[R-sig-Geo] GWR Analysis

Tue Jun 15 12:37:39 CEST 2010

On Tue, 15 Jun 2010, Pinar Aslantas Bostan wrote:

>
>
> Dear Roger,
>
> Thank you for your advice. I performed lm() and predict(),the same problem is not produced that time. I created an attribute (voronoi) and set as factor for station and grid datasets. (I created voronoi polygons for station data. Each station has a voronoi code such a primary key variable. Then I made join operation between voronoi and grid, so each point at the grid data takes the attributes of the voronoi polygon if it falls inside. So that station and grid has the same voronoi codes) After setting voronoi attribute as factor, GWR analysis has worked and gave results. But this time it gives a warning message:
> In points2grid(points, tolerance, round, fuzz.tol) :
>  grid has empty column/rows in dimension 1'.
>
> Is this an important warning, do I have to consider it carefully?

No, this occurs when the representation is changed to grid, and where some 
complete columns or rows are absent. This can occur for good reason, say 
in an archipelago, but the warning is given in case the analyst made a 
mistake, and that absent grid rows or columns were not expected.

Roger

>
> ###########################################################################
>> station$VORONOI<-factor(station$VORONOI)
>> grid$VORONOI<-factor(grid$VORONOI)
>> gridded(grid) <- TRUE
> Warning message:
> In points2grid(points, tolerance, round, fuzz.tol) :
>  grid has empty column/rows in dimension 1
>
>> bw=gwr.sel(PREC~Z+V1+V2,data=station,adapt=T)
>> xx<-gwr(PREC~Z+V1+V2,station,adapt=bw,hatmatrix=TRUE)
>> x <-gwr(PREC~Z+V1+V2,data=station,adapt=bw, fit.points = grid, predict=T, se.fit=T, fittedGWRobject=xx)
> Warning message:
> In points2grid(points, tolerance, round, fuzz.tol) :
>  grid has empty column/rows in dimension 1
>
>> gwrres<-x$SDF
>> names(gwrres)
> [1] "sum.w"              "(Intercept)"        "Z"                  "V1"
> [5] "V2"                 "(Intercept)_se"     "Z_se"               "V1_se"
> [9] "V2_se"              "pred"               "pred.se"            "localR2"
> [13] "(Intercept)_se_EDF" "Z_se_EDF"           "V1_se_EDF"          "V2_se_EDF"
> [17] "pred.se_EDF"
>
> Alinti Roger Bivand <Roger.Bivand at nhh.no>
>
>> On Fri, 11 Jun 2010, Pinar Aslantas Bostan wrote:
>>
>>>
>>>
>>> Dear Roger,
>>>
>>> I tried with data frames instead of SPDF. I gave some details of
>>> names and class types of datasets. Grid data has 31204 rows (first
>>> row is column names) and 5 columns (Z,X,Y,V1,V2). Station data has
>>> 226 rows (first row is column names) and 6 columns
>>> (PREC,Z,X,Y,V1,V2). I gave an example about datasets below.
>>
>> Please do try to read up on what is going on. Find out how variables
>> may be treated as string not numeric on input, and converted into
>> factors. Do report str() of the imported objects as I said last time.
>> This will show if some numeric are really factors - this may happen
>> for example if the locale thinks that ",", not "." is the decimal
>> sign, or if there is any non-numeric character in a column. If need
>> be put your data (or a subset with the same problem on a website and
>> post the link. Please do try to use lm() and predict() on the lm
>> output object to see whether the problem is reproduced there too.
>>
>> Please also be aware that using GWR for prediction has no good basis
>> anywhere for anything - and the standard errors should not be given
>> any credibility. This is not what GWR is for at all.
>>
>> Roger
>>
>>>
>>> Grid data
>>> Z   X   Y   V1  V2
>>> 495 259725.32564500000 5010656.22353999000 44.15584946 53525.25605680000
>>> 621 264725.32564499800 5010656.22353999000 50.67575455 50139.84515760000
>>> 432 274725.32564499800 5010656.22353999000 58.62472534 44295.66734330000
>>> 418 279725.32564499800 5010656.22353999000 45.05175400 41966.14328090000
>>> 361 254725.32564500000 5005656.22353999000 37.61386871 60596.18505359990
>>>
>>> Station data
>>> PREC  Z X   Y   V1  V2
>>> 514.3522848 1039 410428.05168500000 4478786.07400999000 100.69618225
>>> 81321.18533050000
>>> 650.6115948 29 942127.25593400000 4409537.10120000000 7.64556551
>>> 34796.24442280000
>>> 690.1826088 614 1201166.68404000000 4496907.09750000000 39.14785004
>>> 212666.95252600000
>>> 427.7718516 1213 1080214.69930000000 4550293.99677000000 53.82775879
>>> 164788.61213900000
>>> 402.9552240 1197 525196.23761299900 4619486.48904000000 69.23021698
>>> 206901.35695100000
>>>
>>> And from the list Binbin lu send me a message, according to his
>>> advice I don't write "station$" term while performing gwr.sel and
>>> gwr functions. In this way gwr function has worked but gave a
>>> warning message. And when I plot the predicted precipitation values
>>> of the grid there are some negative values which should't be occur!
>>> It didn't calculate predicted errors also.
>>>
>>> What do you mean with 'factor' and I don't know how to run the
>>> function under debug.
>>>
>>> Thanks for your help.
>>> Pinar
>>>
>>> ###########################################################################
>>>> grid<-read.table("D:\\R\\dem.txt", header=TRUE)
>>>> station<-read.table("D:\\R\\station.txt", header=TRUE)
>>>
>>>> class(grid)
>>> [1] "data.frame"
>>>
>>>> class(station)
>>> [1] "data.frame"
>>>
>>>> names(grid)
>>> [1] "Z"  "X"  "Y"  "V1" "V2"
>>>
>>>> names(station)
>>> [1] "PREC" "Z"    "X"    "Y"    "V1"   "V2"
>>>
>>>> coordinates(station) <- c("X", "Y")
>>>> coordinates(grid) <- c("X", "Y")
>>>
>>>> bw=gwr.sel(PREC~Z+V1+V2,data=station,adapt=T)
>>>> gwr <-gwr(PREC~Z+V1+V2,data=station,adapt=bw, fit.points = grid,
>>>> predict=T, se.fit=T, hatmatrix=T)
>>>
>>> Warning message:
>>> In gwr(PREC ~ Z + V1 + V2, data = station, adapt = bw, fit.points = grid,  :
>>> standard errors set to NA, normalised RSS not available
>>>
>>> Alinti Roger Bivand <Roger.Bivand at nhh.no>
>>>
>>>> On Fri, 11 Jun 2010, Pinar Aslantas Bostan wrote:
>>>>
>>>>>
>>>>>
>>>>> Dear all,
>>>>>
>>>>> I want to make GWR analysis to predict precipitation distribution
>>>>> measured from 225 meteorological stations. I have three independent
>>>>> variables (Z, V1, V2). I want to obtain predictions on the grid
>>>>> dataset which has 31203 number of pixels. I gave some details about
>>>>> my datasets below. While performing GWR, I get error message: "new
>>>>> data matrix rows mismatch"
>>>>>
>>>>> How can I solve that problem?
>>>>
>>>> You need to provide the detailed output of traceback(), and probably
>>>> also run gwr() under debug() to see whether this is a data problem or
>>>> a code problem. Does gwr() work when predict=FALSE, and/or
>>>> predict=TRUE, se.fit=FALSE? The error message is generated when the
>>>> number of columns in the matrix of X variables is not the same in
>>>> data and and fit.points. Are the variables in both cases stored in
>>>> the same way? What do str(station.grid) and str(station) look like
>>>> before conversion to SPDF and after? Is one variable a factor in one
>>>> and numeric in the other? Running under debug() will let you check
>>>> what x and predx look like.
>>>>
>>>> If this sounds hard, consider putting your data on a website and
>>>> posting the link.
>>>>
>>>> Hope this helps,
>>>>
>>>> Roger
>>>>
>>>>>
>>>>> Thanks, Pinar.
>>>>>
>>>>> station.grid<-read.table("D:\\R\\dem.txt", header=TRUE)
>>>>> station<-read.table("D:\\R\\station.txt", header=TRUE)
>>>>>
>>>>> grid = SpatialPointsDataFrame(data=station.grid,
>>>>> coords=cbind(station.grid$X, station.grid$Y))
>>>>
>>>>> station = SpatialPointsDataFrame(data=station,
>>>>> coords=cbind(station$X,station$Y))
>>>>>
>>>>>> names(grid)
>>>>> [1] "Z" "X" "Y" "V1" "V2"
>>>>>> names(station)
>>>>> [1] "PREC" "Z" "X" "Y" "V1" "V2"
>>>>>
>>>>> bw=gwr.sel(PREC~station$Z+station$V1+station$V2,data=station,adapt=T)
>>>>
>>>>> gwr <-gwr(PREC ~ station$Z +
>>>> station$V1 + station$V2, data=station, adapt=bw, fit.points = grid,
>>>> predict=TRUE, se.fit=T)
>>>>>
>>>>> Error in gwr(PREC ~ station$Z + station$V1 + station$V2, data = station, :
>>>>> NEW DATA MATRiX ROWS MiSMATCH
>>>>>
>>>>> ----------------------------------------------------------------
>>>>>
>>>>>
>>>>>        [[alternative HTML version deleted]]
>>>>>
>>>>> _______________________________________________
>>>>> R-sig-Geo mailing list
>>>>> R-sig-Geo at stat.math.ethz.ch
>>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>>
>>>>
>>>> --
>>>> Roger Bivand
>>>> Economic Geography Section, Department of Economics, Norwegian School of
>>>> Economics and Business Administration, Helleveien 30, N-5045 Bergen,
>>>> Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
>>>> e-mail: Roger.Bivand at nhh.no
>>>>
>>>>
>>>
>>>
>>>
>>> ----------------------------------------------------------------
>>> This message was sent using IMP, the Internet Messaging Program.
>>>
>>
>> --
>> Roger Bivand
>> Economic Geography Section, Department of Economics, Norwegian School of
>> Economics and Business Administration, Helleveien 30, N-5045 Bergen,
>> Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
>> e-mail: Roger.Bivand at nhh.no
>>
>>
>
>
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no