We have blogged a few times about niche modeling and how to improve it. Below Mohamed Fawzy Farag Nawar briefly highlights what will become a useful resource in this field, Lifemapper (the data for the modeling comes from GBIF), but points out some limitations.
Lifemapper.org is an initiative to implement online some sort of generic model to predict where a species might exists based on where it was collected, or where it was otherwise documented that it lived. Fine. But here is the results of the model on the species Clivipollia pulcher. This is a marine mollusk that was found along the coasts of eastern Australian, Papua New Guinea and the Philippines. You will see on the map of predicted distribution that the model suggests it might be found in various places inland in central Africa and Latin America. That is what you get when essential prior knowledge is not introduced to the model. Something like telling the developer of the model that marine species should be modeled to a different set of environmental data than Worldclim, which should only be used for terrestrial fauna and flora. Agricultural species are included in Lifemapper, though again the predicted distributions will have to be looked at fairly carefully before use.
That’s where species range maps based on literature, and even on the occurrence records themselves need to be used, so that you don’t go predicting high probabilities in areas where a species is not even present.
The quality of geographic information is a fundamental issue to be addressed too, not to mention taxonomic and specimen-verification/identification issues.
And then, when modeling, let’s say, landraces you will also need to consider other drivers than environmental (socio-economic). Niche modeling tools are advancing fast and provide pretty useful knowledge when you have even a few occurrence points of a taxon, but you need to ensure you’re generating the proper information.
True! This is an important warning for uncritical use of niche models. The niche model predictions of a species distribution does not intend to imply that the species would necessarily be expected to be found there. The niche model only calculates a signature of the ecological climate for the specific occurrence data used to calibrate and train the model. If the occurrence data provided to the model calibration step is not representative for the species distribution, then the predicted distribution could be heavily biased. There might also be many barriers and reasons why the species does not arrive at other locations with potential eco-geographical conditions for the species to thrive. In fact many species today are undesired invasive species in location where they obviously thrive, but have previously been hindered from arriving at. One compelling example could be arctic species unable to reach the antarctic (where they might thrive) because they would need to pass through the tropics… Next the WorldClim climate dataset (popular for niche modeling) only includes elevation, temperature (min, max, mean) and precipitation climate layers (the 19 BIOCLIM variables are derived from these). Obviously these limited climate layers might be too limited to efficiently model the ecological niche of a species. A niche model prediction of the species distribution need thus to be taken for what it is – a rough estimate of where the species might be able to thrive. The niche model prediction may of course never replace the needed for verification of species distribution with occurrence data. I agree that niche models seems to be used too often without moving on to verification of the predictions. But when used appropriately the niche models can be a very powerful new tool!
Niche modeling and common sense… and climate change?
I’ve read a recent paper in Science. It’s on the perspectives stuff of the journal, and was quoted a few days ago by the BBC. Some of you might have been noticed via the Crop Wild Relatives Yahoo group.
Although it’s behind a paywall, and I couldn’t access it, I’ve made my way to read the entire paper.
Despite the authors doesn’t perform an specific analysis to determine extinction risks driven by climate change via an specific model, or have performed a sensitivity analysis to determine the influence of spatial scales on niche modeling, they have reviewed some interesting work on those sides, and have stated that involving:
1. Small-scale topography and climates,
2. Increased CO2 effects, and
3. Adaptive and migration capacity of some species (e.g. butterflies, birds, etc.)
Certainly alter predictions of species extinction risks due to climate change. They also say:
“Over 75% of the Earth’s terrestrial biomes now show evidence of alteration as a result of human residence and land use. Yet, recent case studies suggest that even in a highly fragmented landscape, all is not lost for biodiversity”..
So that the scaring numbers of biodiversity loss/threats due to (1) habitat loss, (2) habitat fragmentation, and (3) climate change might change, and thus conclusions derived from all this could change.
And they’re certainly right. We need to go to the field and verify which kind of microclimatic and habitat-change modifications are “acceptable” and wouldn’t cause losses in current global biodiversity. The usage of coarse niche modeling techniques helps a lot in order to derive practical conclusions over big geographic areas, and even to suggest policy making, but we must go downwards and keep focused on local scales, where things could actually happen differently.
Re-design reserves, bring corridors, and let migration happen