Deep learning for supervised classification of temporal data in ecology

Capinha, César; Ceia-Hasse, Ana; Kramer, Andrew M.; Meijer, Christiaan

doi:10.1101/2020.09.14.296251

Cited by 3 publications

(3 citation statements)

References 48 publications

(113 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As mentioned previously, combining inputs from different sources is natural for deep learning and Rammer and Seidl (2019) take advantage of this to predict and map future bark beetle outbreaks based on temporal information on climate, vegetation, and past outbreaks. Capinha et al (2020) proposed a generalized approach to classification and prediction from ecological time series data leveraging automated choice of the best network architecture for the task at hand.…”

Section: Environmental Monitoring and Modelingmentioning

confidence: 99%

Deep learning as a tool for ecology and evolution

Borowiec¹,

Frandsen²,

Dikow³

et al. 2021

Preprint

View full text Add to dashboard Cite

Deep learning is driving recent advances behind many everyday technologies, including those relying on speech and image recognition, natural language processing, and autonomous driving. It is also gaining popularity in biology, where it has been used for automated species identification, environmental monitoring, behavioral studies, DNA sequencing, and population genetics and phylogenetics, among other applications. Deep learning relies on artificial neural networks for predictive modeling and excels at recognizing complex patterns. Operating within the machine learning paradigm, deep learning can be viewed as an alternative to likelihood-based inference methods. It has desirable properties, including good performance and scaling with increasing complexity, while posing unique challenges such as sensitivity to bias in input data. In this review we provide a gentle introduction to deep learning, review its applications in ecology and evolution, and discuss its limitations and efforts to overcome them. We also provide a practical primer for biologists interested in including deep learning in their toolkit and identify its possible future applications.

show abstract

Section: Environmental Monitoring and Modelingmentioning

confidence: 99%

Deep learning as a tool for ecology and evolution

Borowiec¹,

Frandsen²,

Dikow³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Data mining of time series is the process of extracting information from the properties of time series for tasks such as classification, clustering, prediction, and anomaly detection (Esling & Agon, 2012 ). These tasks are common in ecology, for example, clustering time series of parasite counts to identify infection patterns (Marques et al, 2018 ); predicting the emergence of fruiting bodies by classifying time series of environmental drivers (Capinha, 2019 ); identifying insect species by classifying wingbeat frequency signals (Potamitis et al, 2015 ); surveying bird population sizes by classifying recorded calls (Priyadarshani et al, 2020 ); and predicting species distributions based on time series of environmental variables (Capinha et al, 2020 ). These tasks all rely on the notion of (dis‐) similarity.…”

Section: Introductionmentioning

confidence: 99%

A user‐friendly guide to using distance measures to compare time series in ecology

Dove,

Böhm,

Freeman

et al. 2023

Ecology and Evolution

View full text Add to dashboard Cite

Time series are a critical component of ecological analysis, used to track changes in biotic and abiotic variables. Information can be extracted from the properties of time series for tasks such as classification (e.g., assigning species to individual bird calls); clustering (e.g., clustering similar responses in population dynamics to abrupt changes in the environment or management interventions); prediction (e.g., accuracy of model predictions to original time series data); and anomaly detection (e.g., detecting possible catastrophic events from population time series). These common tasks in ecological research all rely on the notion of (dis‐) similarity, which can be determined using distance measures. A plethora of distance measures have been described, predominantly in the computer and information sciences, but many have not been introduced to ecologists. Furthermore, little is known about how to select appropriate distance measures for time‐series‐related tasks. Therefore, many potential applications remain unexplored. Here, we describe 16 properties of distance measures that are likely to be of importance to a variety of ecological questions involving time series. We then test 42 distance measures for each property and use the results to develop an objective method to select appropriate distance measures for any task and ecological dataset. We demonstrate our selection method by applying it to a set of real‐world data on breeding bird populations in the UK and discuss other potential applications for distance measures, along with associated technical issues common in ecology. Our real‐world population trends exhibit a common challenge for time series comparisons: a high level of stochasticity. We demonstrate two different ways of overcoming this challenge, first by selecting distance measures with properties that make them well suited to comparing noisy time series and second by applying a smoothing algorithm before selecting appropriate distance measures. In both cases, the distance measures chosen through our selection method are not only fit‐for‐purpose but are consistent in their rankings of the population trends. The results of our study should lead to an improved understanding of, and greater scope for, the use of distance measures for comparing ecological time series and help us answer new ecological questions.

show abstract

“…Data mining of time series is the process of extracting information from the properties of time series for tasks such as classification, clustering, prediction, and anomaly detection (Esling and Agon, 2012). These tasks are common in ecology, e.g., clustering time series of parasite counts to identify infection patterns (Marques et al ., 2018); predicting the emergence of fruiting bodies by classifying time series of environmental drivers (Capinha, 2019); identifying insect species by classifying wingbeat frequency signals (Potamitis et al ., 2015); surveying bird population sizes by classifying recorded calls (Priyadarshani et al ., 2020); and predicting species distributions based on time series of environmental variables (Capinha et al ., 2020). These tasks all rely on the notion of (dis-) similarity.…”

Section: Introductionmentioning

confidence: 99%

A User-Friendly Guide to Using Distance Measures to Compare Time Series in Ecology

Dove

Böhm

Freeman

et al. 2022

Preprint

View full text Add to dashboard Cite

Time series are a critical component of ecological analysis, used to track changes in biotic and abiotic variables. Information can be extracted from the properties of time series for tasks such as classification, clustering, prediction, and anomaly detection. These common tasks in ecological research rely on the notion of (dis-) similarity which can be determined by using distance measures. A plethora of distance measures have been described in the scientific literature, but many of them have not been introduced to ecologists. Furthermore, little is known about how to select appropriate distance measures and the properties they focus on for time-series related tasks.Here we describe 16 potentially desirable properties of distance measures, test 42 distance measures for each property, and present an objective method to select appropriate distance measures for any task and ecological dataset. We then demonstrate our selection method by applying it to a set of real-world data on breeding bird populations in the UK. We also discuss ways to overcome some of the difficulties involved in using distance measures to compare time series.Our real-world population trends exhibit a common challenge for time series comparison: a high level of stochasticity. We demonstrate two different ways of overcoming this challenge, first by selecting distance measures with properties that make them well-suited to comparing noisy time series, and second by applying a smoothing algorithm before selecting appropriate distance measures. In both cases, the distance measures chosen through our selection method are not only fit-for-purpose but are consistent in their rankings of the population trends.The results of our study should lead to an improved understanding of, and greater scope for, the use of distance measures for comparing time series within ecology, and allow for the answering of new ecological questions.

show abstract

Deep learning for supervised classification of temporal data in ecology

Cited by 3 publications

References 48 publications

Deep learning as a tool for ecology and evolution

Deep learning as a tool for ecology and evolution

A user‐friendly guide to using distance measures to compare time series in ecology

A User-Friendly Guide to Using Distance Measures to Compare Time Series in Ecology

Contact Info

Product

Resources

About