POST
|
Hi Mervyn, Thanks for your answer. I arrived at a similar conclusion using different methods, but would like to put it to the group to see if others find it plausible. My understanding is that the variation in layers that load heavily with a given PC; i.e. layers that have high positive or low negative eigenvectors, is captured by the given PC. Layers with high positive eigenvectors are directly related to the PC; layers with low negative eigenvectors are inversely related. For a given PC, a high pixel is a location where layers with high positive eigenvectors also had high values (direct relationship). A low pixel is a location where layers with low negative eigenvectors has high values (inverse relationship). I tested this by running a PCA on a very small number of input layers so I could easily observe the relationships between inputs and PCs, and it holds true. Does anyone see any flaws in this conclusion?
... View more
10-14-2015
09:06 AM
|
0
|
0
|
1852
|
POST
|
Hi Gordon, Thanks - that is a great link to a very clear and concise explanation of a PCA analysis. But I still don't understand what the raster PC pixel values are, or how they are calculated. PC1 is an axis along the direction of greatest variance (side question: because I'm working with spatial data, are directions true directions; e.g. the direction of greatest variance denoted by PC1 above is inshore -> offshore?), but what does a high pixel value mean vs. a low pixel value? If I were to hazard another guess following: pca - What are principal component scores? - Cross Validated , I'd guess that each pixel value is a PC score. In other words, a given pixel(x)'s value in PC1 is a linear recombination calculated as the product of var1 at pixel(x) * var1 loading for PC1 and then summed across all variables at pixel(x). Following this logic, a pixel w a high value in PC1 likely had original variables with high values that loaded heavily in PC1 (because multiple sums of a big variable value * a high loading would lead to larger numbers). So I guess I would interpret PC1 as representing original variables that were generally low valued inshore, and high valued offshore. Am I on the right track? And if so, it begs a follow up question: why are none of the PC1 raster pixel values negative? A given variable can load negatively w a given PC: if linear recombination is how pixel values are calculated, some could be less than zero. Does Arc include some sort of standardization in the calculation?
... View more
10-08-2015
08:03 AM
|
0
|
2
|
1852
|
POST
|
Hi - I understand that principle components are calculated such that the first explains the greatest amount of variation in a dataset, the second PC explains the greatest amount of remaining variation, and so on. I understand PCs are uncorrelated variables that retain information in the original dataset using reduced dimensions. What I'm unclear on is what PCs mean in terms of the multiband raster output from the Principal Components tool. I can look at my Eigenvalues for each PC and determine how many to use to maintain the variation in my dataset, but I don't know how to interpret the PCs in raster form. For example, here is PC1, with values ranging from 4.6 (red) to 0 (blue). What do these values mean? My best guess is that these values are 'distances' or errors from the PC1 axis.
... View more
10-07-2015
01:26 PM
|
0
|
5
|
6047
|
POST
|
No, I disabled background processing and the tool has now run for over an hour with no progress.
... View more
10-06-2015
07:39 AM
|
0
|
0
|
234
|
POST
|
Hello, I'm trying to use the Dendrogram tool (Spatial Analyst\Multivariate) to create a dendrogram for a signature file produced using the Iso Cluster Unsupervised Classification tool. The tool runs when I unselect Use variance in distance calculations (optional), but fails with no error message when I keep this option selected. Instead I get a pop-up that background processing has encountered a serious error and is unable to continue. Has anyone else encountered this? I was able to run the Dendrogram tool fine in 10.2.1, but now with the upgrade to 10.3.1 it fails. Signature file attached. -Heather
... View more
10-02-2015
08:21 AM
|
0
|
3
|
2470
|
POST
|
I'm interested to try the Select Data tool and see if it allows Mosaic to New Raster to recognize the number of bands. As it is, I found a work around. I fed my sub model 'scan1' into the Mosaic tool with Create Raster Dataset run as a precondition - which allowed me to set the number of bands without the model needing to recognize the number of bands in the input. A little clunky, but it works.
... View more
04-09-2015
01:57 PM
|
0
|
0
|
422
|
POST
|
Hi all, I'm interpolating series of bathymetric points using kriging, and then exporting my data to raster. I'm struggling to determine an appropriate output resolution that is supported by the characteristics of my input point data. Does anyone have any advice? Two factors I'd like to take into account are 1. the density of the input points (such that increased density = increased resolution, and vice versa) and 2. the 'complexity' of the seafloor (or other real-world surface), perhaps calculated as the variance of the z-value of input points (in this case depth). In this regard, a less complex surface (low variance) could be interpolated at a higher resolution because we'd have more confidence that z-values will remain consistent, and a more complex surface (high variance) would be interpolated at a lower resolution to avoid introducing false precision. RE factor 1, I read a great paper - Hengl, T. 2006. Finding the right pixel size. Computers & Geoscience. 32: 1283-1298 - that suggested, "...the grid resolution should be at most half the average spacing between the closest point pairs...". I've been working off this calculation, but would welcome other ideas. RE factor 2, Any ideas for how to use input point variance as a scale factor for resolution size? Has anyone heard of this being done before? I'd be glad to hear of any other factors I should take into consideration. Thank you.
... View more
04-09-2015
09:05 AM
|
0
|
5
|
4366
|
POST
|
Hello, I'm working on a model to iterate through a series of rasters, perform a couple operations to each raster, collect the values, and then mosaic them together into a new raster. My submodel includes Iterate Rasters>Make Raster Layer>Copy Raster>Collect Values, and then my main model has the Mosaic to New Raster. When I try to connect the output from my submodel to Mosaic to New Raster, I get the error that the number of bands from the input raster must match the number in the output, and I cannot run the model. My original data has 4 bands, but during the submodel I reduce the number to 1, and then my output also has 1. Has anyone else run into this problem in model builder? Any ideas how to get around it?
... View more
04-09-2015
06:11 AM
|
0
|
2
|
4397
|
Online Status |
Offline
|
Date Last Visited |
11-11-2020
02:25 AM
|