Loading ...
Sorry, an error occurred while loading the content.

2489RE: [ai-geostats] A novice question

Expand Messages
  • Ted Harding
    Mar 23, 2006
    • 0 Attachment
      On 23-Mar-06 West, Nancy \(DOH\) wrote:
      > Hi all,
      > I am trying to do a simple interpolation of just 17 values
      > with a very wide range from about 0.12 to 6500 over about
      > a city block area. Being a novice, I decided to use a simple
      > method without assumptions and few parameters to set. So, I
      > chose local polynomial interpolation. It turned out about the
      > way we expected. I set it to just four classes for easy
      > identification. BTW, I am using ESRI's Geostatistical Analyst.
      > I was asked to overlay the sample points with the values labeled.
      > It was then pointed out to me that there were sampled values such
      > as 14 that are in the interpolated class range of 900 to 6500.
      > My questions are "Can this be correct?" and "If so, how?" Also,
      > is there a reference anyone can point me to that explains this?
      > Hopefully, I will learn more about this in April at Isobel's 0 to
      > Kriging course. Hi, Isobel!
      > Thanks for any guidance you can give me on this.
      > Nancy

      It *could* be "correct", *if* the observed data include a very
      wide random scatter relative to the values they are supposed
      to measure.

      However, if that is not a plausible interpretation, then one
      is tempted to conclude that your polynomial interpolation is
      not a good model. (And, by the way, I take it that when you
      say "interpolation" you really mean "smoothing", since true
      interpolation exactly fits the observed data where the data
      points occur, and the mismatch you report should not happen).

      Since there are only 17 data points (presumably each with
      [x,y] coordinates and a z value), provided it is acceptable
      from the point of view of whatever confidentiality may apply
      to your investigation, it might be possible to offer more
      considred advice if you showed us the data.

      An alternative view of your data might be that, while they
      are good measurements of the values at the points where the
      variable is measured, that variable varies substantially
      from place to place, so that neighbouring measurements can
      be very different from each other.

      If you want to preserve the point values, but interpolate
      between them to make estimates of values at points where
      you have not made measurements, then that is one thing;
      but if you regard the measured values at one time as a
      realisation of some random field, and you want to estimate
      the underlying intensity of the random field, then that is
      another thing!

      Hoping this helps,

      E-Mail: (Ted Harding) <Ted.Harding@...>
      Fax-to-email: +44 (0)870 094 0861
      Date: 23-Mar-06 Time: 19:18:54
      ------------------------------ XFMail ------------------------------
    • Show all 2 messages in this topic