Complex species distribution models of Goldcrests and Firecrests densities in Poland: are remote sensing‐based predictors sufficient?

Abstract

Species distribution models should identify ecological requirements of species and predict their spatial density. However, data from remote sensing sources are often used alone as predictors in modelling distributions. Such data will only produce accurate models if features that are distinguishable by remote sensing are a good match to the environmental traits that delineate habitat requirements. Both the Goldcrest Regulus regulus and the Firecrest Regulus ignicapilla respond to complex features of habitats that are not described by simple remote sensing data. We tested the usefulness of remote sensing data as a predictor for two Regulus species according to data from 970 study plots sized 1 · 1 km. Predictors were aggregated using the PCAs and related to the Hayne estimator of species density using GAMs. The models based on both remote sensing data and detailed environmental data proved to be better than the model based only on remote sensing data and/or detailed forest structure data. The Goldcrest reached the highest density in areas with a high share of old spruce-dominated forests with a substantial share of the fir, avoiding the pine, and it preferred forests with a low number of tree species. In turn, the Firecrest favoured old forests, dominated by the spruce and the beech, with an admixture of single old fir and larch trees, avoiding the pine, and preferring forests with a high number of tree species. We suggest using not only free data sources, but also more detailed data containing thorough information on forest inventory derived from ground measurements.

Linking species interactions with phylogenetic and functional distance in European bird assemblages at broad spatial scales

“…Indeed, detailed data of environmental conditions might attenuate the coefficients we found. In contrast, it has been shown that even when modelling forest bird species distributions with very detailed forest structure data, the density of a bird species remains a significant predictor of the density of a close relative (Kosicki, Stachura, Ostrowska, & Rybska, ).…”

Are Landscape Configuration Metrics Worth Including When Predicting Specialist and Generalist Bird Species Density? A Case of the Generalised Additive Model Approach

“…The species density in grid cells was expressed as the Hayne estimator, calculated according to the equation below [6,[31][32][33][34]:…”

“…A study conducted on a small spatial scale suggested that habitat specialist species with narrow niches were more likely to be found in the core of patches with a simple shape, while generalist species tended to occur in the buffer zone of a more complicated shape [49,50]. Therefore, SDMs that use only the area of patches may not provide complete information on species' specific habitat preferences [7,24,33]. Indeed, this study confirmed and generalised the pattern on a large spatial scale, because when landscape configuration metrics were used as additional predictors, models turned out to be definitely better than models without them.…”

“…It shows that an inter-seasonal variation of green vegetation also plays an important role. A low level of green vegetation variation is featured by both coniferous forest (except extreme situations green vegetation is constant throughout a year, [33]) and intensively used farmland wherein repeatability of green vegetation conditions is provided by multi-annual fertilisation [46]. On the other hand, a high level of green vegetation variation is characteristic to pastures, especially located near water bodies, and deciduous forest where NDVI depends on floods and the rate of leaf development in a given year.…”

“…The Generalised Additive Model method used in this study is often used as a tool for predictive spatial distribution modelling [6,32,33,35,41]. However, despite the advantages of GAMs described by [41], this procedure-being a tool for creating a predictive spatial density pattern-requires evaluation based on an independent dataset [52].…”