The ongoing biodiversity crisis presents a complex challenge for ecological science. Despite a consensus on general biodiversity decline, identifying clear trends remains difficult due to variability in data, methodologies, and scales of analysis. To enhance our understanding of ongoing biodiversity changes and address discrepancies in biodiversity trend detection, we propose integrating macroecological theory with temporal and trait-based perspectives.

• First, analyzing temporal changes in diversity scaling relationships, such as species accumulation curves or distance decay, can reconcile and synthesize conflicting observations of biodiversity change, enabling quantification of diversity shifts from local to regional spatial scales.
• Second, diversity patterns across scales are linked to three proximate components: abundance, evenness, and spatial aggregation of species. Investigating temporal changes in these components provides deeper insights into how human activities directly influence biodiversity trends.
• Third, incorporating species traits into the analysis of these macroecological patterns improves our understanding of human impacts on biodiversity by elucidating the links between species characteristics and their responses to environmental changes.

We illustrate this integration in a case study of forest and farmland birds in France, highlighting how studying diversity changes across scale, and decomposing temporal change in different components, can help to elucidate the mechanisms driving diversity change.

We discuss the limitations and challenges of this integrative approach and highlight how it offers a comprehensive framework for understanding the drivers of biodiversity change across scales. This framework facilitates a more nuanced understanding of how human activities impact biodiversity, ultimately paving the way for more informed actions to mitigate biodiversity loss across spatial and temporal scales.

Bird community data were from the French Breeding Bird Survey. The French breeding bird survey was designed to monitor population dynamics of common passerine bird species in France. In this survey, skilled volunteer ornithologists count birds at a given site, following a standardized protocol, at the same site, year after year (Jiguet et al. 2012). Species abundances are recorded across 2792 sites, each covering a 4km² area. Volunteers provide their home locality to the national coordinator, and a 2×2 km site is randomly selected from within a 10 km radius (out of 80 possible sites) by the coordinator. Each spring, volunteers carry out 10 point counts separated by at least 300 m within the selected site, for a fixed period of five minutes. Two sampling sessions are carried out from 1 April to 8 May, and then from 9 May to the end of June, to detect both early and late breeders, with a gap of 4–6 weeks between sessions. Counts are repeated annually on approximately the same date (±7 days) and at dawn (1–4 h after sunrise) by the same observer, in the same order. The highest count from these two sessions is used as the measure of point-level species abundance. We sub-selected sites that were monitored between 2002 and 2013, in order to avoid the first year of the restructured monitoring scheme (i.e 2001), and to limit our analyses to linear trends, more likely to characterize decadal dynamics. We used the latest PECBMS classification (https://pecbms.info/) to classify farmland and forest species according to their pre- dominant habitat.

Climatic data were extracted from CHELSA (https://chelsa-climate.org/, v.2.1) for each site and each sampling year. We computed the average daily temperature and precipitation during the bird breeding season (April - August). Land cover data were extracted from CORINE Land Cover (European Environment Agency 2010). Percentage land covers within FBBS site were computed by taking the habitat class area (in square meters) and dividing it by the total area of the site. Because CLC data were available only for 2000, 2006, 2012 and 2018, some FBBS site-year combinations were not covered by the dataset. In this case, we attributed site land cover for the uncovered year to the last year for which we had CLC data available (for example, sites monitored in 2001 were attributed land cover from CLC 2000). More specifically, we focused on two aggregated CLC classes, agricultural areas and forests.