Machine learning of large‐scale spatial distributions of wild turkeys with high‐dimensional environmental data

Farrell, A. Vallely; Wang, Guiming; Rush, Scott A.; Martin, James A.; Belant, Jerrold L.; Butler, Adam; Godwin, Dave

doi:10.1002/ece3.5177

Cited by 51 publications

(32 citation statements)

References 63 publications

(141 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Species distribution models can be used to predict suitable conditions for species survival and to extrapolate areas where the species could occur based on known locations (Franklin ). A promising tool within SDMs is machine‐learning methods, due to their nonparametric approaches and ability to overcome difficulties such as multicollinearity (Farrell et al, ). Through the framework of machine learning and SDMs, we aim to quantify sex‐specific distribution range for a low‐density recolonizing population of American black bears ( Ursus americanus ; Family: Carnivora), evaluating if differences between females and males could influence their conservation planning.…”

Section: Introductionmentioning

confidence: 99%

Conservation implications of sex‐specific landscape suitability for a large generalist carnivore

Gantchoff

Conlee

Belant

2019

Diversity and Distributions

View full text Add to dashboard Cite

Aim: Terrestrial mammal distribution models typically do not differentiate between sexes when making spatial predictions, which could have important conservation implications. As male carnivores are usually more risk tolerant and travel longer distances, male potential range should be larger and include more human-modified areas than female range. To evaluate if differences between females and males could influence their conservation planning, we quantified sex-specific suitable range for a recolonizing population of American black bears (Ursus americanus). Location:We applied this method in Missouri, USA. Methods:We collected telemetry data from 57 females and 43 male bears from 2010 to 2018. We used three machine-learning methods (generalized boosted regression, random forest and maximum entropy) to develop sex-specific distribution models, which were combined into a weighted average ensemble model and converted into a binary (presence/absence) model using an optimized threshold. We used 80% of individuals for each sex as training data, and 20% as evaluation data. Results:All models had high predictive performance; the greatest uncertainty in model predictions corresponded with the periphery of the distributions, whereas there was high agreement in the core distribution. Male suitable range was 66% larger than female range, and males were predicted to occur in more human-modified areas and more likely to be present in developed land and agriculture.Main conclusions: Distribution models based on data from both sexes, or that is male-biased, could overestimate the female (i.e., reproductive) range and potentially misrepresent the biologically relevant species distribution. This bias can potentially lead to misguided decisions and suboptimal use of resources. By improving models when sex-specific data are available, we can better focus resources in areas where there is reproductive potential, leading to more accurate assessments of species' conservation status and better identifying areas vital for species persistence. K E Y W O R D Scarnivore, habitat suitability, mammal, range, sex-specific, spatial ecology, species distribution model, ursid | 1489 GANTCHOFF eT Al.

show abstract

Section: Introductionmentioning

confidence: 99%

Conservation implications of sex‐specific landscape suitability for a large generalist carnivore

Gantchoff

Conlee

Belant

2019

Diversity and Distributions

View full text Add to dashboard Cite

show abstract

“…ReliefF [45] is a multiclass version of the relief algorithm family. The principle of ReliefF is to estimate the importance of features based on how well their values distinguish among instances that are close to each other [12].…”

Section: ) Relieffmentioning

confidence: 99%

A Hybrid Feature Method for Handling Redundant Features in a Sentinel-2 Multidate Image for Mapping Parthenium Weed

Kiala

Mutanga

Odindi

et al. 2020

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Multidate images provide additional spectral information valuable for mapping plant species. However, correlated or redundant variables created from multiple image data and a large set of variables impede accurate and efficient landscape classification. Nevertheless, with the implementation of an appropriate feature selection method, the full potential of multidate images and subsequent increased classification accuracies may be achieved. Feature selection is a process of automatically selecting features in a dataset that contribute the most to the prediction of a target variable. It improves the classification process in terms of computation cost and predictive accuracy. Feature selection algorithms are typically classified into three groups: filters, wrappers, and embedded. Due to the inherent tradeoffs provided by different feature selection approaches, we hypothesize that a hybrid approach could optimize landscape delineation by leveraging on their complementary strengths. In this regard, we propose a new feature selection method that combines a filter ReliefF, a wrapper Support vector machines-backward (SVM-b), and the embedded random forest in mapping the noxious Parthenium weed using a Sentinel-2 multidate image. The new approach was compared to its three constituents based on the size of optimal feature subsets and classification accuracies. To make comparison and assessment of developed models from selected variables more valuable, the split between training and test datasets was implemented in three different ratios, namely, 3:1(Dataset 1), 1:1(Dataset 2); 1:3 (Dataset 3). The new approach was also evaluated against the multidate image without feature selection and a single-date image. Results showed that the new approach yielded the highest overall accuracies with the smallest optimal feature subsets. For instance, on Dataset 3, the OA was 86.6% with 22 optimal features, whereas it was 84.7% with 35 optimal features using SVM-b, which was the second most performing feature selection algorithm. The new approach yielded higher classification accuracies than the multidate image without feature selection and the single-date image. The findings of this study underscore the capability of hybrid methods to select fewer features from multidate images, with higher predictive accuracies than individual feature selection methods.

show abstract

“…However, the presented results suggest that precipitation is not linearly correlated with PPR virus presence. Other approaches, including PCA and stepwise backward elimination, are proven ways to effectively and transparently reduce the number of environmental variables (Farrell et al, 2019;Wubetie, 2019).…”

Section: Dear Editormentioning

confidence: 99%

A general species distribution model does not necessarily contribute to disease control

Gao

Zeng

Wang

et al. 2019

Transbound Emerg Dis

View full text Add to dashboard Cite

The spread of PPR is an active area of research and under close scrutiny of the national government. There has not been significant progress on this issue yet in China, among others, because of a lack of a suitable monitoring method and consequently of information.In the published paper, the presented MaxEnt model seems neither a good geographic predictor of PPR risk nor useful for PPR control. In the concluding coloured map, the reader can easily assess that not all occurrence points are covered by the predicted probability of PPR occurrence ('risk') in China. Many PPR presence points are spread over the full width of the northern half of China in the low-risk zone (blue). Further, the presented model fails to predict the historical PPR presence in Tibet from 2007.Next, we tried to assess the input data quality and the application of the MaxEnt algorithm. Several issues were identified. The most important issue could be that the presence point data do not represent reported PPR point locations, as suggested in Methods section. Instead, each presence point seems to represent a county 1 as described in the last text line of the publication. However, the reader is not informed about how the point is located within each county. Secondly, Methods section informs us that the presence points are 'rarefied at 1 km 2 '. If we assume that the PPR presence points are located in the centre of counties, none of the points might be within a 1-km distance of the centre of the neighbouring county.However, the reader is told that the rarefication has eliminated 15 out of 294 presence points. Further, assuming point locations within the centre of counties, the 1-km distance is too small to achieve substantial rarefication and consequently spatial auto-correlation of presence points has not been reduced. This illustrates that it was improper to accept a standard (1 km 2 ) from another related study to reduce spatial auto-correlation. The prediction is therefore doubtful, because the use of uncorrected spatial auto-correlated input data. The use of the Pearson correlation coefficient for elimination of environmental variables needs clarification as well. Are both variables of a pair of correlated variables eliminated? If so, the pairing procedure needs to be made transparent. The Pearson statistic assumes a normal distribution and a linear correlation. However, the presented results suggest that precipitation is not linearly correlated with PPR virus presence. Other approaches, including PCA and stepwise backward elimination, are proven ways to effectively and transparently reduce the number of environmental variables (Farrell et al., 2019; Wubetie, 2019). Finally, the use of goat and sheep density as environmental predictor variables of risk areas seems doubtful. More appropriate would be to 'cross/overlay' the risk areas obtained by the MaxEnt model based on bioclimatic and ultraviolet variables with the goat and sheep density maps.Based on our above assessment, a scientifically sound PPR risk zoning for China needs multiple addi...

show abstract

Machine learning of large‐scale spatial distributions of wild turkeys with high‐dimensional environmental data

Cited by 51 publications

References 63 publications

Conservation implications of sex‐specific landscape suitability for a large generalist carnivore

Conservation implications of sex‐specific landscape suitability for a large generalist carnivore

A Hybrid Feature Method for Handling Redundant Features in a Sentinel-2 Multidate Image for Mapping Parthenium Weed

A general species distribution model does not necessarily contribute to disease control

Contact Info

Product

Resources

About