Deep learning approach to reconstruct satellite ocean color time series in the global ocean

2021 ◽  
Author(s):  
Joana Roussillon ◽  
Ronan Fablet ◽  
Lucas Drumetz ◽  
Thomas Gorgues ◽  
Elodie Martinez
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Phytoplankton Biomass ◽  
Decadal Variability ◽  
Temporal Structure ◽  
Ocean Color ◽  
Global Scale ◽  
Global Ocean ◽  
Chl A ◽  
Reconstruction Performance

<p>Phytoplankton plays a key role in the carbon cycle and constitutes the basis of the marine food web. Its seasonal and interannual cycles are relatively well-known on a global scale thanks to continuous ocean color satellite observations acquired since 1997. The satellite-derived chlorophyll-a concentrations (Chl-a, a proxy of phytoplankton biomass) time series are still too short to investigate phytoplankton biomass low-frequency variability. However, it is a vital prerequisite before being able to confidently detect anthropogenic signals, as natural decadal variability can accentuate, weaken or even mask out any anthropogenic trends. Machine learning appears as a promising tool to reconstruct Chl-a past signals (including periods before satellite Chl-a era), and deep learning models seem particularly relevant to explore the spatial and/or temporal structure of the data.</p><p>Here, different neural network architectures have been tested on a 18-year satellite and re-analysis dataset to infer Chl-a from physical predictors. Their ability to reconstruct spatial and temporal (seasonal and interannual) variations on a global scale will be presented. Convolutional neural networks (CNN) better capture Chl-a spatial fields than models that do not account for the structure of the data, such as multi-layer perceptrons (MLPs). We also assess how the selection of training period may affect the reconstruction performance. This is a necessary step before being able to reconstruct any past Chl-a multi-decadal time series with confidence, which is the ultimate goal of this work.</p><p>Our study also addresses the carbon footprint associated with the use of GPU resources when training the CNN. GPUs are energy intensive, and their use in geosciences is expected to grow fast. Systematically reporting the computational energy costs in the geoscience community studies would provide an overview of models energy-efficiency on different kinds of datasets and may encourage actions to reduce consumption when possible.</p>

scholarly journals Neural Network Approaches to Reconstruct Phytoplankton Time-Series in the Global Ocean

2020 ◽  
Vol 12 (24) ◽  
pp. 4156
Author(s):  
Elodie Martinez ◽  
Anouar Brini ◽  
Thomas Gorgues ◽  
Lucas Drumetz ◽  
Joana Roussillon ◽  
...  
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Learning Algorithms ◽  
Global Scale ◽  
Machine Learning Algorithms ◽  
Global Ocean ◽  
Time Variability ◽  
Support Vector ◽  
Chl A ◽  
Phytoplankton Variability

Phytoplankton plays a key role in the carbon cycle and supports the oceanic food web. While its seasonal and interannual cycles are rather well characterized owing to the modern satellite ocean color era, its longer time variability remains largely unknown due to the short time-period covered by observations on a global scale. With the aim of reconstructing this longer-term phytoplankton variability, a support vector regression (SVR) approach was recently considered to derive surface Chlorophyll-a concentration (Chl, a proxy of phytoplankton biomass) from physical oceanic model outputs and atmospheric reanalysis. However, those early efforts relied on one particular algorithm, putting aside the question of whether different algorithms may have specific behaviors. Here, we show that this approach can also be applied on satellite observations and can even be further improved by testing performances of different machine learning algorithms, the SVR and a neural network with dense layers (a multi-layer perceptron, MLP). The MLP outperforms the SVR to capture satellite Chl (correlation of 0.6 vs. 0.17 on a global scale, respectively) along with its seasonal and interannual variability, despite an underestimated amplitude. Among deep learning algorithms, neural network such as MLP models appear to be promising tools to investigate phytoplankton long-term time-series.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 1.91% to 6.69%. <div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

scholarly journals Seasonal Cycles of Phytoplankton Expressed by Sine Equations Using the Daily Climatology from Satellite-Retrieved Chlorophyll-a Concentration (1997–2019) Over Global Ocean

2020 ◽  
Vol 12 (16) ◽  
pp. 2662 ◽  
Author(s):  
Zexi Mao ◽  
Zhihua Mao ◽  
Cédric Jamet ◽  
Marc Linderman ◽  
Yuntao Wang ◽  
...  
Keyword(s):  
Time Series ◽  
Chlorophyll A ◽  
Seasonal Variations ◽  
Global Ocean ◽  
Boreal Winter ◽  
Seasonal Cycles ◽  
Chlorophyll A Concentration ◽  
Chl A ◽  
Global Oceans ◽  
Sine Equation

The global coverage of Chlorophyll-a concentration (Chl-a) has been continuously available from ocean color satellite sensors since September 1997 and the Chl-a data (1997–2019) were used to produce a climatological dataset by averaging Chl-a values at same locations and same day of year. The constructed climatology can remarkably reduce the variability of satellite data and clearly exhibit the seasonal cycles, demonstrating that the growth and decay of phytoplankton recurs with similarly seasonal cycles year after year. As the shapes of time series of the climatology exhibit strong periodical change, we wonder whether the seasonality of Chl-a can be expressed by a mathematic equation. Our results show that sinusoid functions are suitable to describe cyclical variations of data in time series and patterns of the daily climatology can be matched by sine equations with parameters of mean, amplitude, phase, and frequency. Three types of sine equations were used to match the climatological Chl-a with Mean Relative Differences (MRD) of 7.1%, 4.5%, and 3.3%, respectively. The sine equation with four sinusoids can modulate the shapes of the fitted values to match various patterns of climatology with small MRD values (less than 5%) in about 90% of global oceans. The fitted values can reflect an overall pattern of seasonal cycles of Chl-a which can be taken as a time series of biomass baseline for describing the state of seasonal variations of phytoplankton. The amplitude images, the spatial patterns of seasonal variations of phytoplankton, can be used to identify the transition zone chlorophyll fronts. The timing of phytoplankton blooms is identified by the biggest peak of the fitted values and used to classify oceans as different bloom seasons, indicating that blooms occur in all four seasons with regional features. In global oceans within latitude domains (48°N–48°S), blooms occupy approximately half of the ocean (50.6%) during boreal winter (December–February) in the northern hemisphere and more than half (58.0%) during austral winter (June–August) in the southern hemisphere. Therefore, the sine equation can be used to match the daily Chl-a climatology and the fitted values can reflect the seasonal cycles of phytoplankton, which can be used to investigate the underlying phenological characteristics.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

scholarly journals Enhanced chlorophyll-<i>a</i> concentration in the wake of Sable Island, eastern Canada, revealed by two decades of satellite observations: a response to grey seal population dynamics?

2021 ◽  
Vol 18 (23) ◽  
pp. 6115-6132
Author(s):  
Emmanuel Devred ◽  
Andrea Hilborn ◽  
Cornelia Elizabeth den Heyer
Keyword(s):  
Time Series ◽  
Phytoplankton Biomass ◽  
Standing Stock ◽  
Satellite Observations ◽  
Trophic Levels ◽  
Leeward Side ◽  
Grey Seal ◽  
Scotian Shelf ◽  
Eastern Canada ◽  
Chl A

Abstract. Elevated surface chlorophyll-a (chl-a) concentration ([chl-a]), an index of phytoplankton biomass, has been previously observed and documented by remote sensing in the waters to the southwest of Sable Island (SI) on the Scotian Shelf in eastern Canada. Here, we present an analysis of this phenomenon using a 21-year time series of satellite-derived [chl-a], paired with information on the particle backscattering coefficient at 443 nm (bbp(443), a proxy for particle suspension) and the detritus/gelbstoff absorption coefficient at 443 nm (adg(443), a proxy to differentiate water masses and presence of dissolved organic matter) in an attempt to explain some possible mechanisms that lead to the increase in surface biomass in the surroundings of SI. We compared the seasonal cycle, 8 d climatology and seasonal trends of surface waters near SI to two control regions located both upstream and downstream of the island, away from terrigenous inputs. Application of the self-organising map (SOM) approach to the time series of satellite-derived [chl-a] over the Scotian Shelf revealed the annual spatio-temporal patterns around SI and, in particular, persistently high phytoplankton biomass during winter and spring in the leeward side of SI, a phenomenon that was not observed in the control boxes. In the vicinity of SI, a significant increase in [chl-a] and adg(443) during the winter months occurred at a rate twice that of the ones observed in the control boxes, while no significant trends were found for the other seasons. In addition to the increase in [chl-a] and adg(443) within the plume southwest of SI, the surface area of the plume itself expanded by a factor of 5 over the last 21 years. While the island mass effect (IME) explained the enhanced biomass around SI, we hypothesised that the large increase in [chl-a] over the last 21 years was partly due to an injection of nutrients by the island's grey seal colony, which has increased by 200 % during the same period. This contribution of nutrients from seals may sustain high phytoplankton biomass at a time of year when it is usually low following the fall bloom. A conceptual model was developed to estimate the standing stock of chl-a that can be sustained by the release of nitrogen (N) by seals. Comparison between satellite observations and model simulations showed a good temporal agreement between the increased abundance of seal on SI during the breeding season and the phytoplankton biomass increase during the winter. We found that about 20 % of chl-a standing stock increase over the last 21 years could be due to seal N fertilisation, the remaining being explained by climate forcing and oceanographic processes. Although without in situ measurements for ground truthing, the satellite data analysis provided evidence of the impact of marine mammals on lower trophic levels through a fertilisation mechanism that is coupled with the IME with potential implications for conservation and fisheries.

A hybrid CNN-LSTM based model for the prediction of sea surface temperature using time-series satellite data.

2020 ◽  
Author(s):  
Pavan Kumar Jonnakuti ◽  
Udaya Bhaskar Tata Venkata Sai
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Surface Temperature ◽  
Sea Surface Temperature ◽  
Satellite Data ◽  
Time Series Data ◽  
Sea Surface ◽  
Machine Learning Techniques ◽  
Global Ocean ◽  
Series Data

&lt;p&gt;Sea surface temperature (SST) is a key variable of the global ocean, which affects air-sea interaction processes. Forecasts based on statistics and machine learning techniques did not succeed in considering the spatial and temporal relationships of the time series data. Therefore, to achieve precision in SST prediction we propose a deep learning-based model, by which we can produce a more realistic and accurate account of SST &amp;#8216;behavior&amp;#8217; as it focuses both on space and time. Our hybrid CNN-LSTM model uses multiple processing layers to learn hierarchical representations by implementing 3D and 2D convolution neural networks as a method to better understand the spatial features and additionally we use LSTM to examine the temporal sequence of relations in SST time-series satellite data. Widespread studies, based on the historical satellite datasets spanning from 1980 - present time, in Indian Ocean region shows that our proposed deep learning-based CNN-LSTM model is extremely capable for short and mid-term daily SST prediction accurately exclusive based on the error estimates (obtained from LSTM) of the forecasted data sets.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Keywords: Deep Learning, Sea Surface Temperature, CNN, LSTM, Prediction.&lt;/strong&gt;&lt;/p&gt;&lt;p&gt;&amp;#160;&lt;/p&gt;

scholarly journals Factors challenging our ability to detect long-term trends in ocean chlorophyll

2013 ◽  
Vol 10 (4) ◽  
pp. 2711-2724 ◽  
Author(s):  
C. Beaulieu ◽  
S. A. Henson ◽  
Jorge L. Sarmiento ◽  
J. P. Dunne ◽  
S. C. Doney ◽  
...  
Keyword(s):  
Climate Change ◽  
Time Series ◽  
Chlorophyll A ◽  
Satellite Data ◽  
Global Climate ◽  
Ocean Color ◽  
Global Scale ◽  
Trend Detection ◽  
Wide Field ◽  
Detection And Attribution

Abstract. Global climate change is expected to affect the ocean's biological productivity. The most comprehensive information available about the global distribution of contemporary ocean primary productivity is derived from satellite data. Large spatial patchiness and interannual to multidecadal variability in chlorophyll a concentration challenges efforts to distinguish a global, secular trend given satellite records which are limited in duration and continuity. The longest ocean color satellite record comes from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS), which failed in December 2010. The Moderate Resolution Imaging Spectroradiometer (MODIS) ocean color sensors are beyond their originally planned operational lifetime. Successful retrieval of a quality signal from the current Visible Infrared Imager Radiometer Suite (VIIRS) instrument, or successful launch of the Ocean and Land Colour Instrument (OLCI) expected in 2014 will hopefully extend the ocean color time series and increase the potential for detecting trends in ocean productivity in the future. Alternatively, a potential discontinuity in the time series of ocean chlorophyll a, introduced by a change of instrument without overlap and opportunity for cross-calibration, would make trend detection even more challenging. In this paper, we demonstrate that there are a few regions with statistically significant trends over the ten years of SeaWiFS data, but at a global scale the trend is not large enough to be distinguished from noise. We quantify the degree to which red noise (autocorrelation) especially challenges trend detection in these observational time series. We further demonstrate how discontinuities in the time series at various points would affect our ability to detect trends in ocean chlorophyll a. We highlight the importance of maintaining continuous, climate-quality satellite data records for climate-change detection and attribution studies.

scholarly journals Temporal variability of 10-year global SeaWiFS time-series of phytoplankton chlorophyll a concentration

2009 ◽  
Vol 66 (7) ◽  
pp. 1547-1556 ◽  
Author(s):  
V. Vantrepotte ◽  
F. Mélin
Keyword(s):  
Time Series ◽  
Chlorophyll A ◽  
Temporal Variability ◽  
Global Scale ◽  
Wide Field ◽  
Seasonal Component ◽  
Ocean Colour ◽  
Chlorophyll A Concentration ◽  
Chl A ◽  
The Stability

Abstract Vantrepotte, V., and Mélin, F. 2009. Temporal variability of 10-year global SeaWiFS time-series of phytoplankton chlorophyll a concentration. – ICES Journal of Marine Science, 66: 1547–1556. The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) global dataset now offers a 10-year time-series of a consistent, well-calibrated, ocean colour record suitable to analyse temporal variability. The relative importance of the seasonal term in the chlorophyll a (Chl a) concentration signal is first assessed using statistical techniques of temporal decomposition. The emphasis is on the Census method II (X-11) approach, which allows year-to-year variations in the seasonal component. The seasonality detected in the SeaWiFS Chl a record is analysed through a generic province-based classification of marine ecosystems and at global scale and is found very variable spatially. Working with 5′-resolution gridded Chl a products, the contribution of the seasonal component derived from X-11 amounts to 64% of the total variance, compared with only 36% if a fixed annual cycle is assumed. The capacity of X-11 to capture interannual variations in seasonality is used to diagnose the stability of the Chl a seasonal cycle. Finally, linear changes in Chl a concentration observed after a decade of continuous ocean colour record agree globally with previous observations on shorter series. Significant changes of both signs are detected in various regions of the world’s oceans, but primarily a general decrease of Chl a in the mid-ocean gyres.

scholarly journals Self-Supervised Pre-Training of Transformers for Satellite Image Time Series Classification

2020 ◽  
Author(s):  
Yuan Yuan ◽  
Lei Lin
Keyword(s):  
Time Series ◽  
Deep Learning ◽  
Large Scale ◽  
Temporal Structure ◽  
Satellite Image ◽  
Fine Tuning ◽  
Small Scale ◽  
Model Parameters ◽  
Learning Approaches ◽  
Wide Range

<div>Satellite image time series (SITS) classification is a major research topic in remote sensing and is relevant for a wide range of applications. Deep learning approaches have been commonly employed for SITS classification and have provided state-of-the-art performance. However, deep learning methods suffer from overfitting when labeled data is scarce. To address this problem, we propose a novel self-supervised pre-training scheme to initialize a Transformer-based network by utilizing large-scale unlabeled data. In detail, the model is asked to predict randomly contaminated observations given an entire time series of a pixel. The main idea of our proposal is to leverage the inherent temporal structure of satellite time series to learn general-purpose spectral-temporal representations related to land cover semantics. Once pre-training is completed, the pre-trained network can be further adapted to various SITS classification tasks by fine-tuning all the model parameters on small-scale task-related labeled data. In this way, the general knowledge and representations about SITS can be transferred to a label-scarce task, thereby improving the generalization performance of the model as well as reducing the risk of overfitting. Comprehensive experiments have been carried out on three benchmark datasets over large study areas. Experimental results demonstrate the effectiveness of the proposed method, leading to a classification accuracy increment up to 2.38% to 5.27%. The code and the pre-trained model will be available at https://github.com/linlei1214/SITS-BERT upon publication.</div><div><b>This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.</b></div>

scholarly journals Phytoplankton Phenology in the Coastal Zone of Cyprus, Based on Remote Sensing and In Situ Observations

2021 ◽  
Vol 14 (1) ◽  
pp. 12
Author(s):  
Monica Demetriou ◽  
Dionysios E. Raitsos ◽  
Antonia Kournopoulou ◽  
Manolis Mandalakis ◽  
Spyros Sfenthourakis ◽  
...  
Keyword(s):  
Remote Sensing ◽  
Time Series ◽  
Phytoplankton Biomass ◽  
Trophic Levels ◽  
Data Availability ◽  
Ocean Colour ◽  
Chl A ◽  
In Situ Data

Alterations in phytoplankton biomass, community structure and timing of their growth (phenology), are directly implicated in the carbon cycle and energy transfer to higher trophic levels of the marine food web. Due to the lack of long-term in situ datasets, there is very little information on phytoplankton seasonal succession in Cyprus (eastern Mediterranean Sea). On the other hand, satellite-derived measurements of ocean colour can only provide long-term time series of chlorophyll (an index of phytoplankton biomass) up to the first optical depth (surface waters). The coupling of both means of observations is essential for understanding phytoplankton dynamics and their response to environmental change. Here, we use 23 years of remotely sensed, regionally tuned ocean-colour observations, along with a unique time series of in situ phytoplankton pigment composition data, collected in coastal waters of Cyprus during 2016. The satellite observations show an initiation of phytoplankton growth period in November, a peak in February and termination in April, with an overall mean duration of ~4 months. An in-depth exploration of in situ total Chl-a concentration and phytoplankton pigments revealed that pico- and nano-plankton cells dominated the phytoplankton community. The growth peak in February was dominated by nanophytoplankton and potentially larger diatoms (pigments of 19’ hexanoyloxyfucoxanthin and fucoxanthin, respectively), in the 0–20 m layer. The highest total Chl-a concentration was recorded at a station off Akrotiri peninsula in the south, where strong coastal upwelling has been reported. Another station in the southern part, located next to a fish farm, showed a higher contribution of picophytoplankton during the most oligotrophic period (summer). Our results highlight the importance of using available in situ data coupled to ocean-colour remote sensing, for monitoring marine ecosystems in areas with limited in situ data availability.

Sign in / Sign up

Export Citation Format

Share Document