library(lme4)
library(devtools)
library(knitr)
library(ggplot2)
library(plyr)
opts_chunk$set(comment=NA)#utility functions, load (source) from my saved gist
source_gist("https://gist.github.com/anarosner/ba285306fc0ce9d812a5", sha1="b25a1b73e02cc2b2d2c590f6c0b2c9c9945fa980")
model_dir<-"C:/ALR/Models/MetToFlow"
model_data_dir<-"C:/ALR/Models_processed_data"
#load (source) this project's general functions
setwd(file.path(model_dir,"A_get_data"))
purl(input="0_general_functions.Rmd",output="0_general_functions.R",documentation=0)
source("0_general_functions.R")
file.remove("0_general_functions.R")
season.names<-c("winter","spring","summer","fall")load(file=file.path(model_data_dir,"flow_timeseries","dseasonal.Rdata"))
dseasonal<-subset(dseasonal,!is.na(HydrologicGroupAB) & !is.na(flow) & !is.na(rain) & !is.na(precip.e.lag2))
# length(unique(dseasonal$site_no))
# #238 unique gages
#
length(unique(dseasonal$site_no[dseasonal$large_barriers==0]))[1] 120
length(unique(dseasonal$site_no[dseasonal$TNC_DamCount==0]))[1] 97
length(unique(dseasonal$site_no[dseasonal$OnChannelWaterSqKM<.5]))[1] 171
length(unique(dseasonal$site_no[dseasonal$OnChannelWaterSqKM<.5 & dseasonal$large_barriers==0]))[1] 110
length(unique(dseasonal$site_no[dseasonal$OnChannelWaterSqKM<.5 & dseasonal$TNC_DamCount==0]))[1] 89
#120 with no LARGE barriers
#97 with no barriers
# 7870 total records
nrow(dseasonal)[1] 7870
# 4140 records from sites w/ no LARGE barriers
nrow(dseasonal[dseasonal$large_barriers==0,])[1] 4140
nrow(dseasonal[dseasonal$OnChannelWaterSqKM<.5,])[1] 6226
# 3879 records from sites w/ no LARGE barriers and <.5 sq km impoundments
nrow(dseasonal[dseasonal$OnChannelWaterSqKM<.5 & dseasonal$large_barriers==0,])[1] 3879
# 3283 records from sites w/ no barriers
nrow(dseasonal[dseasonal$TNC_DamCount==0,])[1] 3283
# 3114 records from sites w/ no barriers and <.5 sq km impoundments
nrow(dseasonal[dseasonal$OnChannelWaterSqKM<.5 & dseasonal$TNC_DamCount==0,])[1] 3114
#subset of unregulated gages
# defined as NO dams from TNC barrier inventory
# AND upstream on-channel open waterbodies < 0.5 sq km
d.unreg<-subset(dseasonal,TNC_DamCount==0 & dseasonal$OnChannelWaterSqKM<.5)# View(d.unreg[,c(1:9,58,59,21:23)])
#choose validation set
#select some gages to leave out for validations
# AND select some years, leave out records from all gages for those year for validation
# so that combined validation set #records is 10-15% of total records
#choose 8 sites w/ more than 4 yrs records
set.seed(933550175)
val.gages<-sample(unique(d.unreg$site_no[d.unreg$qseasonal>12]),size=8,replace=F)
# val.gages
#choose 5 years
set.seed(993889335)
val.year<-sample(unique(d.unreg$year),5,replace=F)
sort(val.year)[1] 1954 1956 1983 1999 2006
#check number of records chosen as part of validation gages and validation years
nrow(d.unreg) #3114[1] 3114
nrow(d.unreg[d.unreg$qseasonal>12,]) #2870[1] 2870
nrow(subset(d.unreg,site_no %in% val.gages)) #291[1] 291
nrow(subset(d.unreg,year(as.Date(date)) %in% val.year)) #212[1] 212
nrow(subset(d.unreg, site_no %in% val.gages & year(as.Date(date)) %in% val.year )) #22[1] 22
#subset and count size of calib and valid data sets
d.calib<-subset(d.unreg,!(site_no %in% val.gages | year(as.Date(date)) %in% val.year))
nrow(d.calib) #2633[1] 2633
d.valid<-subset(d.unreg, site_no %in% val.gages | year(as.Date(date)) %in% val.year)
nrow(d.valid) #481 [1] 481
nrow(d.valid)/nrow(d.unreg)*100 #15.44637 approx 15% of total records[1] 15.45
#save calibration and validation data
setwd(file.path(model_dir,"c_seasonal_means/calib_data"))
save(d.calib,d.valid,file="calib_valid.Rdata")m.fixed<-lm(log(flow) ~
log(drain_area_va) +
log(non.zero(DrainageClass)) +
log(non.zero(PercentSandy))+
# log(non.zero(HydrologicGroupAB))+ log(non.zero(SurficialCoarseC)) +
log(non.zero(Forest)) +
log(non.zero(Agriculture)) +
# log(non.zero(Impervious))+
log(non.zero(OffChannelWaterSqKM)) +
# log(non.zero(OffChannelWetlandSqKM))+
# log(non.zero(OffChannelWaterSqKM+OffChannelWetlandSqKM)) +
# log(non.zero(BasinSlopePCNT))+
log(non.zero(ReachSlopePCNT)) +
log(non.zero(precip.e)) + log(non.zero(precip.e.lag1)) +
# ((precip.e)) + ((precip.e.lag1)) +
# log(non.zero(precip.e.lag2)) +
1,
na.action=NULL, #to ensure we attached fitted values w/ correct site_no
data=d.calib)
#create model for each season
m.fixed.season<-list()
# looping didn't work! model would change when called later on depending on *current* value of i... oh dear
m.fixed.season[["winter"]]<-update(m.fixed, subset=season=="winter")
m.fixed.season$winter<-update(m.fixed.season$winter,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(PercentSandy))
+log(non.zero(Impervious))
-log(non.zero(Forest))
-log(non.zero(precip.e)) -log(non.zero(precip.e.lag1))
+precip.e + precip.e.lag1
# -log(non.zero(ReachSlopePCNT))
# + log(AnnualTminC+20)+ log(non.zero(frozen))+frozen+log(non.zero(gdd)) +gdd
)
m.fixed.season[["spring"]]<-update(m.fixed, subset=season=="spring")
m.fixed.season$spring<-update(m.fixed.season$spring,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(OffChannelWaterSqKM))
# -log(non.zero(ReachSlopePCNT))
# +gdd
+log(non.zero(gdd))
+log(pet)
)
m.fixed.season[["summer"]]<-update(m.fixed, subset=season=="summer")
m.fixed.season$summer<-update(m.fixed.season$summer,formula=.~.
+log(non.zero(precip.e.lag2))
- log(non.zero(PercentSandy))
-log(non.zero(OffChannelWaterSqKM))
+log(non.zero(gdd))
+log(pet)
# -log(non.zero(ReachSlopePCNT)) +log(non.zero(Impervious))-log(non.zero(Forest))
)
m.fixed.season[["fall"]]<-update(m.fixed, subset=season=="fall")
m.fixed.season$fall<-update(m.fixed.season$fall,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(ReachSlopePCNT))
+log(non.zero(gdd))
+log(pet)
)
rm(m.fixed)
# summaries of fixed models
for (i in season.names) {
print(paste("#####",i,"#####"))
print(summary(m.fixed.season[[i]]))
vif(m.fixed.season[[i]])
}[1] "##### winter #####"
Call:
lm(formula = log(flow) ~ log(drain_area_va) + log(non.zero(Agriculture)) +
log(non.zero(OffChannelWaterSqKM)) + log(non.zero(ReachSlopePCNT)) +
log(non.zero(Impervious)) + precip.e + precip.e.lag1, data = d.calib,
subset = season == "winter", na.action = NULL)
Residuals:
Min 1Q Median 3Q Max
-1.8802 -0.1965 0.0269 0.2421 2.1654
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.591345 0.093548 -6.32 4.8e-10
log(drain_area_va) 0.971027 0.021614 44.93 < 2e-16
log(non.zero(Agriculture)) -0.012747 0.004758 -2.68 0.00757
log(non.zero(OffChannelWaterSqKM)) 0.020969 0.006269 3.35 0.00087
log(non.zero(ReachSlopePCNT)) 0.075527 0.018521 4.08 5.1e-05
log(non.zero(Impervious)) -0.026342 0.008213 -3.21 0.00140
precip.e 0.003552 0.000164 21.64 < 2e-16
precip.e.lag1 0.001649 0.000180 9.16 < 2e-16
(Intercept) ***
log(drain_area_va) ***
log(non.zero(Agriculture)) **
log(non.zero(OffChannelWaterSqKM)) ***
log(non.zero(ReachSlopePCNT)) ***
log(non.zero(Impervious)) **
precip.e ***
precip.e.lag1 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.427 on 657 degrees of freedom
Multiple R-squared: 0.824, Adjusted R-squared: 0.822
F-statistic: 440 on 7 and 657 DF, p-value: <2e-16
[1] "##### spring #####"
Call:
lm(formula = log(flow) ~ log(drain_area_va) + log(non.zero(PercentSandy)) +
log(non.zero(Forest)) + log(non.zero(Agriculture)) + log(non.zero(ReachSlopePCNT)) +
log(non.zero(precip.e)) + log(non.zero(precip.e.lag1)) +
log(non.zero(gdd)) + log(pet), data = d.calib, subset = season ==
"spring", na.action = NULL)
Residuals:
Min 1Q Median 3Q Max
-1.4318 -0.1673 0.0308 0.1936 1.5957
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -6.24571 0.53736 -11.62 < 2e-16 ***
log(drain_area_va) 0.95985 0.01490 64.42 < 2e-16 ***
log(non.zero(PercentSandy)) -0.00639 0.00200 -3.19 0.00147 **
log(non.zero(Forest)) 0.25214 0.04075 6.19 1.1e-09 ***
log(non.zero(Agriculture)) -0.01123 0.00299 -3.75 0.00019 ***
log(non.zero(ReachSlopePCNT)) 0.03012 0.01756 1.72 0.08680 .
log(non.zero(precip.e)) 0.66673 0.03923 17.00 < 2e-16 ***
log(non.zero(precip.e.lag1)) 0.11878 0.01866 6.37 3.6e-10 ***
log(non.zero(gdd)) 0.13374 0.03008 4.45 1.0e-05 ***
log(pet) -0.96215 0.24985 -3.85 0.00013 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.311 on 660 degrees of freedom
Multiple R-squared: 0.885, Adjusted R-squared: 0.884
F-statistic: 566 on 9 and 660 DF, p-value: <2e-16
[1] "##### summer #####"
Call:
lm(formula = log(flow) ~ log(drain_area_va) + log(non.zero(DrainageClass)) +
log(non.zero(Forest)) + log(non.zero(Agriculture)) + log(non.zero(ReachSlopePCNT)) +
log(non.zero(precip.e)) + log(non.zero(precip.e.lag1)) +
log(non.zero(precip.e.lag2)) + log(non.zero(gdd)) + log(pet),
data = d.calib, subset = season == "summer", na.action = NULL)
Residuals:
Min 1Q Median 3Q Max
-1.873 -0.337 0.037 0.361 1.740
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -28.14305 1.66340 -16.92 < 2e-16 ***
log(drain_area_va) 0.94399 0.02671 35.34 < 2e-16 ***
log(non.zero(DrainageClass)) -0.55772 0.11058 -5.04 5.9e-07 ***
log(non.zero(Forest)) 0.24277 0.06676 3.64 0.00030 ***
log(non.zero(Agriculture)) -0.04225 0.00494 -8.55 < 2e-16 ***
log(non.zero(ReachSlopePCNT)) -0.10694 0.03075 -3.48 0.00054 ***
log(non.zero(precip.e)) 1.38931 0.07721 17.99 < 2e-16 ***
log(non.zero(precip.e.lag1)) 0.36951 0.06832 5.41 8.9e-08 ***
log(non.zero(precip.e.lag2)) 0.03773 0.03617 1.04 0.29735
log(non.zero(gdd)) 1.84067 0.17853 10.31 < 2e-16 ***
log(pet) -7.01483 0.75970 -9.23 < 2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.534 on 647 degrees of freedom
Multiple R-squared: 0.778, Adjusted R-squared: 0.774
F-statistic: 226 on 10 and 647 DF, p-value: <2e-16
[1] "##### fall #####"
Call:
lm(formula = log(flow) ~ log(drain_area_va) + log(non.zero(PercentSandy)) +
log(non.zero(Forest)) + log(non.zero(Agriculture)) + log(non.zero(OffChannelWaterSqKM)) +
log(non.zero(precip.e)) + log(non.zero(precip.e.lag1)) +
log(non.zero(gdd)) + log(pet), data = d.calib, subset = season ==
"fall", na.action = NULL)
Residuals:
Min 1Q Median 3Q Max
-2.7668 -0.2519 0.0383 0.3090 2.6068
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -17.09324 1.36631 -12.51 < 2e-16
log(drain_area_va) 0.93750 0.02784 33.67 < 2e-16
log(non.zero(PercentSandy)) 0.01217 0.00352 3.46 0.00058
log(non.zero(Forest)) 0.10261 0.04462 2.30 0.02179
log(non.zero(Agriculture)) -0.04540 0.00551 -8.24 1e-15
log(non.zero(OffChannelWaterSqKM)) -0.01687 0.00731 -2.31 0.02139
log(non.zero(precip.e)) 1.73004 0.06640 26.06 < 2e-16
log(non.zero(precip.e.lag1)) 0.72811 0.07212 10.10 < 2e-16
log(non.zero(gdd)) 0.17523 0.09065 1.93 0.05368
log(pet) -0.88661 0.41523 -2.14 0.03313
(Intercept) ***
log(drain_area_va) ***
log(non.zero(PercentSandy)) ***
log(non.zero(Forest)) *
log(non.zero(Agriculture)) ***
log(non.zero(OffChannelWaterSqKM)) *
log(non.zero(precip.e)) ***
log(non.zero(precip.e.lag1)) ***
log(non.zero(gdd)) .
log(pet) *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 0.515 on 630 degrees of freedom
Multiple R-squared: 0.808, Adjusted R-squared: 0.806
F-statistic: 295 on 9 and 630 DF, p-value: <2e-16
# # all DID...have adj-r-sqr around 0.8976 with old calibration set, but that set was too small....
i<-"winter" #0.8224 highest vif 3.00 previous 0.827 0.8321 v 0.8249
i<-"spring" #0.8838 highest vif 4.04 without gdd & pet 0.8776 previous0.8929 v .8697
i<-"summer" #0.7741 highest vif 5.12 without gdd & pet 0.7356 previous 0.7553 v 0.715
i<-"fall" #0.8056 highest vif 6.03 without gdd & pet 0.7833 previous0.7985 v 0.7827
### next addition: add avg temp, and total number of days <0C
# might help account for runoff when ground is frozen in winter,
# and account for ET in summer and fall (when r-sqr values are lower) m.mixed<-lmer(log(flow) ~
log(drain_area_va) +
log(non.zero(DrainageClass)) +
log(non.zero(PercentSandy))+
# log(non.zero(HydrologicGroupAB))+ log(non.zero(SurficialCoarseC)) +
log(non.zero(Forest)) +
log(non.zero(Agriculture)) +
# log(non.zero(Impervious))+
log(non.zero(OffChannelWaterSqKM)) +
# log(non.zero(OffChannelWetlandSqKM))+
# log(non.zero(OffChannelWaterSqKM+OffChannelWetlandSqKM)) +
# log(non.zero(BasinSlopePCNT))+
log(non.zero(ReachSlopePCNT)) +
log(non.zero(precip.e)) + log(non.zero(precip.e.lag1)) +
(1|date)+(1|site_no),
na.action=NULL, #to ensure we attached fitted values w/ correct site_no
data=d.calib)
m.mixed.season<-list()
# m.mixed.season[["winter"]]<-update(m.mixed, subset=season=="winter")
# m.mixed.season$winter<-update(m.mixed.season$winter,formula=.~.
# -log(non.zero(DrainageClass))
# -log(non.zero(PercentSandy)) )
#
#
# m.mixed.season[["spring"]]<-update(m.mixed, subset=season=="spring")
# m.mixed.season$spring<-update(m.mixed.season$spring,formula=.~.
# -log(non.zero(DrainageClass))
# -log(non.zero(ReachSlopePCNT))
# )
#
#
# m.mixed.season[["summer"]]<-update(m.mixed, subset=season=="summer")
# m.mixed.season$summer<-update(m.mixed.season$summer,formula=.~.
# - log(non.zero(PercentSandy))
# -log(non.zero(OffChannelWaterSqKM))
# +log(non.zero(Impervious))
# -log(non.zero(Forest))
# )
#
#
# m.mixed.season[["fall"]]<-update(m.mixed, subset=season=="fall")
# m.mixed.season$fall<-update(m.mixed.season$fall,formula=.~.
# -log(non.zero(DrainageClass))
# -log(non.zero(ReachSlopePCNT))
# )
m.mixed.season[["winter"]]<-update(m.mixed, subset=season=="winter")
m.mixed.season$winter<-update(m.mixed.season$winter,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(PercentSandy))
+log(non.zero(Impervious))
-log(non.zero(Forest))
-log(non.zero(precip.e)) -log(non.zero(precip.e.lag1))
+precip.e + precip.e.lag1
# -log(non.zero(ReachSlopePCNT))
# + log(AnnualTminC+20)+ log(non.zero(frozen))+frozen+log(non.zero(gdd)) +gdd
)
m.mixed.season[["spring"]]<-update(m.mixed, subset=season=="spring")
m.mixed.season$spring<-update(m.mixed.season$spring,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(OffChannelWaterSqKM))
# -log(non.zero(ReachSlopePCNT))
# +gdd
+log(non.zero(gdd))
+log(pet)
)
m.mixed.season[["summer"]]<-update(m.mixed, subset=season=="summer")
m.mixed.season$summer<-update(m.mixed.season$summer,formula=.~.
+log(non.zero(precip.e.lag2))
- log(non.zero(PercentSandy))
-log(non.zero(OffChannelWaterSqKM))
+log(non.zero(gdd))
+log(pet)
# -log(non.zero(ReachSlopePCNT)) +log(non.zero(Impervious))-log(non.zero(Forest))
)
m.mixed.season[["fall"]]<-update(m.mixed, subset=season=="fall")
m.mixed.season$fall<-update(m.mixed.season$fall,formula=.~.
-log(non.zero(DrainageClass))
-log(non.zero(ReachSlopePCNT))
+log(non.zero(gdd))
+log(pet)
)
rm(m.mixed)#create pred df
pred.calib<-d.calib
pred.calib$obs<-log(pred.calib$flow)
pred.calib$pred.fixed<-NA
pred.calib$pred.mixed<-NA
rm(goodness.fixed.calib,goodness.mixed.calib)
#predict and calc goodness of fit metrics of fixed and mixed models, calibration set
for (i in season.names) {
pred.calib[pred.calib$season==i,"pred.fixed"]<-
predict(m.fixed.season[[i]],
newdata=pred.calib[pred.calib$season==i,])
pred.calib[pred.calib$season==i,"pred.mixed"]<-
predict(m.mixed.season[[i]],
newdata=pred.calib[pred.calib$season==i,],
allow.new.levels=T)
print(i)
print(head(pred.calib[pred.calib$season==i,c(1:2,68:70)]))
temp<-goodness(
pred.calib[pred.calib$season==i,c("obs","pred.fixed")]
)
if(!exists("goodness.fixed.calib")){
goodness.fixed.calib<-temp
goodness.fixed.calib$season<-i
goodness.fixed.calib<-goodness.fixed.calib[,c(ncol(goodness.fixed.calib),1:(ncol(goodness.fixed.calib)-1))]
}
else
goodness.fixed.calib[nrow(goodness.fixed.calib)+1,]<-c(i,temp)
temp<-goodness(
pred.calib[pred.calib$season==i,c("obs","pred.mixed")]
)
if(!exists("goodness.mixed.calib")){
goodness.mixed.calib<-temp
goodness.mixed.calib$season<-i
goodness.mixed.calib<-goodness.mixed.calib[,c(ncol(goodness.mixed.calib),1:(ncol(goodness.mixed.calib)-1))]
}
else
goodness.mixed.calib[nrow(goodness.mixed.calib)+1,]<-c(i,temp)
}[1] "winter"
site_no date obs pred.fixed pred.mixed
6 01049221 1978-02-28 2.9619 2.7203 2.69928
10 01049221 1979-02-28 1.9881 2.6532 2.40122
13 01049300 1964-02-28 0.4580 0.3985 0.17880
17 01049300 1965-02-28 0.2263 0.2801 -0.04823
21 01049300 1966-02-28 -0.8406 0.3343 -0.27860
25 01049300 1967-02-28 -0.3188 0.2511 -0.16082
[1] "spring"
site_no date obs pred.fixed pred.mixed
7 01049221 1978-05-28 3.2361 3.6209 3.3658
11 01049221 1979-05-28 3.6801 3.8252 3.6457
14 01049300 1964-05-28 1.0799 0.8626 1.0433
18 01049300 1965-05-28 0.9475 0.6512 0.7077
22 01049300 1966-05-28 1.2241 0.9721 1.1564
26 01049300 1967-05-28 1.3933 1.0629 1.4427
[1] "summer"
site_no date obs pred.fixed pred.mixed
8 01049221 1978-08-28 1.4648 0.9289 1.52273
12 01049221 1979-08-28 1.9010 1.1459 1.58876
15 01049300 1964-08-28 -1.0920 -0.8118 -0.97803
19 01049300 1965-08-28 -1.6607 -1.6080 -1.42808
23 01049300 1966-08-28 -1.1445 -1.2227 -1.23390
27 01049300 1967-08-28 -0.1766 -0.1509 0.01603
[1] "fall"
site_no date obs pred.fixed pred.mixed
9 01049221 1978-11-28 -2.4327 0.3341 -1.346726
16 01049300 1964-11-28 -0.2044 -0.7673 -0.491215
20 01049300 1965-11-28 -0.5933 -0.9947 -0.737250
24 01049300 1966-11-28 0.3026 -0.1065 0.008043
28 01049300 1967-11-28 -0.3520 -0.6402 -0.200381
32 01049300 1968-11-28 -0.4775 -1.1692 -0.861738
#create pred df
pred.valid<-d.valid
pred.valid$obs<-log(pred.valid$flow)
pred.valid$pred.fixed<-NA
pred.valid$pred.mixed<-NA
rm(goodness.fixed.valid,goodness.mixed.valid)
#predict and calc goodness of fit metrics of fixed and mixed models, calibration set
for (i in season.names) {
pred.valid[pred.valid$season==i,"pred.fixed"]<-
predict(m.fixed.season[[i]],
newdata=pred.valid[pred.valid$season==i,])
pred.valid[pred.valid$season==i,"pred.mixed"]<-
predict(m.mixed.season[[i]],
newdata=pred.valid[pred.valid$season==i,],
allow.new.levels=T)
print(i)
print(head(pred.valid[pred.valid$season==i,c(1:2,68:70)]))
temp<-goodness(
pred.valid[pred.valid$season==i,c("obs","pred.fixed")]
)
if(!exists("goodness.fixed.valid")){
goodness.fixed.valid<-temp
goodness.fixed.valid$season<-i
goodness.fixed.valid<-goodness.fixed.valid[,c(ncol(goodness.fixed.valid),1:(ncol(goodness.fixed.valid)-1))]
}
else
goodness.fixed.valid[nrow(goodness.fixed.valid)+1,]<-c(i,temp)
temp<-goodness(
pred.valid[pred.valid$season==i,c("obs","pred.mixed")]
)
if(!exists("goodness.mixed.valid")){
goodness.mixed.valid<-temp
goodness.mixed.valid$season<-i
goodness.mixed.valid<-goodness.mixed.valid[,c(ncol(goodness.mixed.valid),1:(ncol(goodness.mixed.valid)-1))]
}
else
goodness.mixed.valid[nrow(goodness.mixed.valid)+1,]<-c(i,temp)
}[1] "winter"
site_no date obs pred.fixed pred.mixed
89 01049300 1983-02-28 0.4487 -0.005257 -0.02617
112 01049396 1983-02-28 3.0626 2.683441 2.73753
248 01063310 1999-02-28 0.6748 -0.012671 0.21805
276 01063310 2006-02-28 1.2215 1.099568 1.05249
538 01064400 1983-02-28 2.3438 1.586063 1.73208
701 01073000 1954-02-28 3.0298 3.001064 2.89553
[1] "spring"
site_no date obs pred.fixed pred.mixed
90 01049300 1983-05-28 1.817 1.615 1.804
113 01049396 1983-05-28 4.394 3.997 4.136
249 01063310 1999-05-28 1.413 1.102 1.361
277 01063310 2006-05-28 1.078 1.121 1.339
539 01064400 1983-05-28 3.321 3.211 3.274
702 01073000 1954-05-28 4.082 3.920 4.090
[1] "summer"
site_no date obs pred.fixed pred.mixed
91 01049300 1983-08-28 -0.09347 -0.4877 -0.4675
114 01049396 1983-08-28 2.16918 1.5965 1.5707
250 01063310 1999-08-28 -1.16176 -0.7503 -0.4526
278 01063310 2006-08-28 1.01271 0.5530 0.7645
540 01064400 1983-08-28 1.43664 1.2508 1.2546
703 01073000 1954-08-28 2.11206 2.1995 2.1029
[1] "fall"
site_no date obs pred.fixed pred.mixed
251 01063310 1999-11-28 0.9863 0.59900 0.5254
279 01063310 2006-11-28 0.9391 0.93637 0.9527
297 01063452 1999-11-28 -0.5893 -0.05364 -0.1743
541 01064400 1983-11-28 2.2004 1.98137 2.0546
704 01073000 1954-11-28 3.5397 2.90088 2.7322
712 01073000 1956-11-28 1.6046 1.68829 1.5126
goodness.fixed.calib season sample.n mean RMSE NSEff bias percent.bias pearsonR CV.error
1 winter 665 2.001 0.300 0.824 0 0 0.908 0.212
2 spring 670 2.608 0.218 0.885 0 0 0.941 0.118
3 summer 658 1.141 0.374 0.778 0 0 0.882 0.464
4 fall 640 1.469 0.361 0.808 0 0 0.899 0.348
goodness.mixed.calib season sample.n mean RMSE NSEff bias percent.bias pearsonR CV.error
1 winter 665 2.001 0.185 0.933 0 0 0.966 0.131
2 spring 670 2.608 0.108 0.972 0 0 0.986 0.059
3 summer 658 1.141 0.242 0.907 0 0 0.953 0.300
4 fall 640 1.469 0.253 0.906 0 0 0.952 0.244
goodness.fixed.valid season sample.n mean RMSE NSEff bias percent.bias pearsonR CV.error
1 winter 120 1.922 0.274 0.794 -0.126 -6.549 0.904 0.191
2 spring 122 2.279 0.277 0.807 0.077 3.388 0.903 0.169
3 summer 121 1.086 0.382 0.780 -0.066 -6.095 0.887 0.496
4 fall 118 1.370 0.399 0.799 0.135 9.860 0.907 0.401
goodness.mixed.valid season sample.n mean RMSE NSEff bias percent.bias pearsonR CV.error
1 winter 120 1.922 0.272 0.797 -0.147 -7.655 0.910 0.185
2 spring 122 2.279 0.253 0.839 0.040 1.771 0.917 0.157
3 summer 121 1.086 0.348 0.818 -0.018 -1.688 0.906 0.454
4 fall 118 1.370 0.396 0.802 0.077 5.602 0.901 0.406
#create melted data frames for ggplot
#for correlation plots
dd.corr.calib<-melt(
pred.calib[,c("site_no","season","year","obs","pred.fixed","pred.mixed")],
id.vars=c("site_no","season","year","obs"))
dd.corr.calib$season<-factor(capitalize(dd.corr.calib$season),levels=c("Winter","Spring","Summer","Fall"))
dd.corr.valid<-melt(
pred.valid[,c("site_no","season","year","obs","pred.fixed","pred.mixed")],
id.vars=c("site_no","season","year","obs"))
dd.corr.valid$season<-factor(capitalize(dd.corr.valid$season),levels=c("Winter","Spring","Summer","Fall"))
#for hydrographs and other plots
dd.calib<-pred.calib[,c("site_no","FEATUREID","date","year","season","qseasonal","da_sqkm",
"OnChannelWaterSqKM","OnChannelWetlandSqKM","large_barriers","small_barriers",
"precip.e","precip.e.lag1","precip.e.lag2","flow",
"obs","pred.fixed","pred.mixed" )]
dd.calib<-melt(dd.calib,id.vars=c("site_no","FEATUREID","date","year","season",
"qseasonal","da_sqkm","OnChannelWaterSqKM",
"OnChannelWetlandSqKM","large_barriers","small_barriers"))
dd.valid<-pred.valid[,c("site_no","FEATUREID","date","year","season","qseasonal","da_sqkm",
"OnChannelWaterSqKM","OnChannelWetlandSqKM","large_barriers","small_barriers",
"precip.e","precip.e.lag1","precip.e.lag2","flow",
"obs","pred.fixed","pred.mixed" )]
dd.valid<-melt(dd.valid,id.vars=c("site_no","FEATUREID","date","year","season",
"qseasonal","da_sqkm","OnChannelWaterSqKM",
"OnChannelWetlandSqKM","large_barriers","small_barriers"))###View correlation of observed vs predicted values
gg.corr<- ggplot(dd.corr.calib,
aes(x=obs,y=value,colour=variable,pch=variable))+
geom_abline(a=1,b=0,col="black",lty=2)+
theme_bw()+xlab("observed")+ylab("predicted")+
#scale_colour_hue(l=40)
scale_colour_hue(l=40,
name="Model Type",
breaks=c("pred.fixed", "pred.mixed"),
labels=c("Fixed", "Mixed/Random Effects"))+
scale_shape_manual(values=c(16,5))
gg.corr + geom_point() + facet_wrap(~season) + ggtitle("Calibration")
gg.corr %+% dd.corr.valid + geom_point() + facet_wrap(~season) + ggtitle("Validation")
gg.hydrograph<-ggplot(data=subset(dd.valid,
variable %in% c("obs","pred.fixed","pred.mixed") & site_no %in% val.gages),
aes(x=year,y=value,colour=variable,linetype=variable)) +
theme_bw()+scale_colour_brewer(type="qual",palette="Set1") + scale_linetype_manual(values=c("dotted","solid","dashed"))
#pull sample sites from calibration set
set.seed(94802834)
sample.calib<-sample(x=dd.calib[dd.calib$qseasonal>=20,"site_no"],size=10,replace=F)
#calibration hydrographs
for (i in season.names) {
print(
gg.hydrograph %+% subset(dd.calib,site_no %in% sample.calib & variable %in% c("obs","pred.fixed","pred.mixed"))+
geom_line(subset=.(season==i)) +
geom_point(subset=.(season==i),cex=2,pch=1)+
geom_point(subset=.(season==i & year %in% val.year),cex=4,pch=18)+
facet_grid(site_no~.)+
ggtitle(paste("Calibration:",capitalize(i)))
)
}   
#validation hydrographs
for (i in season.names) {
print(
gg.hydrograph +
geom_line(subset=.(season==i)) +
geom_point(subset=.(season==i),cex=2,pch=1)+
geom_point(subset=.(season==i & year %in% val.year),cex=4,pch=18)+
facet_grid(site_no~.)+
ggtitle(paste("Validation:",capitalize(i)))
)
}   
gg.error<-ggplot(data=dd.valid,
aes(x=site_no,y=value,col=variable)) +
geom_boxplot(fill="grey30",col="grey30")+geom_point(pch=15,cex=3)+
scale_colour_brewer(type="qual",palette="Set1")+
facet_grid(season~year, scales="free_x", space="free_x")
#create sample of years from calibration data set
set.seed(96536573)
sample.calib.year<-sample(x=dd.calib[,"year"],size=10,replace=F)
#calibration plot
gg.error %+% subset(dd.calib, variable %in% c("obs","pred.mixed") & year %in% sample.calib.year )+ ggtitle("Calibration")
#validation plot
gg.error %+% subset(dd.valid, variable %in% c("obs","pred.mixed") & year %in% val.year )+ ggtitle("Validation")