Recent reports on local extinctions of arthropod species1 and of massive declines in arthropod biomass 2 point to land-use intensification as a major driver of decreasing biodiversity. However, there are no multi-site time-series of arthropod occurrences across land-use intensity gradients to confirm causal relationships. Moreover, it remains unclear which land-use types and arthropod groups are affected and whether the observed declines in biomass and diversity are linked to one another and continue. Here we analyzed arthropod data on more than 1 million individuals and 2,700 species from standardized inventories from 2008 to 2017 at 150 grassland and 140 forest sites in three regions of Germany. Overall gamma diversity in grasslands and forests decreased over time indicating loss of species across sites and regions. In annually sampled grasslands, biomass, abundance and species number of arthropods declined by 67%, 78%, and 34%, respectively. The decline was consistent across trophic levels, mainly affected rare species, and its magnitude was independent of local land-use intensity. However, sites embedded in landscapes with higher cover of agricultural land showed a stronger temporal decline. In 30 forest sites with annual inventories, biomass and species number, but not abundance, decreased by 41% and 36%, respectively. This was supported by analyses of all forest sites sampled in 3year intervals. The decline affected rare and abundant species and trends differed across trophic levels. Our results show that there are widespread declines in arthropods that concern biomass, abundance and diversity across trophic levels. Declines in forests demonstrate that arthropod loss is not restricted to open habitats. Our results 4 suggest that major drivers of arthropod decline act at larger spatial scales, and are, at least for grasslands, associated with agriculture at the landscape level.This implies that land-use relevant policies need to address the landscape scale to mitigate negative effects of land-use practices. Main textMuch of the debate on the human-induced biodiversity crisis has focused on vertebrates 3 , yet population decline and extinctions may be even more substantial in small organisms such as terrestrial arthropods 4 . Recent studies report declines in biomass of flying insects 2 , diversity of insect pollinators 5,6 , butterflies and moths 1,7-10 , hemipterans 11,12 and beetles 7,13,14 . Owing to the associated negative effects on food webs 15 , ecosystem functioning and ecosystem services 16 , the insect loss has spurred an intense public debate. However, time-series data on arthropods are rather limited and studies have so far focused on a small range of taxa 11,13,14 , few land-use and habitat types 12 or even on single sites 1,17 . In addition, many studies lack species information 2 or high temporal resolution 2,12 . Hence, it remains unclear whether reported declines in arthropods are a general phenomenon that is driven by similar mechanisms across land-use types, taxa and functional groups.The ...

No abstract

Machine learning algorithms find frequent application in spatial prediction of biotic and abiotic environmental variables. However, the characteristics of spatial data, especially spatial autocorrelation, are widely ignored. We hypothesize that this is problematic and results in models that can reproduce training data but are unable to make spatial predictions beyond the locations of the training samples. We assume that not only spatial validation strategies but also spatial variable selection is essential for reliable spatial predictions.We introduce two case studies that use remote sensing to predict land cover and the leaf area index for the "Marburg Open Forest", an open research and education site of Marburg University, Germany. We use the machine learning algorithm Random Forests to train models using non-spatial and spatial cross-validation strategies to understand how spatial variable selection affects the predictions.Our findings confirm that spatial cross-validation is essential in preventing overoptimistic model performance. We further show that highly autocorrelated predictors (such as geolocation variables, e.g. latitude, longitude) can lead to considerable overfitting and result in models that can reproduce the training data but fail in making spatial predictions. The problem becomes apparent in the visual assessment of the spatial predictions that show clear artefacts that can be traced back to a misinterpretation of the spatially autocorrelated predictors by the algorithm. Spatial variable selection could automatically detect and remove such variables that lead to overfitting, resulting in reliable spatial prediction patterns and improved statistical spatial model performance.We conclude that in addition to spatial validation, a spatial variable selection must be considered in spatial predictions of ecological data to produce reliable predictions.

Studies of early human settlement in alpine environments provide insights into human physiological, genetic, and cultural adaptation potentials. Although Late and even Middle Pleistocene human presence has been recently documented on the Tibetan Plateau, little is known regarding the nature and context of early persistent human settlement in high elevations. Here, we report the earliest evidence of a prehistoric high-altitude residential site. Located in Africa’s largest alpine ecosystem, the repeated occupation of Fincha Habera rock shelter is dated to 47 to 31 thousand years ago. The available resources in cold and glaciated environments included the exploitation of an endemic rodent as a key food source, and this played a pivotal role in facilitating the occupation of this site by Late Pleistocene hunter-gatherers.

A new rainfall retrieval technique for determining rainfall rates in a continuous manner (day, twilight, and night) resulting in a 24-h estimation applicable to midlatitudes is presented. The approach is based on satellite-derived information on cloud-top height, cloud-top temperature, cloud phase, and cloud water path retrieved from Meteosat Second Generation (MSG) Spinning Enhanced Visible and Infrared Imager (SEVIRI) data and uses the random forests (RF) machine-learning algorithm. The technique is realized in three steps: (i) precipitating cloud areas are identified, (ii) the areas are separated into convective and advective-stratiform precipitating areas, and (iii) rainfall rates are assigned separately to the convective and advective-stratiform precipitating areas. Validation studies were carried out for each individual step as well as for the overall procedure using collocated ground-based radar data. Regarding each individual step, the models for rain area and convective precipitation detection produce good results. Both retrieval steps show a general tendency toward elevated prediction skill during summer months and daytime. The RF models for rainfall-rate assignment exhibit similar performance patterns, yet it is noteworthy how well the model is able to predict rainfall rates during nighttime and twilight. The performance of the overall procedure shows a very promising potential to estimate rainfall rates at high temporal and spatial resolutions in an automated manner. The near-real-time continuous applicability of the technique with acceptable prediction performances at 3–8-hourly intervals is particularly remarkable. This provides a very promising basis for future investigations into precipitation estimation based on machine-learning approaches and MSG SEVIRI data.