Estimation of functional diversity and species traits from ecological monitoring data

Abstract

The twin crises of climate change and biodiversity loss define a strong need for functional diversity monitoring. While the availability of high-quality ecological monitoring data is increasing, the quantification of functional diversity so far requires the identification of species traits, for which data are harder to obtain. However, the traits that are relevant for the ecological function of a species also shape its performance in the environment and hence, should be reflected indirectly in its spatiotemporal distribution. Thus, it may be possible to reconstruct these traits from a sufficiently extensive monitoring dataset. Here, we use diffusion maps, a deterministic and de facto parameter-free analysis method, to reconstruct a proxy representation of the species' traits directly from monitoring data and use it to estimate functional diversity. We demonstrate this approach with both simulated data and real-world phytoplankton monitoring data from the Baltic Sea. We anticipate that wider application of this approach to existing data could greatly advance the analysis of changes in functional biodiversity.

Keywords: data science; diffusion map; ecological monitoring; functional diversity; phytoplankton.

Fig. 1.

Functional diversity estimation from a monitoring dataset. (A) We use data on the biomass of 516 phytoplankton species in 730 samples collected from 1993 until 2015 from 10 stations in the Baltic Sea (27) (2 species are shown for illustration). (B) We then compute the pairwise similarity between species from the correlation between species abundances over the set of samples. (C) From the similarities, the i-trait space of the species (dots) is constructed using diffusion maps. In this space, the pairwise functional dissimilarity is quantified by the diffusion distance d_ij. (D) Once the distances between species have been determined, the diversity in a specific sample can be quantified by applying the Rao’s index to the species present (highlighted dots). In the sample shown, most of the biomass is concentrated in a small area of trait space, leading to a comparatively low Rao index.

Fig. 2.

Numerical validation of the proposed method. (A) We numerically generate randomly distributed traits of 200 species (colored dots) that fill a triangle in a trait space spanned by three resource requirement ($R^{*}$) parameters. Color indicates the resource ratio preferred by a species. (B) The inferred i-trait space generated by diffusion mapping simulated biomass data. The reconstruction identifies traits that span a space that is qualitatively similar to the ground-truth traits. Colors are the same as in A, illustrating that neighborhood relationships are mostly reconstructed correctly. (C) Rao’s functional diversity calculated from diffusion distances in the reconstructed trait space correlates strongly with the numerical ground truth (based on $R^{*}$ values). Indicated are local diversity (blue dots) in individual patches and regional diversity in the metacommunity (yellow dots). The R² value for regional diversity is 0.92, relative to a cubic regression (green line). These results show that the proposed method can be used to identify traits and robustly estimate functional diversity based on monitoring data.

Fig. 3.

Inferred traits from the monitoring dataset. Shown are species (dots) projected onto the space spanned by the most important i-traits. Color coded are environmental conditions under which the species were observed with high relative abundance (see text). (A) The first i-trait aligns well with NO${}_3^{-}$ concentration separating species by their nitrogen requirements. The water temperature (B) and the day of the year (C) align with the second trait, separating the early from the late species. The PO${}_4^{-}$ concentration is closely aligned with the third reconstructed trait (D).

Fig. 4.

Phytoplankton diversity on the Lithuanian coast. We observe an increase in functional diversity over the measurement period at all of the 10 stations included in the dataset (circles). The station located at the exit of the Curonian Lagoon is marked with a black circle. The fastest increase (warmer colors) is found at some of the coastal stations. The coastal stations are also the most diverse in average (larger diameter).