Convolutional Neural Networks (CNNs) are statistical models suited for learning complex visual patterns. In the context of Species Distribution Models (SDM) and in line with predictions of landscape ecology and island biogeography, CNN could grasp how local landscape structure affects prediction of species occurrence in SDMs. The prediction can thus reflect the signatures of entangled ecological processes. Although previous machine-learning based SDMs can learn complex influences of environmental predictors, they cannot acknowledge the influence of environmental structure in local landscapes (hence denoted “punctual models”). In this study, we applied CNNs to a large dataset of plant occurrences in France (GBIF), on a large taxonomical scale, to predict ranked relative probability of species (by joint learning) to any geographical position. We examined the way local environmental landscapes improve prediction by performing alternative CNN models deprived of information on landscape heterogeneity and structure (“ablation experiments”). We found that the landscape structure around location crucially contributed to improve predictive performance of CNN-SDMs. CNN models can classify the predicted distributions of many species, as other joint modelling approaches, but they further prove efficient in identifying the influence of local environmental landscapes. CNN can then represent signatures of spatially structured environmental drivers. The prediction gain is noticeable for rare species, which open promising perspectives for biodiversity monitoring and conservation strategies. Therefore, the approach is of both theoretical and practical interest. We discuss the way to test hypotheses on the patterns learnt by CNN, which should be essential for further interpretation of the ecological processes at play.

The use of naturalist mobile applications have dramatically increased during last years, and provide huge amounts of accurately geolocated species presences records. Integrating this novel type of data in species distribution models (SDMs) raises specific methodological questions. Presence-only SDM methods require background points, which should be consistent with sampling effort across the environmental space to avoid bias. A standard approach is to use uniformly distributed background points (UB). When multiple species are sampled, another approach is to use a set of occurrences from a Target-Group of species as background points (TGOB). We here investigate estimation biases when applying TGOB and UB to opportunistic naturalist occurrences. We modelled species occurrences and observation process as a thinned Poisson point process, and express asymptotic likelihoods of UB and TGOB as a divergence between environmental densities, in order to characterize biases in species niche estimation. To illustrate our results, we simulated species occurrences with different types of niche (specialist/generalist, typical/marginal), sampling effort and TG species density. We conclude that none of the methods are immune to estimation bias, although the pitfalls are different: For UB, the niche estimate fits tends towards the product of niche and sampling densities. TGOB is unaffected by heterogeneous sampling effort, and even unbiased if the cumulated density of the TG species is constant. If it is concentrated, the estimate deviates from the range of TG density. The user must select the group of species to ensure that they are jointly abundant over the broadest environmental sub-area.

Early detection and rapid response are essential to deal effectively with new introductions of invasive alien species (IAS). Citizen science platforms for opportunistic recording of species are increasingly popular, and there is potential to harvest their data for early detection of IAS, but this has not been tested. We evaluated the potential of data from existing citizen science platforms for early detection of IAS by obtaining 687 first records of species from 30 European countries where there was both an official first record (i.e. published in scientific literature or by a government agency) and a record in a citizen science platform. We tested how the difference between the two (time lag) was related to species traits, popularity in citizen science platforms, public and research attention and regulatory status. We found that for 50% of the time lag records, citizen science platforms reported IAS earlier than or in the same year as the official databases. Although we cannot determine causality (the first official record could have been from a citizen science platform, or contemporaneous with it), this demonstrates that citizen science platforms are effective for IAS early detection. Time lags were largely affected by species traits. Compared with official records, vertebrates were more likely to have earlier records on citizen science platforms, than plants or invertebrates. Greater popularity of the IAS in citizen science platforms and its observation in neighbouring countries resulted in earlier citizen science reporting. In contrast, inclusion in the EU priority list resulted in earlier official recording, reflecting the efficacy of targeted surveillance programmes. However, time lags were not affected by the overall activity of citizen platforms per country. Synthesis and applications. Multi‐species citizen science platforms for reporting nature sightings are a valuable source of information on early detection of IAS even though they are not specifically designed for this purpose. We recommend that IAS surveillance programmes should be better connected with citizen science platforms, including greater acknowledgement of the role of citizen scientists and better data flow from smaller citizen science initiatives into global databases, to support efficient early detection. Read the free Plain Language Summary for this article on the Journal blog.

Building reliable species distribution models (SDMs) from presence‐only information requires a good understanding of the spatial variation in the sampling effort. However, in most cases, the sampling effort is unknown, leading to biases in SDMs. This study proposes a method to jointly estimate the parameters of sampling effort and species densities to avoid such biases. The method is particularly suited to the analysis of massive but highly heterogeneous presence‐only data. The proposed method is based on estimating the variation in sampling effort over units of a spatial mesh in parallel with the environmental density of multiple species using a marked Poisson process model. Based on simulations with realistic settings, we examined the performance and robustness of parameter estimations. We also analysed a large‐scale citizen science dataset with highly heterogeneous sampling (Pl@ntNet), including around 300,000 occurrences of 150 plant species. We found that sampling effort was correctly estimated when the true sampling effort was constant within the cells of a spatial mesh. Estimation bias arose when sampling effort and environmental drivers strongly covaried within cells. Otherwise, the inference was correct and robust to sampling variation within cells. Running the model on real occurrences of 150 plant species provided an estimated map of relative sampling effort for 15% of French territory. We also found that the density estimated for an exotic invasive plant was consistent with prior data. This is the first method jointly estimating species densities depending on environment, and sampling effort as an explicit spatial function, from occurrence data of multiple species. An asset of the method is that a few frequently observed species greatly contribute to correctly estimate sampling effort, thereby improving density estimation of all other species. This approach can thus provide reliable SDM for large opportunistic presence‐only datasets, with broad spatial variation in sampling effort but also many species, such as datasets from citizen science programmes.