1.1 A global database of plant traits—TRY

In 2007, the TRY initiative (https://www.try-db.org) was launched, aiming at developing a global database of plant traits to support biodiversity research, functional biogeography, and modeling of vegetation dynamics. The TRY database initiative received strong support from the ecological community, who contributed many original and collective datasets and has led to multiple updates (Kattge, Díaz, et al., 2011). The current version of the TRY database (version 6), released in October 2022, is based on 696 datasets and contains 15.4 million trait records, accompanied by 43 million ancillary data records, for 2661 traits and 305,000 plant taxa, mostly at the species level. About 6.7 million trait records are georeferenced from about 48,000 measurement sites worldwide (Figure 1). In 2015, some TRY datasets became public, and since 2019 the data are open access under a Creative Commons (CC)-BY license by default (Kattge et al., 2020). As of today, the TRY initiative has served more than 30,000 data requests (Figure 1), releasing over 4.5 billion trait records in combination with 40 billion ancillary data records. The TRY database has thus become a central resource for the ecological community, allowing users from around the globe to retrieve plant trait data based on selected traits and species or request individual datasets via the data portal on the TRY website. Step-by-step instructions on how to register and request data from the TRY database can be found on the GitHub Wiki of ‘rtry’: https://github.com/MPI-BGC-Functional-Biogeography/rtry/wiki/The-TRY-database#request_rtry_data.

Through the data request process, users can navigate the intellectual property guidelines of the database, review the description of the requested traits and species, and ascertain the number of trait measurements before sending out the request. Once the request is approved, users have the option to retrieve the dataset from the portal whenever necessary. The data release notes provided with each data request (https://www.try-db.org/TryWeb/TRY_Data_Release_Notes.pdf) offer information on the generalities, data structure, column headers (Table 1) of the requested dataset, and the identifiers for some of the widely used ancillary data (‘DataID’). Additionally, users can access descriptions and corresponding identifiers of traits (‘TraitName’ and ‘TraitID’) and species (‘AccSpeciesName’ and ‘AccSpeciesID’) on the TRY data explorer (https://www.try-db.org/de/de.php). This information, particularly the identifiers, is invaluable for the data preprocessing tasks.

1.2 Structure of datasets released from TRY

Plant traits provide essential information about plant growth strategies and adaptations to their environment as constrained by genetic characteristics. As a consequence, individual trait values can be broadly explained by multiple interacting factors: macro-level genetics in a phylogenetic context (i.e., evolutionary adaptations), micro-level genetics (i.e., selection), trait–trait correlations, competition, and the abiotic and biotic environmental conditions at provenance (i.e., ontogeny), during growth, and at the time of measurement including phenology (Díaz et al., 2016; Garnier et al., 2017; Kattge, Díaz, et al., 2011; Kattge, Ogle, et al., 2011; Mungall et al., 2010; Violle et al., 2007). Not all of these dependencies are well studied, and their interacting effects on traits are, for most taxa, poorly characterized. For these reasons, the most useful trait data include ancillary data describing the conditions, i.e., under which the plants had grown and traits were measured. Thus, the data structure to represent trait data must include the relevant dependencies and allow for different types of ancillary data.

The structure of TRY data releases is based on the extensible observation ontology (OBOE; Madin et al., 2007) schema, implemented in a generic entity-attribute-value model (Kattge, Ogle, et al., 2011). The TRY database features a long-table structure of trait records and ancillary data, with 27 columns (version 6; Table 1). Different trait records and ancillary data measured on the same entity are linked by a unique identifier (‘ObservationID’; Figure 2). The TRY data release notes (https://www.try-db.org/TryWeb/TRY_Data_Release_Notes.pdf), distributed with each release from the TRY database, provide a more detailed overview of this data structure. Due to the size of the TRY database—15.4 million trait records and 43 million ancillary data—this can result in data releases of up to 58 million rows of trait records and ancillary data. In addition, different attributes within the released datasets are relevant for trait data filtering, i.e., trait names, species names, ancillary data, units, and identifiers for duplicates and outliers. Therefore, the process to obtain all relevant information for further analyses and discard all inconsistent data is not straightforward and there is a high risk that not all information provided for data selection is used to optimize data quality for the downstream analyses.

This paper provides an overview of the ‘rtry’ package and demonstrates its utility from a user perspective, underscoring its potential as a valuable resource for researchers grappling with the complexities of preprocessing plant trait data. By facilitating more efficient and reliable data preprocessing tasks, ‘rtry’ aims to enhance the quality of plant trait datasets for scientific inquiry.

3.1 Dataset import

However, it is impossible to do so for larger datasets, which leads to the next data preprocessing step—dataset exploration.

3.2 Dataset exploration

The second step of the data preprocessing workflow is the exploration of the dataset. Even though the TRY data release notes (https://www.try-db.org/TryWeb/TRY_Data_Release_Notes.pdf) provide an overview of the data structure and column headers (Table 1) of the requested dataset, they do not include the informational content of the trait records and ancillary data, which makes it challenging for preprocessing. The dataset exploration facilitated by the ‘rtry_explore’ function allows users to gain insights into the inherent traits, species, and ancillary data, enabling informed decisions and evaluation of the outcomes during preprocessing. Exploring the datasets proactively before and after each data combination or filtering step is recommended. This practice promotes data integrity and helps prevent the accidental exclusion of valuable data.

The ‘rtry_explore’ function takes four arguments—‘input’, ‘…’, ‘sortBy’, and ‘showOverview’—and organizes the input into a grouped data table based on the specified column names (‘…’). A column displaying the total count within each group is provided as additional information to the exploration. By default, the output is grouped by the first attribute when ‘sortBy’ is not specified.

Data exploration can also be used to obtain the species information for which data are available by including the column headers ‘AccSpeciesID’ and/or ‘AccSpeciesName’ within the argument ‘…’. However, users should be aware that an exploration on species, traits, and sub-traits simultaneously may result in a long list of results due to the potentially diverse dataset.

3.3 Data combination

Given the diverse origins of plant trait data, users frequently encounter the need to manage multiple datasets during preprocessing. To facilitate this, ‘rtry’ provides four data combination functions, namely ‘rtry_bind_col()’, ‘rtry_bind_row()’, ‘rtry_join_left()’, and ‘rtry_join_outer()’. The visual interpretation of these functions is shown in Figure 4.

3.4 Data filtering

A major goal of data preprocessing is data filtering. This functionality is especially crucial for datasets retrieved from the TRY database, as they often contain more information than necessary for user objectives and trait data inconsistent with planned analyses. To avoid incorporating substantial data filtering in their downstream analyses—which is possible but prone to errors and reduces computational efficiency—it is essential to extract (select) relevant information or remove (exclude) irrelevant information beforehand. ‘rtry’ offers six functions to facilitate this data filtering process: ‘rtry_select_col()’, ‘rtry_remove_col()’, ‘rtry_select_row()’, ‘rtry_exclude()’, ‘rtry_select_anc()’, and ‘rtry_remove_dup()’.

3.5 Long-table to wide-table transformation

Trait datasets can be structured in either long- or wide-table formats. The data released from TRY are given in a long-table format, which allows a consistent structure as different traits or ancillary data are stored in separated rows (i.e., simply add or remove rows when needed, instead of having empty columns for missing information). The long-table format keeps this type of data in a denser format and is more flexible for data storage. Yet, a wide-table format is often more convenient for analyses as a tabular view is more straightforward to visually interpret and assess. Therefore, the ‘rtry’ package provides the ‘rtry_trans_wider’ function to transform the preprocessed trait data from long- to wide-table format for further analyses. This function accepts five arguments—a data frame (‘input’), the columns from which the output column names and values are to be obtained (‘names_from’ and ‘values_from’), the optional argument to define the function applied to the output values when necessary (‘values_fn’), and whether to display the dimension of the resulting wide-table (‘showOverview’).

Several preprocessing steps are necessary before performing the long- to wide-table transformation on the TRY dataset. The first step is to select only traits with numerical values and relevant columns (else the attribute in ‘values_fn’ might cause error). Next, users can obtain a list of relevant ancillary data from the original dataset as needed, e.g., georeferencing information like latitude and longitude indicated by ‘DataID's 59 and 60, respectively. The ‘rtry’ package provides the ‘rtry_select_anc’ function to facilitate this step. The ‘rtry_select_anc’ function takes three arguments—an imported data frame (‘input’), a list of ‘DataID's of the ancillary data to be selected (‘…’), and the optional argument ‘showOverview’. This function returns a unique list of ‘ObservationID’ and the corresponding ancillary data of interest. When the ancillary data (latitude and longitude in this case) are extracted, they can be merged to the numerical traits using 'rtry_join_left()' to include the ancillary data in the resulting wide-table.

3.6 Dataset export

4.1 Geocoding and reverse geocoding

Georeferencing is necessary to assess the plausibility of location information, filter data using a common coordinate system, estimate geographic patterns, link to georeferenced—e.g., environmental—data, and address the spatial autocorrelation of the plant trait data.

There are two functions within ‘rtry’ to assist users with geocoding (‘rtry_geocoding()’ derives latitude and longitude for a given location name) and reverse geocoding (‘rtry_revgeocoding()’ derives the location name from provided latitude and longitude) for a list of locations or coordinates in the WGS84 Coordinate System. These functions rely on Nominatim, a search engine for OpenStreetMap (OSM) data. The data provided by the OSM are freely available for any purpose, including commercial use, and are governed by the Open Database License (ODbL; https://wiki.osmfoundation.org/wiki/Licence). Users should note that an absolute maximum of one request per second (no heavy usage) and a valid email address to identify the request are required when using the OSM service as part of the Nominatim Usage Policy (details can be found on: https://operations.osmfoundation.org/policies/nominatim/).

4.2 Preprocessing NEON plant foliar trait data

The National Ecological Observatory Network (NEON) program is a research platform funded by the United States National Science Foundation (NSF) that provides free and long-term data across biomes comprising the continental U.S. and Hawaii on key ecological metrics as a basis to discover and understand the impacts of climate change (NEON, 2023). We have chosen the plant foliar traits dataset (product ID: DP1.10026.001) from the NEON data portal (NEON, 2016) to demonstrate a use case of the ‘rtry’ package outside of plant trait data from TRY. The NEON plant foliar traits dataset contains trait measurements (leaf mass per area, leaf water content, chlorophyll, carbon and nitrogen concentrations and stable isotopes, major and minor elements, and lignin) of sun-lit canopy foliage at either individual (woody plants) or community (herbaceous plants) levels (NEON, 2016).

While the detailed example is available on the GitHub Wiki, this section provides an overview of the preprocessing steps using ‘rtry’ for NEON data (Figure 5). The objective is to demonstrate the versatility of the ‘rtry’ package beyond the TRY database and illustrate how users can seamlessly chain together various functions within the package to suit the needs of cross-cutting and integrative analyses.