develop/compareOldTool/GLPartitioningTestCase.R
In bgctw/REddyProc: Post Processing of (Half-)Hourly Eddy-Covariance Measurements
# Script for running REddyProc with different options on 29 site-years
# same as was used in Wutzler17 for benchmarking daytime data
# install_github(repo = "bgctw/REddyProc",branch = "master")
if (!exists("partitionNEEGL") ) library(REddyProc) # only load library if not sourced its already
.tmp.f <- function(){
library(Rcpp)
library(plyr)
source("R/GeoFunctions.R")
source("R/FileHandling.R")
source("R/DataFunctions.R")
source("R/PartitioningLasslop10.R")
pkg <- 'REddyProc'
loadDll <- function (dynFilenameLocal = file.path( system.file(package = pkg), "libs", "x64"
, paste0(pkg, .Platform$dynlib.ext)), pkg = pkg, isWarningDllNotFound = TRUE
){
if (file.exists(dynFilenameLocal)) {
dyn.load(dynFilenameLocal)
}
else if (isWarningDllNotFound) {
warning(paste("dynLib", dynFilenameLocal, "not found."))
}
}
loadDll(pkg = pkg)
source("R/RcppExports.R")
}
library(dplyr)
library(tibble)
library(foreach)
#library(doMC) #twutz: doMC is not supported for Windows any more, may swith to same functionality with doParallel
library(doParallel)
isCluster<- if (.Platform$OS.type == "windows") FALSE else TRUE
#CLUSTER<-FALSE
#registerDoMC(8)
#doMC::registerDoMC(cores = 48) # or however many cores you have access to
if (isCluster ){
registerDoParallel(min(cores = 28L, detectCores())) # having 28 sites
} else {
registerDoParallel(cores = 2L)
}
## aggregation sd
aggregate.sd <- function(x){ # where x is the sd of the variable
n <- sum(!is.na(x))
agg_sd <- sqrt(sum(x^2,na.rm = T)) / n
return(agg_sd)
}
##############
## settings ##
##############
#scen <- "default"
#scen <- "omitSaturationFilter"
#scen <- "SaturationPenalty50"
#scen <- "SaturationPenalty5"
#scen <- "omitBoundLowerNEEUnc"
#scen <- "filterParSaturationProp50"
#scen <- "omitSmoothTempSens"
#scen <- "neglectNEEUncertaintyOnMissing"
#scen <- "TempLasslop10"
scen <- "Lasslop10" # wait for Gittas answer, how to treat missing sdNEE
nBoot = 3L
#nBoot = 60L # for the default scenario
# uncomment the scenario in quesiton and then run
# cd develop/compareOldTool
# bsub -q mpi_large -M 20971520 -n 28 -R span[hosts=1] R CMD BATCH --vanilla GLPartitioningTestCase.R GLPartitioningTestCase_log.txt
scenarioConfigurations = tibble::tribble(
~scenario, ~ouputPath, ~ctrl, ~comment
,"Lasslop10", "Lasslop10_1709", within(partGLControlLasslopCompatible(), useSolarTime <- FALSE), "most Lasslop compatible"
,"default", "default1709", partGLControl(), ""
,"omitBoundLowerNEEUnc","default_options1709/omitBoundLowerNEEUnc", partGLControl(isBoundLowerNEEUncertainty = FALSE), "omit lower bound on NEE uncertainty allowing for high leverage"
,"filterParSaturationProp50","default_options1709/filterParSaturationProp50", partGLControl(minPropSaturation = 0.5), "filter those windows where GPP prediction at highest PAR is less than 50% of GPP at PAR = 2000"
,"omitSmoothTempSens","default_options1709/omitSmoothTempSens", partGLControl(smoothTempSensEstimateAcrossTime = FALSE), "use raw Temperature nighttime sensitivity estimates instead of smoothing them over time"
,"TempLasslop10","default_options1709/TempLasslop10", partGLControl(smoothTempSensEstimateAcrossTime = FALSE, fixedTRefAtNightTime = 15, isExtendTRefWindow = FALSE), "similar E0 from nighttime as Lallop (RRef = 15, do not extend window, not Smooth E0"
,"neglectNEEUncertaintyOnMissing","default_options1709/neglectNEEUncertaintyOnMissing", partGLControl(neglectNEEUncertaintyOnMissing = TRUE), "neglectNEEUncertaintyOnMissing = TRUE: omit weighting on NA in sdNEE"
)
scenConf <- as.list(subset(scenarioConfigurations, scenario == scen))
scenConf$ctrl[[1]]$nBootUncertainty <- nBoot
str(scenConf, max.level = 3)
date()
####################
## site selection ##
####################
##here is my site selection based on Mirco's mail:
# also remove FR-Fon, as there was an error in pvWave processing (actually processed FR-Lq1)
sitesMircoUsed <- c("BR-Ma2","BR-Sa1","CA-Let","CA-NS7","CA-TP3","CH-Oe2","CN-HaM","DE-Hai","DK-Sor","ES-E1","ES-VDA","FI-Hyy","FI-Kaa","FR-Gri","FR-Hes","FR-Lq1","FR-Lq2","FR-Pue","IE-Dri","IL-Yat","IT-Amp","IT-MBo","IT-Pia","IT-SRo","JP-Tef","PT-Esp","RU-Cok","SE-Nor","TH98_new","US-Ha1","US-MMS","US-SO2","US-Ton","VU-Coc")
sitesRestricted <- c("BR-Ma2","BR-Sa1","CA-Let","CN-HaM","JP-Tef","US-SO2","IT-Pia")
siteMapping = list(TH98_new = "DE-Tha", "ES-E1" = "ES-ES1")
sitesUsed <- setdiff( c( unlist(siteMapping), setdiff(sitesMircoUsed,sitesRestricted)), names(siteMapping))
sites <- sort(sitesUsed)
#bgiDir <- if (.Platform$OS.type == "windows") "M:" else "/Net/Groups/BGI"
bgiDir <- "~/bgi"
path <- file.path(bgiDir,"work_3/REddyProcRelease/Eval_GL_Partitioning")
flist <- list.files(file.path(path,"MR_GL_partitioning"), pattern = "*DataSetafterFluxpart.txt") #[-31]
flist <- flist[is.element(substr(flist,1,6),sites)]
## exclude sites that were not processed in the online tool
sites <- sites[is.element(sites,substr(flist,1,6))]
latLongSites <- rbind(
data.frame(site = "DE-Tha", lat = 51, long = 11, timeOffset = -1 )
,data.frame(site = "IT-MBo", lat = 45.0, long = 1, timeOffset = 0 )
)
tmp <- read.csv(file.path(path,"CommonAnc.csv"), colClasses = c(Site.ID = "character", Latitude = "numeric", Longitude = "numeric", UTC = "numeric"), na.strings = "TBD")
latLongSites <- data.frame(site = tmp$Site.ID, lat = tmp$Latitude, long = tmp$Longitude, timeOffset = floor(-tmp$UTC))
# create output directory if not existing
if (!dir.exists(file.path(path,"Results",scenConf$ouputPath)) ) dir.create( file.path(path,"Results",scenConf$ouputPath) )
###########################
#---- Reading data
# NEE_orig already ustar filtered!! as in Papale_2006 most conservative threshold,
# the seasons are computed in a different manner (not e.g. JFM)
# gapfilling according to Reichstein_2005
iProcSites <- seq_along(sites)
.tmp.inspectSingleSites <- function(){
#s <- grep("CA-TP3",sites)[1]
s <- grep("DE-Hai",sites)[1]
s <- grep("DE-Tha",sites)[1] # large differences with omitSmoothTempSens in month 8
s <- grep("FR-Gri",sites)[1]
s <- grep("ES-VDA",sites)[1]
s <- grep("IL-Yat",sites)[1] # high PAR with low temp-diff, subsetting data does not impar estimates here
s <- grep("PT-Esp",sites)[1] # largest differences with omitSmoothTempSens month 4 (but consider entire data for effect of smoothing)
s <- grep("IT-Amp",sites)[1] # largest differences with pvWave in month 7
siteName <- sites[s]
fileName <- flist[s]
#iProcSites <- 2:3
#try(rm(REddy.mm.all)); try(REddy.yy.all)
}
computeSite <- function(siteName, fileName, scenConf){
year <- as.numeric(substr(fileName, nchar(fileName) - 28, nchar(fileName) - 25))
latLongSite <- unlist(subset(latLongSites, site == siteName)[1,2:4])
#+++ Loading data from MR partitioning and data for running the partitioning
dfall <- fLoadTXTIntoDataframe(fileName, file.path(path,"MR_GL_partitioning"))
.tmp.readPvWave <- function(){
fname.PVwave <- paste(siteName,'.',year,'.','DataSetafterFluxpartGL2010.txt', sep = "")
dfallPv <- read.table(file.path(path,"MR_GL_partitioning",fname.PVwave,sep = ""),skip = 2)
title <- scan(file.path(path,"MR_GL_partitioning",fname.PVwave,sep = ""), nlines = 1, sep = "", strip.white = TRUE,
what = list(rep('character',17)))
names(dfallPv) <- title[[1]]
dfallPv$julday <- dfall$julday
#
iRowEst <- seq(97, 17473, by = 96)
dsE0Pv <- data.frame(E0 = dfallPv$E0[iRowEst], sdE0 = 20, RRef = dfallPv$rb[iRowEst])
dsE0Pv[ dsE0Pv == -9999 ] <- NA
dsE0Pv$E0 <- fillNAForward(dsE0Pv$E0)
dsE0Pv$RRef <- fillNAForward(dsE0Pv$RRef)
}
ctrlOpt <- scenConf$ctrl[[1]]
#ctrlOpt$isNeglectPotRadForNight <- TRUE
#ctrlOpt$fixedTempSens = dsE0Pv
useSolarTime <- if (length(ctrlOpt$useSolarTime) ) ctrlOpt$useSolarTime else TRUE
dfall$PotRadSolar <- as.numeric(fCalcPotRadiation(dfall$julday,dfall$Hour,latLongSite["lat"],latLongSite["long"],latLongSite["timeOffset"], useSolartime.b = TRUE))
dfall$PotRad <- as.numeric(fCalcPotRadiation(dfall$julday,dfall$Hour,latLongSite["lat"],latLongSite["long"],latLongSite["timeOffset"], useSolartime.b = useSolarTime))
#plot( PotRad ~ PotRadSolar, dfall)
# here fake potential radiation to be high on all records where Lasslop Partitioning quantified non-night
dfall$day <- (1 - dfall$night)*100
dfall_posix <- dsMonth <- fConvertTimeToPosix(dfall, 'YMDH', Year.s = 'Year', Month.s = 'Month', Day.s = 'Day', Hour.s = 'Hr')
# dsMonth <- subset(dfall_posix, Month %in% 6:8) #& DateTime >= "2002-08-09 00:00:00" & DateTime <= "2002-08-12 23:30:00")
#
# START - RUN THE REddyProc DT partitioning
#ctrlOpt$isAssociateParmsToMeanOfValids <- FALSE # in order to match exact source data
#ctrlOpt$fixedTempSens = data.frame(E0 = 220, sdE0 = 50, RRef = 2.7)
dsResOpt <- partitionNEEGL(dsMonth,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0
,RadVar.s = "Rg_f"
#,PotRadVar.s = "day"
,PotRadVar.s = "PotRad"
,suffix = "",
controlGLPart = ctrlOpt
)
dsResOpt$DateTime <- dsMonth$DateTime
dsResOpt$julday <- dsMonth$julday
.tmp.debug <- function(){
ctrlDefault <- partGLControl(); ctrlDefault$isAssociateParmsToMeanOfValids <- FALSE; nBootUncertainty = 3L # in order to match exact source data
#ctrlDefault$fixedTempSens <- ctrl$fixedTempSens
dsResDefault <- partitionNEEGL(dsMonth,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0,RadVar.s = "Rg",PotRadVar.s = "day",suffix = "",
controlGLPart = ctrlDefault
)
dsResDefault$DateTime <- dsResOpt$DateTime <- dsMonth$DateTime
dsResOpt$Day <- dsResDefault$Day <- dsMonth$Day
dsResOpt$julday <- dsResDefault$julday <- dsMonth$julday
plot(FP_beta ~ julday, dsResOpt, col = "red", pch = "x", ylim = range(c(dsResOpt$FP_beta, dsResDefault$FP_beta),na.rm = TRUE))
points(FP_beta ~ julday, dsResDefault)
dsResOpt$diffFP_beta <- dsResOpt$FP_beta - dsResDefault$FP_beta
subset( dsResOpt, is.finite(FP_errorcode), c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night"))
subset( dsResDefault, is.finite(FP_errorcode), c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night"))
#plot(diffFP_beta ~ julday, dsRes, col = "red", pch = "x")
#plot(diffFP_beta ~ DateTime, subset(dsRes, julday %in% 200:250), col = "red", pch = "x")
#plot(diffFP_beta ~ julday, subset(dsRes, julday %in% 210:230 & is.finite(FP_beta), col = "red", pch = "x")
.tmp.inspectE0 <- function(){
Sys.setlocale(locale = "C") # for displying English time axis
plot( FP_E0 ~ DateTime, dsResOpt, ylim = range(c(dsResOpt$FP_E0, dsResDefault$FP_E0),na.rm = TRUE), pch = "x", col = "red" )
lines(FP_E0 ~ DateTime, subset(dsResDefault, is.finite(FP_E0)))
plot( dsResOpt$FP_E0 ~ dsResDefault$FP_E0)
isInPeriod <- (dsMonth$DateTime >= "2004-03-01" & dsMonth$DateTime <= "2004-05-01")
.commonCols <- intersect(names(dsResDefault), names(dsResOpt))
dss <- rbind(
cbind( method = "Default", subset( dsResDefault[,.commonCols], isInPeriod & is.finite(FP_E0)))
,cbind( method = scenConf$scen, subset( dsResOpt[,.commonCols], isInPeriod & is.finite(FP_E0)))
)
ggplot( dss, aes(DateTime, FP_E0, color = method, fill = method)) +
#geom_ribbon(aes(ymin = FP_E0-1.96*FP_E0_sd, ymax = FP_E0+1.96*FP_E0_sd), alpha = 0.3) + ylim(-250,750) +
#geom_ribbon(aes(ymin = FP_E0-FP_E0_sd, ymax = FP_E0+FP_E0_sd), alpha = 0.3) + ylim(-50,550) +
geom_line() + theme_bw() + theme(legend.position = "bottom") +
ylab(bquote(E[0]*' (K)')) +
theme(legend.position = "none") +
theme(axis.title.x = element_blank()) + theme( legend.title = element_blank() )
#ggsave("../../eclipse46_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensE0_PT-Esp.pdf", width = 3.27, height = 2.5)
#ggsave("../../eclipse_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensE0_PT-Esp.pdf", width = 3.27, height = 2.5)
isInPeriod <- (dsMonth$DateTime >= "2004-04-01" & dsMonth$DateTime <= "2004-04-20")
dss <- rbind(
cbind( method = "Default", subset( dsResDefault[,.commonCols], isInPeriod))
,cbind( method = scenConf$scen, subset( dsResOpt[,.commonCols], isInPeriod))
)
ggplot( dss, aes(DateTime, GPP_DT, color = method)) + geom_line() + theme_bw() + theme(legend.position = "bottom") +
ylab(bquote(GPP*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme(axis.title.x = element_blank()) + theme( legend.title = element_blank() )
#ggsave("../../eclipse46_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensGPP_PT-Esp.pdf", width = 3.27, height = 2.5)
}
# in partGLFitLRCOneWindow:
#if (as.POSIXlt(dsDay$sDateTime[1])$mday+2L > 27 ) recover()
# save(dsDay, file = "tmp/dsDayDebug.RData")
load(file = "tmp/dsDayDebug.RData")
dsDay <- arrange_(dsDay, ~Rg)
plot( -NEE ~ Rg, dsDay)
DayInspect <- 7#11
(thetaDef <- unlist(subset(dsResDefault, Day == DayInspect & is.finite(FP_beta), c("FP_k","FP_beta","FP_alpha","FP_RRef","FP_E0","FP_RRef_Night") )))
(thetaOpt <- unlist(subset(dsResOpt, Day == DayInspect & is.finite(FP_beta), c("FP_k","FP_beta","FP_alpha","FP_RRef","FP_E0","FP_RRef_Night") )))
lrcFitter <- RectangularLRCFitter()
dsDay$GPPDefault <- lrcFitter$predictGPP( dsDay$Rg, thetaDef["FP_beta"], thetaDef["FP_alpha"])
dsDay$GPPOpt <- lrcFitter$predictGPP( dsDay$Rg, thetaOpt["FP_beta"], thetaOpt["FP_alpha"])
dsDay$NEPDefault <- lrcFitter$predictLRC( thetaDef, dsDay$Rg, dsDay$VPD, dsDay$Temp)$NEP
dsDay$NEPOpt <- lrcFitter$predictLRC( thetaOpt, dsDay$Rg, dsDay$VPD, dsDay$Temp)$NEP
ylim <- range(-dsDay$NEE, na.rm = TRUE); ylim[2] <- max(ylim[2], thetaDef["FP_beta"], thetaOpt["FP_beta"], na.rm = TRUE )*1.1
maxRg <- max(1200,max(dsDay$Rg,na.rm = TRUE))
maxRg <- max(2000,max(dsDay$Rg,na.rm = TRUE))
#
plot( -NEE ~ Rg, dsDay)
lines( NEPDefault ~ Rg, dsDay )
lines( NEPOpt ~ Rg, dsDay, col = "red" )
points(maxRg, thetaDef["FP_beta"]); points(maxRg, thetaOpt["FP_beta"], col = "red", pch = "x")
#
plot( GPPDefault ~ Rg, dsDay, xlim = c(0,maxRg), ylim = ylim)
plot( GPPOpt ~ Rg, dsDay, xlim = c(0,maxRg), ylim = ylim, col = "red")
abline(h = thetaDef["FP_beta"]); abline(h = thetaOpt["FP_beta"], col = "red", pch = "x")
#
RgSeq <- seq(0,maxRg, length.out = 80)
predGPP <- lrcFitter$predictLRC( thetaOpt, Rg = RgSeq, VPD = 0, Temp = NA)$GPP
lines( predGPP ~ RgSeq, col = "red")
}
### add modelled NEE
dsResOpt$NEE_DT <- -(dsResOpt$GPP_DT - dsResOpt$Reco_DT)
## save data frames, but omit list columns
iListColumns <- which( sapply( dsResOpt, is.list ) )
dsResOptNotListCol <- if (length(iListColumns)) dsResOpt[-iListColumns] else dsResOpt
write.table(dsResOptNotListCol,file = file.path(path,"Results",scenConf$ouputPath,paste0(siteName,".txt")),row.names = F,col.names = T)
#
# extract those rows where fitting object is not NULL abd save
dsFits <- as_tibble(dsResOpt)[,iListColumns] %>% mutate_(row = ~1:n()) %>% rowwise() %>% filter_(~!is.null(FP_OPT_VPD[[1]]))
save( dsFits ,file = file.path(path,"Results",scenConf$ouputPath,paste0(siteName,"_fits.RData")))
## the _agg columns are used for aggregation in aggregate() function below
julday_agg <- c(1,dfall$julday[1:(nrow(dsResOpt)-1)])
Month_agg <- c(1,dfall$Month[1:(nrow(dsResOpt)-1)])
Year_agg <- c(dfall$Year[1],dfall$Year[1:(nrow(dsResOpt)-1)])
# Aggregation Monthly per site
df.REddy.mm <- aggregate(dsResOpt,by = list(Year_agg = Year_agg, Month_agg = Month_agg),mean,na.rm = T)
# sd
df.REddy.mm$Reco_DT_SD <- aggregate(dsResOpt$Reco_DT_SD,by = list(Year_agg = Year_agg, Month_agg = Month_agg),aggregate.sd)[,3]
df.REddy.mm$GPP_DT_SD <- aggregate(dsResOpt$GPP_DT_SD,by = list(Year_agg = Year_agg, Month_agg = Month_agg),aggregate.sd)[,3]
df.REddy.mm <- cbind(Site = siteName,df.REddy.mm)
# Aggregation Annual
df.REddy.yy <- aggregate(dsResOpt,by = list(Year_agg = Year_agg),mean,na.rm = T)
# sd
df.REddy.yy$Reco_DT_SD <- aggregate(dsResOpt$Reco_DT_SD,by = list(Year_agg = Year_agg),aggregate.sd)[,2]
df.REddy.yy$GPP_DT_SD <- aggregate(dsResOpt$GPP_DT_SD,by = list(Year_agg = Year_agg),aggregate.sd)[,2]
df.REddy.yy <- cbind(Site = siteName,df.REddy.yy)
#
list(
mm = df.REddy.mm
,yy = df.REddy.yy
)
}
.tmp.f <- function(){
REddy.mm.all$Site <- REddy.mm.all$siteName
save(REddy.mm.all, file = "all_sites_monthly.RData")
#
REddy.yy.all$Site <- REddy.yy.all$siteName
save(REddy.yy.all, file = "all_sites_annual.RData")
}
ansList <- foreach (s = iProcSites, .errorhandling = "pass", .packages = c("REddyProc","plyr","mlegp","logitnorm")) %dopar% {
#ansList <- lapply( iProcSites, function(s){
siteName <- sites[s]
fileName <- flist[s]
message("-------- starting site ",siteName)
computeSite(siteName,fileName, scenConf)
}# end of site loop
#)
str(ansList, max.level = 2)
REddy.mms <- lapply(ansList, "[[" , 1L)
REddy.mm.all <- do.call( rbind, REddy.mms)
REddy.yys <- lapply(ansList, "[[" , 2L)
REddy.yy.all <- do.call( rbind, REddy.yys)
## save as RData:
#save(NT_vs_DT_REddy,DT_REddy_vs_pvwave,file = file.path(path,"Results",scenConf$ouputPath,"/eval_metrics.RData")) # 1) evaluation metrics
outputDir <- file.path(path,"Results",scenConf$ouputPath)
save(REddy.mm.all,file = file.path(outputDir,"all_sites_monthly.RData")) # 2) monthly aggregated results for all sites
save(REddy.yy.all,file = file.path(outputDir,"all_sites_annual.RData")) # 3) annual aggregated results for all sites
# write scenario information
save(scenConf, file = file.path(outputDir,"scenConf.RData"))
# generate readmegen.txt
sink(file.path(outputDir,'readmegen.txt')); {
cat("Generated be REddyProc/inst/compareOldTool/GLPartitioningTestCase.R\n")
cat(date(),"\n\n")
cat("scenConf.RData: list of information on scenario.\n")
cat("all_sites_monthly.RData: data.frame of partitioning outputs across sites and months.\n")
cat("all_sites_annual.RData: data.frame of partitioning outputs across sites for a single year.\n")
cat("<siteName>.txt: csv file of partitioning outputs\n")
cat("<siteName>_fits.RData: tibble of fitted objects and central row number of corresponding window rows\n\n")
cat("scenConf: ")
str(scenConf, max.level = 3)
}; sink()
.tmp.lookForHighPar <- function(){
#duplated column names?
#ds <- dfall_posix[,!(names(dfall) %in% c("Day","Hour"))]
ds <- partGLExtractStandardData(dfall_posix,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0,RadVar.s = "Rg",PotRadVar.s = "day",suffix = "",
controlGLPart = ctrlOpt)
ds$julday <- dfall_posix$julday
.tmp.maxPAR <- function(){
dailyMaxPar <- ds %>% group_by_(~julday) %>% summarize_(Rg = ~max(Rg, na.rm = TRUE))
arrange_(dailyMaxPar, ~desc(Rg))
ggplot( dailyMaxPar, aes(julday, Rg)) + geom_point()
}
# extract data of day 174
julDayInspect <- 214
dss <- subset(ds, julday %in% (julDayInspect+(-2:1)))
isValidDayRecNoVPDConstraint <- !is.na(dss$isDay) & dss$isDay & !is.na(dss$NEE) & !is.na(dss$sdNEE) & !is.na(dss$Temp) & !is.na(dss$Rg)
dsDay <- dss[isValidDayRecNoVPDConstraint,]
p1 <- ggplot( dsDay, aes(Rg,-NEE, col = Temp)) + geom_point(); p1
#p1 <- ggplot( dsDay, aes(Rg,-NEE, col = sdNEE)) + geom_point(); p1
lrcFitter <- RectangularLRCFitter()
#ctrlOpt <- partGLControl( ) # use same as in debug above
dsE0Opt <- subset( dsResOpt, is.finite(FP_beta) & julday == julDayInspect, c("FP_E0","FP_E0_sd","FP_RRef","FP_RRef_Night"))
resOpt <- resOpt0 <- lrcFitter$fitLRC(dsDay
#, E0 = 130, sdE0 = 50, RRefNight = 1 # IL-Yat day 148
#, E0 = 120, sdE0 = 50, RRefNight = 3 # DE-Hai day
, E0 = dsE0Opt$FP_E0, sdE0 = dsE0Opt$FP_E0_sd, RRefNight = dsE0Opt$FP_RRef_Night # from dsRes in debug above
, controlGLPart = ctrlOpt)
(thetaOpt <- resOpt$thetaOpt)
dsE0Default <- subset( dsResDefault, is.finite(FP_beta) & julday == julDayInspect, c("FP_E0","FP_E0_sd","FP_RRef","FP_RRef_Night"))
resDefault <- lrcFitter$fitLRC(dsDay
#, E0 = 130, sdE0 = 50, RRefNight = 1 # IL-Yat day 148
#, E0 = 120, sdE0 = 50, RRefNight = 3 # DE-Hai day
, E0 = dsE0Default$FP_E0, sdE0 = dsE0Default$FP_E0_sd, RRefNight = dsE0Default$FP_RRef_Night # from dsRes in debug above
, controlGLPart = ctrlDefault)
(thetaDefault <- resDefault$thetaOpt)
#(thetaOpt <- resOpt$theta)
dsPred <- data.frame(Rg = seq(0,2000, length.out = 80), Temp = median(dsDay$Temp))
dsPredOpt <- {
dsPred$GPP <- lrcFitter$predictLRC( thetaOpt, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$GPP
dsPred$NEP <- lrcFitter$predictLRC( thetaOpt, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$NEP
cbind(scenario = scenConf$scen, dsPred)
}
dsPredDefault <- {
dsPred$GPP <- lrcFitter$predictLRC( thetaDefault, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$GPP
dsPred$NEP <- lrcFitter$predictLRC( thetaDefault, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$NEP
cbind(scenario = "Default", dsPred)
}
dsPred <- rbind(dsPredOpt, dsPredDefault)
#ggplot( dsPred, aes(Rg, GPP)) + geom_line()
# in paper example DE-Hai, julday %in% (220+(0:4))
p1 + geom_line(data = dsPred, aes(Rg, GPP, linetype = scenario), color = "black") +
xlab(bquote('Rg ('*W~m^-2*')')) +
ylab(bquote(NEP[Obs]*" & "*GPP[mod]*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme_bw(base_size = 9) +
theme(legend.position = c(0.95,0.05), legend.justification = c(1,0))
p1 + geom_line(data = dsPred, aes(Rg, NEP, linetype = scenario), color = "black") +
xlab(bquote('Rg ('*W~m^-2*')')) +
ylab(bquote(NEP[Obs]*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme_bw(base_size = 9) +
theme(legend.position = c(0.95,0.05), legend.justification = c(1,0))
# on period of high PAR-range, e.g. Yatir, experiment with decreasing dataset to lower bounds
# result: similar estimates also from constrained dataset -> not not constrain GPP2000, only filter extrem cases
maxRgs <- c(1200,1000,800,600,400)
thetaOptRg <- cbind( maxRg = maxRgs, as.data.frame(t(sapply( maxRgs, function(maxRg){
dsDay <- subset(dsDay, Rg <= maxRg)
resOpt <- resOpt0 <- lrcFitter$fitLRC(dsDay, controlGLPart = ctrlOpt
, E0 = 120, sdE0 = 50, RRefNight = 3)
c(nRec = nrow(dsDay), resOpt$thetaOpt)
}))))
thetaOptRg
}
.tmp.comparePvWaveREddyProc <- function(){
iRowEst <- which(is.finite(dsResOpt$FP_errorcode))
parR <- dsResOpt[ iRowEst, c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night")]
names(parR) <- c("julday","FP_errorcode","beta","alpha","E0","k","rb","rbN")
parP <- dfallPv[ iRowEst, ]
parP[ parP == -9999] <- NA
plot( parP$E0 ~ parR$E0 ); abline(0,1) # large differences
#plot( Rg_f ~ DateTime, dfall_posix[dfall$night == 1,] )
plot( parP$beta ~ parR$beta ); abline(0,1) # large differences
}
stopImplicitCluster()
bgctw/REddyProc documentation built on March 26, 2024, 11:35 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# Script for running REddyProc with different options on 29 site-years
# same as was used in Wutzler17 for benchmarking daytime data
# install_github(repo = "bgctw/REddyProc",branch = "master")
if (!exists("partitionNEEGL") ) library(REddyProc) # only load library if not sourced its already
.tmp.f <- function(){
library(Rcpp)
library(plyr)
source("R/GeoFunctions.R")
source("R/FileHandling.R")
source("R/DataFunctions.R")
source("R/PartitioningLasslop10.R")
pkg <- 'REddyProc'
loadDll <- function (dynFilenameLocal = file.path( system.file(package = pkg), "libs", "x64"
, paste0(pkg, .Platform$dynlib.ext)), pkg = pkg, isWarningDllNotFound = TRUE
){
if (file.exists(dynFilenameLocal)) {
dyn.load(dynFilenameLocal)
}
else if (isWarningDllNotFound) {
warning(paste("dynLib", dynFilenameLocal, "not found."))
}
}
loadDll(pkg = pkg)
source("R/RcppExports.R")
}
library(dplyr)
library(tibble)
library(foreach)
#library(doMC) #twutz: doMC is not supported for Windows any more, may swith to same functionality with doParallel
library(doParallel)
isCluster<- if (.Platform$OS.type == "windows") FALSE else TRUE
#CLUSTER<-FALSE
#registerDoMC(8)
#doMC::registerDoMC(cores = 48) # or however many cores you have access to
if (isCluster ){
registerDoParallel(min(cores = 28L, detectCores())) # having 28 sites
} else {
registerDoParallel(cores = 2L)
}
## aggregation sd
aggregate.sd <- function(x){ # where x is the sd of the variable
n <- sum(!is.na(x))
agg_sd <- sqrt(sum(x^2,na.rm = T)) / n
return(agg_sd)
}
##############
## settings ##
##############
#scen <- "default"
#scen <- "omitSaturationFilter"
#scen <- "SaturationPenalty50"
#scen <- "SaturationPenalty5"
#scen <- "omitBoundLowerNEEUnc"
#scen <- "filterParSaturationProp50"
#scen <- "omitSmoothTempSens"
#scen <- "neglectNEEUncertaintyOnMissing"
#scen <- "TempLasslop10"
scen <- "Lasslop10" # wait for Gittas answer, how to treat missing sdNEE
nBoot = 3L
#nBoot = 60L # for the default scenario
# uncomment the scenario in quesiton and then run
# cd develop/compareOldTool
# bsub -q mpi_large -M 20971520 -n 28 -R span[hosts=1] R CMD BATCH --vanilla GLPartitioningTestCase.R GLPartitioningTestCase_log.txt
scenarioConfigurations = tibble::tribble(
~scenario, ~ouputPath, ~ctrl, ~comment
,"Lasslop10", "Lasslop10_1709", within(partGLControlLasslopCompatible(), useSolarTime <- FALSE), "most Lasslop compatible"
,"default", "default1709", partGLControl(), ""
,"omitBoundLowerNEEUnc","default_options1709/omitBoundLowerNEEUnc", partGLControl(isBoundLowerNEEUncertainty = FALSE), "omit lower bound on NEE uncertainty allowing for high leverage"
,"filterParSaturationProp50","default_options1709/filterParSaturationProp50", partGLControl(minPropSaturation = 0.5), "filter those windows where GPP prediction at highest PAR is less than 50% of GPP at PAR = 2000"
,"omitSmoothTempSens","default_options1709/omitSmoothTempSens", partGLControl(smoothTempSensEstimateAcrossTime = FALSE), "use raw Temperature nighttime sensitivity estimates instead of smoothing them over time"
,"TempLasslop10","default_options1709/TempLasslop10", partGLControl(smoothTempSensEstimateAcrossTime = FALSE, fixedTRefAtNightTime = 15, isExtendTRefWindow = FALSE), "similar E0 from nighttime as Lallop (RRef = 15, do not extend window, not Smooth E0"
,"neglectNEEUncertaintyOnMissing","default_options1709/neglectNEEUncertaintyOnMissing", partGLControl(neglectNEEUncertaintyOnMissing = TRUE), "neglectNEEUncertaintyOnMissing = TRUE: omit weighting on NA in sdNEE"
)
scenConf <- as.list(subset(scenarioConfigurations, scenario == scen))
scenConf$ctrl[[1]]$nBootUncertainty <- nBoot
str(scenConf, max.level = 3)
date()
####################
## site selection ##
####################
##here is my site selection based on Mirco's mail:
# also remove FR-Fon, as there was an error in pvWave processing (actually processed FR-Lq1)
sitesMircoUsed <- c("BR-Ma2","BR-Sa1","CA-Let","CA-NS7","CA-TP3","CH-Oe2","CN-HaM","DE-Hai","DK-Sor","ES-E1","ES-VDA","FI-Hyy","FI-Kaa","FR-Gri","FR-Hes","FR-Lq1","FR-Lq2","FR-Pue","IE-Dri","IL-Yat","IT-Amp","IT-MBo","IT-Pia","IT-SRo","JP-Tef","PT-Esp","RU-Cok","SE-Nor","TH98_new","US-Ha1","US-MMS","US-SO2","US-Ton","VU-Coc")
sitesRestricted <- c("BR-Ma2","BR-Sa1","CA-Let","CN-HaM","JP-Tef","US-SO2","IT-Pia")
siteMapping = list(TH98_new = "DE-Tha", "ES-E1" = "ES-ES1")
sitesUsed <- setdiff( c( unlist(siteMapping), setdiff(sitesMircoUsed,sitesRestricted)), names(siteMapping))
sites <- sort(sitesUsed)
#bgiDir <- if (.Platform$OS.type == "windows") "M:" else "/Net/Groups/BGI"
bgiDir <- "~/bgi"
path <- file.path(bgiDir,"work_3/REddyProcRelease/Eval_GL_Partitioning")
flist <- list.files(file.path(path,"MR_GL_partitioning"), pattern = "*DataSetafterFluxpart.txt") #[-31]
flist <- flist[is.element(substr(flist,1,6),sites)]
## exclude sites that were not processed in the online tool
sites <- sites[is.element(sites,substr(flist,1,6))]
latLongSites <- rbind(
data.frame(site = "DE-Tha", lat = 51, long = 11, timeOffset = -1 )
,data.frame(site = "IT-MBo", lat = 45.0, long = 1, timeOffset = 0 )
)
tmp <- read.csv(file.path(path,"CommonAnc.csv"), colClasses = c(Site.ID = "character", Latitude = "numeric", Longitude = "numeric", UTC = "numeric"), na.strings = "TBD")
latLongSites <- data.frame(site = tmp$Site.ID, lat = tmp$Latitude, long = tmp$Longitude, timeOffset = floor(-tmp$UTC))
# create output directory if not existing
if (!dir.exists(file.path(path,"Results",scenConf$ouputPath)) ) dir.create( file.path(path,"Results",scenConf$ouputPath) )
###########################
#---- Reading data
# NEE_orig already ustar filtered!! as in Papale_2006 most conservative threshold,
# the seasons are computed in a different manner (not e.g. JFM)
# gapfilling according to Reichstein_2005
iProcSites <- seq_along(sites)
.tmp.inspectSingleSites <- function(){
#s <- grep("CA-TP3",sites)[1]
s <- grep("DE-Hai",sites)[1]
s <- grep("DE-Tha",sites)[1] # large differences with omitSmoothTempSens in month 8
s <- grep("FR-Gri",sites)[1]
s <- grep("ES-VDA",sites)[1]
s <- grep("IL-Yat",sites)[1] # high PAR with low temp-diff, subsetting data does not impar estimates here
s <- grep("PT-Esp",sites)[1] # largest differences with omitSmoothTempSens month 4 (but consider entire data for effect of smoothing)
s <- grep("IT-Amp",sites)[1] # largest differences with pvWave in month 7
siteName <- sites[s]
fileName <- flist[s]
#iProcSites <- 2:3
#try(rm(REddy.mm.all)); try(REddy.yy.all)
}
computeSite <- function(siteName, fileName, scenConf){
year <- as.numeric(substr(fileName, nchar(fileName) - 28, nchar(fileName) - 25))
latLongSite <- unlist(subset(latLongSites, site == siteName)[1,2:4])
#+++ Loading data from MR partitioning and data for running the partitioning
dfall <- fLoadTXTIntoDataframe(fileName, file.path(path,"MR_GL_partitioning"))
.tmp.readPvWave <- function(){
fname.PVwave <- paste(siteName,'.',year,'.','DataSetafterFluxpartGL2010.txt', sep = "")
dfallPv <- read.table(file.path(path,"MR_GL_partitioning",fname.PVwave,sep = ""),skip = 2)
title <- scan(file.path(path,"MR_GL_partitioning",fname.PVwave,sep = ""), nlines = 1, sep = "", strip.white = TRUE,
what = list(rep('character',17)))
names(dfallPv) <- title[[1]]
dfallPv$julday <- dfall$julday
#
iRowEst <- seq(97, 17473, by = 96)
dsE0Pv <- data.frame(E0 = dfallPv$E0[iRowEst], sdE0 = 20, RRef = dfallPv$rb[iRowEst])
dsE0Pv[ dsE0Pv == -9999 ] <- NA
dsE0Pv$E0 <- fillNAForward(dsE0Pv$E0)
dsE0Pv$RRef <- fillNAForward(dsE0Pv$RRef)
}
ctrlOpt <- scenConf$ctrl[[1]]
#ctrlOpt$isNeglectPotRadForNight <- TRUE
#ctrlOpt$fixedTempSens = dsE0Pv
useSolarTime <- if (length(ctrlOpt$useSolarTime) ) ctrlOpt$useSolarTime else TRUE
dfall$PotRadSolar <- as.numeric(fCalcPotRadiation(dfall$julday,dfall$Hour,latLongSite["lat"],latLongSite["long"],latLongSite["timeOffset"], useSolartime.b = TRUE))
dfall$PotRad <- as.numeric(fCalcPotRadiation(dfall$julday,dfall$Hour,latLongSite["lat"],latLongSite["long"],latLongSite["timeOffset"], useSolartime.b = useSolarTime))
#plot( PotRad ~ PotRadSolar, dfall)
# here fake potential radiation to be high on all records where Lasslop Partitioning quantified non-night
dfall$day <- (1 - dfall$night)*100
dfall_posix <- dsMonth <- fConvertTimeToPosix(dfall, 'YMDH', Year.s = 'Year', Month.s = 'Month', Day.s = 'Day', Hour.s = 'Hr')
# dsMonth <- subset(dfall_posix, Month %in% 6:8) #& DateTime >= "2002-08-09 00:00:00" & DateTime <= "2002-08-12 23:30:00")
#
# START - RUN THE REddyProc DT partitioning
#ctrlOpt$isAssociateParmsToMeanOfValids <- FALSE # in order to match exact source data
#ctrlOpt$fixedTempSens = data.frame(E0 = 220, sdE0 = 50, RRef = 2.7)
dsResOpt <- partitionNEEGL(dsMonth,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0
,RadVar.s = "Rg_f"
#,PotRadVar.s = "day"
,PotRadVar.s = "PotRad"
,suffix = "",
controlGLPart = ctrlOpt
)
dsResOpt$DateTime <- dsMonth$DateTime
dsResOpt$julday <- dsMonth$julday
.tmp.debug <- function(){
ctrlDefault <- partGLControl(); ctrlDefault$isAssociateParmsToMeanOfValids <- FALSE; nBootUncertainty = 3L # in order to match exact source data
#ctrlDefault$fixedTempSens <- ctrl$fixedTempSens
dsResDefault <- partitionNEEGL(dsMonth,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0,RadVar.s = "Rg",PotRadVar.s = "day",suffix = "",
controlGLPart = ctrlDefault
)
dsResDefault$DateTime <- dsResOpt$DateTime <- dsMonth$DateTime
dsResOpt$Day <- dsResDefault$Day <- dsMonth$Day
dsResOpt$julday <- dsResDefault$julday <- dsMonth$julday
plot(FP_beta ~ julday, dsResOpt, col = "red", pch = "x", ylim = range(c(dsResOpt$FP_beta, dsResDefault$FP_beta),na.rm = TRUE))
points(FP_beta ~ julday, dsResDefault)
dsResOpt$diffFP_beta <- dsResOpt$FP_beta - dsResDefault$FP_beta
subset( dsResOpt, is.finite(FP_errorcode), c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night"))
subset( dsResDefault, is.finite(FP_errorcode), c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night"))
#plot(diffFP_beta ~ julday, dsRes, col = "red", pch = "x")
#plot(diffFP_beta ~ DateTime, subset(dsRes, julday %in% 200:250), col = "red", pch = "x")
#plot(diffFP_beta ~ julday, subset(dsRes, julday %in% 210:230 & is.finite(FP_beta), col = "red", pch = "x")
.tmp.inspectE0 <- function(){
Sys.setlocale(locale = "C") # for displying English time axis
plot( FP_E0 ~ DateTime, dsResOpt, ylim = range(c(dsResOpt$FP_E0, dsResDefault$FP_E0),na.rm = TRUE), pch = "x", col = "red" )
lines(FP_E0 ~ DateTime, subset(dsResDefault, is.finite(FP_E0)))
plot( dsResOpt$FP_E0 ~ dsResDefault$FP_E0)
isInPeriod <- (dsMonth$DateTime >= "2004-03-01" & dsMonth$DateTime <= "2004-05-01")
.commonCols <- intersect(names(dsResDefault), names(dsResOpt))
dss <- rbind(
cbind( method = "Default", subset( dsResDefault[,.commonCols], isInPeriod & is.finite(FP_E0)))
,cbind( method = scenConf$scen, subset( dsResOpt[,.commonCols], isInPeriod & is.finite(FP_E0)))
)
ggplot( dss, aes(DateTime, FP_E0, color = method, fill = method)) +
#geom_ribbon(aes(ymin = FP_E0-1.96*FP_E0_sd, ymax = FP_E0+1.96*FP_E0_sd), alpha = 0.3) + ylim(-250,750) +
#geom_ribbon(aes(ymin = FP_E0-FP_E0_sd, ymax = FP_E0+FP_E0_sd), alpha = 0.3) + ylim(-50,550) +
geom_line() + theme_bw() + theme(legend.position = "bottom") +
ylab(bquote(E[0]*' (K)')) +
theme(legend.position = "none") +
theme(axis.title.x = element_blank()) + theme( legend.title = element_blank() )
#ggsave("../../eclipse46_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensE0_PT-Esp.pdf", width = 3.27, height = 2.5)
#ggsave("../../eclipse_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensE0_PT-Esp.pdf", width = 3.27, height = 2.5)
isInPeriod <- (dsMonth$DateTime >= "2004-04-01" & dsMonth$DateTime <= "2004-04-20")
dss <- rbind(
cbind( method = "Default", subset( dsResDefault[,.commonCols], isInPeriod))
,cbind( method = scenConf$scen, subset( dsResOpt[,.commonCols], isInPeriod))
)
ggplot( dss, aes(DateTime, GPP_DT, color = method)) + geom_line() + theme_bw() + theme(legend.position = "bottom") +
ylab(bquote(GPP*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme(axis.title.x = element_blank()) + theme( legend.title = element_blank() )
#ggsave("../../eclipse46_R/paper17_REddyProc/figOrig/part_LRC_SmoothTempSensGPP_PT-Esp.pdf", width = 3.27, height = 2.5)
}
# in partGLFitLRCOneWindow:
#if (as.POSIXlt(dsDay$sDateTime[1])$mday+2L > 27 ) recover()
# save(dsDay, file = "tmp/dsDayDebug.RData")
load(file = "tmp/dsDayDebug.RData")
dsDay <- arrange_(dsDay, ~Rg)
plot( -NEE ~ Rg, dsDay)
DayInspect <- 7#11
(thetaDef <- unlist(subset(dsResDefault, Day == DayInspect & is.finite(FP_beta), c("FP_k","FP_beta","FP_alpha","FP_RRef","FP_E0","FP_RRef_Night") )))
(thetaOpt <- unlist(subset(dsResOpt, Day == DayInspect & is.finite(FP_beta), c("FP_k","FP_beta","FP_alpha","FP_RRef","FP_E0","FP_RRef_Night") )))
lrcFitter <- RectangularLRCFitter()
dsDay$GPPDefault <- lrcFitter$predictGPP( dsDay$Rg, thetaDef["FP_beta"], thetaDef["FP_alpha"])
dsDay$GPPOpt <- lrcFitter$predictGPP( dsDay$Rg, thetaOpt["FP_beta"], thetaOpt["FP_alpha"])
dsDay$NEPDefault <- lrcFitter$predictLRC( thetaDef, dsDay$Rg, dsDay$VPD, dsDay$Temp)$NEP
dsDay$NEPOpt <- lrcFitter$predictLRC( thetaOpt, dsDay$Rg, dsDay$VPD, dsDay$Temp)$NEP
ylim <- range(-dsDay$NEE, na.rm = TRUE); ylim[2] <- max(ylim[2], thetaDef["FP_beta"], thetaOpt["FP_beta"], na.rm = TRUE )*1.1
maxRg <- max(1200,max(dsDay$Rg,na.rm = TRUE))
maxRg <- max(2000,max(dsDay$Rg,na.rm = TRUE))
#
plot( -NEE ~ Rg, dsDay)
lines( NEPDefault ~ Rg, dsDay )
lines( NEPOpt ~ Rg, dsDay, col = "red" )
points(maxRg, thetaDef["FP_beta"]); points(maxRg, thetaOpt["FP_beta"], col = "red", pch = "x")
#
plot( GPPDefault ~ Rg, dsDay, xlim = c(0,maxRg), ylim = ylim)
plot( GPPOpt ~ Rg, dsDay, xlim = c(0,maxRg), ylim = ylim, col = "red")
abline(h = thetaDef["FP_beta"]); abline(h = thetaOpt["FP_beta"], col = "red", pch = "x")
#
RgSeq <- seq(0,maxRg, length.out = 80)
predGPP <- lrcFitter$predictLRC( thetaOpt, Rg = RgSeq, VPD = 0, Temp = NA)$GPP
lines( predGPP ~ RgSeq, col = "red")
}
### add modelled NEE
dsResOpt$NEE_DT <- -(dsResOpt$GPP_DT - dsResOpt$Reco_DT)
## save data frames, but omit list columns
iListColumns <- which( sapply( dsResOpt, is.list ) )
dsResOptNotListCol <- if (length(iListColumns)) dsResOpt[-iListColumns] else dsResOpt
write.table(dsResOptNotListCol,file = file.path(path,"Results",scenConf$ouputPath,paste0(siteName,".txt")),row.names = F,col.names = T)
#
# extract those rows where fitting object is not NULL abd save
dsFits <- as_tibble(dsResOpt)[,iListColumns] %>% mutate_(row = ~1:n()) %>% rowwise() %>% filter_(~!is.null(FP_OPT_VPD[[1]]))
save( dsFits ,file = file.path(path,"Results",scenConf$ouputPath,paste0(siteName,"_fits.RData")))
## the _agg columns are used for aggregation in aggregate() function below
julday_agg <- c(1,dfall$julday[1:(nrow(dsResOpt)-1)])
Month_agg <- c(1,dfall$Month[1:(nrow(dsResOpt)-1)])
Year_agg <- c(dfall$Year[1],dfall$Year[1:(nrow(dsResOpt)-1)])
# Aggregation Monthly per site
df.REddy.mm <- aggregate(dsResOpt,by = list(Year_agg = Year_agg, Month_agg = Month_agg),mean,na.rm = T)
# sd
df.REddy.mm$Reco_DT_SD <- aggregate(dsResOpt$Reco_DT_SD,by = list(Year_agg = Year_agg, Month_agg = Month_agg),aggregate.sd)[,3]
df.REddy.mm$GPP_DT_SD <- aggregate(dsResOpt$GPP_DT_SD,by = list(Year_agg = Year_agg, Month_agg = Month_agg),aggregate.sd)[,3]
df.REddy.mm <- cbind(Site = siteName,df.REddy.mm)
# Aggregation Annual
df.REddy.yy <- aggregate(dsResOpt,by = list(Year_agg = Year_agg),mean,na.rm = T)
# sd
df.REddy.yy$Reco_DT_SD <- aggregate(dsResOpt$Reco_DT_SD,by = list(Year_agg = Year_agg),aggregate.sd)[,2]
df.REddy.yy$GPP_DT_SD <- aggregate(dsResOpt$GPP_DT_SD,by = list(Year_agg = Year_agg),aggregate.sd)[,2]
df.REddy.yy <- cbind(Site = siteName,df.REddy.yy)
#
list(
mm = df.REddy.mm
,yy = df.REddy.yy
)
}
.tmp.f <- function(){
REddy.mm.all$Site <- REddy.mm.all$siteName
save(REddy.mm.all, file = "all_sites_monthly.RData")
#
REddy.yy.all$Site <- REddy.yy.all$siteName
save(REddy.yy.all, file = "all_sites_annual.RData")
}
ansList <- foreach (s = iProcSites, .errorhandling = "pass", .packages = c("REddyProc","plyr","mlegp","logitnorm")) %dopar% {
#ansList <- lapply( iProcSites, function(s){
siteName <- sites[s]
fileName <- flist[s]
message("-------- starting site ",siteName)
computeSite(siteName,fileName, scenConf)
}# end of site loop
#)
str(ansList, max.level = 2)
REddy.mms <- lapply(ansList, "[[" , 1L)
REddy.mm.all <- do.call( rbind, REddy.mms)
REddy.yys <- lapply(ansList, "[[" , 2L)
REddy.yy.all <- do.call( rbind, REddy.yys)
## save as RData:
#save(NT_vs_DT_REddy,DT_REddy_vs_pvwave,file = file.path(path,"Results",scenConf$ouputPath,"/eval_metrics.RData")) # 1) evaluation metrics
outputDir <- file.path(path,"Results",scenConf$ouputPath)
save(REddy.mm.all,file = file.path(outputDir,"all_sites_monthly.RData")) # 2) monthly aggregated results for all sites
save(REddy.yy.all,file = file.path(outputDir,"all_sites_annual.RData")) # 3) annual aggregated results for all sites
# write scenario information
save(scenConf, file = file.path(outputDir,"scenConf.RData"))
# generate readmegen.txt
sink(file.path(outputDir,'readmegen.txt')); {
cat("Generated be REddyProc/inst/compareOldTool/GLPartitioningTestCase.R\n")
cat(date(),"\n\n")
cat("scenConf.RData: list of information on scenario.\n")
cat("all_sites_monthly.RData: data.frame of partitioning outputs across sites and months.\n")
cat("all_sites_annual.RData: data.frame of partitioning outputs across sites for a single year.\n")
cat("<siteName>.txt: csv file of partitioning outputs\n")
cat("<siteName>_fits.RData: tibble of fitted objects and central row number of corresponding window rows\n\n")
cat("scenConf: ")
str(scenConf, max.level = 3)
}; sink()
.tmp.lookForHighPar <- function(){
#duplated column names?
#ds <- dfall_posix[,!(names(dfall) %in% c("Day","Hour"))]
ds <- partGLExtractStandardData(dfall_posix,NEEVar.s = "NEE_f",QFNEEVar.s = "NEE_fqc",QFNEEValue.n = 0,NEESdVar.s = "NEE_fs_unc",
TempVar.s = "Tair_f",QFTempVar.s = "Tair_fqc",QFTempValue.n = 0,VPDVar.s = "VPD_f",QFVPDVar.s = "VPD_fqc",
QFVPDValue.n = 0,RadVar.s = "Rg",PotRadVar.s = "day",suffix = "",
controlGLPart = ctrlOpt)
ds$julday <- dfall_posix$julday
.tmp.maxPAR <- function(){
dailyMaxPar <- ds %>% group_by_(~julday) %>% summarize_(Rg = ~max(Rg, na.rm = TRUE))
arrange_(dailyMaxPar, ~desc(Rg))
ggplot( dailyMaxPar, aes(julday, Rg)) + geom_point()
}
# extract data of day 174
julDayInspect <- 214
dss <- subset(ds, julday %in% (julDayInspect+(-2:1)))
isValidDayRecNoVPDConstraint <- !is.na(dss$isDay) & dss$isDay & !is.na(dss$NEE) & !is.na(dss$sdNEE) & !is.na(dss$Temp) & !is.na(dss$Rg)
dsDay <- dss[isValidDayRecNoVPDConstraint,]
p1 <- ggplot( dsDay, aes(Rg,-NEE, col = Temp)) + geom_point(); p1
#p1 <- ggplot( dsDay, aes(Rg,-NEE, col = sdNEE)) + geom_point(); p1
lrcFitter <- RectangularLRCFitter()
#ctrlOpt <- partGLControl( ) # use same as in debug above
dsE0Opt <- subset( dsResOpt, is.finite(FP_beta) & julday == julDayInspect, c("FP_E0","FP_E0_sd","FP_RRef","FP_RRef_Night"))
resOpt <- resOpt0 <- lrcFitter$fitLRC(dsDay
#, E0 = 130, sdE0 = 50, RRefNight = 1 # IL-Yat day 148
#, E0 = 120, sdE0 = 50, RRefNight = 3 # DE-Hai day
, E0 = dsE0Opt$FP_E0, sdE0 = dsE0Opt$FP_E0_sd, RRefNight = dsE0Opt$FP_RRef_Night # from dsRes in debug above
, controlGLPart = ctrlOpt)
(thetaOpt <- resOpt$thetaOpt)
dsE0Default <- subset( dsResDefault, is.finite(FP_beta) & julday == julDayInspect, c("FP_E0","FP_E0_sd","FP_RRef","FP_RRef_Night"))
resDefault <- lrcFitter$fitLRC(dsDay
#, E0 = 130, sdE0 = 50, RRefNight = 1 # IL-Yat day 148
#, E0 = 120, sdE0 = 50, RRefNight = 3 # DE-Hai day
, E0 = dsE0Default$FP_E0, sdE0 = dsE0Default$FP_E0_sd, RRefNight = dsE0Default$FP_RRef_Night # from dsRes in debug above
, controlGLPart = ctrlDefault)
(thetaDefault <- resDefault$thetaOpt)
#(thetaOpt <- resOpt$theta)
dsPred <- data.frame(Rg = seq(0,2000, length.out = 80), Temp = median(dsDay$Temp))
dsPredOpt <- {
dsPred$GPP <- lrcFitter$predictLRC( thetaOpt, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$GPP
dsPred$NEP <- lrcFitter$predictLRC( thetaOpt, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$NEP
cbind(scenario = scenConf$scen, dsPred)
}
dsPredDefault <- {
dsPred$GPP <- lrcFitter$predictLRC( thetaDefault, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$GPP
dsPred$NEP <- lrcFitter$predictLRC( thetaDefault, Rg = dsPred$Rg, VPD = 0, Temp = dsPred$Temp)$NEP
cbind(scenario = "Default", dsPred)
}
dsPred <- rbind(dsPredOpt, dsPredDefault)
#ggplot( dsPred, aes(Rg, GPP)) + geom_line()
# in paper example DE-Hai, julday %in% (220+(0:4))
p1 + geom_line(data = dsPred, aes(Rg, GPP, linetype = scenario), color = "black") +
xlab(bquote('Rg ('*W~m^-2*')')) +
ylab(bquote(NEP[Obs]*" & "*GPP[mod]*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme_bw(base_size = 9) +
theme(legend.position = c(0.95,0.05), legend.justification = c(1,0))
p1 + geom_line(data = dsPred, aes(Rg, NEP, linetype = scenario), color = "black") +
xlab(bquote('Rg ('*W~m^-2*')')) +
ylab(bquote(NEP[Obs]*' ('*mu~ 'mol' ~CO[2]~ m^-2~s^-1*')')) +
theme_bw(base_size = 9) +
theme(legend.position = c(0.95,0.05), legend.justification = c(1,0))
# on period of high PAR-range, e.g. Yatir, experiment with decreasing dataset to lower bounds
# result: similar estimates also from constrained dataset -> not not constrain GPP2000, only filter extrem cases
maxRgs <- c(1200,1000,800,600,400)
thetaOptRg <- cbind( maxRg = maxRgs, as.data.frame(t(sapply( maxRgs, function(maxRg){
dsDay <- subset(dsDay, Rg <= maxRg)
resOpt <- resOpt0 <- lrcFitter$fitLRC(dsDay, controlGLPart = ctrlOpt
, E0 = 120, sdE0 = 50, RRefNight = 3)
c(nRec = nrow(dsDay), resOpt$thetaOpt)
}))))
thetaOptRg
}
.tmp.comparePvWaveREddyProc <- function(){
iRowEst <- which(is.finite(dsResOpt$FP_errorcode))
parR <- dsResOpt[ iRowEst, c("julday","FP_errorcode","FP_beta","FP_alpha","FP_E0","FP_k","FP_RRef","FP_RRef_Night")]
names(parR) <- c("julday","FP_errorcode","beta","alpha","E0","k","rb","rbN")
parP <- dfallPv[ iRowEst, ]
parP[ parP == -9999] <- NA
plot( parP$E0 ~ parR$E0 ); abline(0,1) # large differences
#plot( Rg_f ~ DateTime, dfall_posix[dfall$night == 1,] )
plot( parP$beta ~ parR$beta ); abline(0,1) # large differences
}
stopImplicitCluster()
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.