Changes between Version 23 and Version 24 of PostGIS_Raster_SoC_Idea_2012/Distance_Analysis_Tools/document


Ignore:
Timestamp:
Jul 12, 2012, 11:06:13 PM (12 years ago)
Author:
qliu
Comment:

—

Legend:

Unmodified
Added
Removed
Modified

  • PostGIS_Raster_SoC_Idea_2012/Distance_Analysis_Tools/document

    v23 v24  
    7070 * Producing an intermediate raster is costly if the requested raster resolution is very high (constraint 3, 5b & 5d).
    7171 * ST_Union could be very inefficient at producing the required raster from a large set of geometries and there is no efficient method to produce such a raster right now in PostGIS (constraints 1, 5c & 5d).
    72  * It could be very inefficient to rescan all the source points to find the one nearest to the current pixel (constraint 6, 7, 8 & 12). {{{We still have to see how GRASS does this efficiently.}}}
     72 * It could be very inefficient to rescan all the source points to find the one nearest to the current pixel (constraint 6, 7, 8 & 12). __We still have to see how GRASS does this efficiently.__
    7373 * Only rasters which extent contains all the source points could be produced.
    7474 * This approach does not answer well to the requirement of developing a generic reusable solution for more interpolation needs (constraint 7).
    75 
     75[[BR]][[BR]]
    7676==== 2. The TIN approach ====
    7777
    78 The second approach involve creating a TIN (triangulated irregular network) from the table of source geometries first and then to use this TIN to determine the nearest neighbors for each pixel.
     78  The second approach involve creating a TIN (triangulated irregular network) from the table of source geometries first and then to use this TIN to determine the nearest neighbors for each pixel.
    7979
    80 The TIN is stored temporarily. A function iterating over each pixel determines in which part of the TIN the current pixel falls and determine the triangle corner nearest to the centroid of the pixel.
     80  The TIN is stored temporarily. A function iterating over each pixel determines in which part of the TIN the current pixel falls and determine the triangle corner nearest to the centroid of the pixel.
    8181
    82         s
     82'''Remarks:'''
     83 * Nearest point to pixel could be determined by interesting the pixel centroid with the TIN in order to determine the triangle including the centroid. The nearest point is then determined from the three corner of the triangle.
     84 * It could also be determined by producing the TIN with the set of point including all the pixel centroid. Would this fasten the search for the nearest neighbor? This is still to determine.
     85 * Would this work with a table of line or geometries? How do we produce a TIN from a line or polygon coverage?
     86 * PostGIS might not be ready to produce a TIN from a point coverage.
    8387
    84 Remarks:
    85 •       Nearest point to pixel could be determined by interesting the pixel centroid with the TIN in order to determine the triangle including the centroid. The nearest point is then determined from the three corner of the triangle.
    86 •       It could also be determined by producing the TIN with the set of point including all the pixel centroid. Would this fasten the search for the nearest neighbor? This is still to determine.
    87 •       Would this work with a table of line or geometries? How do we produce a TIN from a line or polygon coverage?
    88 •       PostGIS might not be ready to produce a TIN from a point coverage.
     88'''Pros:'''
     89 * This would work well with a large number of source points (constraints 1, 5c & 5d) since we would not have to  rescan all the source points for each pixel calculation.
     90 * __This approach fulfills constraint 7 in that it would allow easy implementation of other interpolation methods.__
     91{{{
     92Comments from the mentor:
     93Not really. If the raster HAS to be in the extent of the point coverage, as stated in the cons, it means it could not correspond to the arbitrary alignment/extent of an existing coverage. This is probably the main reason the third approach is preferable to the TIN one.
     94}}}
    8995
    90         Pros:
    91 ○       This would work well with a large number of source points (constraints 1, 5c & 5d) since we would not have to  rescan all the source points for each pixel calculation.
    92 ○       This approach fulfills constraint 7 in that it would allow easy implementation of other interpolation methods.
    93 Cons:
    94 ○       We have no idea for now about the algorithm and performance of building TIN from source geometries in PostGIS if the number of source point is very large (Constraint 7 & 8)
    95 ○       Might be very inefficient and a waste of computing TIN in case there is only one point in the source table (Constraint 1) or there are very small number of source points and the requested raster is relatively small too (Constraint 5).
     96'''Cons:'''
     97 * We have no idea for now about the algorithm and performance of building TIN from source geometries in PostGIS if the number of source point is very large (Constraint 7 & 8)
     98 * Might be very inefficient and a waste of computing TIN in case there is only one point in the source table (Constraint 1) or there are very small number of source points and the requested raster is relatively small too (Constraint 5).
    9699
    97100==== 3. The KNN index approach ====