Biology subjects

Ceia-Hasse, A.

Publications and source records attributed to Ceia-Hasse, A..

4 recordsLinked to original sources

bioRxiv2024

Supporting early detection of biological invasions through short-term spatial forecasts of detectability

Early detection of invasive species is crucial to prevent biological invasions. To increase the success of detection efforts, it is often essential to know the phenological stages in which the invasive species are found. This includes knowing, for example, if invasive insect species are in their adult phase, invasive plants are flowering, or invasive mammals have finished their hibernation. Unfortunately, this kind of information is often unavailable or is provided at very coarse temporal and spatial resolutions. On the other hand, opportunistic records of the location and timing of observations of these stages are increasingly available from biodiversity data repositories. Here, we demonstrate how to apply these data for predicting the timing of phenological stages of invasive species. The predictions are made across Europe, at a daily temporal resolution, including in near real time and for multiple days ahead. We apply this to detectability-relevant phenological stages of four well-known invasive species: the freshwater jellyfish, the geranium bronze butterfly, the floating primrose-willow, and the garden lupine. Our approach uses machine learning and statistical-based algorithms to identify the set of temporal environmental conditions (e.g., temperature values and trends, precipitation, snow depth, and wind speed) associated with the observation of each phenological stage, while accounting for spatial and temporal biases in recording effort. Correlation between predictions from models and the actual timing of observations often exceeded values of 0.9. However, some inter-taxa variation occurred, with models trained on several thousands of observation records performing consistently better than those based on a few hundred records. The analysis of daily predictions also allowed mapping EU-wide regions with similar phenological dynamics (i.e., phenoregions). Our results underscore the significant potential of opportunistic biodiversity observation data in developing models capable of predicting and forecasting species phenological stages across broad spatial extents. This information has the potential to significantly improve decision-making in invasion surveillance and monitoring activities.

ecology↗

bioRxiv2023

Predicting the timing of ecological phenomena across regions using citizen science data

Spatial predictions of intra-annual ecological variation enhance ecological understanding and inform decision-making. Unfortunately, it is often challenging to use statistical or machine learning techniques to make such predictions, due to the scarcity of systematic, long-term observational data. Conversely, opportunistic time-stamped observation records, supported by highly informative data such as photographs, are increasingly available for diverse ecological phenomena in many regions. However, a general framework for predicting such phenomena using opportunistic data remains elusive. Here, we introduce a novel framework that leverages the concept of relative phenological niche to model observation records as a sample of temporal environmental conditions in which the represented ecological phenomenon occurs. We demonstrate its application using two distinct, management-relevant, ecological events: the emergence of the adult stage of the invasive Japanese beetle (Popillia japonica), and of fruiting bodies of the winter chanterelle mushroom (Craterellus tubaeformis). The framework accounts for spatial and temporal biases in observation data, and it contrasts the temporal environmental conditions (e.g., in temperature, precipitation, wind speed, etc.) associated with the observation of these events to those available in their occurrence locations. To discriminate between the two sets of conditions, we employ machine-learning algorithms (boosted regression trees and random forests). The proposed approach can accurately predict the temporal dynamics of ecological events across large geographical scales. Specifically, it successfully predicted the intra-annual timing of occurrence of adult Japanese beetles and of winter chanterelle mushrooms across Europe and North America. We further validate the approach by successfully predicting the timing of occurrence of adult Japanese beetles in Northern Italy, a recent hotspot of invasion in continental Europe, and the winter chanterelle mushroom in Denmark, a country with a high number of records of this mushroom. These results were also largely insensitive to temporal bias in recording effort. Our results highlight the potential of opportunistic observation data to predict the temporal variation of a wide range of ecological phenomena in near real-time. Furthermore, the conceptual and methodological framework is intuitive and easily applicable for the large number of ecologists already using machine-learning and statistical-based predictive approaches.

ecology↗

bioRxiv2022

Forecasting the numbers of disease vectors with deep learning

Arboviral diseases such as dengue, Zika, chikungunya or yellow fever are a worldwide concern. The abundance of vector species plays a key role in the emergence of outbreaks of these diseases, so forecasting these numbers is fundamental in preventive risk assessment. Here we describe and demonstrate a novel approach that uses state-of-the-art deep learning algorithms to forecast disease vector numbers. Unlike classical statistical and machine learning methods, deep learning models use time series data directly as predictors and identify the features that are most relevant from a predictive perspective. We demonstrate the application of this approach to predict temporal trends in the number of Aedes aegypti mosquito eggs across Madeira Island for the period 2013 to 2019. Specifically, we apply the deep learning models to predict whether, in the following week, the number of Ae. aegypti eggs will remain unchanged, or whether it will increase or decrease, considering different percentages of change. We obtained high predictive accuracy for all years considered (mean AUC = 0.92 {+/-} 0.05 sd). We also found that the preceding numbers of eggs is a highly informative predictor of future numbers. Linking our approach to disease transmission or importation models will contribute to operational, early warning systems of arboviral disease risk.

ecology↗

bioRxiv2020

Deep learning for time series classification in ecology

Temporal data is ubiquitous in ecology and ecologists often face the challenge of accurately differentiating these data into predefined classes, such as biological entities or ecological states. The usual approach consists of transforming the time series into user-defined features and then using these features as predictors in conventional statistical or machine learning models. Here we suggest the use of deep learning models as an alternative to this approach. Recent deep learning techniques can perform the classification directly from the time series, eliminating subjective and resource-consuming data transformation steps, and potentially improving classification results. We describe some of the deep learning architectures relevant for time series classification and show how these architectures and their hyper-parameters can be tested and used for the classification problems at hand. We illustrate the approach using three case studies from distinct ecological subdisciplines: i) insect species identification from wingbeat spectrograms; ii) species distribution modelling from climate time series and iii) the classification of phenological phases from continuous meteorological data. The deep learning approach delivered ecologically sensible and accurate classifications demonstrating its potential for wide applicability across subfields of ecology.

ecology↗