Sea Surface Temperature Prediction Research Articles

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.

Atmosphere-driven cold SST biases over the western North Pacific in the GloSea5 seasonal forecast system

The predictability of the sea surface temperature (SST) in seasonal forecast systems is crucial for accurate seasonal predictions. In this study, we evaluated the prediction of SST in the Global Seasonal forecast system version 5 (GloSea5) hindcast, particularly focusing on the western North Pacific (WNP), where the SST can modify atmospheric convection and the East Asian weather. GloSea5 has a cold SST bias in the WNP that grows over at least 7 months. The bias originates from the surface net heat flux. At the beginning of model integration, the ocean receives excessive heat from the atmosphere because of the predominant positive bias in the downward shortwave radiation (SW), which rapidly decreased within a few days as cloud cover builds. Then, the negative bias in the latent heat (LH) flux increases over time and induces a negative bias in the surface net heat flux. Although the magnitude of the negative bias in LH flux gradually decreases, it remains the most significant contributor to the negative bias in the net heat flux bias for more than 250 days. Uncoupled ocean model experiments showed that the ocean model is unlikely to be the primary source of the SST bias.

High Precision Sea Surface Temperature Prediction of Long Period and Large Area in the Indian Ocean Based on the Temporal Convolutional Network and Internet of Things.

Impacted by global warming, the global sea surface temperature (SST) has increased, exerting profound effects on local climate and marine ecosystems. So far, investigators have focused on the short-term forecast of a small or medium-sized area of the ocean. It is still an important challenge to obtain accurate large-scale and long-term SST predictions. In this study, we used the reanalysis data sets provided by the National Centers for Environmental Prediction based on the Internet of Things technology and temporal convolutional network (TCN) to predict the monthly SSTs of the Indian Ocean from 2014 to 2018. The results yielded two points: Firstly, the TCN model can accurately predict long-term SSTs. In this paper, we used the Pearson correlation coefficient (hereafter this will be abbreviated as “correlation”) to measure TCN model performance. The correlation coefficient between the predicted and true values was 88.23%. Secondly, compared with the CFSv2 model of the American National Oceanic and Atmospheric Administration (NOAA), the TCN model had a longer prediction time and produced better results. In short, TCN can accurately predict the long-term SST and provide a basis for studying large oceanic physical phenomena.

Seasonal forecast skill of upper-ocean heat content in coupled high-resolution systems

Ocean heat content (OHC) anomalies typically persist for several months, making this variable a vital component of seasonal predictability in both the ocean and the atmosphere. However, the ability of seasonal forecasting systems to predict OHC remains largely untested. Here, we present a global assessment of OHC predictability in two state-of-the-art and fully-coupled seasonal forecasting systems. Overall, we find that dynamical systems make skilful seasonal predictions of OHC in the upper 300 m across a range of forecast start times, seasons and dynamical environments. Predictions of OHC are typically as skilful as predictions of sea surface temperature (SST), providing further proof that accurate representation of subsurface heat contributes to accurate surface predictions. We also compare dynamical systems to a simple anomaly persistence model to identify where dynamical systems provide added value over cheaper forecasts; this largely occurs in the equatorial regions and the tropics, and to a greater extent in the latter part of the forecast period. Regions where system performance is inadequate include the sub-polar regions and areas dominated by sharp fronts, which should be the focus of future improvements of climate forecasting systems.

Sea Surface Temperature Prediction With Memory Graph Convolutional Networks

We develop a memory graph convolutional network (MGCN) framework for sea surface temperature (SST) prediction. The MGCN consists of two memory layers: one graph layer and one output layer. The memory layer captures SST temporal changes via temporal convolution units and gate linear units. The graph layer encodes SST spatial changes in terms of characteristics derived from graph Laplacian. The output layer encapsulates information from the previous layers and produces SST prediction results. The MGCN characterizes both the temporal and spatial changes, rendering a comprehensive SST prediction strategy. We use daily mean SST data for two areas near the Bohai Sea and the East China Sea for experimental evaluations and validate that the MGCN performs better than other traditional machine learning methods for nearshore SST prediction. In addition, we test the MGCN on weekly and monthly mean SST datasets and validate that the MGCN is robust and suitable for SST prediction.

Multiple Granularity Spatiotemporal Network for Sea Surface Temperature Prediction

Sea surface temperature (SST) prediction has an important practical value in marine disaster prevention and mitigation. Most current methods only use the temporal correlation of SST during prediction, but the spatial correlation is not considered, resulting in low prediction accuracies. In addition, the changing trend of SST as reflected by the single granularity feature is unreliable, and the degrees of dependence between historical SST and future SST tend to vary. In order to overcome these issues, the multiple granularity spatiotemporal network (MGSN) is proposed for SST prediction. The proposed method consists of three parts. First, a multibranch network structure is constructed to extract different temporal features of different granularities. Second, a temporal dependence representation module is developed to represent the different degrees of dependence between historical SST and predicted SST in the temporal dimension. Third, the spatiotemporal fusion prediction module is used to achieve a spatiotemporal prediction of the SST and fuse the prediction results of different granular features. Comparative experiments have been conducted. The experimental results show that the root-mean-square error (RMSE) of the proposed method is reduced by 0.1360, 0.1608, and 0.1448 compared with the RMSE of convolutional LSTM (ConvLSTM), when predicting SST for the next one day, three days, and seven days, respectively. Our method has strong spatiotemporal feature modeling capabilities and is suitable for regional SST prediction.

Sea Surface Temperature Forecasting With Ensemble of Stacked Deep Neural Networks

Oceanic temperature has a great impact on global climate and worldwide ecosystems, as its anomalies have been shown to have a direct impact on atmospheric anomalies. The major parameter for measuring the thermal energy of oceans is the sea surface temperature (SST). SST prediction plays an essential role in climatology and ocean-related studies. However, SST prediction is challenging due to the involvement of complex and nonlinear sea thermodynamic factors. To address this challenge, we design a novel ensemble of two stacked deep neural networks (DNNs) that uses air temperature, in addition to water temperature, to improve the SST prediction accuracy. To train our model and compare its accuracy with the state-of-the-art, we employ two well-known datasets from the national oceanic and atmospheric administration as well as the international Argo project. Using DNNs, our proposed method is capable of automatically extracting required features from the input time series and utilizing them internally to provide a highly accurate SST prediction that outperforms state-of-the-art models.

Deep Learning Models to Predict Sea Surface Temperature in Tohoku Region

The prediction of sea surface temperature (SST) is a challenging task, especially for regions with high SST variability. Such predictions are either achieved by physics-based models, which often yield poor predictions and are computationally intensive, or by using data-driven methods, which are skillful and computationally less intensive. However, recent machine learning studies exploring SST prediction have not included the important meteorological parameters governing SST variability. Therefore, in this study, we propose various of deep learning (DL) models trained using past meteorological features to predict day-ahead SST. The proposed models include a deep multilayer perceptron (deep MLP), long short-term memory network (LSTM) and spatial 2-dimensional convolutional neural network (spatial 2D CNN). We explore the potential of the proposed DL models for day-ahead SST prediction across different locations in the Tohoku region (Japan’s east coast), including in situ validation. Evaluation of these DL models’ prediction skills suggests that the spatial 2D CNN’s are highly skillful at coastal locations, whereas at offshore locations, equal prediction skills were noted from the deep MLP and LSTM. We further attempted to improve the spatial 2D CNN by including past SST features, and such improvisation showed very low errors ranging from 0.35°C to 0.75°C and high correlation skill from 0.64 to 0.96. These improved skills were also compared with persistent model (PM) skills using RMSE and Correlation (RC) phase diagram, where we found that improved skills are consistently better than PM skills. In addition, we extracted features from the spatial 2D CNN to understand the reason underlying such improved skills, and we noted that the proposed DL model successfully captured the major meteorological and oceanic features governing SST variability. This led us to conclude that the proposed DL models are capable of producing highly reliable SST predictions, and may be equally applicable to other study regions.

A TMA-Seq2seq Network for Multi-Factor Time Series Sea Surface Temperature Prediction

Sea surface temperature (SST) is closely related to global climate change, ocean ecosystem, and ocean disaster. Accurate prediction of SST is an urgent and challenging task. With a vast amount of ocean monitoring data are continually collected, data-driven methods for SST time-series prediction show promising results. However, they are limited by neglecting complex interactions between SST and other ocean environmental factors, such as air temperature and wind speed. This paper uses multi-factor time series SST data to propose a sequence-to-sequence network with two-module attention (TMA-Seq2seq) for long-term time series SST prediction. Specifically, TMA-Seq2seq is an LSTM-based encoder-decoder architecture facilitated by factor- and temporal-attention modules and the input of multi-factor time series. It takes six-factor time series as the input, namely air temperature, air pressure, wind speed, wind direction, SST, and SST anomaly (SSTA). A factor attention module is first designed to adaptively learn the effect of different factors on SST, followed by an encoder to extract factor-attention weighted features as feature representations. And then, a temporal attention module is designed to adaptively select the hidden states of the encoder across all time steps to learn more robust temporal relationships. The decoder follows the temporal-attention module to decode the feature vector concatenated from the weighted features and original input feature. Finally, we use a fully-connect layer to map the feature into prediction results. With the two attention modules, our model effectively improves the prediction accuracy of SST since it can not only extract relevant factor features but also boost the long-term dependency. Extensive experiments on the datasets of China Coastal Sites (CCS) demonstrate that our proposed model outperforms other methods, reaching 98.29% in prediction accuracy (PACC) and 0.34 in root mean square error (RMSE). Moreover, SST prediction experiments in China’s East, South, and Yellow Sea site data show that the proposed model has strong robustness and multi-site applicability.

Predicting Sea Surface Temperature Based on a Parallel Autoreservoir Computing Approach With Short-Term Measured Data

Sea surface temperature (SST) is an essential parameter for observing the marine environment. It directly reflects the state of heat storage and release in the ocean. The change in SST will cause many phenomena that profoundly affect human production and life. Therefore, predicting SST accurately and efficiently can help us avoid many risks. In this article, we present a method based on neural networks. First, we propose a finite-dimensional description of SST, which investigates SST in the finite-dimensional phase space. Based on phase space reconstruction technology and the autoreservoir neural network (ARNN), the Spatial Parallel ARNN (SPARNN) is proposed for predicting SST. Unlike the previous machine learning methods that require big data, our approach only needs a small amount of locally short-term data to catch the dynamic features of the SST field. It also has excellent parallelism and is easy to run on a large-scale computer platform.

Retrieving Remotely Sensed Data Using Plank Algorithm in the Black Sea and Surrounding Seas (2008-2018)

The Black Sea is the world’s largest inner sea, its only connections with other seas being the Kerch Strait, which leads to the Azov Sea, and the Istanbul Strait, which connects it with the Mediterranean Sea. The main aim of this study processes to image analysis and applied algorithm, to retrieve remote SST data from satellite, to assess the estimated SSTs, to understand the characteristics of SST by presenting in the Black Sea and surrounding seas, to discuss the SSTs changes according to climate change impacts. In this study, Landsat 7 and Landsat 8 Operational Land Imager (OLI)/Thermal Infrared Sensor (TIRS) Level-1 images of TIR bands are used to predict Sea Surface Temperature (SST) changes using the Plank algorithm in the Black Sea and surrounding seas between 2008 and 2018. The SST values were taken from 55 different stations chosen from the study area. The SST result presented the central months of the four hydrological seasons. The increasing temperature data were supported by the literature studies that alien invasive species entered the Black Sea. The climatic fluctuations obtained as a result of this study will increase the invasion level of non-native species and their invasion will continue into the Black Sea and surrounding seas.

Statistical bias correction on the climate model for el nino index prediction

El Nino can harm many sectors in Indonesia by reducing precipitation levels in some areas. The occurrence of El Nino can be estimated by observing the sea surface temperature in Nino 3.4 region. Therefore, an accurate model on sea surface temperature prediction in Nino 3.4 region is needed to optimize the estimation of the occurrence of El Nino, such as ECMWF. However, the prediction model released by ECMWF still consists of some systematic errors or biases. This research aims to correct these biases using statistical bias correction techniques and evaluate the prediction model before and after correction. The statistical bias correction uses linear scaling, variance scaling, and distribution mapping techniques. The results show that statistical bias correction can reduce the prediction model bias. Also, the distribution mapping and variance scaling are more accurate than the linear scaling technique. Distribution mapping has better RMSE in December-March, and variance scaling has better RMSE in April-June also in October and November. However, in July-September, prediction from ECMWF has better RMSE. The application of statistical bias correction techniques has the highest refinement in January-March at the first lead time and in April at the fifth until the seventh lead time.

Improving the performance of sea surface temperature predictions using a revised method

ABSTRACT Sea surface temperature (SST) prediction plays an important role in global and regional ocean-related research. With the rapid development of remote sensing technology, there are plenty of studies for SST prediction with deep learning to pursue better prediction performance and normalization was used for most studies. However, SST always fluctuates widely, so the denormalization scaled up the difference between normalized observed SSTs and normalized predicted SSTs (OnP-DIFF), and it results in a significant decline in the prediction performance. Therefore, a revision method used to revert the scale-up is proposed in this letter. This method uses cube B-spline interpolation to enhance OnP-DIFF, utilizes Long Short Term Memory network (LSTM) to train and predict OnP-DIFF and generate revision tensor (RT), and finally uses RT to revise denormalized predicted SSTs. To our knowledge, it is the first attempt to use the revision method to solve the problem of performance decline for SST prediction because of wide fluctuations of SST. The experiment results indicate that the proposed method can revert the performance back to and even better than the value before denormalization for deep learning models, significantly improving the prediction performance, and is reliable for SST prediction with high performance for a wide time range and large spatial scope.

Time series and neural network to forecast water quality parameters using satellite data

Complex and highly variable parameters of water quality are difficult to predict over time and space. In most cases, predictions of surface chlorophyll-a (Chl-a) and sea surface temperature (SST) are based on coupled ecological-hydrodynamic modeling, which is driven by oceanic and climatic conditions. Since these models are complex, simpler and efficient methods of predictions are required. The objective of this study is to evaluate the performance of stochastic time series models - Seasonal Auto Regressive Integrated Moving Average (SARIMA) and nonlinear neural network (NN) - to predict Chl-a and SST in coastal areas. Additionally, this study examines the sensitivity of the aforementioned models with respect to variation in depth and changes in turbidity. The data used in this study are satellite-derived Chl-a and SST collected for the Arabian Gulf region for a period of 10 years. This region has been selected for the study due to pronounced variability in depth and changes in turbidity in its waters. As a result of this study, it has been shown that SARIMA and NN are able to forecast Chl-a and SST. In the deep & less turbid water, SARIMA and NN models (SARIMA/NN) produced Chl-a R2:RMSE values of (0.5:0.6)/(0.37:0.53) and SST R2:RMSE (0.9:1.36)/(0.8:1.43). However, the NN has performed better than SARIMA in predicting Chl-a in turbid & shallow waters, especially when Chl-a concentrations are high (>2 mgm−3). Based on our analysis, we conclude that SARIMA and NN are capable of forecasting the future values of Chl-a and SST in regions with pre-existing knowledge of the water depth and turbidity.

Identifying and Using Suitable Oceanic Domains for Prediction of the Indian Monsoon Rainfall over Tamil Nadu

Rainfall prediction are vital for agriculture which is one of the primary sectors greatly affected by climate variability and extremes. Agriculture plays a vital role in shaping the economy of India which is often affected by monsoon. Sea surface temperature (SST) plays a vital role in rainfall predictability over the land surface. A total of twelve different domains of oceanic influences of SST on monsoon rainfall over Tamil Nadu were selected for analysis. The SST of different lead times (February, March, April, and May for southwest monsoon (SWM) and June, July, August, and September for northeast monsoon (NEM) from the ERSSTv5 and ECMWF-SEAS5 model with the Canonical Correlation Analysis (CCA) were used in the Climate Predictability Tool (CPT) to identify the best predictor domains for the prediction of SWM and NEM rainfall over Tamil Nadu. The model training utilized the first 40 years (1981-2020) SST and rainfall data and prediction was done for the 2021 seasons. The results of the study revealed from Kendall tau goodness index and CCA score, the predictor domains comprised of a combination of oceanic domains, this were the Indian, Arabian, Bay of Bengal, and Pacific Oceans recorded the best CCA score and the goodness index. Is therefore recommended that, these domains which have the highest overall predictability can be used by the National meteorological services to early warning and monsoon rainfall information over Tamil Nadu.

Improving prediction quality of sea surface temperature (SST) in Niño3.4 region using Bayesian Model Averaging

Prediction of Sea Surface Temperature (SST) in Niño3.4 region (170 W - 120 W; 5S - 5N) is important as a valuable indicator to identify El Niño Southern Oscillation (ENSO), i.e., El Niño, La Niña, and Neutral condition for coming months. More accurate prediction Niño3.4 SST can be used to determine the response of ENSO phenomenon to rainfall over Indonesia region. SST predictions are routinely released by meteorological institutions such as the European Center for Medium-Range Weather Forecasts (ECMWF). However, SST predictions from the direct output (RAW) of global models such as ECMWF seasonal forecast is suffering from bias that affects the poor quality of SST predictions. As a result, it also increases the potential errors in predicting the ENSO events. This study uses SST from the output Ensemble Prediction System (EPS) of ECMWF seasonal forecast, namely SEAS5. SEAS5 SST is downloaded from The Copernicus Climate Change Service (C3S) for period 1993-2020. One value representing SST over Niño3.4 region is calculated for each lead-time (LT), LT0-LT6. Bayesian Model Averaging (BMA) is selected as one of the post-processing methods to improve the prediction quality of SEAS5-RAW. The advantage of BMA over other post-processing methods is its ability to quantify the uncertainty in EPS, which is expressed as probability density function (PDF) predictive. It was found that the BMA calibration process reaches optimal performance using 160 months training window. The result show, prediction quality of Niño3.4 SST of BMA output is superior to SEAS5-RAW, especially for LT0, LT1, and LT2. In term deterministic prediction, BMA shows a lower Root Mean Square Error (RMSE), higher Proportion of Correct (PC). In term probabilistic prediction, the error rate of BMA, which is showed by the Brier Score is lower than RAW. Moreover, BMA shows a good ability to discriminating ENSO events which indicates by AUC ROC close to a perfect score.

A Theory of the Spring Persistence Barrier on ENSO. Part III: The Role of Tropical Pacific Ocean Heat Content

AbstractIn this paper, we investigate the relationship between upper ocean heat content (OHC) and El Niño–Southern Oscillation (ENSO) sea surface temperature (SST) anomalies mainly using the neutral recharge oscillator (NRO) model both analytically and numerically. Previous studies showed that spring OHC, which leads SST by 6–12 months, represents a major source of predictability for ENSO. It is suggested that this seasonality is caused by the seasonally varying growth rate in SST anomalies. Moreover, a shortened ENSO period will lead to a reduced SST predictability from OHC, with the most significant decrease occurring in the latter half of the calendar year. The cross-correlation relationship between OHC and ENSO SST anomalies is further identified in the damped and self-excited version of the recharge oscillator model. Finally, we suggest that the seasonal growth rate of ENSO anomalies is the cause of the seasonality in the effectiveness of OHC as a predictor in ENSO forecasting, particularly as it relates to the boreal spring persistence barrier and associated spring predictability barrier. We also explain the shorter lead time between spring OHC and ENSO SST anomalies after the turn of the twenty-first century in terms of the apparent higher frequency of the ENSO period.

The Short-Term Climate Prediction System FIO-CPS v2.0 and its Prediction Skill in ENSO

The climate model is an important tool for simulating and predicting the mean state and variability of the climate system. The First Institute of Oceanography-Climate Prediction System (FIO-CPS), built on a climate model with the oceanic observation initialization, has been updated from version 1.0 to 2.0, with a finer resolution and more reasonable physical processes. Previous assessments show that the mean state was well simulated in version 2.0, and its influence on the prediction was further analyzed in this study. Hindcast experiments were conducted using FIO-CPS v1.0 and v2.0, and their prediction abilities based on 27 years (1993–2019) experiment data were analyzed. The results show that the sea surface temperature (SST) biases over the eastern Pacific and the Southern Ocean are improved in the initial condition of FIO-CPS v2.0. Moreover, this new system has a higher skill for predicting El Niño-Southern Oscillation (ENSO). The prediction skill represented by the anomaly correlation coefficient (ACC) of the Niño3.4 index is greater than 0.78 at the 6-month lead time, which increases by 11.09% compared to the value of 0.70 in FIO-CPS v1.0. The root mean square error (RMSE) decreases by 0.20, which accounts for 28.59% of the FIO-CPS v1.0 result. Furthermore, the improvement of the prediction skill changes seasonally, featured by the ACC significantly increasing in the boreal winter and early spring. The improvement in the annual mean SST prediction over the Equatorial Pacific mainly contributes to the enhanced ENSO prediction skill in FIO-CPS v2.0. These results indicate that a state-of-the-art climate model with a well-simulated mean state is critical in improving the prediction skill on the seasonal time scale.

Skillful decadal prediction of unforced southern European summer temperature variations

We assess the capability of decadal prediction simulations from the Coupled Model Intercomparison Project phase 6 (CMIP6) archive to predict European summer temperature during the period 1970–2014. Using a multi-model ensemble average, we show that Southern European (SEU) summer temperatures are highly predictable for up to ten years in CMIP6. Much of this predictive skill, is related to the externally forced response: historical simulations explain about 90% of observed SEU summer temperature variance. Prediction skill for the unforced signal of SEU summer temperature is low: initialized model simulations explain less than 10% of observed variance after removing the externally forced response. An observed link between unforced SEU summer temperature and preceding spring Eastern North Atlantic—Mediterranean sea surface temperature (SST) motivates the application of a dynamical-statistical model to overcome the low summer temperature skill over Europe. This dynamical-statistical model uses dynamical spring SST predictions to predict European summer temperature, and significantly increases decadal prediction skill of unforced European summer temperature variations, showing significant prediction skill for unforced Southern European summer temperature 2–9 years ahead. As a result, dynamical-statistical models can benefit the decadal prediction of variables with initially limited skill beyond the forcing, such as summer temperature over Europe.

Sea surface temperature predictability in the North Pacific from multi-model seasonal forecast

Sea surface temperature (SST) prediction based on the multi-model seasonal forecast with numerous ensemble members have more useful skills to estimate the possibility of climate events than individual models. Hence, we assessed SST predictability in the North Pacific (NP) from multi-model seasonal forecasts. We used 23 years of hindcast data from three seasonal forecasting systems in the Copernicus Climate Change Service to estimate the prediction skill based on temporal correlation. We evaluated the predictability of the SST from the ensemble members' width spread, and co-variability between the ensemble mean and observation. Our analysis revealed that areas with low prediction skills were related to either the large spread of ensemble members or the ensemble members not capturing the observation within their spread. The large spread of ensemble members reflected the high forecast uncertainty, as exemplified in the Kuroshio–Oyashio Extension region in July. The ensemble members not capturing the observation indicates the model bias; thus, there is room for improvements in model prediction. On the other hand, the high prediction skills of the multi-model were related to the small spread of ensemble members that captures the observation, as in the central NP in January. Such high predictability is linked to El Niño Southern Oscillation (ENSO) via teleconnection.