Biodiversity data obsolescence and land uses changes

Study region

Navarra is a 10,391 km square region located in the north of Spain, wedged between the western end of the Pyrenean Range and the Ebro Basin (Fig. 1). Both Eurosiberian and Mediterranean bioregions occur in this territory. It gathers a high landscape diversity as a result of its wide variety of climatic and topographic conditions (Nogués-Bravo, 2006). The northern areas, with mean altitudes above 800 m (range 200 m–2,000 m), high precipitations (1,200–1,700 mm of mean annual rain and snowfall) and moderate temperatures, are covered by pastures and natural beech and oak forests as well as pine plantations. Agricultural and Mediterranean landscapes dominate the south of Navarra, with mean altitudes below 400 m and a combination of low precipitation (below 400 mm) and temperature that favour the presence of rich communities of xerophytic shrublands. All this variety of habitats determines the high diversity of the small mammals’ community in Navarra, which comprises 31 of the 34 species of native insectivorous and rodents cited in the Iberian Peninsula (Escala et al., 1997).

Biodiversity data

Data about small mammals for the region were extracted from the Pellet sampling dataset (Escribano et al., 2016) from the Vertebrate Collection of the Museum of Zoology of the University of Navarra. This dataset includes information of small mammals’ records obtained from the analysis of barn owl’s pellets and comprises 75% of the Collection in a time range spanning from 1967 to 2011. At this moment, the dataset constitutes the best comprehensive, publicly accessible collection of small mammals in Spain (Figs. 1 and 2A). As pellets have proved to be very useful for assessing the small mammals’ community in a territory (Torre, Arrizabalaga & Flaquer, 2004; Avenant, 2005), this dataset is a valuable tool for analysing the changes in distribution patterns of diversity in the last four decades.

In order to avoid overestimating the abundance of individuals, only records from skull remains where selected for analyses. Records from species rarely predated by barn owl or records lacking in coordinates and date of the sampling event were also discarded. In total, 59,677 records of 27 species of small mammals were used for the study (see Table S1).

Land use data

Land use data was obtained from the Geographic Information System of Navarra. For our purpose, we chose the land use maps for 1956, 1985, 2000 and 2012 covering the entire time span in our occurrence data. All maps are polygon-based layers that provide information about land cover types in Navarra.

We reclassified the land uses categories into five new categories attending to the structure of the vegetation cover: forest, bushes and meadows, herbaceous crops, woody crops and unproductive areas. In order to standardize the polygon’s information and allow the later comparison of the different maps, each polygon was summarized by a relative abundance matrix of the five new categories considered. In this way, homogeneous polygons were expressed as matrices with a value of 1 in the appropriate category and 0 in the rest, while mixed polygons showed the proportional values for each category (that together equal the unit). For further details on the standardization of the maps see Document S1.

Land use changes and biological data analysis

We used ArcGIS 10 software to perform spatial analyses (ESRI, 2015). We calculated the land uses changes for the three periods delimited by the reference maps (1956–1985, 1985–2000 and 2000–2012) by means of a spatial intersection of the corresponding maps. The three resultant maps of land use changes consisted on a series of polygons that either (1) had not suffered from changes in their land uses (there were no changes in their relative abundance matrices) or (2) had change from one land use to another (their relative abundance matrices were different). A 5 × 5 km grid was then superimposed to the study area, summarizing the percentage of land use changes per cell by dissolving all their polygons (see Document S1). The spatial resolution of the study (25 km sq) was set according to the assumed home range of the barn owl as the mammals’ records come from pellets of this bird. A radius of 3 km from the nest/roosting site (and then an ideal circular area of about 28 km sq) is accepted to be suitable to characterize the habitat of this predator in a territory (Martínez & Zuberogoitia, 2004). The small mammals’ records were also collapsed to the same grid. Cells with no recorded occurrences as well as those with an effective size below 12.5 km sq (mostly located in the border of the study area) were discarded.

In order to evaluate the records’ obsolescence, cells with more of the 25% of their surface affected by land use modifications were labelled as highly modified cell (HMC). The threshold was empirically established based on those cells with sampling series long enough to perform comparative analyses of community and territorial changes (see Document S1 for further details). Then, following the reasoning presented previously, samples collected before or during periods of land use changes were labelled as obsolete.

Temporal variation in data sources

Our analysis of the net changes in the land uses of Navarra showed that 27% of its surface has changed during the entire dataset’s timeframe (1956–2012). Overall, forestry land uses increased by 11% while meadows and shrublands decreased by 12%. Herbaceous and woody crops suffered a small decrease, and unproductive areas increased slightly. Relative to HMC, we found that they were greater in the periods 1956–1985 and 1985–2000, representing more than the 60% of cells. On the contrary, between 2000 and 2012 the number of HMCs decreased and there were only 14 HMC.

After discarding records from species rarely preyed upon by the barn owl and records that did not come from the study area, the dataset provided 59,677 georeferenced records of 27 species of small mammals. Despite the relatively large number of records in a relatively small area, little more than one third of the cells held data (36.02%, 156 cells). Also, the temporal distribution of records from 1966 to 2011 showed an evidently clumped pattern, with a main sampling period centred on the 80’s (Fig. 2A). This period corresponds with the years leading to the publication of a regional atlas about the insectivorous and rodents community (Escala et al., 1997). Although most of the sampling effort was made between 1987 and 1992 (more than 55% of the records come from those years), we can observe another period where records from pellets increased between 1980 and 1987. This inventorying was not large in terms of occurrences, but was remarkably uniform through time. After the publication of the atlas, addition of new records of small mammals decreased.

Regarding to the richness of the dataset, all the 27 species considered in this study were already present in the database by 1991 and richness had almost attained a plateau five years before (Fig. 2A). Even though samples collected after 1985 did not contribute significantly to the total richness of the group, the spatial representativeness of the dataset grew considerably (Fig. 2B). The sampled area duplicated from 1985 to 2000, adding 91 new cells, previously unsampled, and thus increasing the biogeographical information of most of the species.

Obsolescence analysis

A total of 115 of the 156 cells analysed (73.72%) contained obsolete data, that is, some of the samplings were performed before or during periods when the land use changed. If we exclude these outdated occurrences, the dataset information fells down to 14,528 records belonging to 24 species. Although cells holding these non-obsolete records have not suffered from considerable land use changes (<25% of its surface changed) ever since samplings occurred, more than one-half of these cells were not visited again over the last fifteen years (Fig. 3).

Species by species, the overall attrition rate (number of records filtered out by flagging them as obsolete) was generally high, over 70% in 74% of the species. The information loss was critical in species with very low representation in the original dataset, such as Crocidura suaveolens Pallas, 1811, Microtus cabrerae (Thomas, 1906) and Sorex araneus Linnaeus, 1758. Although an overall reduction in record numbers is expected to affect all species to a greater or lesser extent, these species actually disappeared from the dataset (Table 1).

Biodiversity records in a changing territory

The land use analysis showed a generalized increase in forested surface along the considered time, while meadows and shrublands show the opposite tendency. These two changes occurred mainly in the earliest period (from 1956 to 1985). The reduction of meadow land seemed related to an increase in forest cover due mainly to two managerial practices: (1) steady repopulation on bare soils with tree species as an erosion-avoiding strategy, and (2) phasing out of fuel in favour of other energy sources, which occurred mainly between 1970 and 1990 (Gobierno de Navarra, 2000; Nogués-Bravo, 2006). However, our study does not aim to understand how Navarra has shifted its landscape, but to record the main land use changes at the sites where small mammals were recorded. These changes might therefore compromise the information that could be obtained from the biodiversity datasets.

We found that more than half the record-bearing territory area contained records compromised by such changes, and thus rendering these records as potentially obsolete. After applying our criterion of filtering out PBR collected before or during land use changes, we discarded approximately 75% of the small mammals’ records (Fig. 4). As reported in several studies, changes in the management of the territory have led to changes in communities and older records may not reflect the current communities. For example, Torre et al. (2015) resampled historically sampling sites for barn owl pellets finding changes in the frequencies of small mammals and a decrease in the species richness from 14 to 10. Similarly, when we discarded data compromised by land use changes, richness in our ‘Pellet sampling’ dataset lost three of the rarest species. Furthermore, the number of cells in which each species was recorded was reduced in most cases by more than 70%. The loss of information was not concentrated at specific sites, but spread all over the studied territory.

Implications of data obsolescence

Obsolescence could become a critical factor for species whose distribution pattern remains unclear. For example, M. cabrerae is an endemism of the Iberian peninsula, but its distribution is still unclear as information about its presence is patchy (Garrido-García et al., 2013). This knowledge gap ushered this microtine into the UICN Red List as ‘Nearly Threatened’ (see Table S1) and catalogued as ‘Vulnerable’ in Spain (Palomo, Gisbert & Blanco, 2007). However, if the data records of this species are old enough and come from a territory where land use changes have occurred afterwards, we cannot longer guarantee that those records are contributing to a reliable, current picture of its actual distribution. Similarly, other species whose distribution is unknown will suffer from the same lack of confidence.

Obsolescence could also affect the outcomes of SDMs when the habitat descriptors do not match the data frame of the biodiversity records used to calibrate the model, thus giving a biased result. The main challenge is then to find a way of estimating how much this ageing in relation to a desired timeframe, e.g., the ‘now,’ may affect the outcome of the model. As PBR are central to the construction of these models, it is important to considerate this limitation as well as other types of biases.

On the contrary, by highlighting the obsolete information of a dataset we can identify which parts of the territory are not appropriately sampled according to the recent history of human interventions in the area. These areas can then be targeted for surveying. Due to the current budget limitations for inventorying and completing databases with new records, it is imperative to design suitable surveys (Sánchez-Fernández et al., 2011) and monitoring programs (Boero et al., 2015) that optimize the available resources into a significant increase of biodiversity information. Our results showed that records from abundant species are prone to a lower decrease whereas records from rare species show a greater rate of attrition. Together, these two observations can help guide sampling efforts towards more efficient campaigns that may close knowledge gaps (Ariño, Chavan & Otegui, 2016). It might thus be preferable, and accrue more information, to sample those cells were rare species are distributed, where records are old, and where changes in land use have occurred since the locality was last sampled. This would improve survey designs and efficient monitoring of the wildlife, returning maximum information while spending minimum sampling efforts.

However, as we have only considered just one factor affecting obsolescence of the data—land use changes—our results should be interpreted with caution. Other elements should be explored that might be likely playing a role in the obsolescence of data. For example, changes in land uses can affect species differently according to their life histories (De Palma et al., 2015) and, therefore, the biology of a species could likely be a factor affecting obsolescence.

In this study we have explore the obsolescence of biodiversity records related to the land uses changes. We have pointed out that these records can be outdated and might give us an out-of-time picture of species distribution. However, we have only scratched the surface and we still have to explore in depth the implications of obsolescence of data and whether the use of obsolete data can be misleading towards understanding current species distribution patterns.