Validation of species-climate impact models under climate change
read more
Citations
Novel methods improve prediction of species' distributions from occurrence data
Predicting species distribution: offering more than simple habitat models.
Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation
Species Distribution Models: Ecological Explanation and Prediction Across Space and Time
Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS)
References
The measurement of observer agreement for categorical data
A new look at the statistical model identification
The Elements of Statistical Learning
Measuring the accuracy of diagnostic systems
Related Papers (5)
Novel methods improve prediction of species' distributions from occurrence data
Predicting species distribution: offering more than simple habitat models.
Frequently Asked Questions (10)
Q2. What are the future works mentioned in the paper "Validation of species–climate impact models under climate change" ?
The high performance of complex nonlinear techniques suggests that relatively unexplored methodologies such as multivariate adaptive regression splines, adaptive logistic regression ( boosting ) and generalized multiplicative models ( for review see Hastie et al., 2001 ) might deserve future testing. Many studies have used good model fits on nonindependent validation data to support results pertaining to the potential impacts of future climate change on biodiversity ( see references in Table 1 ). There are many reasons, additionally to the effects of autocorrelation in the data, why good model fits on present-day distribution data ( i. e. nonindependent validation data ) do not necessarily translate into good predictions of future ranges. There are clearly limits to the ability of any model to predict the future distribution of species under climate change, and model validation thus becomes a conceptually difficult problem.
Q3. How many different climate parameters were used to calculate the mean values of six different variables in two different?
Average monthly temperature, precipitation and cloud cover of 1416 grid cells covering the area of the UK (71300 E–11400 W and 501N–611N) were used to calculate mean values of six different climate parameters in two different time slices (1967–1972, 1987–1991).
Q4. What is the problem with the resubstituition approach?
A problem with the resubstituition approach is that models may overfit to the calibration data, leaving users unable to judge whether high accuracy on nonindependent data reflect good predictive accuracy on independent data sets.
Q5. What is the way to estimate model accuracy?
As most assessments of model accuracy use nonindependent data, it is useful to estimate the degree to which predictive accuracy measured with nonindependent t1 distribution data provides a good surrogate for accuracy on t2 independent data.
Q6. What factors may affect the prediction of future ranges?
Such factors may include the presence of spurious correlations between response (i.e. species) and predictor (i.e. climate) variables, which may translate into poor predictions on independent validation data (e.g. Guisan & Zimmermann, 2000).
Q7. What is the pattern of performance across modelling techniques?
This pattern of performance across modelling techniques is consistent with previous assessments of performance of species–climate envelope models with nonindependent data (for reviews see Olden & Jackson, 2002; Segurado & Araújo, 2004), and suggests that modelling techniques capable of summarising complex nonlinear relationships are more likely to provide useful projections of species responses to climate change.
Q8. Why does the lack of confidence in the prediction of future ranges lead to poor predictions?
There are many reasons, additionally to the effects of autocorrelation in the data, why good model fits on present-day distribution data (i.e. nonindependent validation data) do not necessarily translate into good predictions of future ranges.
Q9. Why do the authors expect that models’ performance will decrease as observed and modelled events become increasingly independent?
This is because the effect of inflated performance arising from modelling spatially and temporally autocorrelated data should decrease as observed and modelled events become increasingly independent from each other.
Q10. How can the model be fully tested?
It may be argued that the predictive accuracy of species–climate envelope models can only be fully tested by means of validation studies using direct comparison of model predictions with independentempirical observations (Fig. 1c).