Constructing a Nature Relationship Indicator by Latent Spatio-Temporal Process Modelling of Citizen Science Data

Sammanfattning

This thesis explores the use of birding citizen science data and statistical modelling methods to support the construction of indicators for human-nature relationships. Motivated by the broader aspirational framework of the Nature Relationship Index, the study investigates how voluntary bird observation records can be processed into spatial and temporal measures suitable for statistical analysis, from which a human-nature relationship indicator can potentially be derived. Observation events in Sweden from 2019 to 2025 are collected, filtered, spatially joined to a regular grid, and aggregated into yearly event counts, while accounting for administrative land boundaries and lake coverage. The resulting event-count dataset is then used to examine spatial and temporal patterns and to fit Bayesian spatio-temporal models with different combinations of covariates to adjust for biases in reporting frequencies. Using the R package inlabru, which applies Integrated Nested Laplace Approximation (INLA) to efficiently approximate marginal posterior distributions, several Bayesian spatio-temporal models are fitted and compared across different likelihood families, including Poisson, negative binomial, and zero-inflated negative binomial, and across different covariate combinations. Model comparison and checking are conducted using information criteria and posterior predictive diagnostics. The results from two models suggest areas of lower and higher level of human-nature relationship in Sweden. It also shows an increasing trend in the human-nature relationship indicator after 2020. Overall, the study provides a methodological basis for transforming the frequency of species reporting events from citizen science projects into a structured modelling workflow suitable for indicator development.