Available via license: CC BY 4.0
Content may be subject to copyright.
1
AIGD-PFT: The first AI-driven Global Daily gap-free 4 km
Phytoplankton Functional Type products from 1998 to 2023
Yuan Zhang1, Fang Shen1*, Renhu Li1, Mengyu Li1, Zhaoxin Li1, Songyu Chen1, Xuerong Sun2
1 State Key Laboratory of Estuarine and Coastal Research, East China Normal University, Shanghai, China.
2 Centre for Geography and Environmental Science, Department of Earth and Environmental Science, Faculty of Environment, 5
Science and Economy, University of Exeter, Cornwall, United Kingdom.
Correspondence to: Fang Shen (fshen@sklec.ecnu.edu.cn)
Abstract. Long time series of spatiotemporally continuous phytoplankton functional type (PFT) products are essential for
understanding marine ecosystems, global biogeochemical cycles, and effective marine management. In this study, by
integrating artificial intelligence (AI) technology with multi-source marine big data, we have developed a Spatial–Temporal–10
Ecological Ensemble model based on Deep Learning (STEE-DL), and then generated the first AI-driven Global Daily gap-
free 4 km PFTs product from 1998 to 2023 (AIGD-PFT), significantly enhancing the accuracy and spatiotemporal coverage
of quantifying eight major PFTs (i.e., Diatoms, Dinoflagellates, Haptophytes, Pelagophytes, Cryptophytes, Green Algae,
Prokaryotes, and Prochlorococcus). The input data encompass physical oceanographic, biogeochemical, spatiotemporal
information, and ocean color data (OC-CCI v6.0) that have been gap-filled using a Discrete Cosine Transform with a Penalized 15
Least Square (DCT-PLS) approach. The STEE-DL model utilizes an ensemble strategy with 100 ResNet models, applying
Monte Carlo and bootstrapping methods to estimate optimal PFT values and assess model uncertainty through ensemble means
and standard deviations. The model's performance was validated using multiple cross-validation strategies—random, spatial-
block, and temporal-block—combined with in-situ data, demonstrating STEE-DL's robustness and generalization capability.
The daily updates and seamless nature of the AIGD-PFT product capture the complex dynamics of coastal regions effectively. 20
Finally, through a comparative analysis using a triple-collocation (TC) approach, the competitive advantages of the AIGD-
PFT product over existing products were validated. The AIGD-PFT product not only provides the foundation for detailed
analyses of PFT trends, interannual variability, and the impacts of climate change on phytoplankton composition across various
temporal and spatial scales, but also has the potential to facilitate precise quantification of marine carbon flux and enhances
the accuracy of biogeochemical models. A video demonstration is available at https://doi.org/10.5446/67366 (Zhang and Shen, 25
2024a). The complete product dataset (1998-2023) can be freely downloaded at
https://doi.org/10.11888/RemoteSen.tpdc.301164 (Zhang and Shen, 2024b).
1 Introduction
Marine phytoplankton contribute to approximately half of the earth's primary productivity (Field et al., 1998), driving the
operation of marine ecosystems (Beaugrand et al., 2010). These minute organisms are classified into different phytoplankton 30
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
2
functional types (PFTs), playing a crucial role in global biogeochemical cycles, biodiversity, and climate feedbacks (Le Quéré
et al., 2005; Gruber et al., 2019). Comprehensive monitoring and research on the spatiotemporal distribution patterns of PFTs
are foundational for understanding marine ecosystems, predicting the impacts of climate change (Kramer et al., 2024;
Falkowski, 2012). Particularly, for accurately quantification of global ocean carbon fluxes and the improvement of
biogeochemical models (Guidi et al., 2016), long-term, high-resolution PFT data is a scientific priority (Nair et al., 2008). 35
Furthermore, as human reliance on marine resources increases, ensuring the sustainability of fisheries (Chassot et al., 2010),
effective management of coastal areas, and safeguarding against the risks posed by harmful algal blooms (Xi et al., 2023) all
underscore the value of the diversity data of phytoplankton represented by PFTs (Henson et al., 2021).
For the quantification of global PFTs, many analytical techniques and inversion algorithms have been developed in recent
years. Among the field sampling analysis methods for quantifying global phytoplankton community composition from water 40
samples, including optical microscopy (Karlson et al., 2010), flow cytometry (Veldhuis and Kraay, 2000), and recent genomics
(Catlett et al., 2020), the separation of phytoplankton diagnostic pigments through High-Performance Liquid Chromatography
(HPLC) with the assistance of Diagnostic pigment analysis (DPA) or CHEMTAX (Mackey et al., 1996) algorithms remains
the most cost-effective and quality-controlled method to date (Swan et al., 2016). The advent of ocean color satellites has
enabled continuous global observation. In situ HPLC pigment data and ocean color satellite data have laid the foundation for 45
the development of remote sensing inversion methods, primarily including abundance-based and spectral-based approaches
(Mouw et al., 2017; Bracher et al., 2017). Abundance-based indirect methods use chlorophyll-a (Chl-a) concentration as model
input, modelling the statistical relationship between Chl-a concentration and diagnostic pigments to retrieve PFTs globally
(Hirata et al., 2011; Uitz et al., 2006). Spectral-based methods directly construct relationships between remote sensing
reflectance, or absorption spectra, scattering spectra, and the concentrations of different groups, incorporating spectral 50
transformation strategies (such as Principal Component Analysis (Xi et al., 2020), differential spectra (Bracher et al., 2009),
etc.) to improve inversion accuracy (Sun et al., 2022). Considering that marine ecological environmental variables (temperature,
nutrients, etc.) shape the distribution of different groups through their impact on phytoplankton growth, physiology, and
competition, introducing more marine environmental covariates into ecological approaches (Zhang et al., 2023; Raitsos et al.,
2008) has become a current research focus: further introducing other biogeochemical and physical oceanographic data on the 55
basis of ocean color satellite data and integrating advanced machine learning methods like random forests and ensemble
learning can significantly enhance the accuracy of PFTs modelling.
Based on the aforementioned approaches, several global PFT products have been developed (Table 1), such as (1) a global
seasonal surface marine climatology dataset based on CHEMTAX and a global HPLC dataset (Swan et al., 2016); (2) the OC-
PFT product based on abundance (Hirata et al., 2011); (3) the PhytoDOAS product based on phytoplankton differential optical 60
absorption spectroscopy (Bracher et al., 2009) ; (4) the synergistic product SynSenPFT that integrates satellite multispectral
information with retrievals based on high-resolution PFT absorption properties derived from hyperspectral satellite
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
3
measurements (Losa et al., 2017); (5) the EOF-PFT product based on remote sensing reflectance and the empirical orthogonal
functions (EOF) algorithm(Xi et al., 2020), along with its modification, the EOF-SST hybrid algorithm (Xi et al., 2021) which
incorporates sea surface temperature (SST). In addition to these remote sensing products, the NASA Ocean Biogeochemical 65
Model (NOBM, https://gmao.gsfc.nasa.gov/reanalysis/MERRA-NOBM/data/data_description.php) has been developed,
which coupled circulation and radiative models (Gregg and Casey, 2007).
Table 1 Summary of Existing Open-Source PFT Products
Product
Method
Spatial resolution
Time resolution
Reference
CHEMTAX-
PFT
Application of CHEMTAX to a global
climatology of pigment data
1°×1°
global grid points
Seasonal
climatology
Swan et al. (2016)
OC-PFT
Synoptic relationships between Chl-a and
its fractional contribution from PFTs
~4 km
Daily
Hirata et al.
(2011)
PhytoDOAS
Differential Optical
Absorption Spectroscopy (DOAS)
0.5
°
Monthly
Bracher et al.
(2009)
SynSenPFT
Combine synergistically OC-PFT and
PhytoDOAS
~4 km
Daily
Losa et al. (2017)
EOF
Empirical orthogonal functions (EOF),
using CMEMS GlobColour merged
products
~4 km
Monthly
Xi et al. (2020)
EOF-SST
EOF-SST hybrid algorithm
~4 km
Monthly
Xi et al. (2021)
NOBM
NASA Ocean Biogeochemical Model
1.25° longitude
，
2/3° latitude
Daily, Monthly
Gregg and Casey
(2007)
Despite advancements in current algorithms for retrieval PFTs, significant challenges persist in terms of prediction accuracy,
spatial coverage, and spatiotemporal resolution. First, abundance-based methods, which rely on Chl-a remote sensing products 70
and empirical formulas to deduce the composition of various PFTs, are computationally straightforward but suffer from limited
accuracy and robustness globally (Bracher et al., 2017). Spectral-based methods encounter challenges because of the spectral
resolution limitations of current ocean color satellites, which restrict their ability to detect weak phytoplankton signals in
optically complex waters. In such environments, non-algal particulate absorption and significant near-infrared water
reflectance can obscure diagnostic pigment absorption, potentially rendering spectral-based methods ineffective (Nair et al., 75
2008). Another significant limitation is the presence of data gaps due to unfavorable conditions, such as orbital configurations,
cloud cover, sunlight contamination, and large sensor viewing angles (Mikelsons and Wang, 2019). For instance, the
probability of cloud-free conditions over the global ocean for MODIS is only between 25% and 30% (Liu and Wang, 2018).
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
4
Although merging images from different satellite missions (e.g., MODIS, VIIRS, OLCI) into the merged product (such as OC-
CCI products (Sathyendranath et al., 2019) and CMEMS GlobColour merged products (Garnesson et al., 2019)) has effectively 80
reduced data gaps, the issue of data loss remains severe. This not only results in numerous voids in PFT products but may also
introduce biases in trend analysis, obscuring key signals of environmental change and hindering a comprehensive
understanding of marine ecosystem dynamics. Such limitations restrict potential applications in climate change research and
marine health monitoring. Monthly averaging of data can mitigate the issue of missing data to some extent. However, this
approach may conceal significant short-term ecological changes, such as ocean heat waves (Chauhan et al., 2023) and algal 85
blooms (Sadeghi et al., 2012). Additionally, the absence of data also limits the full utilization of on-site data: due to the
incompleteness of remote sensing data, many in-situ data cannot be effectively paired with it. This results in the potential
inability of models to fully utilize on-site sampling data for calibration or optimization, thereby wasting expensive sampling
resources and possibly diminishing the model's generalization capability (Xi et al., 2020). While biogeochemical models offer
a global, spatiotemporally continuous PFT modelling approach, their spatial resolution often lacks the detail necessary to 90
accurately reflect local changes and the dynamic characteristics of marine ecosystems.
In summary, although there have been positive developments, current PFT models and products have an imbalance in accuracy,
spatio-temporal resolution, spatial coverage and temporal span when compared to existing requirements, suggesting that there
is still room for improvement in terms of practicality. The advent of the ocean big data era, coupled with the rise of artificial
intelligence technologies such as machine learning, offers new prospects for overcoming the inherent challenges faced by PFT 95
inversion models that currently rely solely on ocean color satellite data (Zhang et al., 2023). Algorithms for data reconstruction
and the integration of multi-source data can effectively bridge the observational gaps caused by clouds or orbital, enhancing
data utilization efficiency and the continuity of global phytoplankton community monitoring. Furthermore, the application of
machine learning and deep learning technologies has the potential to improve the extraction of useful information from vast
oceanic datasets. These technologies, capable of processing and analysing large-scale datasets to identify complex patterns 100
and trends, hold the promise of developing high-precision PFT products.
Here, we propose a novel Spatial–Temporal–Ecological Ensemble model based on deep learning (STEE-DL), designed to
produce a long time series PFT product. STEE-DL leverages an ensemble of 100 ResNet models, incorporating inputs from
reconstructed missing ocean color data, physical reanalysis, biogeochemical, and spatiotemporal information. Utilizing the
STEE-DL model, we have produced the first AI-driven Global Daily gap-free 4 km resolution Phytoplankton Functional Type 105
products (AIGD-PFT), include eight major PFTs (i.e., Diatoms, Dinoflagellates, Haptophytes, Pelagophytes, Cryptophytes,
Green Algae, Prokaryotes, and Prochlorococcus) from 1998 to 2023. The STEE-DL model's accuracy has been tested through
three types of cross-validation (CV) methods: standard, spatial-block, and temporal-block CV. Moreover, we have performed
a comprehensive comparison and validation of the AIGD-PFT against other products using triple collocation analysis.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
5
2 Methodology 110
2.1 Overall framework
The structure and function of phytoplankton communities are influenced by numerous environmental factors, such as sunlight,
nutrient concentration/supply, temperature, carbon chemistry characteristics, and their fluid dynamic environment. We regard
the inversion process of PFTs as a nonlinear mapping () problem, aiming to overcome the limitations of relying solely on
bio-optical algorithms for predicting the spatial distribution of phytoplankton. This process integrates environmental predictive 115
factors , including bio-optical properties, biogeochemical parameters, physical conditions, and spatio-temporal factors, as
shown in equation (1):
=,,, (1)
Building on the work of Zhang et al. (2023), this study further modifies and constructs a STEE-DL model based on a ResNet
ensemble to establish . An overview of the proposed approach is shown in Figure 1. It specifically includes: (1) based on the
global in-situ HPLC dataset compiled by Zhang et al. (2023), this study has expanded and updated it to increase the quantity 120
and diversity of the in-situ data; (2) to address the issue of missing OC data, we utilized the Discrete Cosine Transform with a
Penalized Least Square (DCT-PLS) method to reconstruct the data and fill in the missing pixel values; (3) We have integrated
multiple sources of marine environmental data as input variables for the regression model; (4) addressing the complex
supervised regression problem encountered in multi-source data processing, we trained an ensemble of 100 ResNet models,
named the STEE-DL model, to generate daily PFT products for the period from 1998 to 2023. 125
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
6
Figure 1 Schematic flow of the methodological approach in this study.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
7
2.2 Input Datasets and Preprocessing
We first compiled and integrated in situ data obtained by high-performance liquid chromatography (HPLC), and then collected
predictor data including ocean color, physical oceanography, and ocean biogeochemistry for model training and product 130
generation.
2.2.1 HPLC Pigment Data
Building upon the updates presented by Zhang et al. (2023), this study integrates additional, newly available HPLC pigment
data collected between 1998 and 2023 (refer to Figure 2 for details). This data was primarily sourced from open-access data
repositories such as SeaBASS (https://seabass.gsfc.nasa.gov/), PANGAEA(https://www.pangaea.de/), the British 135
Oceanographic Data Centre (BODC, https://www.bodc.ac.uk/), the Australian Ocean Data Network (AODN,
https://portal.aodn.org.au/), and Google Dataset Search (https://datasetsearch.research.google.com/). This initiative has
resulted in the acquisition of further HPLC open-source data, leading to the creation of a new global in-situ HPLC pigment
database spanning the years 1998 to 2023. In cases of duplicate samples, whether across spatial or temporal dimensions, the
average of the replicates was calculated. By utilizing an updated Diagnostic Pigment Analysis (DPA) methodology, along with 140
newly adjusted weighting coefficients, we conducted DPA to ascertain in-situ PFT Chl-a concentrations. The adjusted
coefficients for DPA were referenced from Alvarado et al. (2022) and Xi et al. (2023), with specifics available at
https://doi.pangaea.de/10.1594/PANGAEA.954738. From these global HPLC pigment datasets, we selected 6 long-term
observation sites as independent validation data. The locations of these sites are shown in Figure 2.
145
Figure 2 (a) depicts the spatial distribution of in-situ HPLC pigment datasets, with red hexagons and numbers indicating the locations of six
independent long-term time series stations. (b) presents a ridge plot of the probability density distribution for eight types of PFTs.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
8
2.2.2 Ocean Color Data and Missing Value Filling
Satellite ocean color remote sensing data is currently the most important data source for the retrieval of PFT. We obtained
daily merged ocean color data from the Ocean-Colour Climate Change Initiative (OC-CCI, version 6.0, 150
https://www.oceancolour.org/) for the period 1998-2023. This data combines measurements from SeaWiFS, MERIS, MODIS-
Aqua, and VIIRS sensors and has a spatial resolution of 4 km (Sathyendranath et al., 2019).The raw daily OC-CCI dataset
exhibits considerable instances of missing data: Figure 3 illustrates the percentage of valid pixels in the OC-CCI dataset, based
on per-pixel statistics spanning the years 1998 to 2023. The results indicate that the majority of marine areas exhibit less than
50% coverage of valid observations, with pronounced gaps particularly evident in higher latitudes. 155
Figure 3 The percentage of valid pixels in the OC-CCI v6.0 daily dataset.
Given the importance of ocean color data in generating seamless space-time PFT products, it is essential to reprocess missing
pixels to fill gaps, thereby maximizing the availability of in-situ and remote sensing data. Previous studies have developed
various methods for reconstructing missing pixels in remote sensing data, such as DINEOF (Data Interpolation Empirical 160
Orthogonal Function) (Liu and Wang, 2022), Optimal Interpolation (Liston and Elder, 2006), and Kriging (Gunes et al., 2006).
However, these methods are very time-consuming when dealing with large datasets. For long-term and daily product
reconstructions, balancing accuracy and computational efficiency is crucial. Therefore, we adopted the DCT-PLS algorithm,
which was initially proposed for automatic smoothing of multidimensional incomplete data (Garcia, 2010). The primary
advantage of the DCT-PLS is its faster speed, while it requires only a small amount of memory storage, and achieves high 165
reconstruction accuracy, making it suitable for processing large datasets. It has been successfully applied to fill data gaps in
soil moisture (Wang et al., 2012), NDVI (Yang et al., 2022), coastal ocean surface current (Fredj et al., 2016), and Chl-a
(Wang et al., 2022) products. To further improve the computational efficiency of the DCT-PLS algorithm, we modified the
original DCT-PLS code, utilizing the built-in FFT computation in PyTorch for GPU-accelerated DCT operations.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
9
Based on the DCT-PLS algorithm, we designed a gap-fill process (as shown in Figure 4), summarized briefly as follows: (1) 170
Data preparation: The original ocean satellite data (e.g., OC-CCI remote sensing reflectance Rrs, Chl-a concentration, and
diffuse attenuation coefficient Kd490) are stored in a three-dimensional spatiotemporal data cube. To avoid seams, we directly
input the entire global 30-day data cube, with dimensions of 4320×8640×30, representing spatial resolution and a 30-day date-
time span, without using regional segmentation. (2) Normalization: To minimize differences in dimensions and magnitudes of
data across different spatial regions, the dataset is standardized by dividing by the spatial mean. (3) DCT-PLS completion: The 175
DCT-PLS method is used to fill in missing values for the target day. We modified Garcia (2010)’s original code
(https://www.mathworks.com/matlabcentral/fileexchange/27994-inpaint-over-missing-data-in-1-d-2-d-3-d-nd-
arrays?s_tid=prof_contriblnk) to a GPU-accelerated form, significantly improving speed compared to the Matlab-based
original code. The entire 30-day time series data undergo a hundred iteration cycles in the DCT-PLS process to fill in the
missing values for the target date. (4) Rolling filling: To enhance the robustness of the filling effect, we adopt a rolling filling 180
strategy. Specifically, for each target day, a 30-day time window is progressively moved forward day by day until the data
window moves past that day. This process is repeated 30 times for each target day, with the average of these fillings taken as
the final result for the target day. (5) Long time series filling: Following the process described, the entire dataset is traversed
and filled day by day, ultimately resulting in a daily continuous and spatially complete data cube from 1998-2023.
This method effectively utilizes time series information to estimate missing values while avoiding discontinuities that might 185
be introduced by data segmentation. Through iteration and averaging, it further improves the accuracy and stability of the filled
data. Additionally, through GPU acceleration, this method achieves higher efficiency compared to traditional methods (such
as DINEOF). It is important to note that in areas of high latitude with extremely high missing values (exceeding 80%), these
data will be directly removed (as demonstrated in the video example available at https://doi.org/10.5446/67366), because
reconstruction under such conditions is impractical. 190
Figure 4 Gap-fill process with DCT-PLS algorithm.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
10
2.2.3 Ocean Physics, Biogeochemistry data, and spatio-temporal information
Incorporation of physical oceanographic data, including Sea Surface Temperature (SST) and Sea Surface Salinity (SSS),
alongside biogeochemical data (Table 2). was performed. These data are provided by the Copernicus Marine Data Store 195
(https://data.marine.copernicus.eu/products). The SST data are sourced from the ESA SST CCI (Climate Change Initiative)
and C3S (Copernicus Climate Change Service) global Sea Surface Temperature Reprocessed product
(https://doi.org/10.48670/moi-00169). The SSS data are obtained from The Operational Mercator global ocean analysis and
forecast system (https://doi.org/10.48670/moi-00016). Biogeochemical data include nitrate concentration (NC), phosphate
concentration (PC), silicate concentration (SC), and dissolved oxygen (DO). These variables are critical for understanding the 200
nutrient dynamics in marine ecosystems, which are fundamental factors influencing phytoplankton growth and distribution.
The data for these biogeochemical variables are sourced from the global biogeochemical multi-year hindcast products
(https://doi.org/10.48670/moi-00019). All data undergo the following preprocessing steps: (1) resampling, where all data is
resampled to a 4km resolution using the pysample library (https://doi.org/10.5281/zenodo.3372769). This resampling process
may lead to missing pixels, which are then filled using the nearest neighbor method; (2) standardization: For Rrs, L2 norm 205
normalization is performed, meaning each band (i.e., Rrs412, Rrs443, Rrs490, Rrs510, Rrs560, Rrs665) is divided by the square root of
the sum of squares of all bands. For Chl-a and Kd490, as well as NC, PC, SC, DO, SST, and SSS, standardization is carried
out using the “StandardScaler” function from the scikit-learn library (https://scikit-learn.org/).
Spatial and temporal components were quantified using polar coordinates to facilitate the capture of complex environmental
changes. The spatial term is characterized in Euclidean space using three spherical coordinates [,,]to reflect 210
autocorrelation and spatial differences. These coordinates represent a point's position in three-dimensional space, calculated
as follows: (1) describes the component in the east-west direction, calculated by longitude, with the formula =
sin 2lon
; (2) combines longitude and latitude to provide the position in the north-south direction and the vertical
distance from the equator, calculated as =cos 2lon
sin 2lat
; (3) represents the straight-line distance from the
center of the Earth to the point, calculated as =cos 2lon
cos 2lat
. Furthermore, the temporal term (~[,]) is 215
represented by two sine and cosine functions of the day of the year (DOY), enabling the capture of both daily variations and
seasonal patterns of PFT. Here, =cos 2DOY
and =sin 2DOY
, where is the total number of days in the
corresponding year.
220
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
11
Table 2 Predictors and corresponding data products.
Dataset
Abbreviation
Definition
Resolution
Ocean color data
Rrs412-670 Remote sensing reflectance at 412, 443, 490,
510,555 and 670 nm ~4 km,
Daily,
1998.1.1-2023.12.31
K
d
490
diffuse attenuation coefficient at 490 nm
Chl-a Chlorophyll-a concentration
Biogeochemistry data
NC
Nitrate concentration
1/4 °,
Daily,
1998.1.1-2023.12.31
PC
Phosphate concentration
SC
Silicate concentration
DO
Dissolved oxygen
Ocean Physical data
SST sea surface temperature
1/20°,
Daily,
1998.1.1-2023.12.31
SSS sea surface salinity
1/12°,
Daily,
1998.1.1-2023.12.31
Spatio-temporal
information
S1 =sin 2
lon
360
S2 =cos 2
lon
360sin 2
lat
180
S3 =cos 2
lon
360cos 2
lat
180
T1 =cos 2
DOY
T2 =sin 2
DOY
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
12
2.3 Spatial–Temporal–Ecological Ensemble model based on deep learning
2.3.1 Network Architecture
Ensemble learning has emerged as a powerful approach to enhancing prediction performance by combining the outputs of 225
multiple models. STEE-DL Models that use deep ensemble learning combine the advantages of deep learning with those of
ensemble learning to achieve better generalization. STEE-DL model framework introduces an ensemble consisting of N
residual neural networks (ResNet) as its components. The ResNet is known for their shortcut connections, which help in
maintaining a smooth flow of gradients during the learning process. To ensure efficiency, each component model is built with
two residual blocks designed to reduce computational demands while preserving the effectiveness of a deep network. These 230
blocks comprise a fully connected layer, a ReLU activation function, and a shortcut connection for uninterrupted information
transmission. This setup decreases the dimensionality of features from 19 to 16, and then to 10, before a final fully connected
layer maps these features to an output value for predicting the target variable. Research, such as the work by Gen and colleagues,
has shown that ensemble stability improves significantly when the number of component models, N, exceeds 50, but the
marginal gains in reducing standard error diminish after reaching 100 models. Therefore, aiming for a balance between 235
accuracy and computational efficiency, we have chosen an ensemble size of N=100. Based on this architecture, we have
implemented the STEE-DL models using PyTorch (https://pytorch.org/).
2.3.2 Model Ensemble and Uncertainty
Each ResNet within the ensemble focuses on different subsets and features of the training data, The mean (μ) of the outputs
from the 100 independent models is considered the optimal estimation of the target variable 240
=PFT()
100 (2)
The variability among ensemble model outputs, quantified by the standard deviation () of the 100 independent models,
provides a measure of uncertainty in predictions. This uncertainty reflects the variability in predictions due to differences in
training sets, initializations, and learning dynamics. A higher standard deviation indicates greater disagreement among models,
suggesting lower confidence in the prediction.
=PFT()
100 (3)
this approach differs from statistical methods based on error propagation, which evaluate prediction uncertainty by analyzing 245
input data uncertainties (e.g., measurement errors) and their transmission through the model to the outputs. Such methods
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
13
require a clear understanding of input error distributions and typically assume these errors are independent. Given the STEE-
DL model’s reliance on diverse marine and in situ High-Performance Liquid Chromatography (HPLC) data of varying quality
control, accurately applying error propagation for uncertainty measurement is challenging. Our ensemble-based approach
primarily addresses model uncertainty but also indirectly reveals data uncertainties by demonstrating how predictions respond 250
to variations in representation and data subsets.
2.3.3 Training Procedure
To compile the training dataset, we align in-situ HPLC data with reconstructed OC-CCI and environmental data, both spatially
and temporally. This alignment projects the data onto a 4km grid according to the latitude, longitude, and date of the HPLC
measurements. In cases where several HPLC measurements are located within the same 4km grid cell, we average these 255
measurements to consolidate corresponding predictor variables.
The STEE-DL model utilizes a Monte Carlo and bootstrapping ensemble learning approach to boost model stability and
predictive accuracy. By resampling, it randomly selects two-thirds of the total dataset as the training set for each iteration,
repeating this procedure 100 times. This method is designed to create a varied collection of models by multiple rounds of
sampling, significantly improving the model’s ability to generalize. This reduces the model’s reliance on specific data 260
distributions, thereby increasing both the accuracy and the robustness of its predictions.
Throughout the training phase, the model optimization relies on the Adam optimizer, complemented by L1 regularization to
promote sparsity within the model and prevent overfitting. Gradient clipping is applied to manage potential issues with
exploding gradients, thus ensuring a more stable training process. An Exponential Moving Average (EMA) strategy is
employed to stabilize the model weights by averaging them over time, which helps to minimize variations and secure a 265
consistent performance from the final model.
To circumvent the issue of the model predicting unreasonably high values during training, we have crafted a specialized loss
function. This function incorporates the traditional Mean Squared Error (MSE) while imposing extra penalties on predictions
that surpass set thresholds. Not only does this effectively prevent the model from making unrealistic predictions, but it also
guides the model towards more accurate parameter adjustments, assuring that its predictions stay within feasible limits. 270
2.4 Evaluation strategies
To comprehensively test the accuracy and robustness of the model, the evaluation of the STEE-DL model comprises two parts:
first, the model performance is validated using a five-fold cross-validation method in three different ways; second, the
evaluation is based on a tripartite matching analysis algorithm.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
14
2.4.1 Cross-validation Approach 275
Cross validation (CV) is a commonly used method for analyzing model performance, allowing for a comprehensive assessment
of a model's accuracy, stability, and generalization. This study implements three types of CV methods: random five-fold CV,
time-block five-fold CV, and spatial-block five-fold CV, to deeply evaluate the model’s multifaceted performance. Standard
five-fold cross-validation: This method randomly divides all data into five equal-sized subsets. In each round of validation,
one subset is selected as the test set, while the remaining four subsets serve as the training set, ensuring that each data point is 280
used as test data. This method primarily evaluates the model’s performance and generalization on the entire dataset. Time-
block five-fold cross-validation: Data is divided into five consecutive time periods in chronological order. In each iteration,
data from one time period is chosen as the test set, with the data from the remaining periods serving as the training set (as
shown in Figure 5). This method takes into account the continuity and dependency of time series, helping to evaluate the
model’s ability to capture time trends and seasonal variations. 285
Figure 5 Temporal block CV procedure.
Spatial-block five-fold cross-validation: Similar to time-block cross-validation, but data is divided based on spatial location.
A hexagonal grid was created at 20° horizontal and vertical intervals, and regions without sampling points were removed for
hexagonal regions. In each round, data from one geographical block is left out as the testset, while data from other blocks are 290
used for training(as shown in Figure 6). This method prevents potential data leakage due to spatial autocorrelation and helps
to assess the model's spatial prediction capability and its generalization across different geographical locations.
The coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and symmetric mean
absolute percentage error (sMAPE) were utilized to quantify the performance of the model, according to:
= 1 []
[
]
(4)
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
15
RMSE
=1
()
(5)
MAE
=1
||
(6)
sMAPE
=100
||
() 2
(7)
where and are the log10-scaled observed and estimated of each PFT for sample i, N is the number of observations, is 295
the log10-scaled mean of the observed values.
Figure 6 Spatial block CV procedure.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
16
2.4.2 Triple Collocation Analysis
The Triple Collocation Analysis (TCA) method was also utilized for a global evaluation of the AIGD-PFT product. TCA is a 300
technique that allows for the assessment and quantification of error characteristics in three independent data sources without
relying on reference data pre-assumed to be “true”(Mccoll et al., 2014). This method has been widely adopted in the uncertainty
evaluation of remote sensing products across various fields, including soil moisture (Kim et al., 2023), sea surface salinity
(Hoareau et al., 2018), and sea surface temperature (Saleh and Al-Anzi, 2021).
For error statistics based on TCA, we selected the fractional Mean Squared Error (fMSE) and the squared correlation coefficient. 305
These metrics offer direct insights into data precision and accuracy. fMSE, in particular, is beneficial because it quantifies the
relative error in a product, scaling from 0 to 1, where a lower value indicates higher precision. fMSE calculated as follows:
=
=
+
=1
1+ (8)
With =++ , corresponds to three spatially and temporally collocated datasets [,,].
is the TCA-based error
variance of an individual product. and represents the scaling factor and systematic additive biases between the unknown
true signal and the datasets . is the variance of the individual data, is the variance of the true signal, and SNR is the 310
Signal-to-Noise Ratio. The fMSE value below 0.5 suggests that the true signal is a more significant component of the data than
the estimation noise, indicating a precise product. Similarly, the squared correlation coefficient () is defined as:
=
+
=
1+ (9)
The foundational assumptions of TCA are important for its application (Kim et al., 2023): (1) a linear relationship exists
between each dataset and the true signal, (2) the errors among the datasets are orthogonal, and (3) there's no correlation among
the errors of different datasets. These principles ensure the robustness of the TCA method in providing an unbiased error and 315
quality assessment of products.
Several other PFT products were introduced and organized into triads for TCA analysis. First, SynSenPFT
(https://doi.org/10.1594/PANGAEA.875873) and NOBM-daily products were obtained, forming a daily product triplets. Both
SynSenPFT and NOBM-daily contain three PFTs - diatoms, cyanobacteria (prokaryotes), and coccolithophores (main
contributing PFT to Haptophytes). TCA evaluations were conducted separately for these three PFTs. The TCA calculation 320
process selected overlapping time periods of SynSenPFT, NOBM-daily, and the proposed AIGD-PFT products, from August
1, 2002, to March 31, 2012, totaling 3,515 days. All three products were resampled to a 1° resolution. Similarly, we also
obtained EOF-PFT data (https://doi.org/10.48670/moi-00281) and NOBM-monthly product to form a monthly triplets, again
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
17
conducting TCA assessments for diatoms, prokaryotes, and Haptophytes. Before evaluation, the AIGD-PFT products were
merged monthly and resampled to 1° resolution along with EOF-PFT and NOBM-monthly. The temporal span of monthly 325
TCA triplets products was from January 2003 to December 2017, totaling 180 months. NOBM's daily and monthly data are
all obtained from Giovanni website (https://giovanni.gsfc.nasa.gov/). We additionally employed RECCAP2 ocean regions for
regional TCA statistics, as shown in Figure 7.
Figure 7 Map of RECCAP2-ocean regions (Regional Carbon Cycle Assessment and Processes, Canadell et al. (2011), https://reccap2-330
ocean.github.io/regions/), include Arctic (Ar), Subtropical Atlantic (StA), Equatorial Atlantic (EA) , South Atlantic (SA), Subtropical
Pacific (StP) , Equatorial Pacific (EP) , South Pacific (SP), Indian Ocean (IO), Southern Ocean (SO).
3 Result
3.1 Model verification
3.1.1 Three CV Methods 335
To comprehensively assess the performance of the proposed STEE-DL model, three five-fold cross-validation (CV) methods
were implemented: random, temporal-block, and spatial-block CV. The results are shown in Table 3. The random CV analysis
revealed generally high prediction accuracy across all 8 PFTs, as visualized by the scatter plot in Figure 8. Diatoms exhibited
highest performance, achieving R2 of 0.8. This confirms the STEE-DL model's strong capability in Diatom prediction.
Conversely, Pelagophytes displayed the weakest performance, reflected by a R2 of just 0.5. Further examination through the 340
probability distribution histograms and Cumulative Distribution Function (CDF) curves of predicted versus actual values
revealed a good alignment, indicating the model's overall ability to accurately mimic observed data distributions. However, a
notable limitation observed was the STEE-DL model's tendency towards overestimating lower values and underestimating
higher values. This suggests a bias towards predicting smoother values, potentially resulting in less accurate predictions for
extreme high or low actual values. 345
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
18
Table 3 Model performance metrics (R2, MAE, RMSE, and sMAPE, based on random, temporal-block, and spatial-block five-fold CV
procedure)
PFT Metrics
Cross-validation approach
random CV
temporal-block
spatial-block
Diatoms
R
2
0.86
0.82
0.81
MAE
0.26
0.29
0.30
RMSE
0.33
0.37
0.40
sMAPE
51.21
55.53
54.25
Dinoflagellates
R2
0.71
0.62
0.64
MAE
0.26
0.30
0.30
RMSE
0.33
0.39
0.40
sMAPE
23.91
27.16
28.75
Haptophytes
R2
0.60
0.50
0.51
MAE
0.21
0.23
0.23
RMSE
0.26
0.30
0.31
sMAPE
17.73
20.24
20.49
Pelagophytes
R2
0.50
0.39
0.42
MAE
0.23
0.26
0.25
RMSE
0.29
0.33
0.34
sMAPE
11.45
12.83
12.55
Cryptophytes
R
2
0.68
0.57
0.61
MAE
0.29
0.34
0.33
RMSE
0.36
0.43
0.43
sMAPE
26.31
30.55
29.56
Green algae
R2
0.72
0.65
0.64
MAE
0.22
0.25
0.25
RMSE
0.27
0.31
0.33
sMAPE
33.16
36.57
36.11
Prokaryotes
R2
0.68
0.59
0.59
MAE
0.23
0.26
0.26
RMSE
0.28
0.33
0.34
sMAPE
13.82
15.76
15.78
Prochlorococcus
R2
0.55
0.19
0.32
MAE
0.22
0.29
0.28
RMSE
0.28
0.40
0.41
sMAPE
14.71
18.37
17.06
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
19
Figure 8 Scatter diagrams, probability distribution and CDF (based on random five-fold CV procedure) of the predicted vs. measured Chl-350
a concentrations of 8 PFTs.
By comparing the model performance under three different CV strategies, we delved further into the STEE-DL model's
generalization abilities in terms of time and space. Figure 9 reveals that the STEE-DL model's accuracy decreases under
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
20
temporal and spatial cross-validation compared to standard random cross-validation. Notably, the predictive accuracy for
diatoms was minimally affected by the different validation strategies, with R2 values remaining above 0.8 for all three methods. 355
This demonstrates the model's robust generalization capability in both temporal and spatial aspects. Except for the
Prochlorococcus, the decrease in accuracy was modest for other PFTs in spatial cross-validation (with about a 0.1 decrease in
R2 and a 0.5 increase in MAE), suggesting that the STEE-DL model is relatively robust and can accurately estimate regions
lacking in situ observational data. Compared to spatial validation, there was a slight decrease in accuracy for temporal cross-
validation, but it still maintained a good level. Except for a significant drop in temporal generalization for the Prochlorococcus, 360
the temporal cross-validation accuracy for other PFTs remained favorable.
Figure 9 Comparison of the results obtained using different CV methods, including random CV, spatial block CV, and temporal block CV.
Blue indicates variations in the R2 under the three cross-validation methods, while red represents changes in MAE.
During the training process of the STEE-DL model, two types of training data are utilized: “original match” training data and 365
“reconstructed match” training data. The “original match” training data refers to data successfully matched directly from the
in situ HPLC database and the OC-CCI original data; the “reconstructed match” training data refers to matched data obtained
after completing the missing parts of OC-CCI data using the DCT-PLS technique. By comparing the model's prediction
accuracy on these two types of data, we can assess not only the STEE-DL model's adaptability to changes in data completeness
but also verify the effectiveness and accuracy of the DCT-PLS technique in reconstructing missing ocean color data. If the 370
STEE-DL model's performance on the “reconstructed match” data is similar to its performance on the “original match” data,
it not only indicates that the DCT-PLS method is effective and reasonable for reconstructing ocean color data, but also confirms
that the STEE-DL model can provide reliable PFT predictions under varying data quality and completeness conditions.
We calculated the R² between predicted and actual values for both original and reconstructed pixels using the three cross-
validation methods (Figure 10). Except for a significant difference in performance for Prochlorococcus, the accuracy of 375
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
21
reconstructed pixels was generally consistent with that of the original pixels, demonstrating good performance. This indicates
that the reconstructed pixels did not degrade model performance, thus confirming both the high congruency of our data
reconstruction method with actual conditions and the robustness of the STEE-DL model.
Figure 10 Model performance comparison on original (blue dashed), reconstructed (orange dashed), and all pixels (orange solid) using (a) 380
random CV, (b) temporal CV, and (c) spatial CV.
3.1.2 Long-time Series Observations
The effectiveness of the proposed STEE-DL model was validated using data from six independent long-term observation sites.
The results, as shown in the Figure 11, display the correlation coefficients between predicted and actual values at these six
sites. The STEE-DL model demonstrated varying degrees of predictive capability across different sites and PFTs. Firstly, the 385
model achieved high prediction accuracy for key ecological types such as Diatoms, Dinoflagellates, and Green algae, with
significant advantages at certain sites: for instance, at sites 4 and 5, the prediction correlation coefficients for Diatoms were as
high as 0.90 and 0.88, respectively. Sites 5 exhibited high correlations for Dinoflagellates and Green algae predictions, reaching
0.69 and 0.83, respectively, highlighting the model's ability to accurately capture the dynamics of these major ecological types.
However, it is noteworthy that predictions for certain ecological types showed considerable fluctuations at specific sites. For 390
example, site 3 had a prediction correlation coefficient of 0.90 for Pelagophytes but a relatively lower coefficient of 0.48 for
Dinoflagellates. In terms of ecological types like Prokaryotes and Prochlorococcus, the model's predictions were generally
more balanced, with site 2 showing a high correlation coefficient of 0.80 for Prochlorococcus. Overall, despite some
fluctuations and differences, these results emphasize the STEE-DL model's capability to capture the temporal trends of
different PFTs with relative accuracy. 395
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
22
Figure 11 STEE-DL model performance at six independent time series stations. Correlation coefficient (bar chart) and number of
successfully matched pixels (blue dashed line).
3.2 Gap-free PFT product and Uncertainties
Following the validation of the STEE-DL model, it was retrained with the entirety of the data available, enabling the generation 400
of a long time series and spatiotemporally continuous AIGD-PFT product for the period from 1998 to 2023. An example from
this dataset, depicted in Figure 12 for March 10, 2020, demonstrates the results of the AIGD-PFT. Notably, while nearly half
of the original OC-CCI data contained missing values (as shown in Figure 12a), our reconstructed dataset has achieved spatial
completeness with good continuity. Within this dataset, the distribution patterns of the eight PFTs showed significant
variability. For example, diatoms were primarily found in the oceanic regions of mid to high latitudes (30°–60°), thriving in 405
nutrient-rich, cold waters, and areas affected by terrestrial runoff. Dinoflagellates, with a distribution pattern similar to diatoms,
were mostly present in the nutrient-rich upwelling zones of high latitudes and nearshore areas, though their content was
relatively lower. Prokaryotes were noted for maintaining higher concentrations in the nutrient-poor, sunlight-abundant waters
of tropical and subtropical regions (0°–30°), with a significant decrease in biomass at higher latitudes, a characteristic closely
resembling that of Prochlorococcus. Haptophytes and green algae were observed more frequently in the subtropical regions of 410
the Pacific, Atlantic, and the Southern Ocean, reaching into mid to high latitudes. In contrast, Pelagophytes and Cryptophytes
were found to be more prevalent in tropical and subtropical regions, showing lower concentrations in areas of lower latitude.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
23
Figure 12 The global distribution (2020-03-10) of the Chl-a concentration for (a) original OC-CCI, (b) Diatoms, (c) Dinoflagellates, (d)
Haptophytes, (e) Green Algae, (f) Prochlorococcus, (g) Prokaryotes, (h) Pelagophytes and (i) Cryptophytes. The grey areas represent lands. 415
Figure 13 delineated the corresponding uncertainties. Overall, the uncertainty relatively low in the open ocean, suggesting that
the model performs with a high degree of confidence. However, in coastal regions such as the East China Sea and the Amazon
River estuary, uncertainties escalate. This increase likely results from the complex coastal processes and land-sea interactions
prevalent in these areas, which can significantly influence the distribution and concentrations of PFTs, thereby challenging the
model's predictive accuracy. Despite the coastal uncertainties, Figure 13 also reveals that AIGD-PFT maintains globally low 420
uncertainty levels (below 0.1) for Diatoms, Dinoflagellates, Haptophytes, and Prokaryotes, highlighting the model's overall
stability and reliability. Additionally, Prochlorococcus exhibits higher uncertainties in the Southern Ocean, while Cryptophytes
show increased uncertainty in the equatorial Pacific. The reasons for this specific pattern require further investigation.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
24
425
Figure 13 The global distribution (2020-03-10) of the uncertainties for (a) Diatoms, (b) Dinoflagellates, (c) Haptophytes, (d) Green Algae,
(e) Prochlorococcus, (f) Prokaryotes, (g) Pelagophytes and (h) Cryptophytes.
Further, Figure 14 illustrated the AIGD-PFT's ability to capture dynamic coastal processes, such as estuary runoff and coastal
circulations, through time-series images of Diatom distribution in the Amazon River estuary (Figure 11a) and the Gulf of
Mexico (Figure 11b). The high Diatom concentrations near the Amazon River estuary, as shown in Figure 6a, correlated with 430
the area's rich nutrient influx, also capturing the influence of the North Brazil Current (NBC) along the Brazilian coastline on
Diatom dispersion. Figure 6b demonstrated the AIGD-PFT’s efficacy in depicting the characteristics dominated by circulation
and associated eddies in the Gulf of Mexico.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
25
Figure 14 Gap-free Diatoms in (a) Brazil Coast in January, 2014 and (b) Gulf of Mexico in July, 2020. 435
3.3 TCA-based Assessment
As depicted in Figure 15, we conducted a TCA on three daily-scaled PFT products: AIGD-PFT, SynSenPFT, and NOBM-
daily. Figure 15a presents the statistical analysis results of correlation coefficients (R) and mean square error (fMSE) on a
global scale. Meanwhile, Figure 15b, Figure 15c, and Figure 15d detail the comparative assessment results across different
marine regions. Globally, the AIGD-PFT product outperforms the other two, demonstrating the highest median correlation 440
values with actual conditions for Diatoms (0.81), Haptophytes (0.80), and Prokaryotes (0.72), respectively. AIGD-PFT product
also have the lowest fMSE values for all three PFTs, confirming its superiority with values of 0.35, 0.35, and 0.48, respectively.
Comparatively, the SynSenPFT product underperforms relative to NOBM-daily in estimating Diatoms and Prokaryotes, yet
excels in estimating Haptophytes.
The regional analysis (Figure 15b, 15c, and 15d) reveals variation in R and fMSE values across regions and PFTs. AIGD-PFT 445
consistently outperforms in most regions for Diatom estimation but shows a slight increase in fMSE in the equatorial Pacific,
indicating a potential dip in estimation accuracy in this area. In contrast, SynSenPFT registers higher fMSE values for
Haptophytes estimation, particularly in the subtropical and southern Pacific regions. NOBM-PFT, on the other hand, tends to
have lower correlation in Haptophytes estimation across regions, with a notable deficiency near the equatorial Pacific.
Additionally, SynSenPFT demonstrates higher global fMSE values for Prokaryotes compared to the other datasets, and NOBM-450
PFT significantly underperforms in Prokaryotes estimation in the Southern Ocean.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
26
Figure 15 TCA result of three daily product (AIGD-PFT, SynSenPFT, and NOBM-daily).
Further extending our analysis to monthly products (AIGD-PFT, EOF-PFT, NOBM-monthly), detailed in Figure 16. We
observed that AIGD-PFT and EOF-PFT exhibit closely matched performances for Diatoms, with median R values of 0.87 and 455
0.86, and fMSE of 0.24 and 0.25, respectively. Their Cumulative Distribution Function (CDF) curves nearly align perfectly.
Although global assessments for Diatoms are consistent, regional discrepancies exist. For instance, AIGD-PFT and EOF-PFT
perform similarly in the subtropical Pacific and the Indian Ocean, but AIGD-PFT achieves superior correlation in the equatorial
Pacific, Southern Ocean, and subtropical Atlantic. Conversely, EOF-PFT performs better in the South Pacific and equatorial
Atlantic. For Haptophytes and Prokaryotes, in summary, both global and regional assessments suggest that AIGD-PFT is the 460
most effective dataset, offering the lowest median fMSE and highest median R values. It stands out not only on a global scale
but also in most regional evaluations, confirming its overall superiority among the comparative datasets.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
27
Figure 16 TCA result of three monthly product (AIGD-PFT, EOF-PFT, and NOBM-monthly).
4 Discussion 465
Phytoplankton serves as the foundation of marine food chains. Comprehensive monitoring and inversion of the spatiotemporal
distribution patterns of Photosynthetic Functional Types (PFTs) are crucial for a deeper understanding of marine ecosystem
functions, predicting and mitigating climate change, and other aspects. Amidst increasing human reliance on marine resources,
maintaining the sustainability of fisheries and ensuring the stability and health of marine, especially coastal, ecosystems have
become particularly urgent. This necessitates higher quality and more detailed phytoplankton diversity data to assist decision-470
making. However, existing satellite PFT products have significant shortcomings in inversion accuracy, spatiotemporal
resolution, spatial coverage, and temporal span, limiting their application in climate and ocean management research.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
28
Therefore, enhancing the quality and coverage of PFT data, with higher temporal resolution, is essential to reveal the immediate
impacts of environmental changes on PFT distribution. Improved spatial coverage would enable more accurate descriptions of
local changes in marine ecosystems, providing more precise data support for scientific management strategies. Additionally, 475
extending the temporal span would enhance the accuracy of long-term trend analysis, thereby better understanding the
evolution of marine ecosystems.
Multi-source marine big data exhibits complementary advantages in terms of spatial integrity and accuracy. By merging data
from various environmental factors, we can produce improved PFT products. In this study, we selected features including
ocean color data, biogeochemistry, temperature and salinity, and spatiotemporal information. Among these, ocean color data, 480
as a crucial predictor, was seamlessly reconstructed using a GPU-accelerated DCT-PLS algorithm, filling gaps caused by
clouds, orbits, and other factors. Compared to traditional reconstruction algorithms, the DCT-PLS algorithm is faster and
effectively addresses the issue of missing observational data, improving data utilization efficiency and monitoring continuity.
Further, by leveraging the powerful nonlinear modelling capabilities of deep learning, we enhanced the accuracy of PFT
inversion. We developed a spatiotemporal ecological integration model based on deep learning, adapting the method proposed 485
by Zhang et al. (2023) for reconstructing global PFTs from 1998 to 2023. The model, composed of 100 ResNet network models,
demonstrates strong nonlinear modelling capabilities and robustness. Using the Monte Carlo method, we utilized ensemble
means and standard deviations as the optimal estimates and uncertainties, generating a temporally continuous global PFT
product covering the entire period and the corresponding uncertainty fields. The standard deviation reflects the variability of
model predictions, indicating the consistency between model predictions, i.e., the level of uncertainty. 490
We also employed three cross-validation methods to comprehensively validate the accuracy. Standard five-fold cross-
validation focuses on the model's performance across the entire dataset, time-block five-fold cross-validation assesses the
model's handling of time series, and space-block five-fold cross-validation concentrates on the model's ability to capture spatial
distribution patterns. The results show that the STEE model generally exhibits good accuracy, demonstrating excellent
performance and stability in addressing temporal and spatial generalization issues. Notably, the model's high adaptability to 495
reconstructed pixels highlights its potential for handling incomplete or inaccurate data, further proving the effectiveness of
integrating ecological parameters and machine learning techniques. By applying the STEE model to all data from 1998 to 2023,
we achieved accurate and robust monitoring of global high-resolution, spatiotemporally continuous PFT products. The TCA
algorithm was used to compare the AIGD-PFT product with other products, showing that our estimation model achieved
competitive overall accuracy. 500
Despite statistical and correlational analyses throughout the paper confirming the reasonable and reliable estimation of global
PFTs by STEE-DL, some uncertainties and limitations still need to be addressed in further work. Firstly, the variance obtained
through ensemble learning mainly focuses on model prediction variability, but this does not fully capture or explain the actual
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
29
product uncertainties. Real product uncertainties are broader, encompassing incompleteness of actual measurements,
uncertainties in predictors, and limitations in understanding the system. Exploring more comprehensive and precise uncertainty 505
estimation methods to further enhance model reliability and applicability is necessary. Additionally, the current STEE-DL
model is solely based on statistical relationships, lacking simulation of biological processes and therefore unable to explain
mechanisms behind phytoplankton abundance changes. Model interpretability will be a focus of our future work. Incorporating
prior information constraints such as ecological principles, biogeographical distributions, and seasonal changes into the model,
constructing physics-guided neural networks, or achieving a symbiotic integration of physical methods and artificial 510
intelligence, will create models that can accurately predict phytoplankton abundance with high interpretability.
The AIGD-PFT product demonstrates the potential application of artificial intelligence and marine big data in PFT modelling.
This study focuses on the production process and product verification of AIGD-PFT, and a deeper analysis of PFT variations
across different spatial and temporal dimensions will be the next research priority. As the product with the longest current time
span (1998-2023) and continuous space-time coverage, AIGD-PFT has the potential to avoid false multi-year fluctuations and 515
trend artifacts caused by data gaps. It helps in understanding the global and local trends of PFTs more broadly and is likely to
reveal how climate change affects the composition of phytoplankton. This is crucial for predicting changes in marine
ecosystems in the future, assessing the impact of climate change on the marine carbon cycle, and formulating corresponding
conservation and management measures.
5 Data Availability 520
The AIGD-PFT (1998-2023, daily) dataset is stored in NetCDF format and can be accessed directly through:
https://doi.org/10.11888/RemoteSen.tpdc.301164 (Zhang and Shen, 2024a). A video demonstration is available at
https://doi.org/10.5446/67366 (Zhang and Shen, 2024b). In addition, a subset of AIGD-PFT (January 2023) can be downloaded
at: https://doi.org/10.5281/zenodo.10910206 (Zhang and Shen, 2024c).
6 Conclusions 525
Constructing long time series models of global Photosynthetic Functional Types (PFTs) has always been a challenging task,
with existing PFT products facing a variety of issues. To refine the monitoring of global phytoplankton groups, this study
developed a deep learning-based spatiotemporal ecological integration model by combining multi-source marine data and
artificial intelligence technology. This model can utilize a wide range of data sources, including ocean color, reanalysis, and
in situ observations, to retrieve and generate the world's first daily updated, 4km resolution seamless PFT product, covering 530
eight major phytoplankton groups. Cross-validation accuracy assessments show that our method can provide accurate and
temporally consistent PFT predictions, demonstrating good performance in TCA evaluations across different products. As the
first phytoplankton group product covering a 26-year span on a daily basis, the AIGD-PFT data aids in analyzing trends and
interannual variations in phytoplankton time series, with the potential to reveal mechanisms by which phytoplankton
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
30
compositions respond to climate change across multiple time and spatial scales. Additionally, the AIGD-PFT product can 535
facilitate the quantification of marine carbon fluxes and improve the accuracy of biogeochemical models. By deepening our
understanding of these key components of marine ecosystems, we can more effectively address the challenges posed by climate
change, ensuring the health of global ecosystems and the sustainable development of human society.
Author contributions
Conceptualization – Project Administration: Yuan Zhang, Fang Shen; 540
Methodology: Yuan Zhang, Fang Shen, Renhu Li, Mengyu Li, Zhaoxin Li;
Writing – Original: Yuan Zhang;
Writing – Review & Editing: Yuan Zhang, Fang Shen, Renhu Li, Mengyu Li, Zhaoxin Li, Songyu Chen, Xuerong Sun.
Competing interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared 545
to influence the work reported in this paper.
Disclaimer
Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Acknowledgements 550
This study has made use of various in situ observations and databases, and the authors would like to thank the many scientists
and crew involved with collecting and processing these data and making them freely and publicly available. All data used are
properly cited and referred to in the reference list.
Financial support
This study was funded by the National Natural Science Foundation of China [Grant No. 42076187 and No.42271348]. 555
References
Alvarado, L., Soppa, M., Gege, P., Losa, S., Dröscher, I., Xi, H., and Bracher, A.: Retrievals of the main phytoplankton groups
at Lake Constance using OLCI, DESIS, and evaluated with field observations, https://elib.dlr.de/189789, 2022.
Beaugrand, G., Edwards, M., and Legendre, L.: Marine biodiversity, ecosystem functioning, and carbon cycles, P Natl Acad
Sci USA, 107, 10120-10124, https://doi.org/10.1073/pnas.0913855107, 2010. 560
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
31
Bracher, A., Vountas, M., Dinter, T., Burrows, J. P., Röttgers, R., and Peeken, I.: Quantitative observation of cyanobacteria
and diatoms from space using PhytoDOAS on SCIAMACHY data, Biogeosciences, 6, 751-764, https://doi.org/10.5194/bg-6-
751-2009, 2009.
Bracher, A., Bouman, H. A., Brewin, R. J. W., Bricaud, A., Brotas, V., Ciotti, A. M., Clementson, L., Devred, E., Di Cicco,
A., Dutkiewicz, S., Hardman-Mountford, N. J., Hickman, A. E., Hieronymi, M., Hirata, T., Losa, S. N., Mouw, C. B., Organelli, 565
E., Raitsos, D. E., Uitz, J., Vogt, M., and Wolanin, A.: Obtaining Phytoplankton Diversity from Ocean Color: A Scientific
Roadmap for Future Development, Front Mar Sci, 4, 55, https://doi.org/10.3389/fmars.2017.00055, 2017.
Canadell, J. G., Ciais, P., Gurney, K., Le Quéré, C., Piao, S., Raupach, M. R., and Sabine, C. L.: An international effort to
quantify regional carbon fluxes, Eos, Transactions American Geophysical Union, 92, 81-82,
https://doi.org/10.1029/2011EO100001, 2011. 570
Catlett, D., Matson, P. G., Carlson, C. A., Wilbanks, E. G., Siegel, D. A., and Iglesias-Rodriguez, M. D.: Evaluation of accuracy
and precision in an amplicon sequencing workflow for marine protist communities, Limnol Oceanogr-Meth, 18, 20-40,
https://doi.org/10.1002/lom3.10343, 2020.
Chassot, E., Bonhommeau, S., Dulvy, N. K., Mélin, F., Watson, R., Gascuel, D., and Le Pape, O.: Global marine primary
production constrains fisheries catches, Ecol Lett, 13, 495-505, https://doi.org/10.1111/j.1461-0248.2010.01443.x, 2010. 575
Chauhan, A., Smith, P. A. H., Rodrigues, F., Christensen, A., John, M. S., and Mariani, P.: Distribution and impacts of long-
lasting marine heat waves on phytoplankton biomass, Front Mar Sci, 10, 1177571,
https://doi.org/10.3389/fmars.2023.1177571, 2023.
Falkowski, P.: OCEAN SCIENCE The power of plankton, Nature, 483, S17-S20, https://doi.org/10.1038/483S17a, 2012.
Field, C. B., Behrenfeld, M. J., Randerson, J. T., and Falkowski, P.: Primary production of the biosphere: Integrating terrestrial 580
and oceanic components, Science, 281, 237-240, https://doi.org/10.1126/science.281.5374.237, 1998.
Fredj, E., Roarty, H., Kohut, J., and Lai, J. W.: Fast Gap Filling of the coastal ocean surface current in the seas around Taiwan,
Oceans-Ieee, https://doi.org/10.1109/OCEANSAP.2016.7485427, 2016.
Garcia, D.: Robust smoothing of gridded data in one and higher dimensions with missing values, Comput Stat Data An, 54,
1167-1178, https://doi.org/10.1016/j.csda.2009.09.020, 2010. 585
Garnesson, P., Mangin, A., d'Andon, O. F., Demaria, J., and Bretagnon, M.: The CMEMS GlobColour chlorophyll a product
based on satellite observation: multi-sensor merging and flagging strategies, Ocean Sci, 15, 819-830,
https://doi.org/10.5194/os-15-819-2019, 2019.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
32
Gregg, W. W. and Casey, N. W.: Modeling coccolithophores in the global oceans, Deep-Sea Res Pt Ii, 54, 447-477,
https://doi.org/10.1016/j.dsr2.2006.12.007, 2007. 590
Gruber, N., Clement, D., Carter, B. R., Feely, R. A., van Heuven, S., Hoppema, M., Ishii, M., Key, R. M., Kozyr, A., Lauvset,
S. K., Lo Monaco, C., Mathis, J. T., Murata, A., Olsen, A., Perez, F. F., Sabine, C. L., Tanhua, T., and Wanninkhof, R.: The
oceanic sink for anthropogenic CO from 1994 to 2007, Science, 363, 1193-+, https://doi.org/10.1126/science.aau5153, 2019.
Guidi, L., Chaffron, S., Bittner, L., Eveillard, D., Larhlimi, A., Roux, S., Darzi, Y., Audic, S., Berline, L., Brum, J. R., Coelho,
L. P., Espinoza, J. C. I., Malviya, S., Sunagawa, S., Dimier, C., Kandels-Lewis, S., Picheral, M., Poulain, J., Searson, S., 595
Stemmann, L., Not, F., Hingamp, P., Speich, S., Follows, M., Karp-Boss, L., Boss, E., Ogata, H., Pesant, S., Weissenbach, J.,
Wincker, P., Acinas, S. G., Bork, P., de Vargas, C., Iudicone, D., Sullivan, M. B., Raes, J., Karsenti, E., Bowler, C., Gorsky,
G., and Coordinator, T. O. C.: Plankton networks driving carbon export in the oligotrophic ocean, Nature, 532, 465-+,
https://doi.org/10.1038/nature16942, 2016.
Gunes, H., Sirisup, S., and Karniadakis, G. E.: Gappy data: To krig or not to krig?, J Comput Phys, 212, 358-382, 600
https://doi.org/10.1016/j.jcp.2005.06.023, 2006.
Henson, S. A., Cael, B. B., Allen, S. R., and Dutkiewicz, S.: Future phytoplankton diversity in a changing climate, Nat
Commun, 12, 5372, https://doi.org/10.1038/s41467-021-25699-w, 2021.
Hirata, T., Hardman-Mountford, N. J., Brewin, R. J. W., Aiken, J., Barlow, R., Suzuki, K., Isada, T., Howell, E., Hashioka, T.,
Noguchi-Aita, M., and Yamanaka, Y.: Synoptic relationships between surface Chlorophyll-a and diagnostic pigments specific 605
to phytoplankton functional types, Biogeosciences, 8, 311-327, https://doi.org/10.5194/bg-8-311-2011, 2011.
Hoareau, N., Portabella, M., Lin, W. M., Ballabrera-Poy, J., and Turiel, A.: Error Characterization of Sea Surface Salinity
Products Using Triple Collocation Analysis, Ieee T Geosci Remote, 56, 5160-5168,
https://doi.org/10.1109/Tgrs.2018.2810442, 2018.
Karlson, B., Godhe, A., Cusack, C., and Bresnan, E.: Introduction to methods for quantitative phytoplankton analysis, 610
Microscopic and molecular methods for quantitative phytoplankton analysis, 5, 2010.
Kim, H., Crow, W., Li, X. J., Wagner, W., Hahn, S., and Lakshmi, V.: True global error maps for SMAP, SMOS, and ASCAT
soil moisture data based on machine learning and triple collocation analysis, Remote Sens Environ, 298, 113776,
https://doi.org/10.1016/j.rse.2023.113776, 2023.
Kramer, S. J., Bolanos, L. M., Catlett, D., Chase, A. P., Behrenfeld, M. J., Boss, E. S., Crockford, E. T., Giovannoni, S. J., 615
Graff, J. R., Haentjens, N., Karp-Boss, L., Peacock, E. E., Roesler, C. S., Sosik, H. M., and Siegel, D. A.: Toward a synthesis
of phytoplankton communities composition methods for global-scale application, Limnol Oceanogr-Meth,
https://doi.org/10.1002/lom3.10602, 2024.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
33
Le Quéré, C., Harrison, S. P., Prentice, I. C., Buitenhuis, E. T., Aumont, O., Bopp, L., Claustre, H., Da Cunha, L. C., Geider,
R., Giraud, X., Klaas, C., Kohfeld, K. E., Legendre, L., Manizza, M., Platt, T., Rivkin, R. B., Sathyendranath, S., Uitz, J., 620
Watson, A. J., and Wolf-Gladrow, D.: Ecosystem dynamics based on plankton functional types for global ocean
biogeochemistry models, Global Change Biol, 11, 2016-2040, https://doi.org/10.1111/j.1365-2468.2005.01004.x, 2005.
Liston, G. E. and Elder, K.: A meteorological distribution system for high-resolution terrestrial modeling (MicroMet), J
Hydrometeorol, 7, 217-234, https://doi.org/10.1175/Jhm486.1, 2006.
Liu, X. M. and Wang, M. H.: Gap Filling of Missing Data for VIIRS Global Ocean Color Products Using the DINEOF Method, 625
Ieee T Geosci Remote, 56, 4464-4476, https://doi.org/10.1109/Tgrs.2018.2820423, 2018.
Liu, X. M. and Wang, M. H.: Global daily gap-free ocean color products from multi-satellite measurements, Int J Appl Earth
Obs, 108, 10271410. https://doi.org/1016/j.jag.2022.102714, 2022.
Losa, S. N., Soppa, M. A., Dinter, T., Wolanin, A., Brewin, R. J. W., Bricaud, A., Oelker, J., Peeken, I., Gentili, B., Rozanov,
V., and Bracher, A.: Synergistic Exploitation of Hyper- and Multi-Spectral Precursor Sentinel Measurements to Determine 630
Phytoplankton Functional Types (SynSenPFT), Front Mar Sci, 4, 203, https://doi.org/10.3389/fmars.2017.00203, 2017.
Mackey, M. D., Mackey, D. J., Higgins, H. W., and Wright, S. W.: CHEMTAX - A program for estimating class abundances
from chemical markers: Application to HPLC measurements of phytoplankton, Mar Ecol Prog Ser, 144, 265-283,
https://doi.org/10.3354/meps144265, 1996.
McColl, K. A., Vogelzang, J., Konings, A. G., Entekhabi, D., Piles, M., and Stoffelen, A.: Extended triple collocation: 635
Estimating errors and correlation coefficients with respect to an unknown target, Geophys Res Lett, 41, 6229-6236,
https://doi.org/10.1002/2014gl061322, 2014.
Mikelsons, K. and Wang, M. H.: Optimal satellite orbit configuration for global ocean color product coverage, Opt Express,
27, A445-A457, https://doi.org/10.1364/Oe.27.00a445, 2019.
Mouw, C. B., Hardman-Mountford, N. J., Alvain, S., Bracher, A., Brewin, R. J. W., Bricaud, A., Ciotti, A. M., Devred, E., 640
Fujiwara, A., Hirata, T., Hirawake, T., Kostadinov, T. S., Roy, S., and Uitz, J.: A Consumer's Guide to Satellite Remote
Sensing of Multiple Phytoplankton Groups in the Global Ocean, Front Mar Sci, 4, 41,
https://doi.org/10.3389/fmars.2017.00041, 2017.
Nair, A., Sathyendranath, S., Platt, T., Morales, J., Stuart, V., Forget, M. H., Devred, E., and Bouman, H.: Remote sensing of
phytoplankton functional types, Remote Sens Environ, 112, 3366-3375, https://doi.org/10.1016/j.rse.2008.01.021, 2008. 645
Raitsos, D. E., Lavender, S. J., Maravelias, C. D., Haralabous, J., Richardson, A. J., and Reid, P. C.: Identifying four
phytoplankton functional types from space: An ecological approach, Limnol Oceanogr, 53, 605-613,
https://doi.org/10.4319/lo.2008.53.2.0605, 2008.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
34
Sadeghi, A., Dinter, T., Vountas, M., Taylor, B., Altenburg-Soppa, M., and Bracher, A.: Remote sensing of coccolithophore
blooms in selected oceanic regions using the PhytoDOAS method applied to hyper-spectral satellite data, Biogeosciences, 9, 650
2127-2143, https://doi.org/10.5194/bg-9-2127-2012, 2012.
Saleh, A. K. and Al-Anzi, B. S.: Statistical Validation of MODIS-Based Sea Surface Temperature in Shallow Semi-Enclosed
Marginal Sea: A Comparison between Direct Matchup and Triple Collocation, Water-Sui, 13, 1078,
https://doi.org/10.3390/w13081078, 2021.
Sathyendranath, S., Brewin, R. J. W., Brockmann, C., Brotas, V., Calton, B., Chuprin, A., Cipollini, P., Couto, A. B., Dingle, 655
J., Doerffer, R., Donlon, C., Dowell, M., Farman, A., Grant, M., Groom, S., Horseman, A., Jackson, T., Krasemann, H.,
Lavender, S., Martinez-Vicente, V., Mazeran, C., Mélin, F., Moore, T. S., Müller, D., Regner, P., Roy, S., Steele, C. J.,
Steinmetz, F., Swinton, J., Taberner, M., Thompson, A., Valente, A., Zühlke, M., Brando, V. E., Feng, H., Feldman, G., Franz,
B. A., Frouin, R., Gould, R. W., Hooker, S. B., Kahru, M., Kratzer, S., Mitchell, B. G., Muller-Karger, F. E., Sosik, H. M.,
Voss, K. J., Werdell, J., and Platt, T.: An Ocean-Colour Time Series for Use in Climate Studies: The Experience of the Ocean-660
Colour Climate Change Initiative (OC-CCI), Sensors-Basel, 19, 4285, https://doi.org/10.3390/s19194285, 2019.
Sun, X. R., Shen, F., Brewin, R. J. W., Li, M. Y., and Zhu, Q.: Light absorption spectra of naturally mixed phytoplankton
assemblages for retrieval of phytoplankton group composition in coastal oceans, Limnol Oceanogr, 67, 946-961,
https://doi.org/10.1002/lno.12047, 2022.
Swan, C. M., Vogt, M., Gruber, N., and Laufkoetter, C.: A global seasonal surface ocean climatology of phytoplankton types 665
based on CHEMTAX analysis of HPLC pigments, Deep-Sea Res Pt I, 109, 137-156, https://doi.org/10.1016/j.dsr.2015.12.002,
2016.
Uitz, J., Claustre, H., Morel, A., and Hooker, S. B.: Vertical distribution of phytoplankton communities in open ocean: An
assessment based on surface chlorophyll, J Geophys Res-Oceans, 111, C08005, https://doi.org/10.1029/2005jc003207, 2006.
Veldhuis, M. J. W. and Kraay, G. W.: Application of flow cytometry in marine phytoplankton research: current applications 670
and future perspectives, Sci Mar, 64, 121-134, https://doi.org/10.3989/scimar.2000.64n2121, 2000.
Wang, G. J., Garcia, D., Liu, Y., de Jeu, R., and Dolman, A. J.: A three-dimensional gap filling method for large geophysical
datasets: Application to global satellite soil moisture observations, Environ Modell Softw, 30, 139-142,
https://doi.org/10.1016/j.envsoft.2011.10.015, 2012.
Wang, T. H., Yu, P., Wu, Z. L., Lu, W. F., Liu, X., Li, Q. P., and Huang, B. Q.: Revisiting the Intraseasonal Variability of 675
Chlorophyll-a in the Adjacent Luzon Strait With a New Gap-Filled Remote Sensing Data Set, Ieee T Geosci Remote, 60,
4201311, https://doi.org/10.1109/Tgrs.2021.3067646, 2022.
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
35
Xi, H., Bretagnon, M., Losa, S. N., Brotas, V., Gomes, M., Peeken, I., Alvarado, L., Mangin, A., and Bracher, A.: Satellite
monitoring of surface phytoplankton functional types in the Atlantic Ocean over 20 years (2002–2021), State of the Planet, 1,
1-13, 2023. 680
Xi, H. Y., Losa, S. N., Mangin, A., Garnesson, P., Bretagnon, M., Demaria, J., Soppa, M. A., D'Andon, O. H. F., and Bracher,
A.: Global Chlorophyll a Concentrations of Phytoplankton Functional Types With Detailed Uncertainty Assessment Using
Multisensor Ocean Color and Sea Surface Temperature Satellite Products, J Geophys Res-Oceans, 126, e2020JC017127,
https://doi.org/10.1029/2020JC017127, 2021.
Xi, H. Y., Losa, S. N., Mangin, A., Soppa, M. A., Garnesson, P., Demaria, J., Liu, Y. Y., D'Andon, O. H. F., and Bracher, A.: 685
Global retrieval of phytoplankton functional types based on empirical orthogonal functions using CMEMS GlobColour merged
products and further extension to OLCI data, Remote Sens Environ, 240, 111704, https://doi.org/10.1016/j.rse.2020.111704,
2020.
Yang, K. X., Luo, Y. M., Li, M. Y., Zhong, S. Y., Liu, Q., and Li, X. H.: Reconstruction of Sentinel-2 Image Time Series
Using Google Earth Engine, Remote Sens-Basel, 14, ARTN 4395, https://doi.org/10.3390/rs14174395, 2022. 690
Zhang, Y., Shen, F., Sun, X. R., and Tan, K.: Marine big data-driven ensemble learning for estimating global phytoplankton
group composition over two decades (1997-2020), Remote Sens Environ, 294, 113596,
https://doi.org/10.1016/j.rse.2023.113596, 2023.
Zhang, Y. and Shen, F.: Global daily gap-free 4km phytoplankton functional types product from 1998 to 2023. National
Tibetan Plateau / Third Pole Environment Data Center, [data set], https://doi.org/10.11888/RemoteSen.tpdc.301164, 2024a. 695
Zhang, Y. and Shen, F.: AIGD-PFT: The first AI-driven Global Daily gap-free 4 km Phytoplankton Functional Type products
from 1998 to 2023. Copernicus Publications, German National Library of Science and Technology [video].
https://doi.org/10.5446/67366 , 2024b.
Zhang, Y. and Shen, F.: AIGD-PFT: The first AI-driven Global Daily gap-free 4 km Phytoplankton Functional Type products
from 1998 to 2023, Zenodo [data set], https://doi.org/10.5281/zenodo.10910206 , 2024c. 700
https://doi.org/10.5194/essd-2024-122
Preprint. Discussion started: 6 May 2024
c
Author(s) 2024. CC BY 4.0 License.
