Hi, I want to make one or more density maps (GIS) on urban agriculture in Rome, I have 4000 polygons differentiated by types (community garden, residential garden etc.) and uses (horticulture, vineyards, olive groves etc.). I thought firstly to kernel density to represent the area density, anyone have other ideas?
Regards
Giuseppe
thisisthemeasure of the number of agriculture sites per unit area (sq km). the higher the index the greater the concentration of urban agricultute at that area.
Can't you use percentage instead? It's seems to me the more accurate value. For each cell you give the percentage of occupied area by comparison with the full area of the cell.
If you want to differentiate types on the same map you can give a different colormap for each type thus having in truth a mixture between continuous and discrete variables.
I've put a small example (from generic variable) with this reply (image.png). So in the example I have 3 colormaps one for each type (Reds,Greens,Purples). In each of the colormaps you can see the color getting stronger in the areas where the value is stronger also.
Thanks Pedro
I saw that your suggestion is also used in several papers, but the problem is that I have only these poligong (photointerpretation - fully occupied at 100%). I'm thinking of do it on a land use/cover map (Corine Land Cover or Urban Atlas), but I do not know.....
Hi Giuseppe,
Maybe I misinterpret but this kind of technique can be used for polygons such as yours. I'm assuming you have (simple case) a shape. When you overlap a regular grid over it each of those cells may be a) completely filled with the polygon (100%) , b) completely outside a polygon (0%) or c) something in between. I'm putting a illustration of the concept to make sure we are talking about the same thing. So you start with the grid and shape, since you know the area of each cell you just calculate the occupied area for each one, and finally you have a colored density map(check image).
The other hypothesis I'm considering is if you have polygons but not of the exact areas of agriculture but just general shapes throughout the city with far less resolution. Is this the case?
I'm not very familiar with kernel density but I believe it's a kind of smoother. I would advise against since that would generate non-null densities in areas where no agriculture exists (unless you're attempting to build a probability based map).
Hi Giuseppe,
I agree with Pedro. Kernel density will introduce artificial values in areas where there may not be any agriculture at all. Of course this will be reflective of the bandwidth of the kernel you are using. If the polygons you are representing are from field surveys you will lose accuracy along the way. Of course, and as mentioned by Pedro, if you are trying to create a probability suggested map then kernel density or another smoothing operator will work.
Thanks Pedro
The first solution that you suggested seems to be the best. So now I have to create a vector grid, I'll try with a 10 m resolution and see how it works. However, also the kernel map can be interesting if presented as a probability map
my suggestion is to take a look to "ThematicMapping.org"
I find always many stimulating ideas there.
Very interesting is this:
http://www.slideshare.net/Turban/using-kml-for-thematic-mapping-google-tech-talk-presentation?ref=http://thematicmapping.org/
For example, you might consider the value of the area of the polygons and the data of land use as it were a piece of land where the height of the "spine line" of the polygon is given by the size of the area.
For "thorns line" mean the characteristic line of the polygon, also called "Straight Line Skeleton", which would be the center line of the polygon, which reach up to the value of the area in a three-dimensional displays as areas where there were mountains concentration of land use type that interests you.
Complicated? But even suggestive.
Dear all
Another consideration for you. In your opinion what would be the ideal grid size? I thought 10x10 meters, whereas the media value for residential garden is near to 100 sqm. Attached a grid 10x10 for a peri-urban residential area where houses have about 300-350 sqm roof area.
I tried a quick and clumsy color based method from a screen grab of this picture. Replacing all the green areas with bright green, all the grid lines with red and all the roofs with black. Then used ImageJ to calculate the area of each color.
count area total.size percent.area pixels
lines 11610 287157 24.734 16.471 287148.8
green 21553 797417 36.998 45.740 797412.9
buildings 11997 603183 50.278 34.599 603185.1
total 45160 1687757 112.010 96.810 1687746.8
The only value I could check was the area of the grid lines. I over estimated them by 476.19%. Most of the area was in the populated area.
While the problem with the grid can never by totally solved. I can be reduced to a very low level by a mask that looks that 2 pixels on ether side of of where it is known to be and those values instead in its place.
Using better methods to separate the colors and adding an algorithm to judge the size and shape of the fields might make a method that is limited by pixel size.
If you can you get multi-spectral images of the same area over time a study of the spectral reflectance at different times of the year, coupled with the size, shape and location and an occasional trip to ground truth your observations should build a data base that has a great deal more intelligence on the nature of urban farming than any of us would think possible.
I worked for the USDA in the summer when they where ground truthing satellite data in the late 1960's to see if they could use it to replace contractors that inspected farms.. We were seeing weeds in the turn rows and near the road that we could identify with the help of the local knowledge we gained working that ground all our lives but we had no way to translate that to something a computer can use.
I worked with multi-spectral non-imaging sensors for weeds, fertilizer and other problems with crops in the 1990's. We could see more things going on in the fields than we had computer power to use.
I ran into a retired statistics professor that when to grad school with my Genetics professor in 1948. It seems nothing has really changed. We still have more data that we have computer power to process.
- Badges
- Science method
- Similar topics
- Methodology