Sea Surface Temperature Forecasts Research Articles

Global ocean surface and subsurface temperature forecast skill over subseasonal to seasonal timescales

Global ocean surface and subsurface temperature forecast skill over subseasonal to seasonal timescales

Machine learning methods to predict sea surface temperature and marine heatwave occurrence: a case study of the Mediterranean Sea

Abstract. Marine heatwaves (MHWs) have significant social and ecological impacts, necessitating the prediction of these extreme events to prevent and mitigate their negative consequences and provide valuable information to decision-makers about MHW-related risks. In this study, machine learning (ML) techniques are applied to predict sea surface temperature (SST) time series and marine heatwaves in 16 regions of the Mediterranean Sea. ML algorithms, including the random forest (RForest), long short-term memory (LSTM), and convolutional neural network (CNN), are used to create competitive predictive tools for SST. The ML models are designed to forecast SST and MHWs up to 7 d ahead. For each region, we performed 15 different experiments for ML techniques, progressively sliding the training and the testing period window of 4 years from 1981 to 2017. Alongside SST, other relevant atmospheric variables are utilized as potential predictors of MHWs. Datasets from the European Space Agency Climate Change Initiative (ESA CCI SST) v2.1 and the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis from 1981 to 2021 are used to train and test the ML techniques. For each area, the results show that all the ML methods performed with minimum root mean square errors (RMSEs) of about 0.1 °C at a 1 d lead time and maximum values of about 0.8 °C at a 7 d lead time. In all regions, both the RForest and LSTM consistently outperformed the CNN model across all lead times. LSTM has the highest predictive skill in 11 regions at all lead times. Importantly, the ML techniques show results similar to the dynamical Copernicus Mediterranean Forecasting System (MedFS) for both SST and MHW forecasts, especially in the early forecast days. For MHW forecasting, ML methods compare favorably with MedFS up to 3 d lead time in 14 regions, while MedFS shows superior skill at 5 d lead time in 9 out of 16 regions. All methods predict the occurrence of MHWs with a confidence level greater than 50 % in each region. Additionally, the study highlights the importance of incoming solar radiation as a significant predictor of SST variability along with SST itself.

SST Forecast Skills Based on Hybrid Deep Learning Models: With Applications to the South China Sea

We explore to what extent data-driven prediction models have skills in forecasting daily sea-surface temperature (SST), which are comparable to or perform better than current physics-based operational systems over long-range forecast horizons. Three hybrid deep learning-based models are developed within the South China Sea (SCS) basin by integrating deep neural networks (back propagation, long short-term memory, and gated recurrent unit) with traditional empirical orthogonal function analysis and empirical mode decomposition. Utilizing a 40-year (1982–2021) satellite-based daily SST time series on a 0.25° grid, we train these models on the first 32 years (1982–2013) of detrended SST anomaly (SSTA) data. Their predictive accuracies are then validated using data from 2014 and tested over the subsequent seven years (2015–2021). The models’ forecast skills are assessed using spatial anomaly correlation coefficient (ACC) and root-mean-square error (RMSE), with ACC proving to be a stricter metric. A forecast skill horizon, defined as the lead time before ACC drops below 0.6, is determined to be 50 days. The models are equally capable of achieving a basin-wide average ACC of ~0.62 and an RMSE of ~0.48 °C at this horizon, indicating a 36% improvement in RMSE over climatology. This implies that on average the forecast skill horizon for these models is beyond the available forecast length. Analysis of one model, the BP neural network, reveals a variable forecast skill horizon (5 to 50 days) for each individual day, showing that it can adapt to different time scales. This adaptability seems to be influenced by a number of mechanisms arising from the evident regional and global atmosphere–ocean coupling variations on time scales ranging from intraseasonal to decadal in the SSTA of the SCS basin.

On the Role of Indian Ocean SST in Influencing the Differences in Atmospheric Variability Between 2020–2021 and 2021–2022 La Niña Boreal Winters

AbstractThe difference in observed atmospheric anomalies over the Northern Hemisphere winter between 2021–22 and 2020–21 La Niña years indicated a tripole pattern consisting of a Japan cyclone, a Bering Sea anticyclone, and a cyclone over the North American continent. This feature, however, was not replicated in the North American Multi‐Model Ensemble (NMME) forecasts. A set of model sensitivity experiments was performed to better understand the cause of this discrepancy. The results revealed the possible role of the influence of sea surface temperature (SST) anomalies, particularly over the Indian Ocean, on the observed circulation differences that was further modulated by internal atmospheric variability. The failure in predicting circulation changes in NMME was next attributed to the errors in SST predictions over the Indian Ocean and highlights the need for improvements in SST forecasts over this region.

Predictability of marine heatwaves: assessment based on the ECMWF seasonal forecast system

Abstract. Marine heatwaves (MHWs), defined as prolonged period of extremely warm sea surface temperature (SST), have been receiving a lot of attention in the past decade as their frequency and intensity increase in a warming climate. This paper investigates the extent to which the seasonal occurrence and duration of MHWs can be predicted with the European Centre for Medium-Range Weather Forecast (ECMWF) operational seasonal forecast system. The prediction of the occurrence of MHW events, the number of MHW days per season, and their intensity and spatial extent are derived from seasonal SST forecasts and evaluated against an observation-based SST analysis using both deterministic and probabilistic metrics over the 1982–2021 period. Forecast scores show useful skill in predicting the occurrence of MHWs globally for the two seasons following the starting date. The skill is the highest in the El Niño region, the Caribbean, the wider tropics, the north-eastern extra-tropical Pacific, and southwest of the extra-tropical basins. The skill is not as good for other midlatitude eastern basins nor for the Mediterranean, with the forecast system being able to represent the low-frequency modulation of MHWs but showing poor skill in predicting the interannual variability of the MHW characteristics. Linear trend analysis shows an increase in MHW occurrence at a global scale, which the forecasts capture well.

The Rossby Normal Mode as a Physical Linkage in a Machine Learning Forecast Model for the SST and SSH of South China Sea Deep Basin

AbstractMachine learning (ML) has been widely applied in Earth sciences, but its black‐box nature still remains. Existing techniques for interpretable ML developed by deep learning researchers are not satisfactory in understanding physical problems. In this study, our objective is to establish the physical linkage between features extracted from ML model and results derived from physical equations that describe the problem. This approach aims to provide a more comprehensible way of understanding an ML model. We select a less complex ML forecasting model, consisting of traditional statistical algorithms, which effectively forecast sea surface temperature anomaly (SSTA) and sea level anomaly (SLA) of the South China Sea (SCS) in 30 days ahead. Here, we focus on the SLA prediction and detect the physical mechanism of model capability to predict SCS SLA. Previous study has identified Rossby normal modes as the physical mechanism for long‐lived eddies in SCS in previous study. We here demonstrate the relationship between the ML model and Rossby normal modes in SCS in terms of temporal variations and spatial distributions of water mass. The model's skill in predicting Rossby normal modes partially explains its skill in predicting SLA, thereby providing a more transparent and interpretable basis for forecasts in SCS.

Sea surface temperature predictability assessment with an ensemble machine learning method using climate model simulations

Ensemble models, statistical analysis and machine learning (ML) can be used to predict novel conditions in a rapidly changing ocean. Traditionally, ML has been understood as a purely data-driven approach and has been used on both observational and model data to forecast Sea Surface Temperature (SST) anomalies. Here we use ML trained only on climate model simulations to predict regional SST variations, thereby suggesting a novel role for ML as an ensemble model interpolator. We propose a measure of the predictability provided by different ML implementations as well as by standard time series analysis methods. Weighting each forecast by this predictability measure computed on model data only, provides a significant improvement in forecast skill. We demonstrate the performance of this approach for regions around Australia, the Nino3.4 region (central-eastern equatorial Pacific) and in the eastern equatorial Pacific. These analyses show that SST predictability varies as a function of geographical location, area size, seasonality, proximity to the coast and model data quality.

Seasonal forecasting of mussel aquaculture meat yield in the Pelorus Sound

This study develops a novel approach to forecasting anomalies of meat yield from mussel aquaculture in Pelorus Sound, New Zealand, based on the relationships between non-local sea surface temperature (SST) and observations of mussel meat yield over 13 years. Overall, we found strong associations between lagged SSTs in the Tasman Sea region and mussel meat yield, with a noticeable seasonal cycle in these relationships. Results also showed that oceanic variables such as SST correlate more strongly with mussel meat yield than atmospheric variables, such as surface wind and other indices of atmospheric flow. The relationship between SST and mussel meat yield is linked to patterns of surface wind anomalies along the west coast of New Zealand. We identified regions where the SST was most correlated with mussel meat yield anomalies for each season and derived empirical relationships from linear regression. We then applied these empirical relationships to seasonal forecasts of SST from the European Centre for Medium-Range Weather Forecasts to generate seasonal forecasts of mussel meat yield. By validating our mussel meat yield forecasts from 13 years of retrospective forecasts, we find significant skill at lead times of up to 3 months in December–February and 5 months in September–November. During March-August forecasts are only skillful at a lead-time of 1 month. The results of this study have the potential to improve the accuracy and reliability of mussel meat yield forecasts and to provide valuable insights for the mussel industry.

Deep learning approach for forecasting sea surface temperature response to tropical cyclones in the Western North Pacific

Tropical cyclones (TCs) induce sea surface temperature cooling (SSTC), which is important for TC development itself as well as for variations of regional air-sea environment. TC-induced SSTC patterns and its prediction is still a challenge. In this study, a long short-term memory neural network deep learning model is developed to forecast TC-induced SSTC in the western North Pacific (WNP) using TCs during 2002–2016 as training cases and TCs during 2017–2018 as prediction and test cases. The 6-h TC-induced SSTC biases to the right-rear area of TC center, with a maximum of ∼0.6 °C on average, while SSTC is greater in higher latitudes than lower latitudes. The input variables for the deep learning model are surface wind at 10 m (U10 and V10), sea surface height (SSH), sea surface temperature (SST), and temperature at 100 m depth (T100), the output variable is SST 6 h after TCs. The model can predict TC-induced SSTC patterns, with an average mean absolute error of ∼0.081 °C, a root mean square error of ∼0.126 °C and a spatial anomaly correlation coefficient of ∼0.948. This work indicates that post-TC SSTC follows similar physical processes and nonlinear relationships with TC wind, initial SSH and ocean temperature, especially in deep-water regions. Although with some limitations, the deep learning model has the potential to be applied to operational forecasts.

Developing a subseasonal ecological forecast to reduce fisheries bycatch in the Northeast U.S.

Over the past decade, substantial progress has been made in projecting and predicting the spatial distribution of many marine species at seasonal to multidecadal time scales. However, managers and fishers often need to make decisions at much shorter time scales. Subseasonal environmental forecasts, which generate predictions over one to several weeks, can now be combined with species-specific habitat preference data to create ecological forecasts that could facilitate dynamic spatial management. The development of such predictive tools could aid in identifying optimal times and areas for fishers to maximize target catch and avoid nontarget catch. Nontarget catch, or bycatch, can have numerous and potentially severe economic and ecological consequences. Here, we focus on a population of anadromous fish known collectively as river herring (alewife and blueback herring), as they are species of concern and are heavily impacted by bycatch. Using bottom trawl survey data from the Northeast US and subseasonal forecasts of sea surface temperature, we constructed a bycatch risk model to generate probabilistic predictions of river herring distributions in regions frequented by the US mid-water trawl fishery. Assessments of model skill showed that our ecological model performed well in predicting the distribution of river herring and that subseasonal forecasts were effective at 1-week timeframes. There was a clear seasonal effect on forecasted bycatch risk throughout the Northeast US, with particularly high risk in winter and spring months. Importantly, variability in risk was detectable at the weekly timescale and our model identified specific areas and times that fishers should avoid in order to decrease their likelihood of bycatch. The bycatch risk forecast developed in this study is a significant advance from near-real time forecasts and the foundation to build forecast systems by combining species co-occurrence models with subseasonal forecasts. As these subseasonal forecasts are available globally, this approach could be adapted to facilitate the management of other natural resource conflicts around the world.

Interannual variation of summer sea surface temperature in the Amundsen Sea, Antarctica

This paper investigates the interannual variability of January sea surface temperature (SST) in the Amundsen Sea (AS) during the period 1982–2022. SST in the Pine Island Bay (PIB) is found to exhibit the most significant interannual variation, with a standard deviation up to 0.6°. Composite analysis indicates that, in warmer years, the January SST at PIB is approximately 1° higher on average than that in cooler years, and its variation in warmer (cooler) years corresponds to lower (higher) sea ice concentration (SIC) and more (less) surface heat flux; the latter factor is primarily influenced by the albedo of SIC. Further analysis suggests that variability in January SIC is largely dominated by northward sea ice motion during the previous November (r = −0.82), which is consistent with the presence of a contemporaneous northerly 10 m wind anomaly trigged by the Amundsen Sea Low (ASL). The ASL-associated northerly wind anomaly drives northward sea ice motion, reduces SIC, and thus increases the downward heat flux that ultimately results in warmer SST, and vice versa. This study identifies the possible mechanism of anomalous January SST in the PIB, which could provide an important clue for seasonal forecasts of summer SST in the AS.

Deep learning of sea-level variability and flood for coastal city resilience

Due to climate change, it is important to study the relationship between floods and sea-level rise in coastal city resilience. In this research sea surface temperature (SST) from MODIS, wind speed, precipitation, and sea-level rise from satellite altimetry are investigated for dynamic sea-level variability. An annual SST increase of 0.1C° is observed around the Gothenburg coast. Also in the middle of the North Sea, an annual increase of about 0.2C° is evident. The annual sea surface height (SSH) trend is 3 mm on the Gothenburg coast. We have a strong positive spatial correlation between SST and SSH near the Gothenburg coast. In the next step, dynamic sea-level variability is predicted with a convolution neural network and long short term memory. Root mean square error of wind speed, precipitation, SST, and mean sea-level forecasts are ±0.84 m/s, ±48.75 mm, ±3.48C° and ±24 mm, respectively. The 5-year trends of mean seal level show a significant increase from 28 mm/year to 46 mm/year in the last 5 year periods and the rate of increase has doubled. In the final step, the water rise of 5–10 m in Gothenburg city was simulated, and in the worst scenario, more than 50 % of the city will be damaged.

Probabilistic extreme SST and marine heatwave forecasts in Chesapeake Bay: A forecast model, skill assessment, and potential value

We test whether skillful 35-day probabilistic forecasts of estuarine sea surface temperature (SST) are possible and whether these forecasts could potentially be used to reduce the economic damages associated with extreme SST events. Using an ensemble of 35-day retrospective forecasts of atmospheric temperature and a simple model that predicts daily mean SST from past SST and forecast atmospheric temperature, we create an equivalent ensemble of retrospective SST forecasts. We compare these SST forecasts with reference forecasts of climatology and damped persistence and find that the SST forecasts are skillful for up to two weeks in the summer. Then, we post-process the forecasts using nonhomogeneous Gaussian regression and assess whether the resulting calibrated probabilistic forecasts are more accurate than the probability implied by the raw model ensemble. Finally, we use an idealized framework to assess whether these probabilistic forecasts can valuably inform decisions to take protective action to mitigate the effects of extreme temperatures and heatwaves. We find that the probabilistic forecasts provide value relative to a naive climatological forecast for 1-2 weeks of lead time, and the value is particularly high in cases where the cost of protection is small relative to the preventable losses suffered when a heatwave occurs. In most cases, the calibrated probabilistic forecasts are also more valuable than deterministic forecasts based on the ensemble mean and naive probabilistic forecasts based on damped persistence. Probabilistic SST forecasts could provide substantial value if applied to adaptively manage the rapid impacts of extreme SSTs, including managing the risks of catch-and-release mortality in fish and Vibrio bacteria in oysters.

Predicting sea surface temperatures with coupled reservoir computers

Abstract. Sea surface temperature (SST) is a key factor in understanding the greater climate of the Earth, and it is also an important variable when making weather predictions. Methods of machine learning have become ever more present and important in data-driven science and engineering, including in important areas for Earth science. Here, we propose an efficient framework that allows us to make global SST forecasts using a coupled reservoir computer method that we have specialized to this domain, allowing for template regions that accommodate irregular coastlines. Reservoir computing is an especially good method for forecasting spatiotemporally complex dynamical systems, as it is a machine learning method that, despite many randomly selected weights, is highly accurate and easy to train. Our approach provides the benefit of a simple and computationally efficient model that is able to predict SSTs across the entire Earth's oceans. The results are demonstrated to generally follow the actual dynamics of the system over a forecasting period of several weeks.

Bayesian negative binomial regression model with unobserved covariates for predicting the frequency of north atlantic tropical storms

Predicting the annual frequency of tropical storms is of interest because it can provide basic information towards improved preparation against these storms. Sea surface temperatures (SSTs) averaged over the hurricane season can predict annual tropical cyclone activity well. But predictions need to be made before the hurricane season when the predictors are not yet observed. Several climate models issue forecasts of the SSTs, which can be used instead. Such models use the forecasts of SSTs as surrogates for the true SSTs. We develop a Bayesian negative binomial regression model, which makes a distinction between the true SSTs and their forecasts, both of which are included in the model. For prediction, the true SSTs may be regarded as unobserved predictors and sampled from their posterior predictive distribution. We also have a small fraction of missing data for the SST forecasts from the climate models. Thus, we propose a model that can simultaneously handle missing predictors and variable selection uncertainty. If the main goal is prediction, an interesting question is should we include predictors in the model that are missing at the time of prediction? We attempt to answer this question and demonstrate that our model can provide gains in prediction.

Prediction Challenges From Errors in Tropical Pacific Sea Surface Temperature Trends

Models in the North American Multi-Model Ensemble (NMME) predict sea surface temperature (SST) trends in the central and eastern equatorial Pacific Ocean which are more positive than those observed over the period 1982–2020. These trend errors are accompanied by linear trends in the squared error of SST forecasts whose sign is determined by the mean model bias (cold equatorial bias is linked to negative trends in squared error and vice versa). The reason for this behavior is that the overly positive trend reduces the bias of models that are too cold and increases the bias of models that are too warm. The excessive positive SST trends in the models are also linked with overly positive trends in tropical precipitation anomalies. Larger (smaller) SST trend errors are associated with lower (higher) skill in predicting precipitation anomalies over the central Pacific Ocean. Errors in the linear SST trend do not explain a large percentage of variability in precipitation anomaly errors, but do account for large errors in amplitude. The predictions toward a too warm and wet tropical Pacific, especially since 2000, are strongly correlated with an increase in El Niño false alarms. These results may be relevant for interpreting the behavior of uninitialized CMIP5/6 models, which project SST trends that resemble the NMME trend errors.

The Influence of the Trend, Basin Interactions, and Ocean Dynamics on Tropical Ocean Prediction

AbstractThe trend, connections between tropical ocean basins and dynamical processes on sea surface temperature (SST) and sea surface height (SSH) forecast skill are investigated using a linear inverse model (LIM) framework. The warming trend has a strong influence on 6‐month SST forecast skill in the Indian Ocean, Western Pacific, and north tropical Atlantic, but little effect on El Niño‐Southern Oscillation (ENSO) prediction. ENSO strongly impacts the SST forecast skill in all three ocean basins including most of the Indian Ocean and the north tropical Atlantic. Without interactions with the Indian Ocean, the 6‐month SST forecast skill is substantially reduced in the western tropical Pacific and SSH skill decreases in the central Pacific. Atlantic and Indian Ocean interactions with the Pacific enhance ENSO forecast skill at 6–12 months leads. The Indian Ocean influences SSH in the eastern Atlantic, while the tropical Atlantic affects SST forecasts in portions of the Indian Ocean.

Improved of Forecasting Sea Surface Temperature based on Hybrid ARIMA and Support Vector Machines Models

Forecasting is a very effortful task owing to its features which simultaneously contain linear and nonlinear patterns. The Autoregressive Integrated Moving Average (ARIMA) model has been one the most widely utilized linear model in time series forecasting. Unfortunately, the ARIMA model cannot effortlessly handle nonlinear patterns alone. Thus, Support Vector Machine (SVM) model is introduced to solve nonlinear behavior in the datasets with high variance and uncertainty. The purposes of this study are twofold. First, to propose a hybrid ARIMA models using SVM. Secondly, to test the effectiveness of the proposed hybrid model using sea surface temperature (SST) data. Our investigation is based on two well-known real datasets, i.e., SST (modis) and in-situ SST (hycom). Statistical measurement such as MAE, MAPE, MSE, and RMSE are carried out to investigate the efficacy of the proposed models as compared to the previous ARIMA and SVMs models. The empirical results reveal that the proposed models produce lesser MAE, MAPE, MSE, and RMSE values in comparison to the single ARIMA and SVMs models. In additional, ARIMA-SVM are much better than compared to the existing models since the forecasting values are closer to the actual value. Therefore, we conclude that the presented models can be used to generate superior predicting values in time series forecasting with a way higher forecast precision.

Planning ahead: Dynamic models forecast blue whale distribution with applications for spatial management

Abstract Resources in the ocean are ephemeral, and effective management must therefore account for the dynamic spatial and temporal patterns of ecosystems and species of concern. We focus on the South Taranaki Bight (STB) of New Zealand, where upwelling generates productivity and prey to support an important foraging ground for blue whales that overlaps with anthropogenic pressure from industrial activities. We incorporate regional ecological knowledge of upwelling dynamics, physical–biological coupling and associated lags in models to forecast sea surface temperature (SST) and net primary productivity (NPP) with up to 3 weeks lead time. Forecasted environmental layers are then implemented in species distribution models to predict suitable blue whale habitat in the STB. Models were calibrated using data from the austral summers of 2009–2019, and ecological forecast skill was evaluated by predicting to withheld data. Boosted regression tree models skilfully forecasted SST (CV deviance explained = 0.969–0.970) and NPP (CV deviance explained = 0.738–0.824). The subsequent blue whale distribution forecast models had high predictive performance (AUC = 0.889), effectively forecasting suitable habitat on a daily scale with 1–3 weeks lead time. The spatial location and extent of forecasted blue whale habitat were variable, with the proportion of petroleum and mineral permit areas that overlapped with daily suitable habitat ranging from 0% to 70%. Hence, the STB and these forecast models are well‐suited for dynamic management that could reduce anthropogenic threats to whales while decreasing regulatory burdens to industry users relative to a traditional static protected area. Synthesis and applications. We develop and test ecological forecast models that predict sea surface temperature, net primary productivity and blue whale suitable habitat up to 3 weeks in the future within New Zealand's South Taranaki Bight region. These forecasts of whale distribution can be effectively applied for dynamic spatial management due to model foundation on quantified links and lags between physical forcing and biological responses. A framework to operationalize these forecasts through a user‐driven application is in development to proactively inform conservation management decisions. This framework is implemented through stakeholder engagement, allows flexibility based on management objectives, and is amenable to improvement as new knowledge and feedback are received.

A Bayesian Machine Learning Algorithm for Predicting ENSO Using Short Observational Time Series

AbstractA simple and efficient Bayesian machine learning (BML) training algorithm, which exploits only a 20‐year short observational time series and an approximate prior model, is developed to predict the Niño 3 sea surface temperature (SST) index. The BML forecast significantly outperforms model‐based ensemble predictions and standard machine learning forecasts. Even with a simple feedforward neural network (NN), the BML forecast is skillful for 9.5 months. Remarkably, the BML forecast overcomes the spring predictability barrier to a large extent: the forecast starting from spring remains skillful for nearly 10 months. The BML algorithm can also effectively utilize multiscale features: the BML forecast of SST using SST, thermocline, and windburst improves on the BML forecast using just SST by at least 2 months. Finally, the BML algorithm also reduces the forecast uncertainty of NNs and is robust to input perturbations.