The EL-BIOS Earth Observation Data Cube for Supporting Biodiversity Monitoring in Greece

Abstract

In recent years, the need to protect and conserve biodiversity has become more critical than ever before, as a prerequisite for both sustainable development and the very survival of the human species. This has made it a priority for the scientific community to develop technological solutions that provide data and information for monitoring, directly or indirectly, biodiversity and the drivers of change. A new era of satellite earth observation upgrades the potential of Remote Sensing (RS) to support, at relatively low cost, but with high accuracy the extraction of information over large areas, at regular intervals, and over extended periods of time. Also, the recent development of the Earth Observation Data Cubes (EODC) framework facilitates EO data management and information extraction, enabling the mapping and monitoring of temporal and spatial patterns on the Earth’s surface. This submission presents the ELBIOS EODC, specifically developed to support the biodiversity management and conservation over Greece. Based on the Open Data Cube (ODC) framework, it exploits multi-spectral optical Copernicus Sentinel-2 data and provides a series of Satellite Earth Observation (SEO) biodiversity products and spectral indices nationwide.

1. Introduction

Conservation of biodiversity (or biological diversity) is critical to maintaining sustainable ecosystem processes and functions that support human health and well-being [1].

Detailed assessments indicate an alarming loss of biodiversity over the past 50 years, driven by several direct and indirect drivers of change related mainly to human activities and demand for food and natural resources supporting economic development [2]. Further, global warming and extreme climate events exacerbate the impact of human activities, increasing biodiversity loss and affecting species populations as well as distribution patterns and ecosystem function, among others [3].

The increasing impact of such human and natural factors requires systematic biodiversity monitoring to assess its state, mitigate risk of further loss and guide restoration efforts [4,5]. To protect nature and reverse the degradation of ecosystems, strategies and policies are introduced at the national and supra-national levels. Examples include the United Nations Convention on Biological Diversity (CBD), Kunming-Montreal Global Biodiversity Framework and the EU’s Biodiversity Strategy for 2030, all setting strategic goals and specific targets, along with a set of indicators for supporting progress of the implementation. Recognizing the limitations and the challenges related to data collection for these indicators [6,7], the scientific community suggested the use of Remote Sensing (RS) as a complementary or an alternative source. Remote Sensing could support the measurement of these indicators and facilitate the monitoring of biodiversity and associated threats at scales ranging from a global to local scale [8]. The most well-known and widely adopted initiative that incorporates RS for monitoring biodiversity is the Essential Biodiversity Variables (EBV) framework, proposed by the Group on Earth Observations Biodiversity Observation Network (GEO BON) [9].

Apart from indirectly monitoring certain biodiversity dimensions as in the case of EBVs concept, RS, and, in particular, Satellite Remote Sensing (SRS) products [7,10,11], have also been recognized for their key role in conservation and management of biodiversity, providing products for extrapolating in situ measurements in a routine, consistent and cost- effective manner, over extended spatial and temporal extents [12]. A number of decommissioned, active and upcoming satellites allow for the development of multi-decadal time series of earth observations that can be related to biodiversity status and condition changes [3,13,14,15]. Currently, petabytes of ΕO data are being generated and distributed daily [16], including freely available imagery from satellite missions such as NASA’s Landsat and Copernicus’ Sentinel with high revisit frequency and global coverage [17,18]. Yet, the volume of the EO data, and the heterogeneity and complexity arising from the various characteristics of the different remotely sensed instruments commonly employed in long-term monitoring projects [19], makes challenging their efficient use and the information extraction process. The aforementioned character of the extended satellite time-series hinders the usefulness of the traditional individual “scene”-based organization and management of EO data [20]. Robust memory capacities, software, data structures and standard protocols are required to manage the increased Volume, Variety, Velocity, Veracity and Value of Big Earth Data [21,22] and integrate both open and commercial data sources in scientific research and operational contexts [23,24,25].

To address these issues, new paradigms have been assessed for improved EO data access from the RS community. In recent years, EO Data Cubes (EODC), defined as collections of multidimensional arrays, have surfaced as a novel paradigm that is transforming users’ engagement with EO data, revolutionizing the way large-scale, multi-dimensional data are managed and analyzed [26]. Operational examples of EODC demonstrate their global reach, with implementations supporting continental [27,28] and national-scale [29,30,31,32,33] EO-based initiatives. Such an approach offers a potential solution for the administration, organization, storage and analysis of the unprecedented amounts of EO data [19,34], improving the relationships between users, apps and EO data, making it easier to manage, access and use data that is ready for analysis (Analysis Ready Data, ARD) [35]. Analysis-ready data refers to EO data that has been appropriately pre-processed to specific requirements and organized in a way that enables low-effort scientific analysis, facilitating information extraction and usage of the EO from different scientific communities [29].

Different open-source software packages and tools are available to users to facilitate the task of setting up infrastructures providing seamless access to EODC such as Open Data Cube (ODC) [36] or gdalcubes [26]. These infrastructures are deployed both on-premises and in commercial and public cloud infrastructure, providing flexibility based on the user’s needs [37]. On-premises deployments are ideal for low-cost testing and development, while cloud-based solutions provide scalable computing resources, making them cost-effective for large-scale data analysis and projects.

Among the aforementioned open-source software solutions, Open Data Cube, born in 2017 [27], provides the RS community with a range of tools and APIs for high-performance access, management and analysis of large-scale, temporally dense EO data collections [38]. Additionally, the ODC framework allows for a pixel-based approach, which directly compares observations from a specific location over extensive epochs, rather than using the traditional scene-based approach [39]. At present, ODC supports continental, national and regional operational initiatives. There are more than one hundred operating EO EODCs at the country and regional scale, either operational or under development [40]. Some of the most important data cubes built upon ODC framework include Digital Earth Australia [27], Digital Earth Africa [28], Swiss Data Cube [29], Brazil Data Cube [30], semantic EO data cube for Austria [31], Colombian Data Cube [32] and regional cubes in Catalonia [41] and Virginia [42].

The purpose of this study is to present in detail the development of an EODC structure, the data hosted, development of EO workflows from satellite image time series and visualisation as part of the hELlenic BIOodiversity Information System (EL-BIOS). The EL-BIOS is the first national-scale EODC infrastructure, with the aim of advancing EO data and products use for biodiversity management and conservation over Greece. This technical note is an updated and expanded version of our IEEE IGARSS 2024 Conference contribution [43], providing a detailed description of the technical specifications of the ELBIOS EODC and implications of the operational use.

To align with the Digital Earth vision of data integration and accessibility [44,45], the EL-BIOS EODC provides to the end-users six satellite earth observation (SEO) biodiversity products linked to four EBVs (i.e., live cover fraction, primary productivity, physiology, phenology) [43,46]. Indexed products as well the spectral indices used for their development are also available through Open Web Services and a webGIS graphical user interface, which offers visualisation and download capabilities.

2. Materials and Methods

2.1. Study Area

Greece is located in the Mediterranean basin biodiversity hot-spot and boasts a diverse range of ecosystems and species. The uniqueness of Greek nature is characterized by its high species density and a variety of different ecosystems and landscapes. Notably, Greece has a particularly rich and diverse flora, with 5700 recorded species, of which approximately 13.2% are endemic due to its complex landscape and morphology [47]. The fauna of Greece is known for its high rate of endemism, with a total of 23,130 recorded species of terrestrial and freshwater animals, as well as 3500 marine species [47,48]. According to the World Wildlife Fund (WWF), Greece is home to 8% and 40% of Europe’s animal and plant species [49]. The rich biodiversity is reflected in the designation of 446 sites as part of the European Natura 2000 network, covering 27% of the country’s land area and over 19% of its marine territory. These protected areas in Greece are managed by the Natural Environment & Climate Change Agency (NECCA) under the Greek Ministry of Environment and Energy (Figure 1). The Natura 2000 network in Greece includes both areas heavily affected by recent disastrous wildfires (as in the case of the GR 1110002 Dasos Dadias-Soufli site) and remote areas under a rigorous protection regime, with limited human presence (as in the case of the GR 1310002 Valia Kalnta kai Techniti Limni Aoou site).

2.2. Satellite Earth Observation Biodiversity Products

Biodiversity variables and indicators are widely used and are a crucial part of most biodiversity inventories and monitoring systems [50]. The selection of such indicators depends on the scope and the feasibility of indicator monitoring with remote sensing means [51]. To effectively align local/national monitoring priorities with supra-national monitoring and reporting needs, it is necessary to carefully consider biodiversity conservation targets, stakeholders requirements and international policies and frameworks [52]. Furthermore, in order to use SEO biodiversity products in the biodiversity indicator measurements process, is critical to assess parameters such as feasibility, accuracy, relevance and maturity of the satellite EO products [7].

The workflow for prioritizing and developing SEO biodiversity products at a national scale prior to the development of the ELBIOS EODC involved three key steps: review of existing international initiatives, stakeholder engagement, analysis of the technical characteristics of the EO platforms and products and modelling (feasibility) and data requirements for implementing the respective algorithms. This process included a comprehensive review of biodiversity indicators adopted in various frameworks, such as Greece’s National Biodiversity Strategy, the Aichi Biodiversity Targets, SEBI 2020, EEA indicators, the Global Biodiversity Framework and EBVs [3,53]. Stakeholders from 46 national bodies were consulted through discussions and surveys to select and prioritize 15 key indicators, with additional input on climate change and new European regulations [14,43,46]. The integration of EO data, notably from Copernicus and NASA, and cloud-computing platforms such as the Google Earth Engine and Sentinel Hub, were assessed for their effectiveness in EO data provision and the development of automated workflows for national-scale modelling. This led to the creation of an initial pool of 94 potential SRS products related to species, ecosystems, protected areas, water, and other aspects, leading to a selection of 50 selected for higher acceptance based on their maturity and relevance. Finally, six satellite earth observation (SEO) biodiversity products were chosen as most closely related to national and international priorities, taking into account particularly the EBVs concept [7,15,54]. The detailed workflow and findings of the stakeholder engagement process and the focal areas of interest are described in Mallinis et al. [46].

2.3. EL-BIOS EODC

The EL-BIOS EODC infrastructure relies the ODC framework to organize and analyze EO data [55]. It comprises a catalogue of Sentinel-2 Level-2A imagery hosted on AWS. EO workflows have been developed to analyze satellite images time series and extract SEO biodiversity products through ODC APIs. The resulting Cloud-Optimized GeoTIFFs (COGs) are stored and indexed in the EL-BIOS EODC. These COGs can be visualized in a Web GIS interface using Open Web Services (Figure 2). The application’s development was based on open-source software. The following sections provide an analytical description of the process.

2.3.1. Architecture (Database, Indexing, Storage)

The EL-BIOS EODC uses a PostgreSQL database to store metadata of the managed data, under the agdc schema of ODC [27]. The EL-BIOS EODC is developed upon Sentinel-2 Level 2A Cloud Optimised GeoTIFFs hosted in the S3 Bucket (s3://sentinel-cogs/) by AWS Open Data Registry and managed by Element 84 [56]. The data catalogue was designed to cover the wider Greek territory, including both the mainland and the islands. An indexing process was foreseen to be executed on a monthly basis to include new acquisitions in the EODC.

To index data, the description and metadata of a product is initially registered. Following the ODC terminology, a product is a collection of datasets (observations) spanning time and space, with each dataset containing an identical set of measurements and associated metadata. The workflows developed for the selected SEO biodiversity products were designed to generate composites of key statistical metrics, such as maximum, minimum, mean, median, sum and standard deviation, for each dataset within a temporal segment. These statistical metrics were computed and stored in COG format. For each SEO dataset, the metadata defines the measurements, which represent these statistical composites and are mapped to the data cube schema. The corresponding STAC metadata file is stored alongside the COG composite images in the directory of each dataset. After each workflow execution, the generated dataset was indexed in the data cube. The pixel data does not need to be stored in the data cube. When it comes to data orchestration, data movement can be a significant performance bottleneck for big data analytics [38]. To address this issue, files are stored as objects and can be downloaded via Hypertext Transfer Protocol (HTTP), enabling the evolution of the EL-BIOS without disrupting its interface with the user community.

The generation of the SEO biodiversity products involves processing, storage, and indexing, which are conducted through an application. This application includes functions in both Python and R and is executed in isolated containerized environments using docker engine (version 27.1) to build the image (i.e., a self-sufficient software package that includes all the necessary components to run an application, such as code, runtime, system tools, system libraries and settings). The constructed docker image exposes the functionality of the application via a REST API, implemented using the FastAPI framework. The use of an image allows for environment-agnostic application packaging. Containerizing an application facilitates a quick and efficient way to deploy code in both on-premises and cloud infrastructures. Furthermore, this portability enables the backend to continuously develop the service and provide it in lightweight capsules for future services, such as sandboxing the execution of user codes. Finally, a distributed processing system using the Dask framework for parallel processing and scheduling was deployed, utilizing Docker container virtualization technology. This approach allows for efficient computation and memory management by distributing computations to workers.

The archive currently hosts 6TB of SEO biodiversity products and spectral indices that are ready for analysis. This amount is expected to increase by almost 1TB annually with new Sentinel-2 acquisitions. The dataset metadata is written in STAC format and points to the lineage of the baseline Sentinel-2 data, which is hosted in AWS and was used in generating each SEO biodiversity product and spectral index.

2.3.2. Visualization with Web GIS and OWS

The EL-BIOS Web GIS, serving as the primary platform for visualization, comprises three core infrastructure units. The server-side infrastructure offers GIS interfaces and data access, while the second unit links the backend with the end-user graphical interface (GUI), facilitating seamless data interaction. The final unit manages content and features for Earth observation data browsing within the EL-BIOS Data Cube, made accessible through datacube-OWS open-source module [57].

Datacube-OWS integration fosters geospatial data standardization by merging Open Geospatial Consortium (OGC) Web Services (OWS) protocols with ODC technology. This integration empowers users to remotely access EL-BIOS EODC contents, enabling spatiotemporal analysis of ΕO data. While OWS faces performance bottlenecks due to on-the-fly data processing, datacube-OWS ensures interoperable access to indexed data collections, supporting both visualization and download functionalities through OGC standards like WMS, WMTS, and WCS. Additionally, datacube-OWS significantly enhances accessibility and efficiency in managing large-scale spatiotemporal datasets. Configuration is simplified via a command-line tool, facilitating index creation and product specification.

2.4. Satellite Earth Observation Data Pre-Processing

The datacube [37] and xarray [58] Python packages were used to interact with Sentinel-2 L2A satellite data from the EL-BIOS Data Cube. The analysis was limited to data from 2017 onwards, due to sparse and non-systematic coverage of Sentinel-2 L2A across Greece prior to this date from Registry of Open Data on AWS [56]. The Sentinel-2 L2A data are atmospherically corrected using the Sen2Cor processor and PlanetDEM Digital Elevation Model (DEM), enabling consistent spatiotemporal analysis [59]. The pre-processing of the satellite image time series (SITS) was conducted per MGRS tile, as shown in Figure 3.

To better fit the atmospherically corrected surface reflectance annual curve, additional observations of one month were included on both sides of the processed year. Sentinel-2 scenes with a cloud cover percentage lower than 40% were selected and broken datasets were filtered out. For quarters that contained less three images after the querying, no products were calculated. Cloud, cloud shadow, cirrus, and missing or invalid data were removed using the Scene Classification Map (SCL). To obtain a more robust cloud mask subsequent morphological transformations were applied using the scikit-image Python [60] and dea-tools [61] libraries. A closing operation with a structuring element of 2 × 2 pixels was performed to fill small gaps within clouds, followed by dilation with a 5 × 5 pixel structuring element to capture thin cloud edges [62]. For pixels to be calculated in a quarterly dataset should have at least three cloud-free observations with its time range. For quality control, metadata contain the N/A percentage cover of each dataset. A land–sea mask was applied based on the Greek coastline shapefile provided by the European Environmental Agency (EEA). The shapefile was reprojected, buffered by 100 m, rasterised, and applied as a land–sea mask.

A workflow for the preprocessing of the SEO data series was implemented as part of each of the SEO biodiversity product development, as many of the selected products require the use of spectral indices (i.e., NDVI or EVI) as baseline data. After the masking process, the spectral indices time series were computed, resulting in irregular time series that span over 14 months. To reconstruct the irregular time series, the data are resampled into 70 equidistant periods, resulting in 60 observations per Julian year. However, there are some short gaps resulting from the filtering process, which are filled in using a linear interpolation method that considers the nearest cloud-free neighbouring values. Due to the high temporal resolution of the time series, data are susceptible to random noise contamination caused by unmasked cloud, atmospheric or solar variability [63]. Therefore, time series were smoothed using a rolling median filter with a window length of three observations, to reduce short-term variability [64], as it is more robust than moving average to outlier bias. To capture long-term trends, seasonality curves and to handle remaining data variation, the Savitzky–Golay technique was applied [65,66].

The Savitzky–Golay filter approximates the spectrum by polynomial least-square fitting inside a moving window of length $2m + 1$. The half-width kernel ($m$) parameter controls the smoothness of the results, but also reduces spike flatness [67]. Conversely, the degree value ($d$) of the polynomial may result in overfitting, yet it also reduces bias [66]. The Savitzky–Golay filter formula is described as follows:

$$Y_i^r=\left({{\displaystyle \sum }}_{j=-m}^m{C}_j\times Y_{i+j}\right)/\left(2m+1\right)$$

(1)

where Y_i means the i-th original data, Y_r means the reconstructed data, C_j is the coefficient of the j-th point in the sliding window and m is the length of the sliding window. As the SCL mask was insufficient to capture all low-quality pixels, the remaining outliers were smoothed out by using the literature-suggested half-window width and a lower polynomial order to capture low-order variations. Through local scale experiments the optimal (m,d) combination for the project is (4,2).

Following the denoising of the time series, it is truncated to the start and end of the calendar year. Based on the modelling definitions of the SEO biodiversity products, statistical metrics such as mean, median, maximum, minimum, sum, and standard deviation are calculated as temporal composites. This is achieved by resampling the time series into quartiles or directly from the annual curve, utilizing the xarray library. The metadata indicates the lineage of the Sentinel-2 L2A imagery used for the time series construction. Finally, to reduce peak memory usage, all processing was applied in parallel with Dask [68], and outputs were written in cloud optimized GeoTIFF format of integer data type to preserve storage space. The ARD workflows were developed as a suite of Python (3.10) and R (4.2) scripts.

3. Results

3.1. EL-BIOS Data Cube Spatiotemporal Coverage

The EL-BIOS EODC consists of 60 MGRS tiles and spans from January 2017 onwards. Despite the presence of high cloud cover or incomplete datasets, which increased the proportion of invalid data, the developed EO workflows achieved 88.47% dataset coverage (Figure 4). In total, 13,005 of the 14,700 datasets were computed and indexed in the EL-BIOS data cube. More than 225 observations, as the 90th percentile of tile-based coverage, were observed in the MGRS tiles 34SDG, 34SFF, 34SFH, 34TCK, 35SKC, 35SLU and 35SMC, which are mainly located in central Greece. The archive had less than 207 observations, as the 10th percentile, in Eastern Macedonia and Thrace, as evidenced by MGRS tiles 34TFL, 35SMU, 35TKF, 35TKG, 35TLF and 35TMF.

In terms of temporal coverage, the year 2017 was the least represented (Figure 5), mainly due to a low pass frequency and high cloud cover, as Sentinel-2B was not yet in orbit. In terms of processing time, based on experimental results, it is estimated that parallel processing using Dask saved 69.44% of the time required.

3.2. Satellite Earth Observation Biodiversity Products in the EL-BIOS EODC

Six SEO biodiversity products are included in the EL-BIOS EODC (

Table 1. The six SEO biodiversity products in EL-BIOS EODC.

SEO Product	Alias	EBV Proxy	Temporal Resolution
Green Fractional Vegetation Cover	FVC	Live cover fraction	Quarter
Annual net primary productivity	NDVI-I	Physiology	Year
Leaf Area Index	LAI	Physiology	Month
Intra-annual relative range	IARR	Primary productivity	Year
Plant Phenology Index	PPI	Phenology	Quarter
Date of Annual maximum NDVI	DAM	Phenology	Year

) along with three spectral indices (i.e., NDVI, EVI, albedo). Green Fractional Vegetation Cover (FVC) relates to EBV “live cover fraction” [7]. The estimation of the green FVC (also known as fractional cover of photosynthetic vegetation) assumes that a pixel’s value results from the spectral reflectance solely from green vegetation and soil. Annual net primary productivity corresponding to EBV “Physiology” was modelled through the annual NDVI integral (NDVI-I) [69]. Under the same EBV groups falls another SEO biodiversity product selected-namely Leaf Area Index (LAI). This product relies on Enhanced Vegetation Index (EVI) based on the formula of Boegh et al. [70].

The intra-annual relative range ($\mathrm{IARR}$) is expressed by dividing the difference between maximum and minimum NDVI values by NDVI-I, thus characterizing the seasonality of carbon fluxes, linked with EBV “Primary productivity”. Finally, the Plant Phenology Index (PPI) and the Date of Annual Maximum NDVI were selected as a proxy for vegetation phenology, corresponding to EBV “Phenology”. The PPI model relied on the work of Jin and Eklundh [71] and Copernicus’ High-Resolution Vegetation Phenology and Productivity (HR-VPP) ATBD [72].

3.3. Case-Study—Protected Areas Monitoring Following Wildifre and Drought Events

Two instances of the potential use of the ELBIOS EODC are presented. The first one relates to the largest wildfire of the century in Europe that occurred in Thrace, northeast Greece, during the summer of 2023 [73]. The fire caused extensive damage to the area, with over 93,000 hectares of land affected, including 74,200 hectares of forest. The wildfire resulted in 20 human losses and a significant impact on local ecosystems, including the biodiversity loss related to the fire-induced damage in over 2452 hectares of the Dadia-Lefkimi-Soufli Forest National Park. The data can be downloaded either by using the webGIS user GUI or by connecting to the EODC in a Jupyter Notebook.

The impact of this event can be quantified through the differential product of pre- and post-fire LAI monthly median composites of October 2022 and October 2023, respectively, readily available in the ELBIOS EODC (Figure 6). The spatial explicit LAI differences can provide readily available information to management authorities (i.e., Natural Environment Climate & Change Agency, local forest department) for decision-making. The differential LAI can help identifying areas characterized by significant changes in vegetation composition and structure as well as habitats with low recovery rate, suggesting that measures to address food shortage, loss of nesting sites, etc., should be addressed.

Likewise, monitoring changes over time for the selected SEO biodiversity at site level offers a swift, broad understanding of vegetation health and ecosystem stability triggering if needed site specific management measures. Annual net primary productivity, for example, is essential in assessing the state of ecosystems, land-use alterations on biodiversity and revealing impacts of climate change, such as drought and extreme temperatures [74], while minor annual productivity changes could be used to identify areas exhibiting stable vegetation patterns, acting effectively in the conservation of priority species and habitats [75]. Annual productivity (expressed through annual NDVI integral) between 2018 and 2023 within the GR1310002 Valia Kalnta kai Techniti Limni Aoou site, suggest a stable pattern during 2018–2020, followed by a subtle negative change in 2021 net primary productivity due to drought, followed by a positive increase during 2022–2023 (Figure 7a). When assessing change over the GR1110002 Dasos Dadias—Soufli site (Figure 7b) the productivity overall follows a similar pattern. However, a major decline is observed in 2023 as a result of the major event described earlier.

4. Summary and Conclusions

The EL-BIOS EODC provides a comprehensive archive of analysis-ready SEO biodiversity products at national scale, with data available from 2017 onwards at a resolution of 10 m. The information is provided in a synoptic and spatially explicit manner, enhancing the capability to monitor biodiversity status and changes across the country. These data can be provided at pixel level, or aggregated at various administrative levels (municipalities, national parks or Natura 2000 sites). Such an infrastructure as the EL-BIOS Data Cube can contribute to national and EU policies implementation [37,76], support reporting at international initiatives [77], and site or habitat management at local scale.

EL-BIOS EODC is the first biodiversity inventory of EO-derived biodiversity related products in Greece. The EL-BIOS Data Cube is funded on a project-based approach however, a more sustainable funding mechanism is called for, one possibility being through a support for a national digital infrastructure. These ARD datasets were compiled in time series stacks to enable interoperability across multiple datasets. Such an organization also facilitates straightforward, efficient time series information extraction and scientific analysis. The later would be nearly impossible, or extremely challenging to be accomplished, through traditional scene-based analysis approach [28].

Scalability of El-BIOS EODC can be achieved by providing users with their own sandbox environment. Technologies such as Kubernetes and Docker Spawner could provide better management tooling for handling resource usage. Additionally, interoperability could be enhanced by avoiding retrieval of EO data using metadata and focus on spatially explicit semantic content-based information search. This could be achieved by introducing Computer Vision based models [31].

Regarding the limitations, the time series are aggregated to seasonal or annual bins, with masked pixels based on cloud-free observations. This results in data gaps in the time series. With regard to recommendations on the pre-processing of SEO products, a potential avenue for further research would be to improve the outlier detection and gap filling method.

Overall, through the development of the EL-BIOS EODC, the existence of ARD can enhance the ability to address local and national scientific needs and operational requirements for decision-making in biodiversity conservation and restoration. Data curation is research studies are time consuming processes with limited scientific interest. This way the Value of Big EO data can emerge with researchers focusing on image classifications, time series analysis, environmental assessments, land change detections and associated scientific analysis.