High-quality, openly available biodiversity datasets (e.g. species occurrence, abundance, traits) are indispensable for the monitoring of species and ecosystems and to improve the development of conservation and management policies (Wetzel et al. 2015, Wetzel et al. 2018). Biodiversity data under FAIR (findable, accessible, interoperable and reusable) principles also help optimising editorial processes for academic publications accelerating peer review, increasing the visibility of scientific papers and improving citation rates (Costello et al. 2013, Piwowar and Vision 2013). Efforts to build and maintain repositories of FAIR data principles have been undertaken by biodiversity data collectors and curators (Hackett et al. 2019), who strive to organise, standardise and share data from a variety of primary (e.g. fieldwork) and secondary (e.g. literature) sources, for example, global initiatives, such as the "Global Biodiversity Information Facility - GBIF" (https://www.gbif.org/), which provide access to comprehensive biodiversity datasets and facilitate their reuse. This is particularly important in the modern era of big data science, which challenges our ability to organise, filter and analyse large and complex datasets (Cao 2016).

The availability of biodiversity data is influenced by several factors, including geographic region, scientific interest and resource availability (e.g. financial and infrastructure constraints), which can affect the quality and type of data produced (Amano et al. 2016). In addition, biodiversity data are not always publicly available, requiring users to conduct extensive searches in scientific publications, technical reports or by directly contacting the researcher who collected the data (Costello et al. 2013). Therefore, conducting systematic literature reviews is often necessary to compile comprehensive databases focusing on biodiversity research programmes. This method is recommended as it provides a rigorous way to search for relevant literature, allowing for peer replication and ensuring data validity and reliability (Xiao and Watson 2019). However, the retrieved data may be in inconsistent and sometimes confusing formats, requiring additional effort from users, including finding separate metadata files, reorganising and renaming fields, integrating aggregated data and discarding poorly documented or questionable data (Parr et al. 2012, Hunt et al. 2015, Culley 2017). The fundamental goal of the structure and standardisation of biodiversity data is to enable them to be understood and used by anyone and to be continuously updated and integrated with other datasets (Borregaard and Hart 2016). There are several guidelines and standards for biodiversity data, including the Darwin Core (DwC; Wieczorek et al. (2012); Darwin Core Maintenance Group (2014) and the DMPTool (https://dmptool.org/).

Finally, data must be shared and archived to ensure findability, accessibility and reusability. Data sharing can occur in a variety of ways, ranging from private sharing on request to depositing data on a public platform. Often, authors make their data available as supplementary material in scientific publications and post datasets on public websites. Sharing biodiversity data is essential for ecological research, conservation and management, education and policy decision-making (Ganzevoort et al. 2017). In addition, ecologists often use shared data for comparative studies, syntheses (e.g. meta-analyses), model parameterisation and reproducibility testing (Michener 2015). Data archiving is a critical final step that allows data to be reused for further analyses and syntheses to address new questions. Whitlock (2011) outlines optimal procedures for archiving ecological and evolutionary data, including selecting the appropriate repository and ensuring the accuracy of data and metadata. Good archiving practices facilitate long-term preservation of data, making them accessible for future research and applications.

Research programmes from megadiverse regions, such as the Neotropics, face difficulties in retrieving, organising and providing quality data due to their intrinsic complexity, which includes unrecognised species and unresolved taxa complexes and taxonomy. To address these challenges and to improve the reusability of secondary biodiversity data, our study had three main objectives:

1. to create a scheme or "pipeline" to improve the usability of secondary data by locating the data, performing quality control, standardising the data and archiving and sharing it;
2. to test our pipeline through a case study, demonstrating the step-by-step management of secondary data according to the FAIR principles; and
3. to evaluate if and how our approach can improve our understanding of the dynamics of regional biodiversity distribution and conservation, promote new scientific studies and knowledge and enhance our ability to generate hypotheses.

Retrieving biodiversity data is not an easy task, as the use of systematic literature searches alone does not guarantee the quality of the data. Additionally, commonly used pipelines for retrieving, standardising and making secondary data available typically overlook grey literature, despite its potential for biodiversity studies. The proposed pipeline is a combination of data retrieval and data management tools that are typically used separately, such as systematic review and the Darwin Core Standard. Following this Biodiversity Data Retrieval Pipeline ensures that secondary data are cleaned, normalised, shared and archived according to the FAIR Data Principles. Additionally, we discuss the challenges of efficient data retrieval, the potential reuse of secondary data in future studies and its limitations. Finally, we predict that initiatives to collect biodiversity data and make it available for reuse can improve knowledge and advance conservation efforts to protect the species, communities and ecosystems of these regions.

The Biodiversity Data Retrieval Pipeline was built following four stages (Fig. 1): (1) Search; (2) Validate; (3) Standardise; and (4) Share and archive. All the steps are described below.

Figure 1.  

Step-by-step guide for the proposed Biodiversity Data Retrieval Pipeline to retrieve secondary biodiversity data from various sources (e.g. scientific articles, technical reports, theses, dissertations, databases) according to the FAIR principles (findable, accessible, interoperable and reusable).

To test the pipeline, we selected the Golfão Maranhense, a region located in the extreme north of the Amazon (Brazil), due to its richness, ecological diversity and importance as an ecotonal mosaic between the Amazon Forest and the dry ecosystems of eastern South America and because of the scarcity of knowledge about the biodiversity in the region. Although reports on biodiversity in the region exist, they are presented in heterogeneous forms, including scientific articles and non-peer-reviewed technical reports, making it difficult to understand the true distribution of biodiversity richness in the region.

Study area

The study was conducted in the Golfão Maranhense (Maranhão State, Brazil) including 13 municipalities in the surroundings (Fig. 2). The Golfão Maranhense is a vast estuarine complex located in eastern Amazonia (Brazil) and is formed by the São Marcos and São José bays separated by the island of São Luís. This region is an area of high ecological relevance known as the “Macromaré” Mangrove Coast of Amazonia where lies the largest continuous mangrove system in the world, with about 5,414 km of mangroves in north-western Maranhão and 2,177 km in north-eastern Pará (Souza Filho 2005). The climate is tropical humid, with well-defined dry (July to December) and rainy (January to June) seasons and average temperatures around 26^oC. The area is characterised by semi-diurnal macrotidal with average variations of 4 m and maximum of 7 m, with maximum tidal currents exceeding 4 m/s (Rebelo-Mochel 1997). São Marcos and São José Bays are port areas that hold significant importance for maritime activities, trade and transportation within the region.

Figure 2.  

Map showing the Golfão Maranhense region, an estuary in the eastern Amazon (Brazil). This is where the secondary biodiversity data was retrieved.

Step 1: Search- systematic review

Step 2: Validate- data quality and control

Step 3: Standardise

Step 4: Share and archive

Statistical analysis

Step 1: Search- Systematic review

Considering all biotic groups, a total of 161 bibliographical references, including papers and technical reports were included in the systematic review of the literature (Fig. 3). In addition, we included species occurrence records from four public repositories (GBIF, VertNet, Wikiaves, SpeciesLink). Considering only published papers, the group included in the largest number of published papers and reports was plants (n = 59) and the group with the least data sources was benthos (n = 11) (Suppl. material 3 “Preferred Reporting Items for Systematic Reviews and Meta-Analyses – PRISMA”, separated by groups).

Figure 3.  

Flowchart of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) for all groups showing the process of selecting studies throughout systematic review. The selection process includes three stages: (1) identifying the database and choosing the papers; (2) scanning the references and selecting the papers to be included; (3) including the selected papers.

Step 2: Validate- data quality and control

A total of 2,070 occurrence events were obtained from bibliographic references and 43,947 were obtained by public repositories (n = 46,017) from 3,871 taxa. These include birds (Aves, 458 species; three other taxonomic level), amphibians (Amphibia, 55 species; nine to the genus level), reptiles (two Crocodylia; 86 Squamata; 11 Testudines); mammals (Class Mammalia; 101 species; 21 to the genus level), fish (268 species, 74 other taxonomic levels), phytoplankton (370 species; 105 other taxonomic levels), benthos (188 species; 204 other taxonomic levels) and plants (1,624 species; 292 other taxonomic levels) (Suppl. material 4). Most of the taxa were identified to species (81%) and genus (14%) level (Fig. 4). Benthos accounted for the highest number of occurrence events, with 12,510 records and reptiles had the lowest number of occurrence events recorded (570).

Figure 4.  

Proportion of each taxonomic level identified for each biological group in the secondary data recovered from the Golfão Maranhense area.

Data were carefully analysed by specialists in each group to check for inconsistencies in identification, spelling and, as much as possible, potentiality of identification correctness (e.g. check if the geographic locations were within expected known geographic distribution for each taxon, checking vouchers when possible). A total of 93 occurrence events were deleted, including 92 from taxa that were not correctly identified (76 birds and 16 mammals) and one bird specimen that was a victim of animal trafficking.

Step 3: Standardise

Step 4: Share and archive

Statistical analysis

The number of occurrences was dependent on the number of taxa in each sampled group (R² = 0.47, p = 0.03). While amphibians and non-bird reptiles were represented by low numbers of both taxa and occurrences, plants, birds and phytoplankton were highly represented for both occurrences and richness. On the other hand, the group “benthos” had a high number of occurrences and a low number of taxa (Fig. 5).

Figure 5.  

Relationships between numbers of taxa and occurrences gathered through an extensive review of secondary biodiversity data from the Golfão Maranhense area, in the estuarine regions of eastern Amazonia.

We proposed a workflow to improve our ability to recover higher quality biodiversity data using secondary data sources. We were able to extract a large amount of information about the biodiversity of the Golfão Maranhense and transform this unrelated data into organised and re-usable data. This systematic approach ensured data accuracy and reliability, facilitating the potential reuse of information in future studies. A further step that we have begun to take for some groups is the systematic survey of museum collections and analyses, focusing on relevant questions that we have identified along the way (e.g. general patterns of occurrence of migratory birds, sampling biases and gaps for many groups etc.).

Researchers can use existing datasets, such as those obtained through our biodiversity data retrieval method, to conduct a wide range of studies to advance scientific research (Pernat et al. 2024). For example, secondary data can be used to conduct meta-analyses (e.g. Biggs et al. (2020)) for comparative studies across different geographic regions and time, to support ecological modelling of species distributions (e.g. Fletcher et al. (2019)), habitat preferences and potential impacts of environmental change (Bayraktarov et al. 2019) as long as they are used judiciously. However, finding high-quality secondary data can be challenging, as evidenced by a recent survey in which most researchers reported that data finding can be arduous (73%) or difficult (19%) (Gregory et al. 2020). Several initiatives have been launched to collect, standardise, store and make biodiversity data openly available (Costello et al. 2013). For example, international repositories, such as “GBIF” (GBIF 2024) and “Freshwater Biodiversity Data Portal- BioFresh” (http://data.freshwaterbiodiversity.eu/) (for other repositories, see Culina et al. (2018)).

Long-term monitoring datasets can help to understand patterns and changes in ecological variables over time (Costa and Magnusson 2010, Magnusson et al. 2021). This can help identify ecological shifts and potential drivers of biodiversity change (e.g. Marques et al. (2022)). We are currently using our database to increase our knowledge of mammal, bird, fish, amphibian, reptile, marine phytoplankton and benthic species in the Golfão Maranhense. We are also studying the patterns and determinants of floristic variation in the region and the temporal variation of migratory birds in the São Marcos Bay region. While data collected in standardised monitoring programmes, such as LTER (Vanderbilt et al. 2015), can be directly linked to FAIR's standardised data repositories, other secondary data that are not standardised may be important to rescue, treat and use, as they can be thoroughly revised and curated beforehand and stored with some rule of error estimation to build robust hypotheses to investigate and understand biodiversity patterns.

While secondary data can be a valuable resource for scientific research, it is crucial to recognise and address its limitations and ideally estimate the errors within. Common challenges include species identification accuracy, geographic coordinate precision and data entry errors. In addition, datasets from different studies may differ in their sampling methods, data structure and definitions of key variables, making direct comparisons difficult. Finally, some datasets may not be openly accessible, which has implications for data availability and complicates data access and sharing policies.

Other limitations are the sampling and temporal biases, which can arise when working with secondary data, making data interpretation more challenging. Sampling bias occurs when the data sampling disproportionately favours certain species or areas over others, for example, the concentration of specimen records in more easily accessible sites, such as major cities, roads and navigable rivers (Boakes et al. 2010). In addition, logistics and human interference are factors that can explain research probability (e.g. 64% of research probability in Amazon; Carvalho et al. (2023)). Temporal bias, on the other hand, refers to the uneven distribution of data across time periods. Secondary data sources may include data collected over different time spans, reflecting historical variations in research focus, funding availability or changes in data-recording practices. Consequently, certain time periods may be over-represented, while others may be sparsely covered or entirely absent. Additionally, the difficulty of conducting research in regions with limited accessibility introduces challenges that restrict the ability to gather data from remote areas. Thus, remote regions potentially hosting unique biodiversity hotspots are often under-represented or completely absent from the dataset.

In our study, sampling bias is evident in the São Marcos Bay area, where an industrial ship port is located. Thus, most of the data were obtained from environmental monitoring reports in the region linked to the environmental licensing process. These reports conducted in a port area inherently prioritise certain species and ecological aspects more relevant to the licensing process, overlooking other important components of biodiversity. Within our database, it becomes apparent that some species records originate from technical reports that are not easily available. For example, we found 365 species and varieties of phytoplankton in technical reports, but 101 were not previously catalogued on the Brazilian Biodiversity Platform REFLORA (Jardim Botânico do Rio de Janeiro 2024) for the Maranhão region. This underscores the fact that the retrieval of biodiversity data can yield enhancements in the comprehension of species composition existing within the defined geographical area.

The workflow that we employed has facilitated the retrieval of biodiversity data from the ecologically rich and megadiverse Golfão Maranhense region in Maranhão, Brazil. By combining a systematic review approach with standardised worksheets with a Darwin Core base, we were able to effectively search and explore a wide range of scientific articles, technical reports and specialised public repositories. The potential use of secondary data for the advancement of scientific research is significant although it must be taken with care and analysed with precautions observing all bias limitation and filters involved. Many technical survey reports were produced in the Golfão Maranhense linked to environmental licensing process for the port and surrounding activities. By using existing datasets, researchers can carry out a wide range of activities which include meta-analyses, comparative studies, ecological modelling and, most of all, building hypotheses and producing experiment designs to monitor diversity in a standardised base. Our study highlights the value of systematic review methods and the need for an approach to address data limitations and biases. Likewise, this method can facilitate collaboration amongst researchers, enable comparative analyses across different datasets and support evidence-based conservation strategies and policy-making.

We would like to thank the Environmental Management of the Ponta da Madeira Maritime Terminal for their support in developing the project. We are grateful to the reviewer Pedro Cardoso for his suggestions for improving the manuscript and to the collectors of the data we retrieved from the literature. Daniel M. Casali is currently being funded by the grant #2022/00044-7, São Paulo Research Foundation (FAPESP)

Vale Institute of Technology