[BMp] Discrepancies between theory (cross validation results) and graphical output (HSM)

 

Dear all.

I’m working on HS model  for an alpine salamander in a restricted area (less than 900 ha) in northern Italy. I selected 15 EGVs discarding a few for correlation reasons and some others to avoid affecting the model. Quite all EGVs were driven as vector using GIS application by means of photo-interpretation and field work, and then converted as raster using a 50x50 m square grid. Instead, a few EGVs ware derived from DTM.

I developed 8 main models varying algorithms and parameters (combining ENFA / PCA & Geometric Mean / Harmonic Mean / Median / Minimum distance). Then, through the validation process (especially Boyce continuous) I selected the ENFA+GM as the best one (higher B values).

 

Projecting the corresponding HSM I was surprised to find quite an half of presence data set inside very low habitat suitability areas. Particularly the core of population (the bigger aggregation of individuals over many continuous sampling units 50x50m) was in area with lower values, instead of maximum as expected. Furthermore the 8 models return maps also very different among them, and, particularly, the model less credited (PCA+MD) best fits the presence data in comparison to all other.

 

Another obvious / apparent (?) contradiction concerns the most important EGV as resulted from ENFA (the one with the higher and positive values in the score matrix with reference to the first factor that in my case explain 100% of marginality and 100% of specialization, too). Overlapping this EGV map (vector or raster) with the HSM, the EGV seems to be negatively correlated instead of highly positively ! With the exception of some cases, where this EGV has consistent percentage of coverage HSM has frequently low values !

 

In summary I find it difficult to interpret and explain results because of discrepancies between theory (cross validation results) and graphical output. I also tried many different ways to reclassify maps without great benefits.

 

Is it a common situation or it is most probably due to some small problems during the analysis process such as 11 maps over 15  that were not continuous enough and very large eigenvalue encountered during processing ?

 

Thank you for your attention and possible help.
Sincerely,
Paolo Eusebio Bergò

__._,_.___

Reply to sender | Reply to group | Reply via web post | Start a New Topic

Messages in this topic (6)

Recent Activity:

• New Members 1

Visit Your Group

------------------------------
Biomapper FAQ: http://www.unil.ch/faq.html
Wiki-Biomapper: http://biomapper.wikispaces.com/
If you found this message useful, please consider copying it to the Wiki-Biomapper
------------------------------

Switch to: Text-Only, Daily Digest • Unsubscribe • Terms of Use

.

__,_._,___