Dataset Information

Dealing with noisy absences to optimize species distribution models: an iterative ensemble modelling approach.

ABSTRACT: Species distribution models (SDMs) are widespread in ecology and conservation biology, but their accuracy can be lowered by non-environmental (noisy) absences that are common in species occurrence data. Here we propose an iterative ensemble modelling (IEM) method to deal with noisy absences and hence improve the predictive reliability of ensemble modelling of species distributions. In the IEM approach, outputs of a classical ensemble model (EM) were used to update the raw occurrence data. The revised data was then used as input for a new EM run. This process was iterated until the predictions stabilized. The outputs of the iterative method were compared to those of the classical EM using virtual species. The IEM process tended to converge rapidly. It increased the consensus between predictions provided by the different methods as well as between those provided by different learning data sets. Comparing IEM and EM showed that for high levels of non-environmental absences, iterations significantly increased prediction reliability measured by the Kappa and TSS indices, as well as the percentage of well-predicted sites. Compared to EM, IEM also reduced biases in estimates of species prevalence. Compared to the classical EM method, IEM improves the reliability of species predictions. It particularly deals with noisy absences that are replaced in the data matrices by simulated presences during the iterative modelling process. IEM thus constitutes a promising way to increase the accuracy of EM predictions of difficult-to-detect species, as well as of species that are not in equilibrium with their environment.

SUBMITTER: Lauzeral C

PROVIDER: S-EPMC3499418 | biostudies-literature | 2012

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Dealing with noisy absences to optimize species distribution models: an iterative ensemble modelling approach.

Lauzeral Christine C Grenouillet Gaël G Brosse Sébastien S

PloS one 20121115 11

Species distribution models (SDMs) are widespread in ecology and conservation biology, but their accuracy can be lowered by non-environmental (noisy) absences that are common in species occurrence data. Here we propose an iterative ensemble modelling (IEM) method to deal with noisy absences and hence improve the predictive reliability of ensemble modelling of species distributions. In the IEM approach, outputs of a classical ensemble model (EM) were used to update the raw occurrence data. The re ...[more]

PMID: 23166691

Similar Datasets

Project description:AimConservation planning exercises increasingly rely on species distributions predicted either from one particular statistical model or, more recently, from an ensemble of models (i.e. ensemble forecasting). However, it has not yet been explored how different ways of summarizing ensemble predictions affect conservation planning outcomes. We evaluate these effects and compare commonplace consensus methods, applied before the conservation prioritization phase, to a novel method that applies consensus after reserve selection.LocationEurope.MethodsWe used an ensemble of predicted distributions of 146 Western Palaearctic bird species in alternative ways: four different consensus methods, as well as distributions discounted with variability, were used to produce inputs for spatial conservation prioritization. In addition, we developed and tested a novel method, in which we built 100 datasets by sampling the ensemble of predicted distributions, ran a conservation prioritization analysis on each of them and averaged the resulting priority ranks. We evaluated the conservation outcome against three controls: (i) a null control, based on random ranking of cells; (2) the reference solution, based on an expert-refined dataset; and (3) the independent solution, based on an independent dataset.ResultsNetworks based on predicted distributions were more representative of rare species than randomly selected networks. Alternative methods to summarize ensemble predictions differed in representativeness of resulting reserve networks. Our novel method resulted in better representation of rare species than pre-selection consensus methods.Main conclusionsRetaining information about the variation in the predicted distributions throughout the conservation prioritization seems to provide better results than summarizing the predictions before conservation prioritization. Our results highlight the need to understand and consider model-based uncertainty when using predicted distribution data in conservation prioritization.

Project description:Interactions between soil, topography, and climatic site factors can exacerbate and/or alleviate the vulnerability of oak woodland to climate change. Reducing climate-related impacts on oak woodland habitats and ecosystems through adaptation management requires knowledge of different site interactions in relation to species tolerance. In Britain, the required thematic detail of woodland type is unavailable from digital maps. A species distribution model (SDM) ensemble, using biomod2 algorithms, was used to predict oak woodland. The model was cross-validated (50%:50% - training:testing) 30 times, with each of 15 random sets of absence data, matching the size of presence data, to maximize environmental variation while maintaining data prevalence. Four biomod2 algorithms provided stable and consistent TSS-weighted ensemble mean results predicting oak woodland as a probability raster. Biophysical data from the Ecological Site Classification (forest site classification) for Britain were used to characterize oak woodland sites. Several forest datasets were used, each with merits and weaknesses: public forest estate subcompartment database map (PFE map) for oak-stand locations as a training dataset; the national forest inventory (NFI) "published regional reports" of oak woodland area; and an "NFI map" of indicative forest type broad habitat. Broadleaved woodland polygons of the NFI map were filled with the biomod2 oak woodland probability raster. Ranked pixels were selected up to the published NFI regional area estimate of oak woodland and matched to the elevation distribution of oak woodland stands, from "NFI survey" sample squares. Validation using separate oak woodland data showed that the elevation filter significantly improved the accuracy of predictions from 55% (p = .53) to 83% coincidence success rate (p < .0001). The biomod2 ensemble, with masking and filtering, produced a predicted oak woodland map, from which site characteristics will be used in climate change interaction studies, supporting adaptation management recommendations for forest policy and practice.

Project description:The mechanisms generating stably differentiated cell-types from multipotent precursors are key to understanding normal development and have implications for treatment of cancer and the therapeutic use of stem cells. Pigment cells are a major derivative of neural crest stem cells and a key model cell-type for our understanding of the genetics of cell differentiation. Several factors driving melanocyte fate specification have been identified, including the transcription factor and master regulator of melanocyte development, Mitf, and Wnt signalling and the multipotency and fate specification factor, Sox10, which drive mitf expression. While these factors together drive multipotent neural crest cells to become specified melanoblasts, the mechanisms stabilising melanocyte differentiation remain unclear. Furthermore, there is controversy over whether Sox10 has an ongoing role in melanocyte differentiation. Here we use zebrafish to explore in vivo the gene regulatory network (GRN) underlying melanocyte specification and differentiation. We use an iterative process of mathematical modelling and experimental observation to explore methodically the core melanocyte GRN we have defined. We show that Sox10 is not required for ongoing differentiation and expression is downregulated in differentiating cells, in response to Mitfa and Hdac1. Unexpectedly, we find that Sox10 represses Mitf-dependent expression of melanocyte differentiation genes. Our systems biology approach allowed us to predict two novel features of the melanocyte GRN, which we then validate experimentally. Specifically, we show that maintenance of mitfa expression is Mitfa-dependent, and identify Sox9b as providing an Mitfa-independent input to melanocyte differentiation. Our data supports our previous suggestion that Sox10 only functions transiently in regulation of mitfa and cannot be responsible for long-term maintenance of mitfa expression; indeed, Sox10 is likely to slow melanocyte differentiation in the zebrafish embryo. More generally, this novel approach to understanding melanocyte differentiation provides a basis for systematic modelling of differentiation in this and other cell-types.

Dataset Information

Dealing with noisy absences to optimize species distribution models: an iterative ensemble modelling approach.

Publications

Dealing with noisy absences to optimize species distribution models: an iterative ensemble modelling approach.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets