One of the main inputs into a niche model is the environmental variables. Optimizing the choice of variables is important for many reasons, primarily interpretation and subsequent accuracy on independent test data.
In almost all cases to date, annual climate averages have been used in modeling species distributions. Where models have been developed and annual averages of climate compared with monthly variables and others such as vegetation, improvements in the accuracy were attributed to the monthly climate data sets (i.e. greater temporal resolution).
The table below from Improving ecological niche models by data mining large environmental datasets for surrogate models (Ecological Modeling) lists the variables selected in the models for the 6 species and clearly shows variables other than annual temperature averages are selected. In this case, a variable called treecover is selected most frequently (3 times), while mgvc188, the first eigenvector of vegetation, is selected next most frequently (2 times) and the rest were monthly climate variables.
|
Name |
Description |
Native Resolution |
Resolution 0.05 |
|
treecover |
Continuous field data - treecover |
0.0083 |
3 |
|
mgvc188 |
1988 MGV PCA Component 1 |
0.167 |
2 |
|
lwmpr04 |
Legates & Willmott April Measured Precipitation |
0.5 |
2 |
|
lctmp12 |
Leemans and Cramer December Temperature (0.1C) |
0.5 |
2 |
|
lwmsd08 |
Legates & Willmott August Measured Precipitation |
0.5 |
1 |
|
owe13a |
Olson World Ecosystems Version 1.3A |
0.5 |
1 |
|
mgvc388 |
1988 MGV PCA Component 3 |
0.5 |
1 |
|
lwmsd09 |
Legates & Willmott September Measured Precipitation |
0.5 |
1 |
|
lwmpr00 |
Legates & Willmott Annual Measured Precipitation |
0.5 |
1 |
|
lwmpr10 |
Legates & Willmott October Measured Precipitation |
0.5 |
1 |
|
lwtsd08 |
Legates & Willmott August Temperature (std. dev.) |
0.5 |
1 |
|
lwmpr06 |
Legates & Willmott June Measured Precipitation |
0.5 |
1 |
|
lcprc05 |
Leemans and Cramer May Precipitation (mm/month) |
0.5 |
1 |
|
fnocrdg |
Navy Terrain Data–Number of Significant Ridges |
0.167 |
1 |
|
macult |
Matthews Cultivation Intensity |
1 |
1 |
|
lccld06 |
Leemans and Cramer June Cloudiness (% Sunshine) |
0.5 |
1 |
|
lccld08 |
Leemans and Cramer August Cloudiness (% Sunshine) |
0.5 |
1 |
|
lwcpr02 |
Legates & Willmott February Corrected Precipitation |
0.5 |
1 |
|
mgv0001 |
Average January Generalized Global Vegetation Index |
0.167 |
1 |
|
lwcpr05 |
Legates & Willmott May Corrected Precipitation |
0.5 |
1 |
Only one annual variables was selected. These are typical of all the results I get these days, where annual climate averages are rarely selected as predictor variables. It makes sense. Why would a niche species be well modeled by a generic variable? Rather a niche variable should be the most predictive.
2 responses so far ↓
I agree with David, but one should not forget that other factors than climatic ones might have an influence on species response curves and distribution… Take plants as an example where most published studies only consider climatic predictors, disregarding important soil factors.
See Coudun et al. (in press) in Journal of Biogeography for an illustration of the importance of soil factors (pH, C/N) to explain ecological response and geographic distribution of Acer campestre in French forests.
Yes, soils too. I look forward to your paper. I didn’t exclude them. The edaphic variable Mathews Cultivation Intensity was included in the table above as are topo variables. My argument is that results show that monthly variables are surprisingly good among all variables, and especially much better than annual averages. It is possible that other non-climate variables could be better predictors (particularly in a marine environment). I think most studies have failed to recognize the importance of selection of variables from a wide range of possible variables.
You must also take into account that most soil important variables like pH and C/N are not really available in extensive fine scaled spatial grids, and so until they are, are not available for prediction of species distributions.
Leave a Comment