• species inventory
• species diversity
• spatial and temporal uncertainties
• sampling optimisation
• occurrence records
• NDVI
• models
• GIS
• floristic richness prediction
• flora
• biodiversity patterns
• algorithms
• Species-Area Relationship

Floristic inventories are an essential part of basic and applied researches in botany. Despite a long history in floristic investigations, they are still conducted following a purposive approach, without substantial improvements in recent decades. Accordingly, final outputs are affected by several biases which affect the reliability and the interpretation of results, and the possibility to perform statistical analyses. To the best of my knowledge, there are very few contributions exploring quantitative and standardised approaches from a floristic point of view, and my PhD work intends to push forward the knowledge in this direction.
More specifically, my aims are: (i) to explore new uses of the Species-Area Relationship applied to the floristic research (Chapter 1), (ii) to elaborate new tools of floristic cartography and methods to objectively manage floristic knowledge available for a given area (Chapter 2), and (iii) to design and test new sampling algorithms aimed to optimise inventory efficiency, i.e. to maximise species detection reducing sampling effort, and at the same time ensuring objectivity and reproducibility of methods (Chapter 3).
The focus of Chapter 1 is to unravel patterns of floristic richness in Tuscany, using a new modeling approach based on the Species-Area Relationship (SAR), namely the increase in the species number with the increase of the area. Residuals in a SAR model, i.e. divergences among observed and predicted values, reflect the actual floristic richness after the removal of the area effect. However, residual values are not able to elucidate, per se, the contribution of environmental variables in causing their divergence from predicted values. To overcome this limitation, I applied to the study area a spatially explicit modeling technique, (i) to quantify how each environmental variable affects SAR residuals, and (ii) to improve species richness predictions, by adjusting the SAR model according to local environmental features. Firstly, I collated 67 floras published after 1970 across Tuscany. For each flora, I extracted the area extent and the number of total, alien, and native taxa inventoried. Then, I georeferenced the study area boundaries to sample a plethora of environmental variables. These variables were used as predictors in a Generalised Linear Model having the SAR residual as dependent variable and the spatial autocorrelation removed. Results show that the area alone explains 87% of the variance of total species richness (86% for native and 56% for alien species, respectively); the total species richness expected for 1 km2 is 303.4 taxa, 12.0 of which are estimated to be aliens. The model for all species, adjusted by environmental predictors, allows to explain more than 47% of deviance of SAR’s residuals. The predictors of species richness at Tuscan level are ‘insularity’, ‘topographic heterogeneity’, ‘spatial heterogeneity of temperature annual range’, and ‘annual precipitation’. This approach pushes forward the floristic research on both theoretical and practical grounds. Firstly, it allows an innovative use of floristic data to quantitatively assess patterns of species richness. Secondly, it represents an operative tool to make adjusted a priori species number predictions for a given territory.
The Chapter 2 is based on the quantitative management of the ‘floristic ignorance’, defined here as a composite condition of complete lack of data, few data, and/or data with high uncertainties. In a floristic perspective, knowledge is essentially based on occurrence records. The availability of huge amount of records as available in the epoch of ‘big data’ poses new challenges for reliable analyses and correct interpretation of results. Indeed, to safely deal with occurrence records, we must consider their uncertainty, which can introduce biases within analyses. I developed an objective framework coded in R programming language, to explicitly include spatial and temporal uncertainties during the mapping and listing of plant occurrence records for a given study area. My workflow returns a ‘Map of Floristic Ignorance’ (MFI), which represents the spatial distribution of floristic ignorance across a study area, and a ‘Virtual Floristic List’ (VFL), i.e. a list of taxa potentially occurring in that area, showing a probability of occurrence for each taxon. Uncertainty cannot be avoided, but it may be incorporated into biodiversity analyses through appropriate methodological approaches and innovative spatial representations. This contribution introduces a workflow which pushes forward the analytical capacities to deal with uncertainty in biological occurrence records, allowing to produce more reliable outputs. From a practical point of view, both MFI and VFL are useful tools to plan and carry out field sampling activities, allowing (i) to detect areas needing additional sampling effort, and (ii) to draft a list of taxa already recorded for a given study site, providing a probability of occurrence for each taxon.
The main focus of the Chapter 3 is to explore the drafting of a floristic inventory by means of probabilistic approaches, based on geostatistical designs. I planned, carried out, and then compared two different sampling strategies: (i) a stratified random sampling design based solely on a coarse environmental stratification followed by a spatial optimization criterion (‘basic strategy’, no prior information is available), and (ii) a sampling design based on the maximisation of the spectral heterogeneity among sampling units, quantified in terms of Normalized Difference Vegetation Index values (‘advanced strategy’). The strategy that maximises collected floristic information was assessed basing on a combination of descriptive and quantitative statistics, such as the completeness of the floristic inventory, the steepness of the rarefaction curves, and the plots contributions to the total β diversity. The advanced strategy detected more taxa than the basic strategy in all the sampling sites investigated. In addition, the rarefaction curve obtained with advanced strategy is steeper in accumulating taxa respect to the basic strategy. The analysis of the contribution of each plot to the total β diversity showed that the advanced strategy selects sampling units having a more homogeneously distributed contribution among plots (i.e. higher complementarity among plots), and that they are better spatially arranged across the study area to capture environmental peculiarities of sampling sites. Accordingly, the advanced strategy is more effective than the basic one in drafting a species inventory, in the face of just a little more effort in the design of the sampling strategy. The algorithm to perform the advanced strategy – to the best of my knowledge proposed here for the first time – can be profitably and freely applied to every geographic area, showing flexibility for several ecological and vegetational contexts.
As conclusion of my work, all single parts developed during the three years are integrated and summarised in a flowchart, finalised in a protocol to draft a flora in an objective way.