025, before the definitive period, and fell by 25% in January–June 2026, most strongly in fertilisers and aluminium. By July–August 2026 the difference had disappeared, and the net effect from October 2025 to August 2026 is small and insignificant. Among liable partners, aluminium imports shifted away from countries with carbon-intensive electricity, whereas sourcing of steel, fertilisers and cement did not respond to carbon intensity. The early effects of CBAM are therefore dominated by the timing of imports and by selective re-sourcing, not by a general retreat from carbon-intensive suppliers.

Keywords: Carbon border adjustment; CBAM; Carbon leakage; International trade; Synthetic difference-in-differences; Poisson pseudo-maximum-likelihood

Disruptions at maritime chokepoints change the lead times of global supply chains for months or years. This study measures these regimes with vessel tracking data and evaluates how inventory policies should respond. Weekly transits at six chokepoints from January 2019 to November 2026 are analysed with Bayesian online change-point detection and hidden Markov models. Route-specific lead times for three trade lanes are derived from sea-route distances and used in a base-stock model with stochastic lead times, simulated over the observed regime path. The diversion of container shipping from the Suez Canal to the Cape of Good Hope was detected within one to two weeks and formed a regime with an expected duration of 137 weeks; disruptions at the Panama Canal, the Strait of Hormuz and the Bosporus followed. The diversion raised the required base stock by 15% on the Asia–North Europe lane and by 36% on the Asia–Mediterranean lane, and a policy calibrated on pre-crisis lead times lowered fill rates to 94% and 86%. Because the base-stock level is concave in the share of diverted ships, partial diversion required most of the inventory of full diversion. Updating the policy with the route mix observed in tracking data reduced costs at the onset by 5% and 33% relative to learning from realised lead times, and filtering the data through regime probabilities avoided overreaction in normal times. Chokepoint disruptions should be managed as persistent changes in lead-time regimes.

Keywords: Supply chain disruption; Stochastic lead time; Base-stock policy; Hidden Markov model; Change-point detection; Maritime chokepoints; AIS

Hydropower supplies about a seventh of global electricity, and its output varies with precipitation. When droughts reduce hydropower, other sources must fill the gap, often fossil plants. This study quantifies this chain for 57 countries with a hydropower share of at least 5%, using monthly electricity data to October 2026 and standardised precipitation indices weighted by the location of hydropower capacity. A Bayesian hierarchical model estimates country-specific responses of hydropower to precipitation anomalies at three- and twelve-month scales; instrumental-variable regressions with a precipitation-predicted instrument estimate how the shortfall is replaced. A one-standard-deviation drought on both scales lowers hydropower by 10% in the median country (10–90% range across countries 2–23%). Across all countries, each terawatt-hour of precipitation-driven shortfall is replaced by 0.41 TWh of fossil generation and 0.66 TWh of net imports; in Asia, Africa and Oceania fossil plants replace it almost one-for-one, whereas European countries import most of it. Precipitation-driven shortfalls added 127 Mt of CO₂ in 2023, mainly in China, whereas wet years after 2023 avoided emissions. Lagged precipitation adds little to forecasts beyond the persistence of hydropower anomalies.

Keywords: Hydropower; Drought; Standardised precipitation index; Bayesian hierarchical model; Instrumental variables; Power-sector emissions

Business bankruptcies in the European Union rose well above pre-pandemic levels after 2022, a period of record energy prices. Using quarterly bankruptcy and registration indices for eight sectors in 19 European countries from 2015 to mid-2026, this study tests whether energy-intensive sectors failed more often when energy prices rose. Local projections with a shift-share exposure and country-by-quarter and sector-by-quarter fixed effects find no differential bankruptcy response over two years; the estimates rule out effects larger than about 10% for a sector one standard deviation more energy-intensive during the 2022 shock. Country differences explain 77% of the variation in post-2022 bankruptcies across country-sectors, sectors only 5%. Energy-intensive sectors saw fewer business registrations after price increases, although this result is sensitive to the pandemic years. Europe's bankruptcy wave is therefore a country phenomenon rather than an energy shock.

Keywords: Business failure; Bankruptcy; Energy prices; Firm entry; Local projections; Europe

Global mean sea level is the average of the ocean surface; seaborne trade, by contrast, is concentrated at a small number of coastal sites, often on subsiding deltas and reclaimed land. This study matches port-level cargo volumes derived from ship tracking with tide-gauge records, satellite altimetry and GNSS vertical velocities to measure the relative sea-level trend experienced by world port trade and to establish how much of that measurement rests on observation. The 2019–2025 mean cargo volume of 2,015 ports is used as the weight. Only 42.8% of port trade lies within 100 km of a tide gauge covering 1993–2024; the share is 73% in high-income and 15% in upper-middle-income countries, and 17 of the thirty largest ports are not covered. At ports observed by tide gauges the trade-weighted relative trend is 3.91 mm/yr (95% confidence interval 3.51–4.35), 21% above the global mean (3.23 mm/yr). About two thirds of the excess comes from land subsidence and the rest from ports being located in sea areas rising faster than average; 66% of the excess comes from ten ports. A composite estimate extended with altimetry and GNSS covers 99% of trade and gives 3.77 mm/yr; however, land motion is unknown for 45% of trade, and each 1 mm/yr of subsidence at these ports raises the global trade-weighted trend by 0.46 mm/yr. The findings indicate that the main gap in port adaptation planning is not sea-level projection but the observation of land motion at port scale.

Keywords: Relative sea level; Ports; Land subsidence; Tide gauge; Satellite altimetry; GNSS; Observation gap; Seaborne trade

Argo profiling floats are the main source of knowledge on the heat and salt content of the ocean, and the continuity of the array depends entirely on annual deployments. This study treats the sustainability of the array as a reliability and renewal problem. The lifetimes of 20,619 floats deployed between 2000 and 2026 are extracted from the profile index of the global data assembly centre; survival is analysed with Kaplan–Meier, Weibull and Cox models, the monthly coverage of the 3°×3° design cells is compared with random and perfect placement benchmarks, and fleet size is projected with a renewal model. Median lifetime rose from 3.0 years for 2000–2004 deployments to 5.0 years for 2015–2019 deployments, whereas the loss rate within the first year has stayed in the 10–13% range for twenty years. The fleet has been around four thousand floats since 2016, but the share of design cells between 60°S and 60°N occupied by a float at any one time has levelled off at 65%, and the share with at least one profile in a month at 75%. Snapshot coverage is below the value expected if the same number of floats were placed at random (69%); placement efficiency has been negative since 2005 (−0.10), i.e. the array is slightly more clustered than a random placement. 58% of active floats are handled by a single data centre. The deployment rate of the last three years (781 floats per year) is roughly enough to maintain the present fleet; the 4,700-float target requires 838–913 deployments per year. A 20% drop in deployments would reduce the fleet to 3,727 floats in eight years. The findings indicate that the way to raise coverage is the management of deployment locations and the reduction of early losses rather than fleet size.

Keywords: Argo; Ocean observing system; Survival analysis; Weibull; Renewal model; Observation coverage; Fleet sustainability

The debate on whether port systems concentrate rests on the question of whether growth depends on port size. This study tests three regularities from the literature on firm and city growth for world ports: the independence of growth from size (Gibrat's law), the scaling of growth volatility with size, and the distribution of growth rates. Using daily estimates of port calls and cargo derived from satellite-based ship tracking, 5.9 million port-days at 2,065 ports over 2019–2025 are examined. To remove the spurious mean reversion caused by measurement error in size, days are split into two interleaved half-samples; size is measured from one half and growth from the other. The elasticity of growth with respect to size is −0.008 (standard error 0.002): a port ten times larger grows about 1.9 percentage points more slowly per year. The departure is statistically significant but small; it is two fifths of the value found when size and growth are measured from the same half-sample (−0.021), the remainder being measurement error. The standard deviation of growth rates declines with size with an exponent of −0.23 (95% confidence interval 0.21–0.25), clearly below the value of 0.5 expected for a sum of independent units and close to the range reported for firms. Growth rates follow a fat-tailed distribution close to the Laplace rather than the normal. The size distribution is close to lognormal in the body and thinner than lognormal among the largest ports. Ranks are highly persistent: 88% of the hundred largest ports of 2019 are still in the top hundred in 2025. The findings indicate that ports share the statistical growth regularities of firms and that there is no marked tendency towards concentration or deconcentration in the global port system.

Keywords: Port growth; Gibrat's law; Volatility scaling; Laplace distribution; Size distribution; Port concentration; Ship-tracking data

Daily data on port activity are increasingly used for the real-time monitoring of trade, yet how these data vary with the calendar is not systematically known. Using daily port calls derived from satellite-based ship tracking, this study examines 5.9 million port-days at 2,065 ports between 1 January 2019 and 31 October 2026. The weekly rhythm is measured with indices relative to the weekly mean; the effect of holidays is measured with Poisson event-study models that control for day of week, year and seasonality, and with a local-baseline estimator that does not depend on the seasonal model. Worldwide, the number of calls on weekend days is 9.2% lower than on weekdays (95% confidence interval 8.9–9.5); Mondays are 7.3% above the weekly mean. The effect is strongest for general cargo ships and weakest for container ships; it is about twice as large at ports in high-income countries as in the other income groups and has weakened slowly over seven years. The local weekend is decisive: in countries where Friday is a day of rest the dip shifts to Friday, and at Israeli ports Friday calls are 45% below the mean. Lunar New Year reduces calls at Chinese ports by 13% on average over two weeks and container exports by 28% in the following two weeks; about four weeks later this gap appears as a 25% fall in container imports at US West Coast ports. Calls fall by 34% in Europe on Christmas Day and by 68% in Japan at New Year; Eid al-Fitr leads to a 31% fall in Indonesia, whereas no significant effect is found in the Gulf countries. Day of week and holidays explain 64% of the non-seasonal variability in world daily port calls. The findings indicate that daily port data should not be interpreted without removing calendar effects.

Keywords: Port calls; Weekend effect; Calendar effects; Lunar New Year; Ship-tracking data; Real-time trade monitoring; Event study

Whether coral reefs can recover after mass bleaching events depends on the time between two events. This study treats the recurrence of heat stress as an event-history problem. Using the daily degree heating week (DHW) series of the 214 regional virtual stations of the US National Oceanic and Atmospheric Administration's Coral Reef Watch programme, 8,560 station-years between 1986 and 2025 are examined. The annual prevalence of bleaching-level (DHW ≥ 4 °C-weeks) and severe (DHW ≥ 8 °C-weeks) events, the time between events and the stress-free recovery windows are estimated with the Kaplan–Meier method; the trend in event probability is estimated with logistic models. The annual share of stations experiencing bleaching-level heat stress rose from 15% in 1986–1995 to 60% in 2016–2025; the share with severe stress rose from 3.6% to 37%. The median time from a severe event to the next fell from 7 years for intervals starting in 1986–1999 to 2 years for those starting in 2010–2019. In 2016–2024 the probability that a station with severe stress experienced severe stress again in the following year was 49%. Between 2023 and 2025, 91% of stations experienced severe stress at least once and 66% reached their record value in these three years. A sensitivity analysis with DHW reconstructed from daily hotspot series shows that thermal tolerance would have to rise by 0.16–0.20 °C per decade to keep event frequency at its 1986–1995 level; this is about twice the tolerance gain reported in field studies. The findings indicate that, in terms of the frequency of heat stress, the recovery window is closing in most reef regions.

Keywords: Coral bleaching; Degree heating week; Return interval; Survival analysis; Recovery window; Thermal tolerance; Marine heatwaves

Year-to-year variability in marine catches directly affects food security and livelihoods in countries that depend on fisheries. It is widely expected that a catch relying on many species will be more stable than one relying on a single species, yet the size and source of this portfolio effect at national scale have not been measured systematically. Using the global capture database of the Food and Agriculture Organization of the United Nations, this study examines the marine catch of 168 countries over three twenty-five-year periods between 1950 and 2024 (403 country-periods). The variability of the total catch is decomposed into the product of the weighted mean variability of taxa and the synchrony among taxa; a second portfolio measure that accounts for mean–variance scaling is computed, and the determinants of variability are examined by regression. In the median country the variability of the total catch is 61% lower than the weighted mean variability of its taxa. The source of this effect is not compensation among species: the median of the mean correlation among taxa is 0.003 and indistinguishable from zero; the portfolio effect arises from the statistical averaging of independently fluctuating taxa. A doubling of the effective number of taxa reduces synchrony by 24% but raises weighted variability by 13%, because small taxa are relatively more variable; the net effect is a 13% reduction in total variability. The effect of catch size is stronger: a catch ten times larger is 29% less variable. The share of small pelagic fish raises variability; a single species, the Peruvian anchoveta, accounts for about 71% of the variance of annual changes in the world total catch. Following the same countries, the portfolio effect rose from 57% in 1950–1974 to 69% in 2000–2024 and the variability of total catch declined. The findings indicate that diversification provides stability, but that its effect is limited and the decisive factors are the scale and composition of the catch.

Keywords: Portfolio effect; Catch variability; Species diversity; Synchrony; Taylor's law; Small pelagic fish; Food security

Port call counts derived from ship-tracking data are increasingly used as a real-time indicator of seaborne trade. This use rests on the assumption that the number of calls moves together with the cargo carried. In this study the assumption is tested with common factors extracted from the monthly activity of world ports. Daily port records are aggregated to monthly totals, global factors are estimated from the year-on-year growth rates of 1,361 ports for January 2020–October 2026, and the significance of the factors is tested against a null distribution that shifts each port's series randomly in time. The assumption holds for non-container ships: the correlation between the growth of calls and of tonnage in the world total is 0.87. It does not hold for container ships: the same correlation is 0.23, and the global factor extracted from container calls is the mirror image of the global factor extracted from container tonnage (correlation −0.81). In 2021 container calls changed by −2.3% while tonnage rose by 4.7%; in 2023 calls rose by 8.7% while tonnage rose by 2.4%. The container call factor is strongly and negatively related to the supply chain pressure index (−0.62) and lags the general port cycle by about a year. The finding shows that in container shipping the number of calls measures the throughput capacity of the system rather than demand. The study also presents a hierarchical variance decomposition of port activity: the global factor explains only 4.2% of variability at the median port, but 49% in the world total.

Keywords: Port calls; Container shipping; Port congestion; Common factor; Real-time trade indicators; Variance decomposition

It is widely accepted that climate modes broadly affect fisheries, and the literature is full of significant relationships found between individual stocks and climate indices. This study asks the question in reverse: what remains when the same method is applied to all countries and all major climate modes? The detrended marine catch of 146 countries from the FAO capture database is related to eleven climate indices (three measures of ENSO, PDO, NP, PNA, NAO, AMO, two tropical Atlantic indices and the Indian Ocean Dipole); significance is tested against a null distribution obtained by shifting the indices in time, which preserves autocorrelation. Three results stand out. First, the footprint of no climate mode in national catches is significant at the field level: for the strongest candidate, the eastern Pacific El Niño index, a significant relationship is found in 17 countries, whereas 12 are expected under the null distribution (p = 0.09). Second, a naive screening shows 47% of countries to be sensitive to at least one climate mode; this share does not differ from that expected by chance (45% expected, p = 0.31), and only 4 of 1,606 tests survive control for multiple testing. For autocorrelated indices, conventional tests raise the false-positive rate from five to eleven per cent. Third, the footprint of ENSO in the world total catch belongs to the eastern Pacific type only: a one-standard-deviation increase in the eastern Pacific index reduces the world catch by 1.9% (t = −3.2), whereas the effect of the central Pacific index cannot be distinguished from zero (t = 0.7). This footprint comes entirely from the Peruvian anchoveta. Together the eleven modes explain only 1.7 percentage points more of catch variability in the median country than expected by chance. The results limit the assumption that climate modes can be used in national catch forecasts and show that stock-level climate–catch relationships are a selected sample.

Keywords: Climate modes; El Niño–Southern Oscillation; Marine catch; Field significance; Multiple testing; Eastern and central Pacific El Niño; Statistical power

The prediction that mass coral bleaching will cause reef fisheries to collapse is one of the main arguments in the debate on reef conservation; yet this prediction has not been tested at global scale with realised catch data. In this study the test is carried out with two independent catch data sets: national catches reported to FAO (51 countries, 1986–2024) and the reconstructed catches of the Sea Around Us initiative, which include subsistence and small-scale fisheries (95 exclusive economic zones, 1986–2019). Both data sets are matched with the heat-stress records of the regional stations of NOAA Coral Reef Watch, and the change in the catch of reef-associated fish over the ten years following heat-stress years is estimated by local projections with unit and year fixed effects. The two data sets give the same result: there is no decline in catch after heat stress. In reported catches the estimated change five years after severe heat stress is −4% (95% confidence interval −13% to 7%); in reconstructed catches it is 5% (−6% to 18%). In the reconstructed data, catches rose by 7% on average three to five years after moderate heat stress (permutation p = 0.01); however, catches of other species in the same zones also rose and the share of reef fish did not change, so the increase is not specific to reefs. Pre-event trends are zero in both data sets. The negative sign seen after repeated events in the reported data is not replicated in the reconstructed data. Over the five years following severe heat stress, the reconstructed data rule out a decline in reef fish catch of more than seven per cent and the reported data one of more than thirteen per cent. Catches continued to rise over the same period. The result indicates that the effect of bleaching on fisheries should be sought in catch composition, effort and at local scale rather than in total catch, and that catch volume cannot be used as an indicator of reef health.

Keywords: Coral bleaching; Degree heating week; Reef fisheries; Reconstructed catch; Local projection; Statistical power

It is often argued that the growth in ship size increases peak loads at ports, but this claim has not been tested on a global sample of ports and against a theoretical benchmark. The compound Poisson model, in which ships arrive at random intervals and carry cargo of random size, gives a clear prediction: the elasticity of the coefficient of variation of daily cargo volume should be −0.5 with respect to cargo volume and +0.5 with respect to cargo per call. In this study the prediction is tested with daily port records derived from satellite-based ship tracking, in an annual panel of 1,264 ports for 2019–2025 with port and year fixed effects. The estimated elasticities are −0.49 (standard error 0.015) and 0.43 (0.035). The median ratio of the variance to the mean of the daily number of calls is 0.93, so ship arrivals cannot be distinguished from a Poisson process; the excess variance comes from the inequality of call sizes. The counter-intuitive implication of the model is also confirmed: with the number of calls held constant, an increase in call size does not affect bunching (elasticity −0.06); bunching rises only when the same cargo arrives in fewer calls. In the world total, cargo per call rose by 7.2% between 2019 and 2025 and accounted for between 49% and 75% of the growth in cargo volume, depending on the end year; yet only 29% of ports fit the pattern of fewer and larger calls. The growth in call size raised daily bunching by 1.8% at the median port and by more than five per cent at 37% of ports, while in the aggregate it was offset by the smoothing effect of the growth in cargo volume. The peak-load effect of large ships is real, can be computed with a simple rule and is concentrated at ports whose calls grow without an increase in cargo volume.

Keywords: Ship size; Peak load; Compound Poisson; Port calls; Cargo per call; Port planning

The tropical cyclone risk of ports is known mostly from major damage events and from modelled hazard maps; how much activity is lost in an ordinary encounter, when a storm passes close to a port, has not been measured at global scale. In this study the storm tracks of the international best track archive for 1980–2025 are interpolated to hourly resolution and matched with 2,065 ports. A total of 13,061 encounters are identified in which a system of at least tropical-storm strength came within 100 km of a port; 44% of ports and 55% of cargo volume were affected at least once in forty-six years. Cargo-weighted exposure (1.89 encounters per decade) is above the port average (1.37). No statistically significant trend is found in the number of encounters, in hurricane-force encounters or in the mean latitude of encounters. An event study of 1,321 encounters in 2019–2026, built on daily port calls derived from satellite-based ship tracking, shows that calls fall to 28% of their normal level on the day of closest approach, and to 9% in hurricane-force storms. Activity returns to normal within two days; about one third of the loss of the first four days is recovered in the following five. Over a seventeen-day window the net loss averages 0.85 port-days (95% confidence interval 0.57–1.15): 0.66 in tropical storms, 1.16 in hurricanes and 2.03 in major hurricanes. The distribution of losses is skewed: 10% of events cause a loss of more than five days and account for 42% of the total loss. Combining exposure with the loss per event gives an expected annual loss of 0.13 days at the average port and 0.17 days weighted by cargo; it exceeds one day per year at only 24 ports. The findings indicate that port risk is concentrated in rare and severe events and not in frequent short interruptions.

Keywords: Tropical cyclone; Port disruption; Event study; Port calls; Climate risk; Resilience

Global compilations of marine microplastic measurements are used to map contamination and infer trends, yet they pool different sampling devices and reporting conventions. We examined the 15,534 water samples (56 studies) of the NOAA NCEI Marine Microplastics archive. First, we propose a consistency check: the reciprocal of the smallest non-zero concentration in a study is the minimum sample volume that could have produced it. For two studies based on one-litre grab or bottle samples (1,470 records, 9.5% of water samples) the implied volume is about 1,000 m³, and the archived values, multiplied by 1,000, reproduce the per-litre statistics of the source publications. The archived values of these records are therefore lower than the published concentrations by a factor of 10⁶. Second, co-located sample pairs show that net tows from different studies differ by a median factor of 2.5, indistinguishable from pairs within one study (2.2), whereas pairs from different method classes differ by a factor of 221. Third, the pooled archive suggests a 3.4-fold increase per decade in net-tow concentrations, but 80% of this slope reflects the changing mix of studies; within studies the change is 1.3-fold (95% CI 0.9–1.7) for the geometric mean of non-zero tows and 2.0-fold (0.9–3.1) for the arithmetic mean. The archive can neither demonstrate nor exclude a global increase, and records should be screened for unit consistency before reuse.

Keywords: Microplastics; Data quality; Harmonisation; Sampling methods; Temporal trend; Open data

Non-parametric richness estimators are widely used to judge how complete regional species inventories are, but their predictions are rarely confronted with what is later found. We used the Ocean Biodiversity Information System to reconstruct, for 66 large marine ecosystems and 100 IHO sea areas, the animal species inventory as it stood with records collected to 1999, 2009 and 2019, and compared estimates made from the earlier inventories with the species actually recorded by 2026. Inventories built from records to 1999 appeared 75% complete by the Chao1 estimator (median). By 2026 the number of recorded species exceeded the 1999 estimate in 76% of areas and its upper 95% confidence bound in 64%; the 1999 inventories held at most 66% of the species now known. Four alternative estimators were exceeded in 52–79% of areas. When the 2026 inventories were randomly thinned to the 1999 number of records, Chao1 was exceeded in 55% of areas, so the excess failure reflects how records accumulate through new places, methods and taxa, not sample size. The estimators still carried information: predicted and realised increases were correlated (Spearman 0.36), and a recalibrated model combining the estimate with record growth explained 51% of ten-year inventory growth out of sample, against a negative R² for the uncorrected estimate. Current estimates suggest that inventories are 80% complete in the median large marine ecosystem; such figures should be read as upper bounds and used for ranking, not as absolute measures of what remains to be found.

Keywords: Species richness estimation; Inventory completeness; Chao1; OBIS; Large marine ecosystems; Knowledge shortfalls

Seafood is among the most heavily traded foods, and import dependence is often described as a source of vulnerability to production failures abroad. How much of the supply lost in such failures importers actually replace has not been measured across the trade as a whole. We use bilateral trade in 105 seafood products between 1995 and 2023 to identify 4,811 episodes in which an exporter's shipments of a product fell by at least half, and follow 116,512 importer–product markets for up to five years. Replacement depends sharply on how much of the market the failing exporter supplied. Where it held 25–50% of pre-shock imports, 84% of the lost quantity was made up from other sources within a year and total imports returned to baseline. Where it held 50–75% the replacement rate was 37%, and above 75% it was 10%; in these markets imports were still 28% and 60% below baseline three years later and unit values rose by up to 31%. Replacement came from suppliers already present in the market, not from new ones, and the number of alternative suppliers had no effect once the share lost was accounted for. More than half of all importer–product markets obtain at least 50% of their supply from one exporter. The results indicate that the exposure that matters is the share held by the largest supplier, and that it is common.

Keywords: Seafood trade; Supply shocks; Import dependence; Diversification; Trade resilience; BACI

Since 1 May 2025 ships in the Mediterranean Sea must use fuel with at most 0.10% sulphur. Model studies made before the decision projected lower sulphur dioxide (SO₂) in coastal cities, but no assessment based on measurements made after the rule took effect is available. We combined hourly SO₂ from the European air quality reporting system for 2016–2026 with hourly winds from nearby meteorological stations at 330 sampling points located within 5 km of 154 ports, and computed for each point and month the port-sector excess: mean SO₂ when the wind blows from the port minus the mean when it blows from the opposite sectors. At Mediterranean monitors the excess fell from 1.45 µg m⁻³ in 2016 to 0.88 in 2019 and 0.35 in 2020, the year the global 0.50% cap began, and stood at 0.40 in 2024. After the emission control area it was 0.33 in 2025 and 0.30 in 2026. Relative to ports in the North Sea and Baltic, where the limit did not change, the change after May 2025 was -0.11 µg m⁻³ (95% confidence interval -0.54 to 0.31), against -0.57 (-1.28 to 0.14) for the 2020 cap. A comparison of mean concentrations at port and inland monitors, the design most often used for such evaluations, returned no effect for 2025 and for 2020 an apparent effect confined to industrial sites in one country. Little ship-related SO₂ remained to be removed at the monitors by 2025, and the fixed network can only bound the effect of the new limit at about half a microgram per cubic metre.

Keywords: Ship emissions; Sulphur dioxide; Emission control area; Mediterranean Sea; IMO 2020; Wind-sector analysis

The Mediterranean and the Black Sea are among the fastest-warming marine regions, yet analyses that combine temperature, heat content and the carbonate system are rarely published in a form that others can re-run. We analysed daily NOAA OISST v2.1 sea surface temperature (SST) for four boxes (Eastern Mediterranean, Aegean, Black Sea and Western Mediterranean), Copernicus Marine ocean heat content (0–700 m) and SOCAT surface fCO₂ for the Mediterranean, all retrieved from the open GOMD platform through a notebook that stops when a series is missing instead of substituting data. Annual mean SST increased significantly in all boxes over 1982–2025, by 0.402 °C per decade in the Eastern Mediterranean, 0.394 in the Aegean, 0.509 in the Black Sea and 0.386 in the Western Mediterranean (Mann–Kendall p < 10⁻⁶). A Pettitt change point in 1998 separates two Eastern Mediterranean regimes whose means differ by 0.91 °C. With a fixed 1982–2011 threshold, marine-heatwave days increased from 2.5 to 213 per year in the Eastern Mediterranean and from 4.4 to 167 in the Aegean between 1982–1991 and the last ten years, showing that the region now spends most of the year above its former 90th percentile. In a standardised matrix of 28 years, the first principal component explained 80.8% of the joint variance, a lag-1 Ridge model reduced the out-of-sample forecast error of Eastern Mediterranean SST from 0.283 °C (persistence) to 0.230 °C, and transfer entropy from SST to fCO₂ was 0.65 bits. The results agree with independent satellite analyses and show how a fixed baseline turns transient extremes into a quasi-permanent state.

Keywords: Sea surface temperature; Marine heatwaves; Eastern Mediterranean; Aegean Sea; Black Sea; Ocean heat content; fCO₂; Reproducibility

Shipping decarbonisation is regulated through ships, operators and ports, whereas national statistics allocate fleets and emissions by flag, ownership, residence or fuel sales. We combined UNCTAD fleet statistics, experimental OECD estimates of monthly maritime CO₂ allocated by the residence principle, and Climate Watch bunker-fuel emissions for Türkiye, Greece, Italy, Spain, France, Germany, Norway and China, retrieved from the open GOMD platform with a notebook that refuses to substitute missing data. Türkiye's flag fleet in 2026 was 6.9 million deadweight tonnes (DWT), of which 26.5% tankers, 23.5% bulk carriers and 14.9% container ships, with an average age of 25.5 years; its nationally owned fleet (47.7 million DWT) was about seven times larger. Residence-based CO₂ attributed to Türkiye rose from 13.3 Mt in 2019 to 19.0 Mt in 2025. Dividing 2025 emissions by the owned fleet gives a crude intensity proxy that ranged from 220 t CO₂ per thousand DWT for Greece and 262 for China to 394 for Türkiye and more than 1,100 for Spain, France and Italy. In a country fixed-effects panel (N = 56), one additional million DWT of owned fleet was associated with 0.141 Mt more CO₂ per year (t = 11.2, within R² = 0.73). Türkiye's international bunker-fuel CO₂ rose by 0.395 Mt per year over 1990–2023, with a change point in 2003; the 2023 value (15.3 Mt) was 2.6 times the 2008 value. We discuss why these accounting bases differ, why the proxy is not the IMO Carbon Intensity Indicator, and what the figures imply for exposure to the EU Emissions Trading System and FuelEU Maritime.

Keywords: Merchant fleet; Shipping emissions; Residence principle; Bunker fuels; IMO GHG strategy; EU ETS; FuelEU Maritime; Türkiye

Blue-economy finance needs indicators that link wealth, climate risk and the protection of marine ecosystems, but such indicators are scattered across institutions and often mislabelled when re-published. Using the open GOMD platform and a notebook that retrieves each series live and stops when a series is missing, we analysed World Bank natural-capital accounts for Türkiye, Eurostat/EEA climate-related economic losses and OECD marine protected area (MPA) statistics for Türkiye, Greece, Italy, Spain and France, and the share of insurance and financial services in trade in services. Türkiye's adjusted net savings rose from 5.8% of GDP in 1990 to 12.65% in 2020 (Sen slope +0.39 points per year, p < 10⁻⁶), while the estimated damage from CO₂ emissions doubled from 1.1% to 2.34% of GDP between 1990 and 2021; energy and mineral depletion were small (0.12% and 0.27%). Climate-related losses per person were highest in Italy (75.9 € per year since 2005), Spain (63.0 €), France (50.4 €) and Greece (44.2 €) and increased significantly in Italy (+1.19 € per person per year, p = 0.0003), France (+0.89, p = 0.036) and Greece (+0.13, p = 0.002). In 2024, MPAs covered 0.15% of the exclusive economic zone (EEZ) extent used by OECD for Türkiye, against 4.67% for Greece, 7.43% for Italy, 12.73% for Spain and 38.72% for France; reaching the Kunming-Montreal 30% target would require about 76,800 km² of additional protected area for Türkiye on this basis. We discuss reporting differences, the limits of country aggregates and the indicator safeguards needed for finance and insurance applications.

Keywords: Natural capital; Adjusted net savings; Climate-related economic losses; Insurance protection gap; Marine protected areas; 30x30; Blue finance; Türkiye

The Sea of Marmara links the Black Sea and the Aegean through the Istanbul and Çanakkale Straits and suffered a severe mucilage outbreak in 2021. We examined surface temperature, wind and biogeochemical indicators for this system using NASA MERRA-2 reanalysis, NOAA OISST v2.1, a Copernicus Marine rim-current index and BGC-Argo floats, all retrieved live from the open GOMD platform by a notebook that stops when a series is unavailable. Over 1996–2025, annual mean skin temperature rose by 0.610 °C per decade over the Sea of Marmara, 0.641 over the Istanbul Strait, 0.418 over the Çanakkale Strait and 0.434 over the Aegean (p ≤ 0.0002). In the reanalysis, the winter preceding the 2021 outbreak (December 2020–March 2021) was 1.30 °C warmer than the 1997–2020 mean, the fifth warmest of 29 winters, whereas April–June 2021 was 0.12 °C below the 1996–2020 mean (18th of 30). The same spring had 37 calm days with maximum wind below 5 m/s, the sixth-highest number in 30 years. Between the two ends of the Straits, the annual SST difference between the Aegean and Black Sea boxes had a Sen slope of −0.149 °C per decade over 1982–2025 (p = 0.0039), because the Black Sea is warming faster than the Aegean. Black Sea surface oxygen and chlorophyll-a from BGC-Argo co-varied positively over 76 common months (r = 0.43). The results support the view that warm antecedent conditions and calm weather accompanied the 2021 event, while showing that coarse reanalysis does not capture the local spring anomalies reported from station data; they do not measure mucilage or establish causation.

Keywords: Sea of Marmara; Turkish Straits; Black Sea; Mucilage; MERRA-2; Sea surface temperature; BGC-Argo; Reproducibility

Attacks on merchant ships in the Red Sea from November 2023 diverted many Asia–Europe liner services around the Cape of Good Hope. We estimated the effect of the crisis on the UNCTAD Liner Shipping Connectivity Index (LSCI) of three Eastern Mediterranean gateway countries, Egypt, Türkiye and Greece, using synthetic difference-in-differences (SDID) with nine donor countries not tied to the Suez route and placebo-in-space inference. All data were retrieved live from the open GOMD platform by a notebook that stops when a series is missing. With a pre-period of February 2019–August 2023 (19 quarterly observations) and a post-period of February 2024–May 2026 (10 observations), SDID estimates were positive for Egypt (+0.135 log points, about +14%) and Türkiye (+0.151, about +16%) and negative for Greece (−0.081, about −8%), with placebo p-values of 0.2, 0.2 and 0.4. The estimates changed little with the regularisation parameter, and unit weights were nearly uniform. Plots of observed and synthetic series showed that the LSCI of Egypt and Türkiye was already rising steeply before the crisis, so the full-period averages carry a pre-existing divergence. A short-window comparison of August 2023 with the first half of 2024 gave connectivity changes relative to the donor mean of −6.2% for Egypt (p = 0.20), −1.4% for Türkiye (p = 0.70) and −7.1% for Greece (p = 0.10). The direction agrees with the fall in Suez Canal transits reported by UNCTAD, but with nine donors the evidence remains weak. We also report Marine Regions EEZ polygon areas for the littoral states and show how the notebook handles a jurisdiction series that has not yet been loaded. The study illustrates simple safeguards for platform-based causal claims.

Keywords: Red Sea crisis; Liner Shipping Connectivity Index; Synthetic difference-in-differences; Placebo inference; Parallel trends; Suez Canal; Eastern Mediterranean; Maritime jurisdiction

Fisheries in the Mediterranean and Black Sea combine a long history of overexploitation with rapid growth of aquaculture. Using FAO, World Bank and ICCAT data retrieved live from the open GOMD platform by a notebook that stops when a series is missing, we analysed Türkiye's capture and aquaculture production (1970–2024), Türkiye's catch by species and total capture in FAO area 37 for Türkiye, Italy, Greece, Spain and Egypt (1950–2024), marine aquaculture in four countries, and ICCAT catches of Atlantic bluefin tuna in the Mediterranean and Black Sea. Aquaculture's share of Türkiye's production rose from 0.2% in 1970 to 61.8% in 2024 (577 of 933 thousand tonnes); a Pettitt change point in 1996 separates a period with a mean share of 1.1% from one with 31.3%. Türkiye's anchovy catch peaked at 388 thousand tonnes in 1995 and was 153 thousand tonnes in 2024, with a coefficient of variation of 0.37 after 2005. Total capture in FAO area 37 declined after 2000 in all five countries, by 8.6 thousand tonnes per year in Türkiye, 7.5 in Italy, 3.0 in Spain, 1.2 in Greece and 0.6 in Egypt. Within countries, one thousand tonnes more marine aquaculture was associated with 0.43 thousand tonnes less capture (t = −6.2, within R² = 0.22, N = 140), a correlation that may reflect shared trends. Mean annual bluefin tuna catches fell by 42–67% between 1997–2006 and 2008–2017, after the ICCAT recovery plan, and returned to 77–125% of earlier levels after 2018. The results agree with regional assessments of declining fishing pressure and expanding aquaculture and show how open notebooks can track fisheries indicators against management milestones.

Keywords: Fisheries; Aquaculture; Black Sea; Mediterranean; FAO area 37; European anchovy; Atlantic bluefin tuna; ICCAT; Türkiye

The ocean absorbs about a quarter of anthropogenic CO₂ emissions and most of the excess heat of the climate system, while becoming more acidic. We combined the Global Carbon Budget 2025 (GCB) with four independent observational records, Mauna Loa atmospheric CO₂, surface pH at Station ALOHA, upper-ocean heat content and September Arctic sea-ice extent, all retrieved live from the open GOMD platform by a notebook that stops when a series is missing. The GCB series close exactly: fossil and land-use emissions equal atmospheric growth plus ocean, land and cement-carbonation sinks plus the budget imbalance in every year (maximum residual below 0.001 GtC per year). In 2020–2024, fossil emissions averaged 10.17 GtC per year, land-use emissions 1.27, atmospheric growth 5.70, the ocean sink 3.24 and the land sink 2.35 GtC per year. The ocean took up 28.3% of total emissions, compared with 25.9% in the 1960s; the ocean-sink fraction increased by 0.0043 per decade (p = 0.007) and the airborne fraction by 0.016 per decade (p = 0.044). Total emissions and the ocean sink are cointegrated (slope 0.288, ADF = −3.63). Annual mean CO₂ at Mauna Loa rose from 315.98 ppm in 1959 to 427.35 ppm in 2025, with a record increase of 3.52 ppm in 2024; converted to carbon mass, its growth correlates with the GCB atmospheric term at r = 0.88. Surface pH at Station ALOHA decreased by 0.0195 units per decade (1988–2024) and correlated with Mauna Loa CO₂ at r = −0.98. Upper-ocean (0–700 m) heat content increased by 5.38 × 10²² J per decade since 1979, and September Arctic sea-ice extent decreased by 0.74 million km² per decade, with a change point in 2001. The independent records are mutually consistent and reproduce published budget values.

Keywords: Global carbon budget; Ocean carbon sink; Airborne fraction; Ocean acidification; Mauna Loa; Station ALOHA; Ocean heat content; Arctic sea ice; Cointegration

Maritime chokepoints and enclosed seas concentrate traffic, so weather conditions there affect safety, schedules and port operations far beyond the local area. Using data retrieved live from the open GOMD platform by a notebook that refuses to substitute missing series, we analysed daily maximum significant wave height (Hs) and its swell and wind-sea components in ten basins and chokepoints over 93 days (17 July–17 October 2026), and 30 years (1996–2025) of NASA MERRA-2 daily maximum 10 m wind at twelve sites. Median daily maximum Hs ranged from 0.20 m at Hormuz to 1.20 m in the Aegean; the highest values were 4.52 m in the Ionian Sea and 4.46 m in the Aegean, which had nine days with Hs of at least 2.5 m. Swell dominated in the Levantine Sea, the Western Mediterranean and the Red Sea (median swell ratio 0.87–0.94), whereas wind sea dominated in the Aegean and the Marmara (0.45–0.48). Within basins, Hs followed the wind-sea component (β = 0.91, R² = 0.86) more closely than swell (β = 1.14, R² = 0.48). Because three months cannot support extreme-value estimates, return levels were computed from 30 annual maxima of reanalysis wind with a Gumbel fit. The 100-year daily maximum wind was 26.1 m/s over the Aegean, 25.6 m/s over the Black Sea, 24.3 m/s at Mersin, 21.8 m/s at the Çanakkale Strait and 17.6 m/s at the Istanbul Strait; the Aegean averaged 2.5 gale days per year. No site showed a significant trend in annual maximum wind (all p > 0.15). Reanalysis at about 50 km resolution underestimates winds channelled through narrow straits, so the return levels are regional screening values, not design values.

Keywords: Significant wave height; Swell; Wind sea; Extreme winds; Gumbel distribution; Return levels; MERRA-2; Maritime chokepoints; Aegean Sea

Onshore power supply (OPS) lets ships switch off their auxiliary engines at berth, but its climate benefit depends on the carbon intensity of the electricity that replaces them. Using data retrieved live from the open GOMD platform by a notebook that stops when a series is missing, we analysed Türkiye's sectoral emissions (JRC EDGAR), electricity generation and grid carbon intensity (Ember) in comparison with Greece, Italy, Spain, France, Germany and Egypt, and global shipping emissions, activity and energy use (IEA). Türkiye's greenhouse-gas emissions excluding LULUCF rose from 102 Mt CO₂e in 1970 to 580 Mt CO₂e in 2024; power and heat generation contributed 140.9 Mt CO₂ (24.3%) and transport 99.3 Mt (17.1%), with transport's share rising from 10.1% in 1970 and a change point in 2012. Wind and solar generation grew from 2.9 TWh in 2010 to 76.7 TWh in 2025, and low-carbon sources supplied 43.3% of electricity, yet grid carbon intensity fell only from 505 to 475 gCO₂/kWh between 2015 and 2025, compared with 604 to 315 in Greece and 328 to 154 in Spain. Across the seven countries, a one-point increase in the low-carbon share was associated with about 2.6% lower power-sector CO₂ (t = −13.9, within R² = 0.45, N = 245). Relative to an auxiliary engine burning marine gas oil at an assumed 220 g/kWh (about 705 gCO₂/kWh), OPS at 2025 grid intensities would cut direct at-berth CO₂ by 94% in France, 78% in Spain, 60% in Italy, 55% in Greece, 53% in Germany, 33% in Türkiye and 20% in Egypt. Globally, shipping CO₂ rose from 0.810 to 0.847 Gt between 2015 and 2024 while CO₂ per tonne-km fell by 18% and CO₂ per unit of energy stayed at 75 g/MJ. We discuss what these numbers imply for port investment under the EU's FuelEU Maritime and AFIR regulations.

Keywords: Onshore power supply; Cold ironing; Grid carbon intensity; Energy transition; Sectoral emissions; Shipping decarbonisation; FuelEU Maritime; Türkiye

Türkiye's ports link Europe, the Black Sea and the Middle East, and their integration into liner networks shapes the country's trade. Using data retrieved live from the open GOMD platform by a notebook that stops when a series is missing, we analysed Türkiye's container port traffic, seaborne cargo and port throughput, and compared merchandise trade, the UNCTAD Liner Shipping Connectivity Index (LSCI) and the World Bank Logistics Performance Index (LPI) with Greece, Italy, Spain, Egypt, Germany and the Netherlands; port-level connectivity (PLSCI) was examined for 14 ports. Container traffic grew from 5.74 million TEU in 2010 to 13.51 million TEU in 2024 (Sen slope +0.55 million TEU per year), and the gross weight of goods handled in Turkish ports rose from 305 to 545 million tonnes between 2008 and 2025. The ratio of seaborne goods loaded to discharged rose from 0.24 in 2000 to 0.44 in 2024, so Türkiye remains a net importer by tonnage. In June 2026 Türkiye's LSCI was 304.6, 32% above its 2019 mean, compared with 415.6 for Spain, 378.8 for the Netherlands and 174.2 for Greece. Ambarlı had the highest port connectivity in Türkiye (501.7 in April 2026) and ranked second of the 14 ports after Algeciras. Between the last quarter of 2023 and the third quarter of 2024, connectivity fell by 17.1% at Port Said, 12.9% at Piraeus, 12.7% at Mersin and 26.4% at İzmir, but only 5.1% at Ambarlı, while Algeciras gained 3.9%. Within countries, a 1% higher LSCI was associated with only 0.46% higher merchandise trade (t = 3.05, within R² = 0.07, N = 140), and container traffic and exports were not cointegrated over 15 years. The results show strong growth and resilience at Türkiye's main hub alongside vulnerable secondary ports, and they caution against reading national connectivity indices as direct drivers of trade.

Keywords: Liner Shipping Connectivity Index; Port connectivity; Container port traffic; Maritime trade; Red Sea crisis; Logistics Performance Index; Türkiye; Eastern Mediterranean

The goal of becoming a maritime nation is often measured by fleet size or port traffic, yet generating seaborne cargo and being able to carry it are different things. This study proposes a two-layer index that can be computed from open data: a cargo-base index (shares of seaborne trade and container port throughput) and a core maritime capability index (shares of shipbuilding, ownership, flag registration and seafarer supply). Their ratio gives maritime intensity. The index is computed for 238 economies for 2014–2016 and 2023–2025; the sensitivity of the ranking to construction choices is measured with 20,000 simulations, the elasticity of capability with respect to the cargo base is estimated, and the physical equivalent of moving up one rank in each indicator is calculated for Türkiye. Türkiye's cargo-base share rose from 1.15% to 1.40% and its rank from 24th to 20th. Its core capability index, however, fell from 1.14 to 0.98 and its rank from 19th to 22nd; maritime intensity dropped from 1.00 to 0.70. Intensity also declines when seafarer data are excluded. The 5–95% range of Türkiye's rank is 6–25, and 68% of this uncertainty stems from a single choice: whether ship recycling is included in the index. Across countries the elasticity of capability with respect to the cargo base is close to one, but the cargo base explains only about a third (37%) of the variation in capability. Türkiye's weakest link is the flag: if half of the Turkish-owned tonnage under foreign flags returned to the national registry, Türkiye's rank would move from 22nd to 15th on that change alone.

Keywords: Maritime capability; Composite index; Uncertainty analysis; Cargo base; Ship registry; Seafarers; Türkiye

Shipbuilding, ownership, registration and recycling are studied separately in the literature; there is no tool that shows on a single scale in which stage of a ship's life cycle a country specialises. This study proposes a ship life-cycle position index inspired by the upstreamness measure of the global value chain literature and applies it, together with a cross-stage balance measure, to 38 countries. It also decomposes the change in a country's recycling volume into market-size and market-share effects, measures the cyclicality of recycling and tests whether Türkiye's market position broke in 2019. The data are UNCTAD's 2014–2025 shipbuilding and ship recycling series together with fleet ownership and registration data. Türkiye's position index rose from 3.30 to 3.61; after Bangladesh, Pakistan and India, Türkiye is the fourth most end-of-life-specialised country and the most balanced across stages within this group. While world recycling volume fell from 22.6 million GT to 8.5 million GT, Türkiye's share rose from 4.1% to 12.2%: the market contraction took 0.96 million GT from Türkiye and the share gain added 1.06 million GT. Türkiye's volume relative to South Asia roughly doubled after 2019, and its volume was the least volatile series among the major recycling countries. In shipbuilding, its world share remained around 0.2%. Türkiye is strong at the end of the ship's life cycle and weak at its beginning.

Keywords: Ship recycling; Shipbuilding; Life cycle; Position index; LMDI decomposition; Structural break; Türkiye

The flag under which a country's nationals operate their ships is a key indicator of its capacity to be a maritime nation; yet the literature studies this choice at ship level and as a national–foreign dichotomy, and offers no measure that is comparable across countries. This study proposes four methodological innovations: a flag home-bias index that adapts the home-bias measure of international portfolio theory to ship registration; a logarithmic mean Divisia decomposition that splits the change in nationally flagged tonnage without residual; a three-component shift-share analysis that separates the change in the registry portfolio into fleet growth, the global registry trend and a country-specific shift; and a maximum-entropy flow matrix that estimates net registry transfers from aggregated data alone. The methods are applied to UNCTAD's flag-by-beneficial-ownership matrix for 35 owner countries over 2014–2026. While the Turkish-owned fleet grew from 28.9 million DWT to 47.7 million DWT, Turkish-flagged tonnage fell from 8.3 million DWT to 6.3 million DWT; the home-bias index dropped from 0.284 to 0.140, and Türkiye fell from 18th to 26th among 35 countries. Fleet growth could have added 3.6 million DWT to the Turkish flag, whereas the fall in the retention rate removed 5.6 million DWT. The shift cannot be explained by the global trend: while the Maltese flag held its share worldwide, Turkish owners withdrew 12.7 million DWT from this registry in a country-specific shift, and while Panama declined worldwide, Turkish-owned tonnage moved towards it. 40.7% of the registry portfolio changed. The problem is not fleet investment but the competitiveness of the registry.

Keywords: Flagging out; Ship registry; Home bias; LMDI decomposition; Shift-share analysis; Maximum entropy; Turkish merchant fleet

Port efficiency studies rely on infrastructure inputs such as quay length, number of cranes and area; in most countries these data are not openly and comparably available. This study transfers two concepts established in electric power systems, the load factor and the coincidence factor, to ports and computes them solely from daily throughput estimates derived from ship tracking data. To separate how much of the load factor is due to scale, a benchmark model based on unconstrained random ship arrivals is built. The method is applied to Türkiye's 38 ports and to the ports of seventeen Mediterranean and Black Sea countries for 2019–2025. The random-arrival model explains 77% of the variation in port load factors; the high load factors of large ports are largely a statistical consequence of the number of ships. In 82% of port-year observations, observed throughput is more volatile than random arrivals would imply, and there is no trace of a capacity ceiling at weekly resolution. Once scale is controlled for, the load factors of Türkiye's ports do not differ from those of the comparison ports. Port peaks are mutually independent: the coincidence factor is 0.72 and equals its expected value under independence; the sum of individual port peaks exceeds the system's coincident peak by 28%. After the earthquake of 6 February 2023, İskenderun's throughput fell to 20% of its pre-earthquake level for four weeks and was still at 73% of that level in 2025; no measurable shift of cargo to neighbouring ports was observed. The findings indicate that in port planning the pooling and substitution capability of the system matters more than the capacity of individual ports.

Keywords: Port capacity utilisation; Load factor; Coincidence factor; Ship arrival process; AIS data; Port resilience; Türkiye