Ecological Indicators 91 (2018) 241–248

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Original Articles

KnowBR: An application to map the geographical variation of survey effort T

and identify well-surveyed areas from biodiversity databases
⁎
Jorge M. Loboa, , Joaquín Hortala, José Luís Yelab, Andrés Millánc, David Sánchez-Fernándezd,
Emilio García-Rosellóe, Jacinto González-Dacostae, Juergen Heinee, Luís González-Vilasf,
Castor Guisandef
a
Department of Biogeography and Global Change, Museo Nacional de Ciencias Naturales, CSIC, Madrid, Spain
b
Facultad de Ciencias Ambientales y Bioquímica, Universidad de Castilla-La Mancha, Campus Tecnológico de la Fábrica de Armas, 4507 Toledo, Spain
c
Departamento de Ecología e Hidrología, Universidad de Murcia, Campus de Espinardo, 30100 Murcia, Spain
d
Instituto de Ciencias Ambientales (ICAM), Universidad de Castilla-La Mancha, Campus Tecnológico de la Fábrica de Armas, 45071 Toledo, Spain
e
Departamento de Informática, Edificio Fundición, Universidad de Vigo, Campus Lagoas-Marcosende, 36310 Vigo, Spain
f
Facultad de Ciencias de Mar, Universidad de Vigo, Lagoas-Marcosende, 36200 Vigo, Spain

A R T I C LE I N FO A B S T R A C T

Keywords: Biodiversity databases are typically incomplete and biased. We identify their three main limitations for char-
Spatial bias acterizing the geographic distributions of species: unknown levels of survey effort, unknown absences of a
Data limitations species from a region, and unknown level of repeated occurrence of a species in different samples collected at the
Database records same location. These limitations hinder our ability to distinguish between the actual absence of a species at a
Geographic distribution
given location and its (erroneous) apparent absence as consequence of inadequate surveys. Good practice in
Survey completeness
Wallacean shortfall
biodiversity research requires knowledge of the number, location and degree of completeness of relatively well-
surveyed inventories within territorial units. We herein present KnowBR, an application designed to simulta-
neously estimate the completeness of species inventories across an unlimited number of spatial units and dif-
ferent geographical extents, resolutions and unit expanses from any biodiversity database. We use the number of
database records gathered in a territorial unit as a surrogate of survey effort, assuming that such number cor-
relates positively with the probability of recording a species within such area. Consequently, KnowBR uses a
“record-by-species” matrix to estimate the relationship between the accumulated number of species and the
number of database records to characterize the degree of completeness of the surveys. The final slope of the
species accumulation curves and completeness percentages are used to discriminate and map well-surveyed
territorial units according to user criteria. The capacity and possibilities of KnowBR are demonstrated through
two examples derived from data of varying geographic extent and numbers of records. Further, we identify the
main advances that would improve the current functionality of KnowBR.

1. Introduction on the geographical distribution of biodiversity is limited and, often,

inaccurate (Rocchini et al., 2011; Ladle and Hortal, 2013), so our
Current development of information technology and biodiversity knowledge on species distributions is typically incomplete (the so-
informatics allows storing, retrieving, sharing, filtering and manip- called Wallacean shortfall; Lomolino, 2004; Whittaker et al., 2005).
ulating massive datasets such as those on species distributions (Bisby, Consequently, rather than providing accurate descriptions of species
2000; Godfray, 2002; Soberón and Peterson, 2004; Graham et al., 2004; geographic ranges, the extant databases are typically characterized by
Guralnick et al., 2007). Global initiatives such as the Global Biodi- incompleteness and biases (e.g., Dennis and Hardy, 1999; Soberón
versity Information Facility (GBIF) provide support for these big data et al., 2000; Zaniewski et al., 2002; Anderson, 2003; Martínez-Meyer,
(Saarenmaa and Nielsen, 2002) that can provide critical information for 2005; Dennis et al., 2006; Lobo et al., 2007; Hortal et al., 2008; Stropp
large-scale environmental questions (Hampton et al., 2013). However, et al., 2016).
even these comprehensively compiled databases suffer from a number Three limitations of the information from biodiversity databases are
of problems and shortfalls (Hortal et al., 2015). In fact, available data particularly important when characterizing the geographic

⁎
Corresponding author at: Departamento de Biogeografía y Cambio Global, Museo Nacional de Ciencias Naturales (MNCN-CSIC), c/José Gutiérrez Abascal 2, 28006 Madrid, Spain.
E-mail address: [email protected] (J.M. Lobo).

https://doi.org/10.1016/j.ecolind.2018.03.077
Received 18 September 2017; Received in revised form 26 March 2018; Accepted 26 March 2018
1470-160X/ © 2018 Elsevier Ltd. All rights reserved.
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

distributions of species: are insufficiently consistent (Hortal and Lobo, 2005; Ladle and Hortal,
2013; Hortal et al., 2015; Ruete, 2015; Meyer et al., 2015; Meyer et al.,
1. Unknown survey effort, a lack of knowledge of the effort devoted to 2016).
survey each territorial unit that is due to most occurrence records The effects of uneven levels of sampling effort have been tradi-
lacking any associated measure of the effort carried out to obtain tionally addressed through species richness estimators and species ac-
them. cumulation curves (Soberón and Llorente, 1993; Colwell and
2. Unknown absences, as almost all the available information involves Coddington, 1994; Hortal and Lobo, 2005). This is done under the as-
only species occurrences (i.e., the localities in which a species has sumption that they allow comparing the values of species richness and
been collected), without any indication of the likelihood that a other aspects of biodiversity between sites surveyed with different le-
species is actually absent from the localities where it was not col- vels of effort. Indeed, Chao and Jost (2012) and Colwell et al. (2012)
lected (whether these have been surveyed or not). recently demonstrated that it is more appropriate to compare estimated
3. Unknown recurrence, which results from the incomplete compilation species richness values between sites showing similar rates of species
of species occurrences in many biodiversity databases, as multiple accumulation with survey effort than between sites surveyed with the
records of the same species in the same site or territorial unit are same intensity. That is, estimates can be reliably compared when the
considered redundant and not reported (Hortal et al., 2007). This slopes of the relationships between observed number of species and the
prevents teasing apart occasional records from the continued pre- amount of survey effort are similar (i.e., standardizing by survey cov-
sence of the species in an area. erage sensu Chao and Jost, 2012). This implies that estimating survey
coverage is crucial when we aim to identify those locations with
These three limitations are mutually interrelated, so only when all probable reliable inventories.
known occurrences are comprehensively compiled it is possible to es- Despite the widely recognized importance of evaluating data quality
timate survey effort with some reliability, thereby helping to differ- and completeness as a preliminary step in any biodiversity study, this
entiate the absence of evidence from the evidence of absence. process is often neglected. Arguably, this is in part because such eva-
Therefore, a biodiversity database that compiles exhaustively all luation process is highly time-consuming, it requires the use of several
available information on the identity and distribution of a group of software applications and/or R packages, and repeating the same pro-
species would enable both identifying well-surveyed areas (e.g. Hortal cess for each one of the territorial units or sites considered (or, in
and Lobo, 2005) and obtaining estimates of the repeated occurrence general, for any type of spatial unit). Here we present KnowBR, a freely
and/or the probability of absence of particular species (e.g. Guillera- available R package to estimate the survey completeness of species
Arroita et al., 2010). inventories across an unlimited number of territorial units or sites si-
An important consequence of data limitations for biogeographical multaneously. Starting with any biodiversity database, KnowBR calcu-
and conservation analyses is the impossibility of distinguishing whether lates the survey coverage per spatial unit as the final slope of the re-
the apparent lack of occurrence of a target species in a given location lationship between the number of collected species and the number of
reflects its actual absence or is the result of insufficient survey effort. As database records, which is used as a surrogate of the survey effort.
a result, maps of observed species richness are often suspiciously similar KnowBR calculate the accumulation curve in each spatial unit according
to maps of the number of records per territorial unit (Hortal et al., to the exact estimator of Ugland et al. (2003) (default estimator), as well
2007). Species Distribution Models (SDMs) are commonly used to offset performing 200 permutations of the observed data (random estimator)
such data incompleteness. Briefly, SDMs relate the available occurrence to obtain a smoothed accumulation curve that is subsequently adjusted
data with a number of environmental variables (often via sophisticated to four different asymptotic accumulation functions. These functions
modelling techniques). The model created during this training phase is allow to obtain a completeness percentage (the percentage representing
then projected into the geographical space to predict the probable, al- the observed number of species against the predicted one) that also may
beit unknown, distribution of species (Guisan and Zimmermann, 2000). be used to estimate the territorial units with probable complete in-
Such predicted distribution, whether potential or realized, is often ventories.
larger than the range documented by occurrence data (Soberón and With KnowBR we aim to provide a tool to assess the levels of survey
Nakamura, 2009). Most SDM techniques rely on absence data to limit completeness across a territory, rather than an application for com-
the geographical response of the species, so they are particularly sen- paring species richness between sites by the use of the analytic rar-
sitive to the unknown absences limitation. However, common usage of efaction and extrapolation techniques developed by Chao and Jost
SDMs promotes an almost-universal use of random pseudo-absences (2012) and Colwell et al. (2012). KnowBR therefore estimates the de-
(a.k.a. background absences) to include absences into the training data gree of completeness of the inventories of all the territorial units within
used to derive the predictive function. This practice comes from the a given territory and, through that, allows identifying those spatial
classic procedure followed in Resource-Selection Functions (Johnson, units that can be considered well surveyed (herein, WSsus) at a given
1980). Use of background absence data is, however, inadequate for resolution and extent, according to the information gathered in any
estimating the probability of occurrence of a species (Hastie and biodiversity database. KnowBR allows performing all these time-con-
Fithiam, 2013), because it only reflects the intensity of the collection suming analyses in a very simple way, and simultaneously for a large
process that led to the data used to train the model (Aarts et al., 2012). number of spatial units both regular (cell option) and irregular (polygon
Hence, complex SDM algorithms calibrated with data containing option).
background absences yield poor and inconsistent predictions, a fact that
often passes unnoticed due to the use of inadequate evaluation methods 2. Installation and data entry
(Hijmans, 2012).
Employing statistical shortcuts on data with unknown levels of error KnowBR can be used as a regular R add-on package in both Linux
and bias can generate unreliable results. Consequently, good practice in and Mac OS by installing the file KnowBR.tar.gz (package source), as
biodiversity informatics requires knowledge about the number, location well as in RGUI for Windows by installing the file KnowBR.zip
and degree of completeness of surveys for those territorial units that (Windows binaries). Both files are available on CRAN (Development
have been, at least relatively, well inventoried. Such knowledge would Core Team R, 2016) and also at the web site http://www.ipez.es/
facilitate identifying localities where the lack of records for a target RWizard, in the download section. However, KnowBR can also be used
species can be reliably assumed to correspond to its actual absence. as a regular application as a plug-in of RWizard, an easy-to-use gra-
Nonetheless, it can be used to guide the location of future surveys and/ phical user interface for the R environment (Guisande et al., 2014).
or determine uncertain or ignorance areas in which biodiversity data RWizard is an open-source interface under GNU General Public License

242
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

Table 1
Format of the two CSV files that can be used as input data in KnowBR. In the two cases the matrices reflect the occurrence of a species in different localities. Note that
the two first records correspond to the same locality and species because they represent different sampling events (e.g. collections made on different dates and/or by
different collectors). The same value (count = 1) is given to each record independently of the number of collected specimens. Each database record is thus a pool of
specimens from a single species with identical information in different fields as location, altitude, date of capture, type of habitat, food resource or collector.
A B

Species Longitude Latitude Counts Longitude Latitude spp1 spp2 spp3

spp1 −4.694 34.800 1 −4.694 34.800 1

spp1 −4.641 34.741 1 −4.694 34.800 1
spp1 −2.244 34.893 1 −2.244 34.893 1
spp2 −1.443 35.098 1 −1.443 35.098 1
spp3 −4.713 36.140 1 −4.713 36.140 1
spp3 −3.224 38.675 1 −3.224 38.675 1
spp3 −3.213 38.787 1 −3.213 38.787 1

that has been developed in C# on the Net platform. It is designed as an by a “records-by-species” matrix containing only four values: the name
interface to facilitate the interaction with R (see video demonstration at of the species, the longitude and latitude of each record in decimal
http://www.ipez.es/RWizard). The only requirement for the installa- degrees, and a count value (see Type A format in Table 1 and the de-
tion of RWizard is that R and Net Framework 4.0 must be already in- scription of data in the KnowBR PDF manual). The data can be also
stalled, and then the file Setup.RWizard.V2.3.exe must be installed. included in KnowBR as a CSV file in which each column represents the
Although KnowBR can be used without RWizard, this interface increases occurrence of a species and each file is a database record (see Type B
the ease of use. KnowBR can be installed into RWizard from http:// format in Table 1 and the description of data in the KnowBR PDF
www.ipez.es/RWizard, following the menu “download” → “RWizard manual, available at http://www.ipez.es/RWizard/ and https://CRAN.
applications” → “Install KnowBR”. This way, KnowBR will be added as R-project.org/package=KnowBR).
another RWizard applications (Guisande et al., 2015; Guisande,
2016a,b,c). 3. Identifying well surveyed territories
The primary data matrix used in KnowBR must be derived from an
exhaustive database including all available georeferenced information. KnowBR aggregate the occurrence data in the spatial units (cells or
This implies that all the records of the study group have been gathered polygons). In the case of cells, a grid is built considering the resolution
without discarding apparently redundant data resulting from the re- selected by the user and the minimum and maximum latitude and
peated occurrences of any target species in the same locality (see Hortal longitude of the database records. Subsequently, all the occurrences
et al., 2007). Such redundancies can be divided among alternate da- included within each cell were discriminated for subsequent analyses,
tabase records that despite pertaining to the same species and having being each cell identified by its central latitude and longitude. In the
been gathered in the same place, differ in any other collection condition case of polygons, the algorithm used the function in.out of the package
(i.e. date of capture, food source, collector, type of microhabitat, etc.). mgcv (Wood, 2018) in order to find the database records belonging to
Here, any difference in the value of any database field yields a new each polygon.
record, regardless of the sex and/or number of individuals captured (see KnowBR uses the species accumulation curve that describes the re-
e.g., Lobo and Martín-Piera, 2002). lationship between the accumulated number of species and the surro-
Most biodiversity data can originate from heterogeneous sources gate of survey effort (database records) to characterize the rate of in-
with different collecting methodologies. Because of this and due to the crease with survey effort in each spatial unit (Clench, 1979; Soberón
inability to obtain a universal sampling effort measure, KnowBR uses and Llorente, 1993; Hortal and Lobo, 2005) and to determine WSsus.
the number of database records as a surrogate of sampling effort (Hortal This accumulation curve is established in KnowBR both analytically
and Lobo, 2005; Soberón et al., 2007; Lobo, 2008). This practice as- (exact estimator) and by randomization (random estimator). The
sumes that the probability of recording a species as occurring in a given equation of Ugland et al. (2003) is used in the first case (default option)
territorial unit correlates positively with the number of database re- as provided in the specaccum function of the vegan R package (Oksanen
cords gathered for that unit. This assumption is not always true. For et al., 2014). Alternatively, the user may perform 200 random permu-
example, the accumulation of database records in a locality can be the tations of the original data with replacement thus smoothing the species
consequence of surveys focused on the collection of one or a few species accumulation curve in order to avoid potential spurious effects re-
of particular interest within a large biological group (e.g., a diverse sulting from the order of addition of the records. The user must take
genus), rather than being due to the even survey of all the species from into account that the random procedure may generate slightly different
that group inhabiting the site. High numbers of database records final slopes depending on the selected random samples. The com-
gathered in such a biased fashion could falsely indicate that a particular pleteness of each territorial unit can be decided based on the final slope
territorial unit is well-surveyed for all of the species in the taxon of of these analytical or smoothed curves (the slope between the two last
interest. In light of this potential problem, it is advisable to assess the steps of the so-generated species accumulation curves). This final slope
degree of coverage of the original surveys, to ascertaining whether (and is thus calculated by KnowBR for each spatial unit. This final slope can
where) they were directed to collect all the potential species from the be considered a measure of the degree of completeness of the inventory
focal taxon or, on the contrary, the collection bias could lead to falsely in each spatial unit (the final rate of increase in the number of species
identifying certain unevenly sampled territorial units as well-surveyed. attained with the survey effort developed so far; see Hortal and Lobo,
KnowBR allows implementing a standard approach for assessing the 2005). Previously, the user has to decide the minimum ratio between
completeness of biodiversity data to represent the geographical dis- the number of database records and the number of species necessary to
tribution of the species diversity of any given biological group. This proceed with the calculations. A particular slope cutoff value can be
approach is particularly appropriate for poorly surveyed groups and/or discretionarily chosen by the user to decide which spatial units are well-
regions lacking sufficient information to correct the unequal sampling sampled enough so as to be flagged as WSsus (e.g. Hortal and Lobo,
efforts resulting from otherwise standardized survey protocols. When 2005).
this rationale is applied, the typical “species-by-sites” matrix is replaced In addition, the obtained species accumulation curves are adjusted

243
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

to four different species-accumulation functions with three or less et al., 2013, 2014) to download and clean all GBIF data available for
parameters: the Michaelis-Menten equation used by Clench (1979) Apoidea (downloaded 4 July 2014). In total, 137,809 records for 5,836
(Clench option), a negative exponential model (N-exponential option), a species names (≈30% of total species) have georeferenced distribu-
classic sigmoid saturation model (saturation option) and the rational tional data in GBIF. By using ModestR we exported all these data di-
function (rational option) which is used as default (see Soberón and rectly to the format required in KnowBR at a resolution of one-degree
Llorente, 1993; Flather, 1996; Mora et al., 2008). The extrapolated grid cells. We then calculated the final slope of the species accumula-
asymptotic values of all these curves can be used, according to user's tion curve with the random estimator for all the considered cells in
selection, to estimate the probable number of species in each spatial 67 min and 53 s for a computer with 8 Gb RAM and 3.40 GHz 2-nuclei
unit when the number of records will tend to infinite, calculating sub- processor. The results (Fig. 1) show that only c.18% of all terrestrial
sequently the completeness (i.e. the percentage of species that has been world one-degree cells (n = 2,776) have georeferenced data for bees,
inventoried). Again, a completeness value chosen by the user can be and that only 9% of all Apoidea species have more than 10 records.
selected to decide the WSsus. Both final slopes and completeness values Approximately 14% of the cells with georeferenced data (i.e. 386 cells)
can be calculated simultaneously for an unlimited number of territorial have twice the number of records than recorded species, only 0.8%
units across the geographical extent defined by the user. These terri- (128 cells) have a final slope less than or equal to 0.1 (i.e. one new
torial units can be regular cells of any resolution (cell option) but also species added to the inventory each 10 database records), and only
irregular polygons (polygon option) according to user preferences. 0.5% (74 cells) have a final slope lower than or equal to 0.01 (i.e. one
RWizard include in the “Area” argument the possibility of select ad- new species per 100 records) (Fig. 1). These results exemplify the
ministrative spatial units (countries and/or provinces) or rivers basins general scarceness, and large degree of bias, in the georeferenced in-
of different levels in which to perform the calculations. Instead of using formation available among databases dealing with the biodiversity of
the polygons available in RWizard, the user may also include any many groups, even as economically important as bees. Notably, in this
shapefile containing the desired irregular polygons (e.g. protected case survey effort has been distributed in an extremely uneven fashion,
areas, countries, etc) by means of the “shape” argument. For further with most records gathered in western North America and central and
details, see PDF manual of KnowBR_1.5 (https://CRAN.R-project.org/ northern Europe, while Australia is the region harboring a compara-
package=KnowBR). tively higher number of well-surveyed cells (Fig. 1).
Classical non-parametric estimators and extrapolated richness va- The second example of the utility of KnowBR for assessing the
lues available in other packages as vegan (Oksanen et al., 2014) are not completeness of distributional information at a regional extent is based
used here for discriminating WSsus. When the data come from hetero- on the use of BANDASCA (Lobo and Martín-Piera, 2002; Hortal and
geneous biodiversity data the use of these methods is undesirable. In all Lobo, 2011). This database currently contains 15,142 records for the 54
the formerly mentioned calculations each database record is treated as species of the dung beetle family Scarabaeidae (Coleoptera) on the
a sampling unit or independent observation. This means that for each Iberian Peninsula and adjoining islands. In this example, final slopes for
record only one species is present. However, in the usual incidence and four grid cells of varying spatial resolutions (5, 15, 30 and 60 min) were
abundance based procedures (Colwell et al., 2012), the number of calculated in 1 min and 20 s. This regional database illustrates some of
species in each observation is a random variable (i.e., cannot not be the caveats and difficulties of applying this kind approach to detect
fixed to be one in advance). According to Robert K. Colwell and Anne areas with reliable inventories at varying scales. Changing the spatial
Chao (personal communication), the use of non-parametric estimators resolution in the analysis notably affects the proportion of cells that can
is undesirable when the data is structured in the format of database be considered to be well-surveyed (Fig. 2; see also the green broken line
records. in Fig. 3A). Only 2% of all Iberian cells (138 cells) can be considered
In order to propose key metrics able to identify WSsus the new well-surveyed at a 5-min resolution, with the consequent reduction in
function SurveyQ (Survey Quality) was included in the package KnowBR the probability of reflecting the whole spectrum of environmental
to identify and plot the well-sampled, moderated-sampled and poorly- conditions present in the Iberian Peninsula, compared to larger cell
sampled cells or polygons. Stated briefly (see PDF manual of sizes. Further, the resolution of the analyses can also affect our capacity
KnowBR_1.5 for further details), using the file called “Estimators” ob- to reflect the “true” faunistic composition of the obtained inventories.
tained from KnowBR which includes the final slope values of the ac- Some well-surveyed cells detected at low resolution (60') contain a
cumulation curves, the obtained completeness values and the ratio large number of well-surveyed cells at the highest resolution (5'),
between the number of records and the observed species (R/S ratio), showing that their well-surveyed inventories at the “regional” scale are
SurveyQ provides a map with the location of well-sampled, moderated- the consequence of a number of complete “local” inventories (Fig. 3B).
sampled and poorly-sampled spatial units. SurveyQ also provides a 3D In some extreme cases, however, a 60' cell would be considered well-
graph showing the distribution of these three parameters in all the surveyed even when none of the 5' cells it contains presents complete
considered spatial units (cells or polygons). The default values used to inventories. This probably arises because many heterogeneous and in-
distinguish well-sampled, moderated-sampled and poorly-sampled cells complete “local” inventories may be capable of providing an, appar-
or polygons are: slope < 0.02, completeness > 90% and R/S ently, reliable “regional” species inventory. It follows that a low
ratio > 15 for well-sampled spatial units, and slope > 0.3, com- number of reliable “local” inventories can give rise high completeness
pleteness < 50% and R/S ratio < 3 for poorly-sampled spatial units. values at coarse resolutions. We suspect that the apparent completeness
In the case of polygons, a plot representing the values of the three of these “regional” inventories may actually underrepresent their local
parameters is depicted by using polar coordinates (Van Sickle, 2010). variability, at least in environmentally heterogeneous areas. Thus,
carrying out an exhaustive survey effort at a locality may imply that
4. Two practical examples larger territorial units containing this locality also appear as well-
sampled, whereas their internal heterogeneity has not been adequately
The first example is the assessment of the completeness of world- surveyed. The results provided by this example illustrate the need for
wide distributional data on bees. Taxonomy is based on the World Bee caution when comparing estimates of inventory completeness carried
Checklist Project, downloaded from the webpage of the Integrated out at different resolutions. It is therefore advisable to conduct ex-
Taxonomic Information System (www.itis.gov; last accessed in ploratory analyses at varying spatial scales to determine whether this
December 2016). This checklist includes 19,508 species names of kind of effect may be hampering the completeness of the inventories in
world-wide Apoidea (Insecta, Hymenoptera) (Ascher and Pickering, coarse territorial units, at least in highly heterogeneous regions. What is
2014), one of the most important insect groups from an applied point of the most appropriate resolution in each case? We do not believe there is
view. Subsequently, we used the software ModestR (see García-Roselló an easy answer to this question. Be that as it may, the used resolution

244
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

Fig. 1. (A) Number of database records included in GBIF for the Apoidea of the World (Ascher and Pickering, 2014) in one-degree grid cells. (B) Grid cells with final
slopes of the relationship between the accumulated number of species and the number of database-records between 0 and 0.1. These cells are preliminarily qualified
as well-sampled (see main text). (C) Final slopes of accumulated richness relationships for all world countries.

should be the one closest to the average home range of the considered distribution of biodiversity, and to study the survey process in detail. To
species”. increase the current functionality of KnowBR, future R applications
should incorporate tools: i) to identify the spatial units that would be
5. Future prospects most appropriate for surveying to maximize the spatial and environ-
mental coverage provided by the set of well-surveyed territorial units
KnowBR has the potential to become a standard tool not only to (such as Medina et al., 2013); ii) to identify reliable absences for any
assess survey completeness, but also to provide fair estimates of the focal species and model species distributions using techniques that take

245
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

Fig. 2. Geographical distribution of the well-surveyed grid cells with reliable inventories for dung beetles (Coleoptera, Scarabaeidae) on the Iberian Peninsula and
adjoining islands at multiple spatial resolutions. Here, well-surveyed grid cells are defined as those with completeness values ≥ 80% of the value predicted by the
exact estimator (see text). The maps depict the results for four grid cell resolutions, 60, 30, 15 and 5 min, which correspond approximately to UTM cells of 14,400,
3,600, 900 and 100 km2.

advantage of both occurrence and probabilistic absence data (Lobo available, the results of the models may extrapolate to unknown con-
et al., 2010); iii) to calculate the degree of uncertainty associated to the ditions or areas (Austin and Heyligers, 1989; Ferrier, 2002; Ferrier
results of these models in poorly surveyed areas, based on their degree et al., 2002a,b), thereby diminishing the accuracy and reliability of
of completeness and the distance to well-surveyed areas (i.e. maps of model projections (Kadmon et al., 2004, Hortal and Lobo, 2011). To
ignorance sensu Rocchini et al., 2011); and iv) to describe the biases in avoid these misleading extrapolations, it is critical to provide the tools
the spatial distribution of sampling effort, and explore the factors be- necessary to calculate the environmental coverage provided by the
hind these biases. data, to identify the territorial units that would likely result in better
These improvements are aimed to allow developing the protocol for coverage once surveyed, and to generate predictions of species dis-
mapping biodiversity attributes described in Hortal et al. (2007) as the tributions that take into account the uncertainty in their projections to
concatenated use of several KnowBR modules. If important gaps in areas and domains with less environmental and spatial coverage.
species distribution information become apparent once well-surveyed In sum, KnowBR aims to improve the reliability of the results of
territories have been identified, additional surveys will be required. biodiversity informatics applications, such as species distribution
These surveys should focus on covering as much of the spatial and models. Good predictions require good biological data. The use of in-
environmental variability of the studied territory as possible while formation from sites with inventories close to complete facilitates
minimizing resource expenditure and efforts (Ferrier, 2002; Hortal and generating reliable predictions, and can also be used to design more
Lobo, 2005; Rocchini et al., 2011). Once well-surveyed areas provide efficient surveys that optimize data coverage for the analysis of biodi-
enough coverage of the region, the future incorporation of species versity patterns. Good applications of biodiversity data will come only
distribution modeling tools in KnowBR will allow filling in the gaps in from assessing the reliability of data and accounting for its actual
the known distribution of species without resorting to the impractical quality and accuracy. Of course, KnowBR does not allow overcoming
task of exhaustively surveying the entire region. A practical con- the intrinsic limitations of the used data due to natural dynamics of
sequence of the use of data from well-surveyed territories is that these ecological systems or survey difficulties. Rare, vagrants or difficult to
models can provide reliable interpolations when these units represent survey species as well as collection bias may generate inadequate in-
the full range of environmental and spatial variation across the region ventories (Dennis et al., 2006; Lobo et al., 2007; Hortal et al., 2008)
of interest. On the contrary, when insufficient well surveyed units are that may add a supplementary uncertainty to the discrimination of well

246
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

20q?guide=Apoidea_species.
Austin, M.P., Heyligers, P.C., 1989. Vegetation survey design for conservation: gradsect
sampling of forests in north-eastern New South Wales. Biol. Conserv. 50, 13–32.
Bisby, F.A., 2000. The quiet revolution: biodiversity informatics and the internet. Science
289, 2309–2312.
Chao, A., Jost, L., 2012. Coverage-based rarefaction and extrapolation: standardizing
samples by completeness rather than size. Ecology 93, 2533–2547.
Clench, H., 1979. How to make regional lists of butterflies: some thoughts. J. Lepid. Soc.
33, 216–231.
Colwell, R.K., Coddington, J.A., 1994. Estimating terrestrial biodiversity through extra-
polation. Philos. Trans. R. Soc. B 345, 101–118.
Colwell, R.K., et al., 2012. Models and estimations linking individual-based and sample-
based rarefaction, extrapolation and comparison of assemblages. J. Plant Ecol. 5,
3–21.
Dennis, R.L.H., Hardy, P.B., 1999. Targeting squares for survey: predicting species rich-
ness and incidence for a butterfly atlas. Glob. Ecol. Biogeogr. 8, 443–454.
Dennis, R.L.H., et al., 2006. The effects of visual apparency on bias in butterfly recording
and monitoring. Biol. Conserv. 128, 486–492.
Development Core Team R, 2016. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna.
Ferrier, S., 2002. Mapping spatial pattern in biodiversity for regional conservation
planning: where to from here? Syst. Biol. 51, 331–363.
Ferrier, S., et al., 2002a. Extended statistical approaches to modelling spatial pattern in
biodiversity in northeast New South Wales. I. Species-level modelling. Biodivers.
Conserv. 11, 2275–2307.
Ferrier, S., et al., 2002b. Extended statistical approaches to modelling spatial pattern in
biodiversity in northeast New South Wales. II. Community-level modelling. Biodivers.
Conserv. 11, 2309–2338.
Flather, C.H., 1996. Fitting species-accumulation functions and assessing regional land
use impacts on avian diversity. J. Biogeogr. 23, 155–168.
García-Roselló, E., et al., 2013. ModestR: a software tool for managing and analyzing
species distribution map databases. Ecography 36, 102–1207.
García-Roselló, E., et al., 2014. Using ModestR to download, import and clean species
distribution records. Meth. Ecol. Evol. 5, 708–713.
Godfray, C., 2002. Challenges for taxonomy. Nature 417, 17–19.
Graham, C.H., et al., 2004. New developments in museum-based informatics and appli-
cations in biodiversity analysis. Trends Ecol. Evol. 19, 497–503.
Guillera-Arroita, G., et al., 2010. Design of occupancy studies with imperfect detection.
Meth. Ecol. Evol. 1, 131–139.
Guisande, C., et al., 2014. RWizard software, http://www.ipez.es/RWizard. University of
Vigo. Spain.
Fig. 3. (A) Variation in the number of grid cells (in logarithms) with reliable Guisande, C., et al., 2015. FactorsR: an RWizard application for identifying the most likely
inventories for dung beetles in the Iberian Peninsula and adjoining islands causal factors in controlling species richness. Diversity 7, 385–396.
(continuous lines and triangles) and their percentages (broken lines and circles) Guisande, C., 2016a. Niche estimation. R package version. 1:3 Available at < http://
CRAN.Rproject.org/package=EnvNicheR > .
with regard to the total number of terrestrial grid cells (black continuous line).
Guisande, C., et al., 2016b. SPEDInstabR: an algorithm based on a fluctuation index for
The numbers and percentages of the cells with at least one database record are selecting predictors in species distribution modeling. Ecol. Inform. 37, 18–23.
represented in red, those cells with twice as many records as species are re- Guisande, C., 2016c. An algorithm for morphometric characters selection and statistical
presented in blue, and cells with completeness values ≥ 80% according to the validation in morphological taxonomy. R package version. 1:1 Available at < http://
exact estimator of the accumulation curves are represented in green. (B) CRAN.R-project.org/package=VARSEDIG > .
Guisan, A., Zimmermann, N.E., 2000. Predictive habitat distribution models in ecology.
Comparison of well-surveyed cells at 60 (grey filled circles) and 5 min (red dots)
Ecol. Model. 135, 147–186.
resolution. Guralnick, R.P., et al., 2007. Towards a collaborative, global infrastructure for biodi-
versity assessment. Ecol. Lett. 10, 663–672.
Hampton, S.E., et al., 2013. Big data and the future of ecology. Front. Ecol. Environ. 11,
surveyed spatial units. If data weaknesses and shortfalls are known in 156–162.
advance, and the analyses take the associated limitations into account Hastie, T., Fithian, W., 2013. Inference from presence-only data; the ongoing controversy.
in a balanced approach, the conclusions obtained will increase in ro- Ecography 36, 864–867.
Hijmans, R.J., 2012. Cross-validation of species distribution models: removing spatial
bustness and confidence. This kind of conclusions will go much farther sorting bias and calibration with a null model. Ecology 93, 679–688.
in countering skeptics about the extent of the biodiversity crisis and/or Hortal, J., et al., 2008. Historical bias in biodiversity inventories affects the observed
the impacts of global change on biodiversity. realized niche of the species. Oikos 117, 847–858.
Hortal, J., Lobo, J.M., 2005. An ED-based protocol for the optimal sampling of biodi-
versity. Biodivers. Conserv. 14, 2913–2947.
Acknowledgements Hortal, J., Lobo, J.M., 2011. Can species richness patterns be interpolated from a limited
number of well-known areas? Mapping diversity using GLM and kriging. Braz. J. Nat.
Conserv. 9, 200–207.
The development of KnowBR has been supported by the projects Hortal, J., et al., 2007. Limitations of biodiversity databases: case study on seed-plant
BIOWEB (CGL2011-15622-E BOS) and BANDENCO (POII11-0277- diversity in Tenerife (Canary Islands). Conserv. Biol. 21, 853–863.
5747) founded by Spanish Ministerio de Ciencia e Innovación and the Hortal, J., et al., 2015. Seven shortfalls that beset large-scale knowledge of biodiversity.
Ann. Rev. Ecol. Evol. Syst. 46, 523–549.
Consejería de Educación, Ciencia y Cultura-Junta de Comunidades de Johnson, D.H., 1980. The comparison of usage and availability measurements for eval-
Castilla-La Mancha, respectively. DSF was supported by a post-doctoral uating resource preference. Ecology 61, 65–71.
contract funded by Universidad de Castilla-La Mancha and the Kadmon, R., et al., 2004. Effect of roadside bias on the accuracy of predictive maps
produced by bioclimatic models. Ecol. Appl. 14, 401–413.
European Social Fund (ESF).
Ladle, R., Hortal, J., 2013. Mapping species distributions: living with uncertainty. Front.
Biogeogr. 5, 8–9.
References Lobo, J.M., 2008. Database records as a surrogate for sampling effort provide higher
species richness estimations. Biodivers. Conserv. 17, 873–881.
Lobo, J.M., Martín-Piera, F., 2002. Searching for a predictive model for Iberian dung
Aarts, G., et al., 2012. Comparative interpretation of count, presence–absence and point beetle species richness based on spatial and environmental variables. Conserv. Biol.
methods for species distribution models. Meth. Ecol. Evol. 3, 177–187. 16, 158–173.
Anderson, R.P., 2003. Real vs. artefactual absences in species distributions: tests for Lobo, J.M., et al., 2007. How does the knowledge about the spatial distribution of Iberian
Oryzomys albigularis (Rodentia: Muridae) in Venezuela. J. Biogeogr. 30, 591–605. dung beetle species accumulate over time? Divers. Distrib. 13, 772–780.
Ascher, J.S., Pickering, J., 2014. Discover Life bee species guide and world checklist Lobo, J.M., et al., 2010. The uncertain nature of absences and their importance in species
(Hymenoptera: Apoidea: Anthophila). Available at http://www.discoverlife.org/mp/ distribution modelling. Ecography 33, 103–114.

247
J.M. Lobo et al. Ecological Indicators 91 (2018) 241–248

Lomolino, M.V., 2004. Conservation biogeography. In: Lomolino, M.V., Heaney, L.R. prediction of species richness. Conserv. Biol. 7, 480–488.
(Eds.), Frontiers of Biogeography: New Directions in the Geography of Nature. Soberón, J., Nakamura, M., 2009. Niches and distributional areas: concepts, methods, and
Sinauer Associates, pp. 293–296. assumptions. Proc. Natl. Acad. Sci. U.S.A. 106, 19644–19650.
Martínez-Meyer, E., 2005. Climate change and biodiversity: some considerations in Soberón, J., Peterson, A.T., 2004. Biodiversity informatics: managing and applying pri-
forecasting shifts in species potential distributions. Biodivers. Inform. 2, 42–55. mary biodiversity data. Philos. Trans. R. Soc. Lond. 359, 689–698.
Medina, N., et al., 2013. Designing bryophyte surveys for an optimal coverage of diversity Soberón, J.M., et al., 2000. The use of specimen-label databases for conservation pur-
gradients. Biodivers. Conserv. 22, 3121–3139. poses: an example using Mexican papilionid and pierid butterflies. Biodivers.
Meyer, C., et al., 2015. Global priorities for an effective information basis of biodiversity Conserv. 9, 1441–1466.
distributions. Nat. Comm. 6 8221. Soberón, J., et al., 2007. Assessing completeness of biodiversity databases at different
Meyer, C., et al., 2016. Multidimensional biases, gaps and uncertainties in global plant spatial scales. Ecography 30, 152–160.
occurrence information. Ecol. Lett. 19, 992–1006. Stropp, J., et al., 2016. Mapping ignorance: 300 years of collecting flowering plants in
Mora, C., et al., 2008. The completeness of taxonomic inventories for describing the Africa. Glob. Ecol. Biogeogr. 25, 1085–1096.
global diversity and distribution of marine fishes. Proc. R. Soc. B 275, 149–155. Ugland, K.I., et al., 2003. The species-accumulation curve and estimation of species
Oksanen, J., et al., 2014. Community Ecology Package. ─ R package version 2.0-10. richness. J. Anim. Ecol. 72, 888–897.
Available at: http://CRAN.R-project.org/package=vegan. Van Sickle, J., 2010. Basic GIS Coordinates. CRC Press, Boca Raton, Florida.
Ruete, A., 2015. Displaying bias in sampling effort of data accessed from biodiversity Whittaker, R.J., et al., 2005. Conservation biogeography: assessment and prospect.
databases using ignorance maps. Biodivers. Data J. 3, e5361. Divers. Distrib. 11, 3–23.
Rocchini, D., et al., 2011. Accounting for uncertainty when mapping species distributions: Wood, S., 2018. Mixed GAM Computation Vehicle with Automatic Smoothness
the need for maps of ignorance. Prog. Phys. Geogr. 35, 211–226. Estimation. R package version. 1.8-23 Available at < https://cran.r-project.org/web/
Saarenmaa, H., Nielsen, E.S., 2002. Towards a global biological information infra- packages/mgcv/index.html > .
structure. Challenges, opportunities, synergies, and the role of entomology. European Zaniewski, A.E., et al., 2002. Predicting species spatial distributions using presence-only
Environment Agency, Copenhagen. data: a case study of native New Zealand ferns. Ecol. Model. 157, 261–280.
Soberón, J., Llorente, B.J., 1993. The use of species accumulation functions for the

248