Sea Surface Temperature Prediction Research Articles

The prediction of two‐dimensional intelligent ocean temperature based on deep learning

AbstractAn important data in intelligent ocean is the sea surface temperature (SST). Most of the previous works on SST prediction deal with independent spatial point, ignoring the spatial correlation of two‐dimensional intelligent ocean, which leads to instability of prediction accuracy. Therefore, in this paper, a deep learning model based on convolutional gated recurrent unit is proposed for the SST prediction of two‐dimensional intelligent ocean. Input and output of the proposed model are both spatiotemporal SST data, which means the model directly process spatiotemporal data. This method adds the spatial information of the intelligent ocean temperature data, thereby improving the accuracy. From the experiments, it turns out that the proposed model shows superior performance compared to another three prevailing deep learning models.

An Ensemble Kalman Filter Data Assimilation Method for the Sea Surface Temperature in the China Seas: Implementation and Simulation Experiments

Data assimilation refers to a method of integrating observation data in the dynamic operation of numerical models on the basis of considering the temporal and spatial distribution of data and the error of observation field and background field. The Ensemble Kalman filter (EnKF) as a technology that has been widely used in the field of atmosphere and ocean has been applied to the ROMS (the Regional Ocean Modeling System) for predicting Sea surface temperature in Yellow, and East China Seas. In order to explore the applicability and effectiveness of the EnKF method for improving the accuracy of marine numerical model, the Sea surface temperature (SST) gained from buoy were used to conduct data assimilation process with EnKF method. Twin experiments have been performed to analysis the sensitivity of this system to the ensemble size and errors in model simulation and observations and a real data assimilation scheme has been conducted to hindcast the SST at the Yellow, and East China Seas during the July of year 2014. The updated results after data assimilation indicate that the model simulation fits observation better when the forecast was updated by observations. The result show that EnKF can effectively reduce the simulation error of complex numerical marine models.

Quantifying the Benefits of Altimetry Assimilation in Seasonal Forecasts of the Upper Ocean

AbstractSatellite altimetry measurements of sea surface height provide near‐global ocean state observations on sub‐monthly time scales, which are not always utilized by seasonal climate forecasting systems. As early as the mid‐1990s, attempts were made to assimilate altimetry observations to initialize climate models. These experiments demonstrated improved ocean forecasting skill, especially compared to experiments that did not assimilate subsurface ocean temperature information. Nowadays, some operational climate forecasting models utilize altimetry in their assimilation systems, whereas others do not. Here, we assess the impact of altimetry assimilation on seasonal prediction skill of ocean variables in two climate forecasting systems that are from the European Centre for Medium‐Range Weather Forecasts (SEAS5) and the Australian Bureau of Meteorology (ACCESS‐S). We show that assimilating altimetry improves the initialization of subsurface ocean temperatures, as well as seasonal forecasts of monthly variability in upper‐ocean heat content and sea level. Skill improvements are largest in the subtropics, where there are typically less subsurface ocean observations available to initialize the forecasts. In the tropics, there are no noticeable improvements in forecast skill. The positive impact of altimetry assimilation on forecast skill related to the subsurface ocean does not seem to affect predictions of sea surface temperature. Whether this is because current forecasting systems are close to the potential predictability limit for the ocean surface, or perhaps altimetry observations are not fully exploited, remains a question. In summary, we find that utilizing altimetry observations improves the overall global ocean forecasting skill, at least for upper‐ocean heat content and sea level.

The CNN-GRU model with frequency analysis module for sea surface temperature prediction

The CNN-GRU model with frequency analysis module for sea surface temperature prediction

On Evaluating the Predictability of Sea Surface Temperature Using Entropy

Sea surface temperature (SST) has important impacts on the global ecology, and having a good understanding of the predictability, i.e., the possibility of achieving accurate prediction, of SST can help us monitor the marine environment and climate change, and guide the selection and design of SST prediction methods. However, existing studies for analyzing SST mostly measure the rising or falling trends of SST. To address this issue, we introduce a temporal-correlated entropy to quantify the predictability of SST series from both global coarse-grained and local fine-grained aspects, and make SST prediction with multiple deep learning models to prove the effectiveness of such predictability evaluation method. In addition, we explore the dynamics of SST predictability by dividing the time range of interest into consecutive time periods, evaluating the corresponding predictability of SST for each time period, and analyzing the stability of the predictability of SST over time. According to the experiments, the SST predictability values near the poles and equator are really high. The average SST predictability values of the East China Sea, Bohai Sea, and Antarctic Ocean are 0.719, 0.706, and 0.886, respectively, and the size relationship of the SST predictability in the three local sea areas is consistent with our prediction results using multiple representative SST prediction methods, which corroborates the reliability of the predictability evaluation method. In addition, we found that the SST predictability in the Antarctic Ocean changes more dramatically over time than in the East China Sea and the Bohai Sea. The results of SST predictability and its dynamic analysis indicate that global warming, ocean currents, and human activities all have significant impacts on the predictability of SST.

Prediction of Sea Surface Temperature in the South China Sea Based on Deep Learning

Sea surface temperature is an important physical parameter in marine research. Accurate prediction of sea surface temperature is important for coping with climate change, marine ecological protection, and marine economic development. In this study, the SST prediction performance of ConvLSTM and ST-ConvLSTM with different input lengths, prediction lengths, and hidden sizes is investigated. The experimental results show that: (1) The input length has an impact on the prediction results of SST, but it does not mean that the longer the input length, the better the prediction performance. ConvLSTM and ST-ConvLSTM have the best prediction performance when the input length is set to 1, and the prediction performance gradually decreases as the input length increases. (2) Prediction length affects SST prediction. As the prediction length increases, the prediction performance gradually decreases. When other parameters are kept constant and only the prediction length is changed, the ConvLSTM gets the best result when the prediction length is set to 2, and the ST-ConvLSTM gets the best result when the prediction length is set to 1. (3) The setting of the hidden size has a great influence on the prediction ability of the sea surface temperature, but the hidden size cannot be set blindly. For ST-ConvLSTM, although the prediction performance of SST is better when the hidden size is set to 128 than when it is set to 64, the consequent computational cost increases by about 50%, and the performance only improves by about 10%.

Study of the Decadal Predictability of Mediterranean Sea Surface Temperature Based on Observations

Abstract Sea surface temperature (SST) changes in the Mediterranean Sea have profound impacts on both the Mediterranean regions and remote areas. Previous studies show that the Mediterranean SST has significant decadal variability that is comparable with the Atlantic multidecadal variability (AMV). However, few studies have discussed the characteristics and sources of the decadal predictability of Mediterranean SST based on observations. Here for the first time we use observational datasets to reveal that the decadal predictability of Mediterranean SST is contributed by both external forcings and internal variability for both annual and seasonal means, except that the decadal predictability of the winter mean SST in the eastern Mediterranean is mostly contributed by only internal variability. Besides, the persistence of the Mediterranean SST is quite significant even in contrast with that in the subpolar North Atlantic, which is widely regarded to have the most predictable surface temperature on the decadal time scale. After the impacts of external forcings are removed, the average prediction time of internally generated Mediterranean SST variations is more than 10 years and closely associated with the multidecadal variability of the Mediterranean SST that is closely related to the accumulated North Atlantic Oscillation forcing. Significance Statement Decadal prediction of sea surface temperature (SST) in the Mediterranean Sea is very important due to its profound climate impacts. Understanding the spatial/seasonal characteristics and sources of the decadal predictability of the Mediterranean SST are crucial for skillful decadal prediction of the Mediterranean SST and related variations. This study for the first time shows that the decadal predictability of the Mediterranean SST is quite significant even in comparison with the AMV. The decadal predictability of the Mediterranean SST is both region and season dependent. The internally generated decadal predictability of the Mediterranean SST is most probably related to the local ocean dynamics.

Impact of Westerly Wind Bursts (WWBs) on ENSO based on a Hybrid Coupled Model: Part II – ENSO Prediction

ABSTRACT With a westerly wind burst (WWB) parameterization scheme introduced into a hybrid coupled model (HCM), we investigated in Part 1 of this study the impact of WWBs on El Nino – Southern Oscillation (ENSO) simulation features, including asymmetry, phase locking, and diversity. In the second part, we investigate the impact of WWBs on ENSO prediction skills. To achieve this, two ensemble experiments, one with WWBs and one without WWBs, are performed to evaluate the predictions of sea surface temperature (SST) anomalies. The results show that both experiments can predict the SST anomalies in the equatorial central and eastern Pacific up to lead times of 12 months. The correlation coefficient between the model and observations shows that the WWB experiment produces better prediction skills than the experiment without WWBs, especially at lead times longer than four months during El Niño events. This result is consistent with the expectation that the WWB parameterization scheme plays an important role in describing physical processes, which is indicated in Part 1. We also presented a predictability analysis for Central Pacific (CP) and Eastern Pacific (EP) El Niño events. The prediction of both types of El Niño events is also improved by the WWB parameterization scheme at long lead times.

Deep-learning model for sea surface temperature prediction near the Korean Peninsula

Recently, sea surface temperatures (SSTs) near the Korean Peninsula have increased rapidly due to global warming; this phenomenon can lead to high water temperatures and extensive damage to Korean fish farms. To reduce such damage, it is necessary to predict high water temperature events in advance. In this study, we developed a method for predicting high water temperature events using time series SST data for the Korean Peninsula obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 product and a long short-term memory (LSTM) network designed for time series data prediction. First, the SST prediction model was used to predict SSTs. Predicted SSTs exceeding 28 °C, which is the Korean government standard for issuing high water temperature warnings, were designated as high water temperatures. To evaluate the prediction accuracy of the SST prediction model, 1-to 7-day predictions were evaluated in terms of the coefficient of determination (R2), root mean square error (RMSE), and mean absolute percentage error (MAPE). The R2, RMSE, and MAPE values of the 1-day prediction SST model were 0.985, 0.14 °C, and 0.38%, respectively, whereas those of the 7-day prediction SST model were 0.574, 0.74 °C, and 2.26%, respectively. We also calculated F1 scores to evaluate high water temperature classification accuracy. The F1 scores of the 1- and 7-day SST prediction models were 0.963 and 0.739, respectively.

Skillful decadal prediction skill over the Southern Ocean based on GFDL SPEAR Model-Analogs

The Model-Analogs technique is used in the present study to assess the decadal sea surface temperature (SST) prediction skill over the Southern Ocean (SO). The Model-Analogs here is based on reanalysis products and model control simulations that have ∼1° ocean/ice (refined to 0.5° at high latitudes) components and 100 km atmosphere/land components. It is found that the model analog hindcasts show comparable skills with the initialized retrospective decadal hindcasts south of 50°S, with even higher skills over the Weddell Sea at longer lead years. The high SST skills primarily arise from the successful capture of SO deep convection states. This deep ocean memory and the associated decadal predictability are also clearly seen when we assess the Model-Analogs technique in a perfect model context. Within 30°S–50°S latitudinal band, the model analog hindcasts show low skills. When we include the externally forced signals estimated from the large ensemble simulations, the model analog hindcasts and initialized decadal hindcasts show identical skills. The Model-Analogs method therefore provides a great baseline for developing future decadal forecast systems. It is unclear whether such analog techniques would also be successful with models that explicitly resolve ocean mesoscale eddies or other small-scale processes. This area of research needs to be explored further.

Penggunaan Metode Rolling Mosaic Untuk Mendukung Pengembangan Peta Prakiraan Daerah Penangkapan Ikan Wilayah Pesisir

Predicted fishing ground maps (PPDPI) which made using satellite image data is often constrained by clouds, causing its production not too optimal. Rolling mosaic methods examined here is expected to reduce cloud cover so the information about oceanographic conditions can be more visible and can increasing PPDPI production. In July, the percentage of sea surface temperature data can increase from 15.3%-30.29% using 1-day mosaic data, to 40.46%-56.75% using 3-day mosaic, it increases to 72.24%-77.88% using 7-day mosaic data and increase to 84.19%-89.07% using 14-day mosaic. While the percentage of sea surface temperature data in December can be increased from around 4.93%-13.03% to 41.48%- 51.60%. In general, at July and December, the relationship between 1-day mosaic and 3-days mosaic data, 7-days and 14-days are very strong, but the strength of the relationship will decrease (the correlation coefficient gets smaller) along with the increasing of the time range used to mosaicking the data. The RMSE shows that the RMSE between the 1-day mosaic with 3-days mosaic is 0.288 (July), 0.263 (December); RMSE between 1-day mosaic and 7-days mosaic is 0.388 (July), 0.387 (December) and RMSE between 1-day mosaic and 14-days mosaic is 0.471 (July), 0.477 (December). This RMSE values shows that the longer time range used to construct the mosaic, the error value will also increase. Scoring analysis using percentage of data, correlation coefficient and RMSE as a parameters indicate that the 7-days mosaic method has the highest score so it is considered as the best method to be used to predict sea surface temperature with minimum cloud cover.

Deep-learning model for sea surface temperature prediction near the Korean Peninsula

Recently, sea surface temperatures (SSTs) near the Korean Peninsula have increased rapidly due to global warming; this phenomenon can lead to high water temperatures and extensive damage to Korean fish farms. To reduce such damage, it is necessary to predict high water temperature events in advance. In this study, we developed a method for predicting high water temperature events using time series SST data for the Korean Peninsula obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 product and a long short-term memory (LSTM) network designed for time series data prediction. First, the SST prediction model was used to predict SSTs. Predicted SSTs exceeding 28 °C, which is the Korean government standard for issuing high water temperature warnings, were designated as high water temperatures. To evaluate the prediction accuracy of the SST prediction model, 1-to 7-day predictions were evaluated in terms of the coefficient of determination (R2), root mean square error (RMSE), and mean absolute percentage error (MAPE). The R2, RMSE, and MAPE values of the 1-day prediction SST model were 0.985, 0.14 °C, and 0.38%, respectively, whereas those of the 7-day prediction SST model were 0.574, 0.74 °C, and 2.26%, respectively. We also calculated F1 scores to evaluate high water temperature classification accuracy. The F1 scores of the 1- and 7-day SST prediction models were 0.963 and 0.739, respectively.

A comparison of seasonal rainfall forecasts over Central America using dynamic and hybrid approaches from Copernicus Climate Change Service seasonal forecasting system and the North American Multimodel Ensemble

AbstractSeasonal rainfall forecasts provide information several months ahead to support decision making. These forecasts may use dynamic, statistical, or hybrid approaches, but their comparative value is not well understood over Central America. This study conducts a regional evaluation of seasonal rainfall forecasts focusing on two of the leading dynamic climate ensembles: the Copernicus Climate Change Service seasonal forecasting system (C3S) and the North American Multimodel Ensemble (NMME). We compare the multimodel ensemble mean and individual model predictions of seasonal rainfall over key wet season periods in Central America to better understand their relative forecast skill at the seasonal scale. Three types of rainfall forecasts are compared: direct dynamic rainfall predictions from the C3S and NMME ensembles, a statistical approach using the lagged observed sea surface temperature (SST), and an indirect hybrid approach, driving a statistical model with dynamic ensemble SST predictions. Results show that C3S and NMME exhibit similar regional variability with strong performance in the northern Pacific part of Central America and weaker skill primarily in eastern Nicaragua. In the northern Pacific part of the region, the models have high skill across the wet season. Indirect forecasts can outperform the direct rainfall forecasts in specific cases where the direct forecasts have lower predictive power (e.g., eastern Nicaragua during the early wet season). The indirect skill generally reflects the strength of SST associations with rainfall. The indirect forecasts based on Tropical North Atlantic SSTs are best in the early wet season and the indirect forecasts based on Niño3.4 SSTs are best in the late wet season when each SST zone has a stronger association with rainfall. Statistical predictions are competitive with the indirect and direct forecasts in multiple cases, especially in the late wet season, demonstrating how a variety of forecasting approaches can enhance seasonal forecasting.

Robust Prediction of Sea Surface Temperature Based on SSPGAN

Robust Prediction of Sea Surface Temperature Based on SSPGAN

Physical Knowledge-Enhanced Deep Neural Network for Sea Surface Temperature Prediction

Traditionally, numerical models have been deployed in oceanography studies to simulate ocean dynamics by representing physical equations. However, many factors pertaining to ocean dynamics seem to be ill-defined. We argue that transferring physical knowledge from observed data could further improve the accuracy of numerical models when predicting Sea Surface Temperature (SST). Recently, the advances in earth observation technologies have yielded a monumental growth of data. Consequently, it is imperative to explore ways in which to improve and supplement numerical models utilizing the ever-increasing amounts of historical observational data. To this end, we introduce a method for SST prediction that transfers physical knowledge from historical observations to numerical models. Specifically, we use a combination of an encoder and a generative adversarial network (GAN) to capture physical knowledge from the observed data. The numerical model data is then fed into the pre-trained model to generate physics-enhanced data, which can then be used for SST prediction. Experimental results demonstrate that the proposed method considerably enhances SST prediction performance when compared to several state-of-the-art baselines.

Global Spatiotemporal Graph Attention Network for Sea Surface Temperature Prediction

Accurately predicting sea surface temperature (SST) plays an important role in the study of marine ecosystems and global climate. The SST prediction problem is usually formulated as a time series regression problem, i.e., the future SST is predicted based on the historical SST. However, existing methods are typically devoted to modeling the highly nonlinear temporal correlations in SST data. They often ignore the dynamic spatial correlations that exist. This can limit the performance of these models, making accurately predicting SST challenging. To address this challenge, by combining graph neural networks that have a clear advantage in modeling spatial correlations, we propose a global spatio-temporal graph attention network (GSTGAT). Specifically, we capture the global dynamic spatial correlations of nodes through a global graph attention network module that fuses the static adjacency matrix learned adaptively by the graph learning module with the dynamic attention coefficients. A gated temporal convolutional network module is used to capture the nonlinear temporal correlations. Then, the above modules are integrated into a unified neural network to predict SST. We conduct experiments on multiple time-scale datasets in the Bohai Sea and the South China Sea. The experimental results show that GSTGAT achieves the best performance and consistently outperforms other methods for different prediction horizons in the two sea areas.

Seven-day sea surface temperature prediction using a 3DConv-LSTM model

Due to the application demand, users have higher expectations for the accuracy and resolution of sea surface temperature (SST) products. Recent advances in deep learning show great advantages in exploiting massive ocean datasets, and provides opportunities for investigating regional SST predictions in an efficiency approach. However, for deep learning-based SST prediction to be adopted by users, the output must be accurate. This paper investigates the 7-day SST prediction over the China seas and their adjacent waters at a 0.05° spatial resolution. To improve the prediction’s accuracy, we designed a deep learning model combining the three-dimensional convolution and long short-term memory under multi-input multi-output strategy. The Operational SST and Sea Ice Analysis (OSTIA) SST anomaly was used as training data. To test the model prediction ability, we verified the predicted results with the Sub-seasonal to Seasonal (S2S) prediction data from 2015 to 2019. Validation of the predicted SSTs using the OSTIA test datasets show that the root-mean-square error increases from 0.27°C to 0.53°C during the 1- to 7-day lead time, with predictability decreases from southeast to northwest in the study area. Furthermore, the comparison of predicted SST and S2S data with Argo shows that our model is slightly more accurate, which can achieve -0.08°C bias, with a standard deviation of 0.35°C for a 1-day lead time and -0.07°C bias, with a standard deviation of 0.59°C for a 7-day lead time. The results indicate that the proposed deep learning model is accurate and can be applied in regional daily SST prediction.

A deep learning approach to predict sea surface temperature based on multiple modes

Sea surface temperature (SST) is an essential physical parameter and plays a vital role in ocean-atmosphere interactions. Accurate SST prediction relies on numerical model methods, which require the understanding of complex dynamical and thermal processes. Data-driven SST prediction methods based on deep learning have been studied to obtain quick results. However, recent deep learning models can hardly accurately capture and simulate complex SST patterns, and thus give only lower resolution predictions. Here, SST map prediction is regarded as a spatiotemporal sequence prediction task. The memory in memory (MIM) model and variational mode decomposition (VMD) are combined (VMD-MIM) to accurately detect variations in SST patterns. The original SST data is firstly decomposed into different modes with different frequencies using VMD. For each mode extracted by VMD, MIM is then applied to obtain the corresponding prediction maps separately. SST data from the South China Sea are used to estimate the VMD-MIM method, in which we enter the ten-day SST maps and predict the next seven days. The results show that VMD-MIM can noticeably improve prediction skills by accurately presenting the fine structures (1/10°) of SST maps, which provides a valuable pathway for fast and lightweight short-term spatiotemporal SST predictions.

Assessment of the spatiotemporal prediction capabilities of machine learning algorithms on Sea Surface Temperature data: A comprehensive study

Spatiotemporal time series prediction plays a crucial role in a wide range of applications. However, in most of the studies, spatial information was ignored and predictions were carried out either on a few points or on average values. In this study, 37 different configurations of 4 traditional ML models and 3 Neural Network (NN) based models were utilized to provide a comprehensive comparison and evaluate the spatiotemporal data prediction capabilities of the ML models. Additionally, to reveal the importance of spatial data for the time series prediction process, the best configuration of each ML model was evaluated with and without using spatial information. The utilized models were: (i) Linear Regression (LR), (ii) K-Nearest Neighbors (KNN), (iii) Decision-Trees (DT), (iv) Support Vector Machine (SVM), (v) Multi-Layer Perceptron (MLP), (vi) Long Short-Term Memory (LSTM), and (vii) Gated Recurrent Unit (GRU). The study was performed on the Sea Surface Temperature (SST) data collected by satellite radiometers via infrared measurements. The models were evaluated according to their one-month ahead spatiotemporal SST prediction performance over the southern coasts of Turkey, and the effects of spatial information on model performance were presented. Results reveal that the spatial information increased the prediction performance by approximately 25%, in terms of RMSE. Additionally, acquired results show that the LSTM model outperforms all other ML models and gives the smallest prediction errors in all metrics.

Impact of ocean data assimilation on climate predictions with ICON-ESM

We develop a data assimilation scheme with the Icosahedral Non-hydrostatic Earth System Model (ICON-ESM) for operational decadal and seasonal climate predictions at the German weather service. For this purpose, we implement an Ensemble Kalman Filter to the ocean component as a first step towards a weakly coupled data assimilation. We performed an assimilation experiment over the period 1960–2014. This ocean-only assimilation experiment serves to initialize 10-year long retrospective predictions (hindcasts) started each year on 1 November. On multi-annual time scales, we find predictability of sea surface temperature and salinity as well as oceanic heat and salt contents especially in the North Atlantic. The mean Atlantic Meridional Overturning Circulation is realistic and the variability is stable during the assimilation. On seasonal time scales, we find high predictive skill in the tropics with highest values in variables related to the El Niño/Southern Oscillation phenomenon. In the Arctic, the hindcasts correctly represent the decreasing sea ice trend in winter and, to a lesser degree, also in summer, although sea ice concentration is generally much too low in both hemispheres in summer. However, compared to other prediction systems, prediction skill is relatively low in regions apart from the tropical Pacific due to the missing atmospheric assimilation. Further improvements of the simulated mean state of ICON-ESM, e.g. through fine-tuning of the sea ice and the oceanic circulation in the Southern Ocean, are expected to improve the predictive skill. In general, we demonstrate that our data assimilation method is successfully initializing the oceanic component of the climate system.