I am new user to this forum and also to GWR. I watched the training video and read the some tutorials on GWR. I want to apply GWR to the air quality data (fine particles (PM2.5)) to check their spatial distribution and also want to predict these particles with help of meteorological parameters (temperature, precipitation, wind speed and wind direction, etc.). Seasonal Data of PM2.5 data from the five station is available in the study area (1000 Sq. Km). Please see attached map of the study area with PM2.5 stations.[ATTACH]6027[/ATTACH]
when I applied GWR using one station data then it was also applied on the other station. Please download the GWR result file from the given link.
That's a great question, and brings up a very important requirement of GWR. Unfortunately, Geographically Weighted Regression does work best when you have lots of observations with geographic coordinate systems. Ideally, you'll have at least 80 or so features. The dataset that you have seems very interesting, but with only five features for analysis GWR is not going to be the right tool for the job.
thanks for reply and valuable information. So its mean we cannot apply GWR if we have very few number of observation geographically.
I have further questions and hope you will reply me.
1) Does GWR is doing interpolation? 2) How much geographical area covered by GWR? For example: If we put the geographical coordinates corresponds to more than 80 observations for particular area and apply GWR then how much neighbour area is covered by GWR. 3) Is it possible that we can use constant geographical coordinates for particular station/site? For example: we have 5 stations and 30 number of observations for each station (N=150 for all stations). If we use same/constant geographical coordinate of one station for 30 observation then change the coordinates for the according to other station for next 30 observations and so on. So this method will work or not.
I hope you will understand my question and if not then please let me know.I will try to explain you.
You are right, you do need at least 80 observations to use GWR (the more the better). Unfortunately, having 30 observations at each of the 5 stations does not do the trick. You really need to have more stations. And the geographic extent of the GWR analysis is always going to be the same as the geographic extent of the input features. GWR is not an interpolation method. If you want to do interpolation, you may want to think about something like Kriging, in either Spatial Analyst or Geostatistical Analyst.