Effects of Training Dataset Characteristics on Species Distribution Models for Western Corn Rootworm
Author Information
Author(s): Dupin Maxime, Reynaud Philippe, Jarošík Vojtěch, Baker Richard, Brunel Sarah, Eyre Dominic, Pergl Jan, Makowski David
Primary Institution: INRA, UR Zoologie Forestière, Orléans, France
Hypothesis
How do the characteristics of training datasets affect the performance of species distribution models for Diabrotica virgifera virgifera?
Conclusion
The performance of species distribution models for the western corn rootworm is highly sensitive to the size of the training dataset, the stage of biological invasion, and the choice of input variables.
Supporting Evidence
- Model performance was highly sensitive to the geographical area used for calibration.
- Principal Component Analysis helped reduce the number of input variables for poorly performing models.
- Models performed better with larger training datasets corresponding to later stages of invasion.
Takeaway
This study looked at how different training data can change the predictions of models that forecast where a pest might spread. It found that using the right data is really important for getting accurate predictions.
Methodology
Nine species distribution models were assessed using various training datasets of different sizes and characteristics to predict the distribution of the western corn rootworm.
Potential Biases
The use of pseudo-absence data may have influenced the accuracy of the model predictions.
Limitations
The models showed substantial misclassification rates, and their performance was not consistently high across all conditions tested.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website