[R-sig-Geo] sp.correlogram

Wed Jun 30 18:30:42 CEST 2010

On Mon, 28 Jun 2010, Robert Pazur wrote:

> Dear all,
>
> according to perfect recommendations I perform moran.test  from "spdep"
> package. My datasets - csv X,Y tables has 90 000 rows representing 90 000
> regular points. In order to obtain morans correlogram I did a batch run and
> my plan is to obtain autocorrelation values up to 3000 meters
> distance.  However according to my hardware limitations I  just came to to
> 170 meters value of the dnearneigh -neighbours of region points by Euclidean
> distance function (dnr_170 <- dnearneigh(pointd_r, 0, 170). After this I
> receive well know memory allocation error problem. R in my machine runs on
> 32-bit Windows xp version, with just 2 GB of RAM. Is there some way to
> overcome this issue?

Are you perhaps not deleting the previous large objects that you have 
created? Look at ls(), and use rm() and gc() to get rid of unneeded 
objects. If you are careful, you may do save() before rm(), to have a copy 
of the nb or listw object for each step. If I recall, your points are in a 
grid 20m apart, so you might also consider using other approaches to this.

Roger

> Thanks in advance.
>
> Robert Pazur
>
>
> -------------------------------------------------------
> Robert Pazur
> PhD student
> Institute of Geography
> Slovak Academy Of Sciences
>
> Mobile : +421 948 001 705
> Skype  : ruegdeg
>
>
> 2010/6/24 Robert Pazur <pazurrobert at gmail.com>
>
>> Thank you very much for wonderful reply! Its exactly what i was looking for
>> last couple of hours.
>> Its incredible smart tool!
>> Kind regards,
>> Robert.
>>
>>
>>
>>
>>
>>
>> 2010/6/24 Roger Bivand <Roger.Bivand at nhh.no>
>>
>> On Thu, 24 Jun 2010, Robert Pazur wrote:
>>>
>>>  Dear all,
>>>>
>>>> I would like to perform Moran'I correlogram (sp.correlogram method in
>>>> spdep
>>>> package) based on euclidian fixed distances  but I have following
>>>> problem:
>>>> I created an artificial table, containing long and lati of regular points
>>>>
>>>>> points <-read.table("http://www.scandinavia.sk/data/moran5.csv",
>>>>> sep=",",
>>>>>
>>>> header=T)
>>>> following the manual I also identified neighbours of region
>>>>
>>>>> dnb <- dnearneigh(as.matrix(points$long, points$lati), 0, 20, longlat=T)
>>>>>
>>>>
>>> No, from your helpful link to the data, you have projected coordinates,
>>> not geographical. In addition, your use of as.matrix() instead of cbind()
>>> has bad consequences:
>>>
>>> str(as.matrix(points$long, points$lati))
>>> str(cbind(points$long, points$lati))
>>>
>>> dnearneigh() will be revised to trap this.
>>>
>>> Had you said:
>>>
>>> coordinates(points) <- c("long", "lati")
>>>
>>> then:
>>>
>>> proj4string(points) <- CRS("+proj=longlat")
>>>
>>> you would have seen the problem, because the sp classes check for the
>>> bounds on objects.
>>>
>>> So after doing:
>>>
>>>
>>> points <-read.table("http://www.scandinavia.sk/data/moran5.csv", sep=",",
>>>  header=T)
>>> coordinates(points) <- c("long", "lati")
>>> dnb <- dnearneigh(points, 0, 20)
>>>
>>> you are good to go. Next step - how to replicate the ArcGIS Moran's I - is
>>> easy with the correct dnb:
>>>
>>> moran.test(points$GRID_CODE, listw=nb2listw(dnb, style="B"))
>>>
>>> You might use correlog() in pgirmess for distance bins, but you'll have
>>> more control over the bin boundaries by makin new sets of neighbours for
>>> your chosen bin thresholds.
>>>
>>> Hope this helps (and thank you for reverting to the list after writing to
>>> me directly 70 minutes earlier. List is always best).
>>>
>>> Roger
>>>
>>>
>>>  neighbours list
>>>>
>>>>> ME200.listw <- nb2listw(dnb, style="W", zero.policy=T)
>>>>>
>>>> but if I perform sp.correlogram function:
>>>>
>>>>> correl<-sp.correlogram(dnb, points$GRID_CODE, order = 2, method = "I",
>>>>>
>>>> style = "W", randomisation = TRUE, zero.policy = TRUE, spChk=NULL)
>>>> my results are :
>>>> Spatial correlogram for points$GRID_CODE
>>>> method: Moran's I
>>>>   estimate expectation   variance standard deviate Pr(I) two sided
>>>> 1 -0.0029855  -0.0344828  0.0019674           0.7101          0.4776
>>>> 2 -0.0044436  -0.0344828  0.0022585           0.6321          0.5273
>>>>
>>>> and if i perform this part of this task in Arcgis for the same point
>>>> shapefile Moran Calculation for Fixed distance band, Euclidian distance a
>>>> and 20m threshold, result of Moran coefficient is
>>>> (SpatialAutocorrelation moran GRID_CODE false "Fixed Distance Band"
>>>> "Euclidean Distance" None 20 # 0 0 0) results are:
>>>> Global Moran's I Summary
>>>> Moran's Index:   0.746511
>>>> Expected Index:  -0.003521
>>>> Variance:        0.001827
>>>> Z Score:         17.545122
>>>> p-value:         0.000000
>>>>
>>>> I would like to perform the same task like in Arcgis but for multiple
>>>> distances. However Arcgis cannot deal with large data with multiple
>>>> points,
>>>> thatswhy I
>>>> would like to use R. Its seems to me much better software, but
>>>> unfortunatelly I never use it (but I really want)
>>>> If you could give me some advice i will be very happy.
>>>>
>>>> Robert.
>>>>
>>>>
>>>> -------------------------------------------------------
>>>> Robert Pazur
>>>> PhD student
>>>> Institute of Geography
>>>> Slovak Academy Of Sciences
>>>>
>>>> Mobile : +421 948 001 705
>>>> Skype  : ruegdeg
>>>>
>>>>        [[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
>>>
>>>
>>
>

-- 
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