Widespread warming of Earth's estuaries

Water temperature responses to climate change may vary across Earth's estuaries. To understand how climate change influences estuarine surface water temperature, we need global, long‐term records of estuarine temperature. Here, we generated surface water temperature data over 1060 estuaries globally using Landsat 5, 7, and 8 from 1985 to 2022 and compared water warming rates with local air temperature warming rates. Forty‐seven percent of Earth's estuaries are warming, with a global average warming rate of 0.070 ± 0.004°C yr−1 (median = 0.060°C yr−1). Estuaries at higher latitudes showed rapid warming. A 1°C increase in air temperature could lead to a 0.81°C increase in estuarine surface water warming and 1.3°C increase in estuaries above 60.5°N. We inferred the potential influences over estuarine warming based on distinct global spatial patterns in water and air warming and discussed the effects of warming water temperature on estuarine metabolism and water quality.

, fish fecundity, growth rates, and habitat availability (Handeland et al. 2008;O'Gorman et al. 2016;McQueen and Marshall 2017;Paul et al. 2022).The warming of estuaries may therefore have cascading impacts that change ecosystem structure, increase the duration or severity of hypoxia and cyanobacterial harmful algal blooms (Rabalais et al. 2009;Paerl 2014), and alter carbon cycling potentially exacerbating climate change (Canuel et al. 2012;Simone et al. 2021).
Global climate change is already affecting estuarine water temperature (James et al. 2013;Robins et al. 2016;Bashevkin et al. 2022;Filho et al. 2022).In the near term (2030)(2031)(2032)(2033)(2034)(2035), global warming is projected to reach 1.5 C (IPCC 2022).Historical data show that the global warming of air and sea surface temperature from 2011 to 2020 was approximately 0.95-1.2C (IPCC 2022).In Australia, the warming of estuarine surface water was 1.4 times faster than air warming and 2 times faster than ocean warming between 2007 and 2019 (Scanes et al. 2020).However, we do not know how many of Earth's estuaries are warming, how fast, and how water temperature changes are related to air temperature changes globally.
Water temperature response to climate change may differ from estuary to estuary around the world.Estuarine surface temperature is influenced by complex interactions among estuarine geomorphology, incoming solar radiation, air temperature, ocean temperature, water exchange with rivers and oceans, and wind patterns (Brown et al. 2016).Heat exchange among rivers, the coastal ocean, and estuaries occurs primarily through advection and tidal dispersion (Uncles and Stephens 2001;van Aken 2008), which is modified by estuarine shape and bathymetry with longer water residence times and shallower waters typically being more susceptible to warming.
To begin to disentangle estuarine water temperature response to climate change, and its controls, and identify vulnerable estuaries, we first need comparable, long-term surface water temperature records from estuaries around the world.Here, we generated globally consistent, spatially explicit, estuarine surface water temperature records from the Landsat missions (Landsat 5, 7, and 8) from 1985 to 2022 by calibrating to in situ temperature observations from around the world.Landsat is widely used for long-term studies and was the best available data source and satellite for our questions given the multi-decadal continuous observations, sufficient spatial resolution (30 m) to measure smaller estuaries ($ 1 km 2 ), and a globally consistent method.Our questions were (1) what proportion of Earth's estuaries are warming, and (2) how fast are estuarine warming rates relative to local climate warming?We discuss potential causes of global trends in estuarine warming and the implications for estuarine carbon cycling.

In situ and satellite data
Generating a global database of estuarine water temperature included two major steps: (1) matching in situ with satellite observations to calibrate Landsat observations using an empirical model, and (2) pulling the entire Landsat records over the water area represented by global estuaries.Our study area included all estuaries > 1 km 2 , 1060 in total (Supporting Information Fig. S1), from the spatial database in Alder (2003).The metadata associated with this database references the definition provided by Cameron and Pritchard (1963): "A semi-enclosed embayment of the coast in which fresh run-off water mixes with saline water entering from the ocean."The database was designed to encompass all estuaries associated with major rivers of the world, while also being mindful of smaller systems.To representatively sample each estuary and generate tabular time series for analysis, we generated a point grid where the number of points within each estuary was proportional to the estuary size (Fig. 1).Within these grid points, we extracted surface reflectance and temperature values from Landsat Collection 2, Level 2, Tier 1 over high confidence estuarine water pixels using the Google Earth Engine Platform and the Python API (Gorelick et al. 2017).High-confidence open-water pixels were identified using Dynamic Surface Water Extent (Jones 2019) and Fmask algorithms (Foga et al. 2017) masking out clouds, cloud shadow, snow, ice, and vegetated water (Gardner et al. 2021) using Landsat daytime scenes with cloud cover less than 30%.See the data quality controls in the supplementary files.
The Landsat thermal infrared sensors record the long wave energy between 10.4 and 12.5 μm emitted from the earth's surface (Parastatidis et al. 2017).The amount of long wave thermal infrared energy emitted from the earth's surface is proportional to its heat.However, Landsat Collection 2 products provide raw surface temperature data instead of heat energy fluxes.A linear model was recommended for calibrating to in situ surface water temperature (Dyba et al. 2022;Herrick et al. 2023).We adjusted Landsat temperature data using a Deming linear regression (Cornbleet and Gochman 1979) developed from a matchup database of 19,244 paired in situ and Landsat temperature observations, split into 80% for training and 20% for hold-out validation.We applied the regression over the validation data to evaluate model performance using the mean absolute error (MAE).

Data analysis
Our final estuarine water temperature database included over 64.3 million unique records across 195,566 individual sites within 1060 estuaries around the globe from 1985 to 2022 (Prum 2023) after removing approximately 10.9 million (13.81% of 74.7 million) records due to outliers and lowquality data (Supporting Information Fig. S2).To calculate surface water temperature trends and compare them with warming rates in other estuaries, we computed the spatial, mean annual temperature across each estuary that had time series longer than 10 yr.The spatial, mean annual temperature was computed from the mean seasonal temperature in each year, and only in years with observations spread across all three and four seasons for estuaries above and below 60.5 latitude, respectively.Only a few estuaries above 60.5 latitudes had observation across four seasons per year due to the cloud cover (Supporting Information Fig. S3).Also, only observations covering at least 70% of the estuary surface (calculated from the proportion of grid points identified as water per season to the maximum number of grid points identified as water of the respected season over the 37-yr records) were included in the trend analysis.These criteria were chosen to calculate representative spatial, annual mean temperatures, reducing biases caused by varying observation frequency and timing due to regional differences in cloud cover and Landsat data collection.As a result, 737 estuaries (562 estuaries in the Northern Hemisphere) met our criteria.The 737 estuaries were analyzed for trends using the Mann-Kendall test (alpha = 0.05) and Sen's slopes to quantify the long-term trend or warming rate, in estuarine surface water with the "Pymannkendall" Python package version 1.4.3.To compare estuarine warming rates with local air temperature warming rates, we used the fifth generation of ECMWF atmospheric re-analyses of the global climate monthly aggregate dataset (Copernicus Climate Change Service [C3S], 2017) and calculated spatial, mean annual temperatures over estuary boundaries.Finally, we used the same Mann-Kendall test and Sen's slope statistics to determine significant air temperature trends and warming rates.

Validation
Landsat Collection 2 surface water temperature data could accurately measure estuarine water temperature (Fig. 2).The comparison between raw Landsat surface temperature data and in situ observations showed a MAE of 1.71 C, while after the adjustment with the regression model, the MAE improved to 1.63 C. Adjusting Landsat temperature shows better agreement with in situ data across all three Landsat missions used here and reduces the error across all months, particularly in the Northern hemisphere summer months when temperatures are highest (Fig. 2A).However, clouds could have a major effect on both the frequency of observation and the accuracy of Landsat temperature data (Supporting Information Fig. S3).Removing Landsat pixels close to the clouds likely improves accuracy but our conservative approach limits the number of observations in the study areas.

Trends in estuarine surface water temperature
Mean annual trends in satellite-derived surface temperatures over 737 estuaries across the globe showed that 48% (346 estuaries) of estuaries are warming (p-value < 0.05), less than 3% (two estuaries) were cooling, and 49% (389 estuaries) had no significant trend from 1985 to 2022 (p-value >0.05) (Fig. 3; Supporting Information Fig. S4).The estuarine warming rate of all statistically significant trends ranged from À0.084 to 0.333 C yr À1 , with an average rate of 0.070 AE 0.004 C yr À1 (median is 0.060 C yr À1 ) (Supporting Information Table S1).Including the nonstatistically significant mean annual trends, the estuarine warming rates ranged from À0.220 C yr À1 to 0.415 C yr À1 with a mean of 0.053 C yr À1 and a median of 0.048 C yr À1 .Smaller estuaries had higher surface water temperature warming rates compared to larger estuaries (Supporting Information Fig. S5).
Estuaries at northern latitudes were warming faster (Mann-Whitney U-test, p-value < 0.01).Approximately, 80% (279) of estuaries with warming trends was in the northern hemisphere with a median warming rate of 0.067 C yr À1 , whereas estuaries in the southern hemisphere had a median warming rate of 0.004 C yr À1 , including significantly decreasing and increasing trends.While there was considerable variability, warming rates generally increased from south to north with an increase in warming rate of 0.0007 C yr À1 per latitude degree using a linear regression model (p-value < 0.01).
Estuaries across continents had significantly different warming rates (Kruskal-Wallis H-test tests, p-value < 0.01).The median warming rate of estuaries in Europe (0.088 C yr À1 ) was the highest followed by Asia (0.080 C yr À1 ) and South America (0.060 C yr À1 ).The estuaries in North America (0.053 C yr À1 ) had a median warming rate higher than Africa (0.047 C yr À1 ), while Australia exhibited the lowest warming rate (0.039 C yr À1 ).In Australia, most rapidly warming trends estuaries were located along the east coast (median of west coast 0.036 C yr À1 vs. median of east coast 0.044 C yr À1 ) based on Mann-Whitney U-test with p-value < 0.01.See Supporting Information Table S1 for details.

Influencing factors over estuary warming or cooling
The surface water temperature had a strong correlation with air temperature (Fig. 4A).In addition, the linear regression model shows that 1 C changes of trends in air temperature were associated with 0.81 C changes of trends in water temperature ( p-value < 0.01) for estuaries below 60.5 N latitude.This result was similar to the influence of warming air temperature on lake surface water temperature (Schmid et al. 2014).However, higher latitude estuaries (above 60.5 N degree) showed a faster surface water warming of 1.30 C per 1 C change in air temperature (p-value < 0.01).
Seventy-five percent (263/348 estuaries) of estuaries with a significant trend in water temperature also had a significant trend in air temperature.Estuaries were more likely to have increasing water temperature trends when air temperature warming rates were higher (Fig. 4B).The median of water warming rates from estuaries with significant trends in air temperate warming rate (0.065 C yr À1 ) was higher than the median of water temperature warming rates from estuaries with nonsignificant air temperate warming rate (0.049 C yr À1 ) (Mann-Whitney U-test, p-value <0.01).
The remaining 25% (85/348 estuaries) of estuaries had no significant trend in air temperature yet showed a surface water warming trend.These estuaries were located mostly in the west coast of the United States, North of Australia, and West bays of Portugal in a consistent, global spatial pattern (Fig. 4C) that suggests the influence of oceanic water exchange on estuary warming.Furthermore, estuaries without trends in air temperature had a slightly stronger correlation between warming rate and the ratio of freshwater discharge to estuary surface area (r = 0.33, p-value < 0.01), compared to estuaries with significant trends in air temperature (r = 0.28, p-value < 0.05).While the correlation was weak (Supporting Information Fig. S6), it may suggest the influence of local factors such as freshwater inflow and/or estuary size over estuarine warming trends.However, in regions with significant air warming trends, local factors had no detectable impact on estuarine warming.Recent studies showed a widespread warming of water temperature in lakes (Woolway et al. 2020) and rivers globally (Zhi et al. 2023), which could impact downstream estuaries.Among the two estuaries with cooling trends in surface water temperature, one estuary in Asia showed warming air temperatures, and the other in South America had no air temperature trend.Surface water cooling in these estuaries occurred recently within the last 10 yr.

Rising estuarine temperature
Global estuarine surface water temperature is warming at an average rate of 0.070 AE 0.004 C yr À1 (median is 0.060 C yr À1 ).
While this global average warming rate seems high, our results include many estuaries without temperature records, and the average is skewed by estuaries with extreme warming, located largely near the Arctic, the most rapidly warming region on Earth (Rantanen et al. 2022).The agreement between our warming rates from individual estuaries and site-specific studies gives confidence in our global analysis.Shi and Hu (2022) reported a warming rate of 0.73 C decade À1 for estuaries in South Florida using MODIS satellite images between 2000 and 2023, showing a similar agreement with our results (0.57 C decade À1 for that same region).In addition, our estimation of the warming rate at Chesapeake Bay at 0.50 C decade À1 is close to the warming rate on average of eight stations in Chesapeake Bay at 0.65 C decade À1 (Ding and Elmore 2015).
Not surprisingly, rising air temperature is a first-order control of warming estuaries compared to freshwater inflow or estuary size.Estuarine water temperature is increasing approximately 81% and 130% on average compared to warming air temperature for estuaries below and above 60.5 latitudes, respectively.Freshwater inflows and estuary size are potential factors over surface water warming in estuaries without  S1.
warming air trends because of the weak positive correlation between the ratio of discharge to estuary size and warming rate, and the faster warming rate in smaller estuaries compared to larger estuaries.
However, estuarine heat budgets are more complex than simple air-water transfer as evident from our results that not all estuaries are warming despite global air temperature increases.Estuaries will respond differently to global atmospheric warming due to the interaction between freshwater inputs from land and ocean exchange which contribute to estuarine circulation patterns that move and distribute heat as a function of estuary size, morphology, and salinity gradients (Brown et al. 2016).In particular, global patterns in ocean warming appear to coincide with patterns of estuary warming rates when we compared our results with spatial patterns in the European Space Agency-Climate Change Initiative Sea Surface Temperature trends (C3S, n.d.).Sea surface temperature (SST) increases are higher in the northern than southern hemisphere, similar to estuary trends.The direction and magnitude of SST trends are predictive of estuary warming over most of Europe, Australia, and North America.SST and coastal water temperature show high warming rates in Europe such as the North Sea, Baltic Sea, Black Sea, and the Mediterranean Sea.Except for the North Sea, these European seas have long residence times and lower tidal mixing than other estuarine regions which may contribute to higher warming.Both ocean and estuary warming rates show that the East coast of the United States, and Southeast of Australia are more susceptible to warming with global climate change compared to the West coast of the United States and Northwest of Australia.Despite the relatively small global footprint of estuary area, estuaries are remarkably important to coastal carbon budgets (Najjar et al. 2018).The increasing estuarine temperatures will likely impact estuarine carbon cycling and water quality because of the temperature dependency of metabolic rates and the solubility of dissolved oxygen in water.A 4 C increase in water temperatures may lead to a 43% increase in heterotrophic respiration and a 20% increase in gross primary production (Harris et al. 2006).Such changes in respiration and gross primary production could lead to relatively more CO 2 emissions to the atmosphere for warming surface waters and associated mixed layers in estuarine ecosystems.This increased respiratory demand also consumes oxygen, and when combined with warmer waters holding less oxygen due to temperature-driven gas solubility, estuaries may become more susceptible to hypoxic and anoxic conditions.These regime shifts in estuaries are likely to have cascading effects on carbon cycling, response to regional management of nutrient pollution, and impacts on higher trophic levels.

Conclusions
This study is the first to estimate surface water warming rates for estuaries globally.The remotely sensed surface water data represents a cost-effective way to study $ 30-yr trends in estuary water temperature, particularly in remote regions or locations lacking historical data.Almost half of estuaries had warming trends in surface water temperature, and our global synthesis shows the known rapid warming across the Arctic is also occurring in high latitude estuaries.Global atmospheric and oceanic warming may influence estuary warming, but ocean exchange and/or local factors such as freshwater discharge and estuary size may play a role in estuaries without detectable trends in atmospheric warming.The suggested significant and nonsignificant warming trends can be studied at estuary level with spatially explicit time series made accessible as tabular data (over 64.3 million points from 1985 to 2022).These temperature data can be used to develop conceptual and statistical models that will help predict the relative sensitivity of different estuarine ecosystems to increasing water temperatures and inform strategies for protecting estuaries that are most susceptible to climate change.

Fig. 1 .
Fig. 1. (A) an example of the sampling grid over the Mississippi Delta.The distances between points were generated according to estuary size (e.g., 300, 800, and 2000 m for estuaries with sizes less than 10 km 2 , between 10 and 100 km 2 , and greater than 100 km 2 , respectively).(B) Full time series of water temperature derived from adjusted Landsat from 1985 to 2022 at one location (the red point) in Fig. 1A.(C) Time series of the spatial, mean annual temperatures of Mississippi Delta calculated over 3330 points from 1985 to 2022 (Sen's slope: 0.023 C yr À1 , and p-value < 0.05).

Fig. 2 .
Fig. 2. The improvement of Landsat temperature data after calibration with field surface water temperature.(A) The monthly MAE of raw Landsat and adjusted Landsat temperature data compared to in situ data across all satellite/in situ matchups.(B) The mean of raw Landsat, and adjusted Landsat, and in situ data across different Landsat sensors.(C) Validation of adjusted Landsat temperature comparing with in situ data.The red dashed line is 1:1 line.

Fig. 3 .
Fig. 3. (A) Map of trends in mean annual estuary surface water temperatures from 348 estuaries >1 km 2 .(B) The box plots of Sen's slope of trends in mean annual estuary surface water temperatures, grouped by latitudes.The points outside the boxplot are estuaries with extreme high or low warming rates among the groups (1.5 times below and above the upper quantiles and lower quantiles).The vertical line between the box is the median.(C) The distribution of warming rates across estuaries as a density histogram.There are 42 estuaries between 60 and 30 South, 27 estuaries between 30 and 0 South, 75 estuaries between 0 and 30 North, 181 estuaries between 30 and 60 North and 23 estuaries between 60 and 90 North.See Supporting Information TableS1.

Fig. 4 .
Fig. 4. (A)The density scatter plot between annual mean water and air temperature.The side plots represent the histogram of annual mean water and air temperature.The black dash line is 1 : 1 line.(B) Scatter plot of warming rates with significant Sen's slopes of annual average estuarine surface water vs. warming rates in air temperature with colors showing if the air warming was significant (blue) or not (orange).The solid black line is the linear relationship between them (slope = 1.266).(C) Map of trends in air surface temperature of 348 estuaries that have significant trends in warming surface water temperature.The orange and blue colors represent estuaries with nonsignificant (orange) and significant trends (blue) in warming air temperature, respectively.