I was wondering how to do the sensitivity analysis on ArcGis? I see there's a guide for the version 9.2 ( ArcGIS Desktop Help 9.2 - Performing sensitivity analysis ) but I don't understand much the steps because the explanation gives for granted that I made a ModelBuilder which is not my case. I made a suitability model for urban abandoned spaces that can be reused for projects of environmental remediation and now I want to check its reliability. As I understood, the best way to do this is the sensitivity analysis but I didn't find any document or web page (apart from the one above) that show how to perform that with ArcGis specifically. So my questions is: can someone kindly explain to a beginner how to perform the sensitivity analysis, please?
Otherwise, if any of you know any other approach to check the validity of a suitability model I'd really appreciate any suggestion or tip you may have
You imply you have a model, albeit conceptual. Can you describe it?
In its simplest form, sensitivity analysis, in the gis sense of the term, could involve varying a variable/term in the model within a range to assess the change in outcomes. For instance, you model may have a factor 'X' which according to the literature is supposed to have an impact on 'Y'. You may find that varying the range that 'X' has within the model, you can view the change in 'Y' to see if this is correct. Quantitative models are best suited to this sort of analysis because simple cases of X affects Y don't look at the combined impact of other variables, some which may mitigate X's affect on Y.
So what do you have?
What is the form of the data? (raster, vector, a combination)
Software availability and version?
Familiarity with other non-gis software?
Familiarity with statistics? (both spatial and inferential)
anything else you can add
Thanks a lot for your answer. So my suitability model counts so far 6 different variables: slope, population density, distance to public transports, to communities venue, to schools and universities as well. They were previously vector data that I transformed into raster when I calculated the distance (except for the slope which was already a raster obtained from the elevation naturally). Afterward, I run the reclassify tool in order to find those vacant lands close to areas with high density of population, little slope and close to all the amenities I listed above. Finally, I used the weighted overlay. I already got a shapefile containing the polygons of the current vacant lands but I sincerely don't know how to insert that variable in the suitability model without affect the results. I plan to overcome that by doing a 'selection by location' on the areas with the highest suitability range obtained from the model I described you above.
I got the software license thanks to the university so I can access to all the range of products inside ArcGis but I don't know how to use other software apart from Envi. I'm not prepared at all on statistics but, if I'll just need the understanding of unadvanced topics, I can always have a look of the literature I guess.
Following the definition that you also wrote, it would be too simplistic if I just change the % of influence when I'm doing the weighted overlay? I read somewhere that when performing the SA we should pick the variables which in our opinion have the highest importance. In this case for me were the proximity to high-density population areas to schools and transports. Although perhaps I would need to figure out from how to calculate the statistics in order to show what I did...
I'm a bit lost..
well, you do have one important bit of information... the land that is not vacant ...
You could begin to look to see whether any of your other variables played a part in their occupation. For example, in a simplistic sense, examine slope... if you find that 'occupied' land occurs on flat as well as steep areas and the vacant land also has a mixture of flat and steep slopes, perhaps slope isn't an issue at all. On the other hand, if the remaining vacant land is predominantly on steep sloped areas, even though there is occupies land on steep slopes, there may be a transitional history going on. Flat land being occupied first, then steep as a last resort.
This is the type of thing that you can do at the exploratory stage. There is no 'right' model, just one that is presented in a logical fashion. There is little to be gained in producing a model to identify most suitable... or most likely to be occupied next... until you have examined the past. Perhaps a variable has not been accounted for... like the view (viewshed analysis)... the variables that influence decisions need not be site specific, nor even proximal. It is in their examination that one can begin to understand how a landscape evolves, since they rarely evolve in a predictable fashion