1
|
###################################################################################
|
2
|
### R code to aquire and process MOD06_L2 cloud data from the MODIS platform
|
3
|
|
4
|
|
5
|
## connect to server of choice
|
6
|
#system("ssh litoria")
|
7
|
#R
|
8
|
|
9
|
library(sp)
|
10
|
library(spgrass6)
|
11
|
library(rgdal)
|
12
|
library(reshape)
|
13
|
library(ncdf4)
|
14
|
library(geosphere)
|
15
|
library(rgeos)
|
16
|
library(multicore)
|
17
|
library(raster)
|
18
|
library(lattice)
|
19
|
library(rgl)
|
20
|
library(hdf5)
|
21
|
library(rasterVis)
|
22
|
library(heR.Misc)
|
23
|
library(car)
|
24
|
|
25
|
X11.options(type="Xlib")
|
26
|
ncores=20 #number of threads to use
|
27
|
|
28
|
setwd("/home/adamw/acrobates/projects/interp")
|
29
|
|
30
|
psin=CRS("+proj=sinu +lon_0=0 +x_0=0 +y_0=0 +a=6371007.181 +b=6371007.181 +units=m +no_defs")
|
31
|
|
32
|
## get MODLAND tile information
|
33
|
tb=read.table("http://landweb.nascom.nasa.gov/developers/sn_tiles/sn_bound_10deg.txt",skip=6,nrows=648,header=T)
|
34
|
tb$tile=paste("h",sprintf("%02d",tb$ih),"v",sprintf("%02d",tb$iv),sep="")
|
35
|
save(tb,file="modlandTiles.Rdata")
|
36
|
|
37
|
tile="h11v08" #can move this to submit script if needed
|
38
|
#tile="h09v04" #oregon
|
39
|
|
40
|
tile_bb=tb[tb$tile==tile,] ## identify tile of interest
|
41
|
roi_ll=extent(tile_bb$lon_min,tile_bb$lon_max,tile_bb$lat_min,tile_bb$lat_max)
|
42
|
#roi=spTransform(roi,psin)
|
43
|
#roil=as(roi,"SpatialLines")
|
44
|
|
45
|
dmod06="data/modis/mod06/summary"
|
46
|
|
47
|
|
48
|
##########################
|
49
|
#### explore the data
|
50
|
|
51
|
months=seq(as.Date("2000-01-15"),as.Date("2000-12-15"),by="month")
|
52
|
|
53
|
getmod06<-function(variable){
|
54
|
d=brick(list.files(dmod06,pattern=paste("MOD06_",tile,".nc",sep=""),full=T),varname=toupper(variable))
|
55
|
projection(d)=psin
|
56
|
setZ(d,format(as.Date(d@z$Date),"%m"),name="time")
|
57
|
# d@z=list(months)
|
58
|
layerNames(d) <- as.character(format(as.Date(d@z$Date),"%b"))
|
59
|
return(d)
|
60
|
}
|
61
|
|
62
|
cer=getmod06("cer")
|
63
|
cld=getmod06("cld")
|
64
|
cot=getmod06("cot")
|
65
|
|
66
|
pcol=colorRampPalette(c("brown","red","yellow","darkgreen"))
|
67
|
#levelplot(cer,col.regions=pcol(20))
|
68
|
|
69
|
## load WorldClim data for comparison (download then uncompress)
|
70
|
#system("wget -P data/worldclim/ http://biogeo.ucdavis.edu/data/climate/worldclim/1_4/grid/cur/prec_30s_bil.zip",wait=F)
|
71
|
#system("wget -P data/worldclim/ http://biogeo.ucdavis.edu/data/climate/worldclim/1_4/grid/cur/alt_30s_bil.zip",wait=F)
|
72
|
|
73
|
### load WORLDCLIM data for comparison
|
74
|
wc=stack(list.files("data/worldclim/prec_30s_bil/",pattern="bil$",full=T)[c(4:12,1:3)])
|
75
|
projection(wc)=CRS("+proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0")
|
76
|
wc=crop(wc,roi_ll)
|
77
|
wc[wc==55537]=NA
|
78
|
wc=projectRaster(wc,cer)#crs=projection(psin))
|
79
|
setZ(wc,months,name="time")
|
80
|
wc@z=list(months)
|
81
|
layerNames(wc) <- as.character(format(months,"%b"))
|
82
|
writeRaster(wc,file=paste("data/tiles/",tile,"/worldclim_",tile,".tif",sep=""),format="GTiff")
|
83
|
|
84
|
### load WORLDCLIM elevation
|
85
|
dem=raster(list.files("data/worldclim/alt_30s_bil/",pattern="bil$",full=T))
|
86
|
projection(dem)=CRS("+proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0")
|
87
|
dem=crop(dem,roi_ll)
|
88
|
dem[dem>60000]=NA
|
89
|
dem=projectRaster(dem,cer)
|
90
|
writeRaster(dem,file=paste("data/tiles/",tile,"/dem_",tile,".tif",sep=""),format="GTiff")
|
91
|
|
92
|
|
93
|
### get station data, subset to stations in region, and transform to sinusoidal
|
94
|
dm=readOGR(paste("data/tiles/",tile,sep=""),paste("station_monthly_",tile,"_PRCP",sep=""))
|
95
|
xyplot(latitude~longitude|month,data=dm@data)
|
96
|
dm2=spTransform(dm,CRS(projection(cer)))
|
97
|
dm2@data[,c("x","y")]=coordinates(dm2)
|
98
|
|
99
|
### extract MOD06 data for each station
|
100
|
stcer=extract(cer,dm2,fun=mean);colnames(stcer)=paste("cer_mean_",as.numeric(format(as.Date(cer@z$Date),"%m")),sep="")
|
101
|
#stcer20=extract(cer20,st2)#;colnames(stcer)=paste("cer_mean_",1:12,sep="")
|
102
|
stcot=extract(cot,dm2);colnames(stcot)=paste("cot_mean_",as.numeric(format(as.Date(cot@z$Date),"%m")),sep="")
|
103
|
stcld=extract(cld,dm2);colnames(stcld)=paste("cld_mean_",as.numeric(format(as.Date(cld@z$Date),"%m")),sep="")
|
104
|
stdem=extract(dem,dm2)
|
105
|
mod06=cbind.data.frame(station=dm$station,stcer,stcot,stcld)
|
106
|
mod06l=melt(mod06,id.vars=c("station"));colnames(mod06l)[grep("value",colnames(mod06l))]="mod06"
|
107
|
mod06l[,c("variable","moment","month")]=do.call(rbind,strsplit(as.character(mod06l$variable),"_"))
|
108
|
mod06l=unique(mod06l)
|
109
|
mod06l=cast(mod06l,station+moment+month~variable,value="mod06")
|
110
|
mod06l=merge(dm2@data,mod06l,by=c("station","month"))
|
111
|
mod06l=mod06l[!is.na(mod06l$cer),]
|
112
|
|
113
|
#mod06l=melt(mod06,id.vars=c("station","longitude","latitude","elevation","month","count","value"))
|
114
|
#mod06l[,c("variable","moment","month2")]=do.call(rbind,strsplit(as.character(mod06l$variable),"_"))
|
115
|
#mod06l=as.data.frame(cast(mod06l,station+longitude+latitude+month~variable,value="value.1"))
|
116
|
#mod06l=mod06l[mod06l$month==mod06l$month2&!is.na(mod06l$value.1),]
|
117
|
mod06l=mod06l[order(mod06l$month),]
|
118
|
|
119
|
xyplot(value~cer|month,data=mod06l,scales=list(relation="free"),pch=16,cex=.5)
|
120
|
xyplot(value~cer|station,data=mod06l[mod06l$count>400,],pch=16,cex=.5)
|
121
|
|
122
|
xyplot(cot~month,groups=station,data=mod06l,type="l")
|
123
|
|
124
|
|
125
|
## explore fit of simple model
|
126
|
m=11
|
127
|
cor(mod06l[mod06l$month==m,c("value","cer","cld","cot")])
|
128
|
lm1=lm(value~latitude+longitude+elevation+cer+cld+cot,data=mod06l[mod06l$month==m,])
|
129
|
summary(lm1)
|
130
|
crPlots(lm1)
|
131
|
|
132
|
plot(mod06l$value[mod06l$month==m],as.vector(predict(lm1,data=mod06l[mod06l$month==m,])))
|
133
|
|
134
|
### draw some plots
|
135
|
gq=function(x,n=10,cut=F) {
|
136
|
if(!cut) return(unique(quantile(x,seq(0,1,len=n+1),na.rm=T)))
|
137
|
if(cut) return(cut(x,unique(quantile(x,seq(0,1,len=n+1),na.rm=T))))
|
138
|
}
|
139
|
|
140
|
### add some additional variables
|
141
|
mod06s$month=factor(mod06s$month,labels=format(as.Date(paste("2000",1:12,"15",sep="-")),"%b"))
|
142
|
mod06s$lppt=log(mod06s$ppt)
|
143
|
mod06s$glon=cut(mod06s$lon,gq(mod06s$lon,n=5),include.lowest=T,ordered=T)#gq(mod06s$lon,n=3))
|
144
|
mod06s$glon2=cut(mod06s$lon,breaks=c(-125,-122,-115),labels=c("Coastal","Inland"),include.lowest=T,ordered=T)#gq(mod06s$lon,n=3))
|
145
|
mod06s$gelev=cut(mod06s$elev,breaks=gq(mod06s$elev,n=3),labels=c("Low","Mid","High"),include.lowest=T,ordered=T)
|
146
|
mod06s$gbin=factor(paste(mod06s$gelev,mod06s$glon2,sep="_"),levels=c("Low_Coastal","Mid_Coastal","High_Coastal","Low_Inland","Mid_Inland","High_Inland"),ordered=T)
|
147
|
mod06s$LWP_mean=(2/3)*mod06s$CER_mean*mod06s$COT_mean
|
148
|
|
149
|
## melt it
|
150
|
mod06sl=melt(mod06s[,!grepl("lppt",colnames(mod06s))],id.vars=c("id","lon","lat","elev","month","ppt","glon","glon2","gelev","gbin"))
|
151
|
levels(mod06sl$variable)=c("Effective Radius (um)","Very Cloudy Days (%)","Cloudy Days (%)","Optical Thickness (%)","Liquid Water Path")
|
152
|
|
153
|
###################################################################
|
154
|
###################################################################
|
155
|
|
156
|
bgyr=colorRampPalette(c("blue","green","yellow","red"))
|
157
|
|
158
|
X11.options(type="cairo")
|
159
|
pdf("output/MOD06_summary.pdf",width=11,height=8.5)
|
160
|
|
161
|
# % cloudy maps
|
162
|
title="Cloudiness (% cloudy days) "
|
163
|
at=unique(quantile(as.matrix(cld),seq(0,1,len=100),na.rm=T))
|
164
|
p=levelplot(cld,xlab.top=title,at=at,col.regions=bgyr(length(at)))+layer(sp.lines(roil, lwd=1.2, col='black'))
|
165
|
print(p)
|
166
|
#bwplot(cer,main=title,ylab="Cloud Effective Radius (microns)")
|
167
|
|
168
|
# CER maps
|
169
|
title="Cloud Effective Radius (microns)"
|
170
|
at=quantile(as.matrix(cer),seq(0,1,len=100),na.rm=T)
|
171
|
p=levelplot(cer,xlab.top=title,at=at,col.regions=bgyr(length(at)))+layer(sp.lines(roil, lwd=1.2, col='black'))
|
172
|
print(p)
|
173
|
#bwplot(cer,main=title,ylab="Cloud Effective Radius (microns)")
|
174
|
|
175
|
# COT maps
|
176
|
title="Cloud Optical Thickness (%)"
|
177
|
at=quantile(as.matrix(cot),seq(0,1,len=100),na.rm=T)
|
178
|
p=levelplot(cot,xlab.top=title,at=at,col.regions=bgyr(length(at)))+layer(sp.lines(roil, lwd=0.8, col='black'))
|
179
|
print(p)
|
180
|
#bwplot(cot,xlab.top=title,ylab="Cloud Optical Thickness (%)")
|
181
|
|
182
|
###########################################
|
183
|
#### compare with PRISM data
|
184
|
|
185
|
at=quantile(as.matrix(subset(m01,subset=1)),seq(0,1,len=100),na.rm=T)
|
186
|
p1=levelplot(subset(m01,subset=1),xlab.top="Effective Radius (um)",at=at,col.regions=bgyr(length(at)),margin=F,
|
187
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
188
|
at=quantile(as.matrix(subset(m01,subset=2)),seq(0,1,len=100),na.rm=T)
|
189
|
p2=levelplot(subset(m01,subset=2),xlab.top="Cloudy days (%)",at=at,col.regions=bgyr(length(at)),margin=F,
|
190
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
191
|
at=quantile(as.matrix(subset(m01,subset=3)),seq(0,1,len=100),na.rm=T)
|
192
|
p3=levelplot(subset(m01,subset=3),xlab.top="Optical Thickness (%)",at=at,col.regions=bgyr(length(at)),margin=F,
|
193
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
194
|
at=quantile(as.matrix(subset(m01,subset=4)),seq(0,1,len=100),na.rm=T)
|
195
|
p4=levelplot(subset(m01,subset=4),xlab.top="PRISM MAP",at=at,col.regions=bgyr(length(at)),margin=F,
|
196
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
197
|
|
198
|
print(p1,split=c(1,1,2,2))
|
199
|
print(p2,split=c(1,2,2,2),new=F)
|
200
|
print(p3,split=c(2,1,2,2),new=F)
|
201
|
print(p4,split=c(2,2,2,2),new=F)
|
202
|
|
203
|
### compare COT and PRISM
|
204
|
print(p3,split=c(1,1,2,1))
|
205
|
print(p4,split=c(2,1,2,1),new=F)
|
206
|
|
207
|
|
208
|
### sample to speed up processing
|
209
|
s=sample(1:nrow(bd),10000)
|
210
|
|
211
|
## melt it to ease comparisons
|
212
|
bdl=melt(bd[s,],measure.vars=c("cer","cld","cot"))
|
213
|
|
214
|
combineLimits(useOuterStrips(xyplot(prism~value|variable+month,data=bdl,pch=16,cex=.2,scales=list(y=list(log=T),x=list(relation="free")),
|
215
|
ylab="PRISM Monthly mean precipitation (mm)",xlab="MOD06 metric",main="PRISM vs. MOD06 (mean monthly ppt)")))+
|
216
|
layer(panel.abline(lm(y~x),col="red"))+
|
217
|
layer(panel.text(0,2.5,paste("R2=",round(summary(lm(y~x))$r.squared,2))))
|
218
|
|
219
|
|
220
|
### Comparison at station values
|
221
|
at=quantile(as.matrix(cotm),seq(0,1,len=100),na.rm=T)
|
222
|
p=levelplot(cotm, layers=1,at=at,col.regions=bgyr(length(at)),main="Mean Annual Cloud Optical Thickness",FUN.margin=function(x) 0)+
|
223
|
layer(sp.lines(roil, lwd=1.2, col='black'))+layer(sp.points(st2_sin, pch=16, col='black'))
|
224
|
print(p)
|
225
|
|
226
|
### monthly comparisons of variables
|
227
|
#mod06sl=melt(mod06s,measure.vars=c("ppt","COT_mean","CER_mean","CER_P20um"))
|
228
|
#bwplot(value~month|variable,data=mod06sl,cex=.5,pch=16,col="black",scales=list(y=list(relation="free")),layout=c(1,3))
|
229
|
#splom(mod06s[grep("CER|COT|CLD|ppt",colnames(mod06s))],cex=.2,pch=16,main="Scatterplot matrix of MOD06 products")
|
230
|
|
231
|
### run some regressions
|
232
|
#plot(log(ppt)~COT_mean,data=mod06s)
|
233
|
#summary(lm(log(ppt)~COT_mean*month,data=mod06s))
|
234
|
|
235
|
## ppt~metric with longitude bins
|
236
|
xyplot(ppt~value|variable,groups=glon,data=mod06sl,
|
237
|
scales=list(y=list(log=T),x=list(relation="free",log=F)),
|
238
|
par.settings = list(superpose.symbol = list(col=bgyr(5),pch=16,cex=.5)),auto.key=list(space="top",title="Station Longitude"),
|
239
|
main="Comparison of MOD06_L2 and Precipitation Monthly Climatologies",ylab="Mean Monthly Station Precipitation (mm)",xlab="MOD06_L2 Product",layout=c(5,1))+
|
240
|
layer(panel.text(9,2.5,label="Coastal stations",srt=30,cex=1.3,col="blue"),columns=1)+
|
241
|
layer(panel.text(13,.9,label="Inland stations",srt=10,cex=1.3,col="red"),columns=1)+
|
242
|
layer(panel.abline(lm(y~x),col="red"))+
|
243
|
layer(panel.text(0,0,paste("R2=",round(summary(lm(y~x))$r.squared,2)),pos=4,cex=.5,col="grey"))
|
244
|
|
245
|
|
246
|
## ppt~metric with longitude bins
|
247
|
#CairoPNG("output/COT.png",width=10,height=5,units="in",dpi=300,pointsize=20)
|
248
|
#png("output/COT.png",width=10,height=5,units="in",res=150)
|
249
|
#trellis.par.set("fontsize",12)
|
250
|
at=quantile(as.matrix(subset(m01,subset=3)),seq(0,1,len=100),na.rm=T)
|
251
|
p1=levelplot(subset(m01,subset=3),xlab.top="Optical Thickness (%)",at=at,col.regions=bgyr(length(at)),margin=F,
|
252
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))+layer(sp.points(st2, cex=.5,col='black'))
|
253
|
at=quantile(as.matrix(subset(m01,subset=3)),seq(0,1,len=100),na.rm=T)
|
254
|
|
255
|
at=quantile(as.matrix(subset(m01,subset=4)),seq(0,1,len=100),na.rm=T)
|
256
|
p2=levelplot(subset(m01,subset=4),xlab.top="PRISM MAP",at=at,col.regions=bgyr(length(at)),margin=F,
|
257
|
)+layer(sp.lines(roi_geo, lwd=1.2, col='black'))+layer(sp.points(st2, cex=.5, col='black'))
|
258
|
|
259
|
p3=xyplot(ppt~value,groups=glon,data=mod06sl[mod06sl$variable=="Optical Thickness (%)",],
|
260
|
scales=list(y=list(log=T),x=list(relation="free",log=F)),
|
261
|
par.settings = list(superpose.symbol = list(col=bgyr(5),pch=16,cex=.3)),auto.key=list(space="right",title="Station \n Longitude"),
|
262
|
ylab="Mean Monthly Station Precipitation (mm)",xlab="Cloud Optical Thickness from MOD06_L2 (%)",layout=c(1,1))+
|
263
|
layer(panel.text(9,2.6,label="Coastal stations",srt=10,cex=1.3,col="blue"),columns=1)+
|
264
|
layer(panel.text(13,.95,label="Inland stations",srt=10,cex=1.3,col="red"),columns=1)
|
265
|
|
266
|
p4=xyplot(ppt~value,groups=glon,data=mod06sl[mod06sl$variable=="Very Cloudy Days (%)",],
|
267
|
scales=list(y=list(log=T),x=list(relation="free",log=F)),
|
268
|
par.settings = list(superpose.symbol = list(col=bgyr(5),pch=16,cex=.3)),auto.key=list(space="right",title="Station \n Longitude"),
|
269
|
ylab="Mean Monthly Station Precipitation (mm)",xlab="Proportion days with Cloud Effective Radius >20um from MOD06_L2 (%)",layout=c(1,1))+
|
270
|
layer(panel.text(9,2.6,label="Coastal stations",srt=10,cex=1.3,col="blue"),columns=1)+
|
271
|
layer(panel.text(13,.95,label="Inland stations",srt=10,cex=1.3,col="red"),columns=1)
|
272
|
|
273
|
#save(p1,p2,p3,file="plotdata.Rdata")
|
274
|
#load("plotdata.Rdata")
|
275
|
|
276
|
#CairoPDF("output/MOD06_Summaryfig.pdf",width=11,height=8.5)
|
277
|
print(p3,position=c(0,0,1,.5),save.object=F)
|
278
|
#print(p4,position=c(0,0,1,.5),save.object=F)
|
279
|
print(p1,split=c(1,1,2,2),new=F)
|
280
|
print(p2,split=c(2,1,2,2),new=F)
|
281
|
#dev.off()
|
282
|
#system("convert output/MOD06_Summaryfig.pdf output/MOD06_Summaryfig.png")
|
283
|
#dev.off()
|
284
|
|
285
|
## with elevation
|
286
|
# xyplot(ppt~value|variable,groups=gbin,data=mod06sl,
|
287
|
# scales=list(y=list(log=T),x=list(relation="free")),
|
288
|
# par.settings = list(superpose.symbol = list(col=c(rep("blue",3),rep("red",3)),pch=rep(c(3,4,8),2),cex=.5)),auto.key=list(space="right",title="Station Longitude"),
|
289
|
# main="Comparison of MOD06_L2 and Precipitation Monthly Climatologies",ylab="Precipitation",xlab="MOD06_L2 Product",layout=c(3,1))+
|
290
|
# layer(panel.text(9,2.5,label="Coastal stations",srt=30,cex=1.3,col="blue"),columns=1)+
|
291
|
# layer(panel.text(13,.9,label="Inland stations",srt=10,cex=1.3,col="red"),columns=1)
|
292
|
|
293
|
## with elevation and longitude bins
|
294
|
combineLimits(useOuterStrips(xyplot(ppt~value|variable+gbin,data=mod06sl,
|
295
|
scales=list(y=list(log=T),x=list(relation="free")),col="black",pch=16,cex=.5,type=c("p","r"),
|
296
|
main="Comparison of MOD06_L2 and Precipitation Monthly Climatologies",ylab="Precipitation",xlab="MOD06_L2 Product")))+
|
297
|
layer(panel.xyplot(x,y,type="r",col="red"))
|
298
|
|
299
|
## *** MOD06 vars vs precipitation by month, colored by longitude
|
300
|
combineLimits(useOuterStrips(xyplot(ppt~value|month+variable,groups=glon,data=mod06sl,cex=.5,pch=16,
|
301
|
scales=list(y=list(log=T),x=list(relation="free")),
|
302
|
par.settings = list(superpose.symbol = list(pch =16, col=bgyr(5),cex=1)),auto.key=list(space="top",title="Station Longitude"),
|
303
|
main="Comparison of MOD06_L2 and Precipitation Monthly Climatologies",ylab="Precipitation",xlab="MOD06_L2 Product")),
|
304
|
margin.x=1,adjust.labels=F)+
|
305
|
layer(panel.abline(lm(y~x),col="red"))+
|
306
|
layer(panel.text(0,0,paste("R2=",round(summary(lm(y~x))$r.squared,2)),pos=4,cex=.5,col="grey"))
|
307
|
|
308
|
|
309
|
xyplot(ppt~CLD_mean|id,data=mod06s,panel=function(x,y,group){
|
310
|
panel.xyplot(x,y,type=c("r"),cex=.5,pch=16,col="red")
|
311
|
panel.xyplot(x,y,type=c("p"),cex=.5,pch=16,col="black")
|
312
|
} ,scales=list(y=list(log=T)),strip=F,main="Monthly Mean Precipitation and % Cloudy by station",
|
313
|
sub="Each panel is a station, each point is a monthly mean",
|
314
|
ylab="Precipitation (mm, log axis)",xlab="% of Cloudy Days")+
|
315
|
layer(panel.text(.5,.5,round(summary(lm(y~x))$r.squared,2),pos=4,cex=.75,col="grey"))
|
316
|
|
317
|
xyplot(ppt~CER_mean|id,data=mod06s,panel=function(x,y,group){
|
318
|
panel.xyplot(x,y,type=c("r"),cex=.5,pch=16,col="red")
|
319
|
panel.xyplot(x,y,type=c("p"),cex=.5,pch=16,col="black")
|
320
|
} ,scales=list(y=list(log=T)),strip=F,main="Monthly Mean Precipitation and Cloud Effective Radius by station",sub="Each panel is a station, each point is a monthly mean",ylab="Precipitation (mm, log axis)",xlab="Mean Monthly Cloud Effective Radius (mm)")+
|
321
|
layer(panel.text(10,.5,round(summary(lm(y~x))$r.squared,2),pos=4,cex=.75,col="grey"))
|
322
|
|
323
|
xyplot(ppt~COT_mean|id,data=mod06s,panel=function(x,y,group){
|
324
|
panel.xyplot(x,y,type=c("r"),cex=.5,pch=16,col="red")
|
325
|
panel.xyplot(x,y,type=c("p"),cex=.5,pch=16,col="black")
|
326
|
} ,scales=list(y=list(log=T)),strip=F,main="Monthly Mean Precipitation and Cloud Optical Thickness by station",
|
327
|
sub="Each panel is a station, each point is a monthly mean \n Number in lower right of each panel is R^2",
|
328
|
ylab="Precipitation (mm, log axis)",xlab="Mean Monthly Cloud Optical Thickness (%)")+
|
329
|
layer(panel.text(10,.5,round(summary(lm(y~x))$r.squared,2),pos=4,cex=.75,col="grey"))
|
330
|
|
331
|
xyplot(ppt~LWP_mean|id,data=mod06s,panel=function(x,y,group){
|
332
|
panel.xyplot(x,y,type=c("r"),cex=.5,pch=16,col="red")
|
333
|
panel.xyplot(x,y,type=c("p"),cex=.5,pch=16,col="black")
|
334
|
} ,scales=list(y=list(log=T)),strip=F,main="Monthly Mean Precipitation and Liquid Water Path by station",
|
335
|
sub="Each panel is a station, each point is a monthly mean \n Number in lower right of each panel is R^2",
|
336
|
ylab="Precipitation (mm, log axis)",xlab="Mean Monthly Liquid Water Path")+
|
337
|
layer(panel.text(10,.5,round(summary(lm(y~x))$r.squared,2),pos=4,cex=.75,col="grey"))
|
338
|
|
339
|
### Calculate the slope of each line
|
340
|
mod06s.sl=dapply(mod06s,list(id=mod06s$id),function(x){
|
341
|
lm1=lm(log(x$ppt)~x$CER_mean,)
|
342
|
data.frame(lat=x$lat[1],lon=x$lon[1],elev=x$elev[1],intcpt=coefficients(lm1)[1],cer=coefficients(lm1)[2],r2=summary(lm1)$r.squared)
|
343
|
})
|
344
|
mod06s.sl$cex=gq(mod06s.sl$r2,n=5,cut=T)
|
345
|
mod06s.sl$cer.s=gq(mod06s.sl$cer,n=5,cut=T)
|
346
|
|
347
|
### and plot it on a map
|
348
|
xyplot(lat~lon,group=cer.s,data=mod06s.sl,par.settings = list(superpose.symbol = list(pch =16, col=bgyr(5),cex=1)),auto.key=list(space="right",title="Slope Coefficient"),asp=1,
|
349
|
main="Slopes of linear regressions {log(ppt)~CloudEffectiveRadius}")+
|
350
|
layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
351
|
|
352
|
### look for relationships with longitude
|
353
|
xyplot(cer~lon,group=cut(mod06s.sl$elev,gq(mod06s.sl$elev,n=5)),data=mod06s.sl,
|
354
|
par.settings = list(superpose.symbol = list(col=bgyr(5),pch=16,cex=1)),auto.key=list(space="right",title="Station Elevation"),
|
355
|
ylab="Slope of lm(ppt~EffectiveRadius)",xlab="Longitude",main="Precipitation~Effective Radius relationship by latitude")
|
356
|
|
357
|
|
358
|
############################################################
|
359
|
### simple regression to get spatial residuals
|
360
|
m="01"
|
361
|
mod06s2=mod06s#[mod06s$month==m,]
|
362
|
|
363
|
lm1=lm(log(ppt)~CER_mean*month*lon,data=mod06s2); summary(lm1)
|
364
|
mod06s2$pred=exp(predict(lm1,mod06s2))
|
365
|
mod06s2$resid=mod06s2$pred-mod06s2$ppt
|
366
|
mod06s2$residg=gq(mod06s2$resid,n=5,cut=T)
|
367
|
mod06s2$presid=mod06s2$resid/mod06s2$ppt
|
368
|
|
369
|
for(l in c(F,T)){
|
370
|
## all months
|
371
|
xyplot(pred~ppt,groups=gelev,data=mod06s2,
|
372
|
par.settings = list(superpose.symbol = list(col=bgyr(3),pch=16,cex=.75)),auto.key=list(space="right",title="Station Elevation"),
|
373
|
scales=list(log=l),
|
374
|
ylab="Predicted Mean Monthly Precipitation (mm)",xlab="Observed Mean Monthly Precipitation (mm)",main="Predicted vs. Observed for Simple Model",
|
375
|
sub="Red line is y=x")+
|
376
|
layer(panel.abline(0,1,col="red"))
|
377
|
|
378
|
## month by month
|
379
|
print(xyplot(pred~ppt|month,groups=gelev,data=mod06s2,
|
380
|
par.settings = list(superpose.symbol = list(col=bgyr(3),pch=16,cex=.75)),auto.key=list(space="right",title="Station Elevation"),
|
381
|
scales=list(log=l),
|
382
|
ylab="Predicted Mean Monthly Precipitation (mm)",xlab="Observed Mean Monthly Precipitation (mm)",main="Predicted vs. Observed for Simple Model",
|
383
|
sub="Red line is y=x")+
|
384
|
layer(panel.abline(0,1,col="red"))
|
385
|
)}
|
386
|
|
387
|
## residuals by month
|
388
|
xyplot(lat~lon|month,group=residg,data=mod06s2,
|
389
|
par.settings = list(superpose.symbol = list(pch =16, col=bgyr(5),cex=.5)),
|
390
|
auto.key=list(space="right",title="Residuals"),
|
391
|
main="Spatial plot of monthly residuals")+
|
392
|
layer(sp.lines(roi_geo, lwd=1.2, col='black'))
|
393
|
|
394
|
|
395
|
dev.off()
|
396
|
|
397
|
|
398
|
####################################
|
399
|
#### build table comparing various metrics
|
400
|
mods=data.frame(
|
401
|
models=c(
|
402
|
"log(ppt)~CER_mean",
|
403
|
"log(ppt)~CLD_mean",
|
404
|
"log(ppt)~COT_mean",
|
405
|
"log(ppt)~CER_mean*month",
|
406
|
"log(ppt)~CLD_mean*month",
|
407
|
"log(ppt)~COT_mean*month",
|
408
|
"log(ppt)~CER_mean*month*lon",
|
409
|
"log(ppt)~CLD_mean*month*lon",
|
410
|
"log(ppt)~COT_mean*month*lon",
|
411
|
"ppt~CER_mean*month*lon",
|
412
|
"ppt~CLD_mean*month*lon",
|
413
|
"ppt~COT_mean*month*lon"),stringsAsFactors=F)
|
414
|
|
415
|
mods$r2=
|
416
|
do.call(rbind,lapply(1:nrow(mods),function(i){
|
417
|
lm1=lm(as.formula(mods$models[i]),data=mod06s2)
|
418
|
summary(lm1)$r.squared}))
|
419
|
|
420
|
mods
|
421
|
|
422
|
|
423
|
|
424
|
|
425
|
|
426
|
|
427
|
|
428
|
|
429
|
load("data/modis/pointsummary.Rdata")
|
430
|
|
431
|
|
432
|
dsl=melt(ds,id.vars=c("id","date","ppt","lon","lat"),measure.vars= c("Cloud_Water_Path","Cloud_Effective_Radius","Cloud_Optical_Thickness"))
|
433
|
|
434
|
dsl=dsl[!is.nan(dsl$value),]
|
435
|
|
436
|
|
437
|
|
438
|
|
439
|
####
|
440
|
## mean annual precip
|
441
|
dp=d[d$variable=="ppt",]
|
442
|
dp$year=format(dp$date,"%Y")
|
443
|
dm=tapply(dp$value,list(id=dp$id,year=dp$year),sum,na.rm=T)
|
444
|
dms=apply(dm,1,mean,na.rm=T)
|
445
|
dms=data.frame(id=names(dms),ppt=dms/10)
|
446
|
|
447
|
dslm=tapply(dsl$value,list(id=dsl$id,variable=dsl$variable),mean,na.rm=T)
|
448
|
dslm=data.frame(id=rownames(dslm),dslm)
|
449
|
|
450
|
dms=merge(dms,dslm)
|
451
|
dmsl=melt(dms,id.vars=c("id","ppt"))
|
452
|
|
453
|
summary(lm(ppt~Cloud_Effective_Radius,data=dms))
|
454
|
summary(lm(ppt~Cloud_Water_Path,data=dms))
|
455
|
summary(lm(ppt~Cloud_Optical_Thickness,data=dms))
|
456
|
summary(lm(ppt~Cloud_Effective_Radius+Cloud_Water_Path+Cloud_Optical_Thickness,data=dms))
|
457
|
|
458
|
|
459
|
#### draw some plots
|
460
|
#pdf("output/MOD06_summary.pdf",width=11,height=8.5)
|
461
|
png("output/MOD06_summary_%d.png",width=1024,height=780)
|
462
|
|
463
|
## daily data
|
464
|
xyplot(value~ppt/10|variable,data=dsl,
|
465
|
scales=list(relation="free"),type=c("p","r"),
|
466
|
pch=16,cex=.5,layout=c(3,1))
|
467
|
|
468
|
|
469
|
densityplot(~value|variable,groups=cut(dsl$ppt,c(0,50,100,500)),data=dsl,auto.key=T,
|
470
|
scales=list(relation="free"),plot.points=F)
|
471
|
|
472
|
## annual means
|
473
|
|
474
|
xyplot(value~ppt|variable,data=dmsl,
|
475
|
scales=list(relation="free"),type=c("p","r"),pch=16,cex=0.5,layout=c(3,1),
|
476
|
xlab="Mean Annual Precipitation (mm)",ylab="Mean value")
|
477
|
|
478
|
densityplot(~value|variable,groups=cut(dsl$ppt,c(0,50,100,500)),data=dmsl,auto.key=T,
|
479
|
scales=list(relation="free"),plot.points=F)
|
480
|
|
481
|
|
482
|
## plot some swaths
|
483
|
|
484
|
nc1=raster(fs$path[3],varname="Cloud_Effective_Radius")
|
485
|
nc2=raster(fs$path[4],varname="Cloud_Effective_Radius")
|
486
|
nc3=raster(fs$path[5],varname="Cloud_Effective_Radius")
|
487
|
|
488
|
nc1[nc1<=0]=NA
|
489
|
nc2[nc2<=0]=NA
|
490
|
nc3[nc3<=0]=NA
|
491
|
|
492
|
plot(roi)
|
493
|
plot(nc3)
|
494
|
|
495
|
plot(nc1,add=T)
|
496
|
plot(nc2,add=T)
|
497
|
|
498
|
|
499
|
dev.off()
|
500
|
|
501
|
|