In bgctw/RespChamberProc: Processing Data from Respiration Chambers

# rmarkdown::render("vignettes/severalCycles.Rmd")
library(knitr)
opts_chunk$set(
    #out.extra = 'style = "display:block; margin: auto"'
    #, fig.align = "center"
    fig.width = 4.3, fig.height = 3.2, dev.args = list(pointsize = 10)
    , message = FALSE
    , results = 'hold'
    )
knit_hooks$set(spar = function(before, options, envir) {
    if (before) {
        par( las = 1 )                   #also y axis labels horizontal
        par(mar = c(2.0,3.3,0,0) + 0.3 )  #margins
        par(tck = 0.02 )                          #axe-tick length inside plots             
        par(mgp = c(1.1,0.2,0) )  #positioning of axis title, axis labels, axis
     }
})

Processing several measurement cycles

isDevelopMode <- TRUE
library(twDev)
setwd('..');loadPkg()

if (!exists("isDevelopMode")) library(RespChamberProc)
set.seed(0815)      # for reproducible results

Determine subsets of single measurement cycles

First, the data is loaded. Here, directly from zipped logger-output.

fName <- system.file(
  "genData/SMANIE_Chamber1_26032015.zip", package = "RespChamberProc")
if (nzchar(fName) ) { 
  ds <- readDat(
    unz(fName, filename = unzip(fName, list = TRUE)[1,"Name"] ),tz = "UTC") }
head(ds)
plot( CO2_LI840 ~ TIMESTAMP, ds, ylab = "CO2 (ppm)", xlab = "Time")

The dataset contains several measurement cycles of light and dark chambers with increasing or decreasing concentations respectively.

First, we correct the pressure to standard units and correct the CO2 concentrations for water vapour.

ds$Pa <- ds$AirPres * 100  # convert hPa to Pa
ds$CO2_dry <- corrConcDilution(ds, colConc = "CO2_LI840", colVapour = "H2O_LI840")
ds$H2O_dry <- corrConcDilution(ds, colConc = "H2O_LI840", colVapour = "H2O_LI840")
ds$VPD <- calcVPD( ds$SurTemp, ds$Pa, ds$H2O_LI840)

In order to process each measurement cycle independently, we first determine parts of the time series that are contiguous, i.e. without gaps and without change of an index variable, here variable collar.

dsChunk <- subsetContiguous(ds, colTime = "TIMESTAMP", colIndex = "Collar") 
head(dsChunk)

The new modified dataset contains a new variable, iChunk, holding a factor that changes with different measurement cycles. This factor can be used to select subset of single measurement cycles.

dss <- subset(dsChunk, iChunk == 15)
plot( CO2_dry ~ TIMESTAMP, dss, ylab = "CO2 (ppm)", xlab = "time (Minute:Second)")

Computing the flux

Function calcClosedChamberFluxForChunks helps with subsetting the data and applying function calcClosedChamberFlux to each subset.

# for demonstration use only the first 20 cycles
dsChunk20 <- subset(dsChunk, as.integer(iChunk) <= 20) 
chamberVol = 0.6*0.6*0.6        # chamber was a cube of 0.6m length
surfaceArea = 0.6*0.6

resChunks1 <- calcClosedChamberFluxForChunks(
  dsChunk20, colTemp = "AirTemp"
  # linear and saturating shape
  , fRegress = c(lin = regressFluxLinear, tanh = regressFluxTanh)   
  , debugInfo = list(omitEstimateLeverage = TRUE)   # faster
  , volume = chamberVol
  , area = surfaceArea
)
head(resChunks1)

The results are similar as for calcClosedChamberFlux, unless there are several rows identified by additional key column iChunk.

Plotting faceted data and fits

Plot the results to dectect problems.

library(ggplot2)
plots <- plotCampaignConcSeries( dsChunk20, resChunks1, plotsPerPage = 64L)  
print(plots$plot[[1]]) # print the first page

If argument fileName is provided to plotCampaignConcSeries. All plots are written to a pdf. If there are more cycles, i.e. plots, than argument plotsPerPage(default 64) there will be several pages in the pdf.

Inspecting lag-times

Lag times between closing the chamber and the start of the concentration increase, i.e. when the gas arrives at the sensor, is by default estimated by a breakpoint detection method. This method is not robust to fluctuations, early saturation, or other possible pecularities of the concentration time series. In other to detect those subsets, where lag-time detection has failed, on can inspect the inferred lag-times for outliers.

For a campaign where all the measurement cycles were performed with similar conditions, the lag-time should not differ much.

table(resChunks1$tLag)

We infer that for this campaign a lag-time of about 15 seconds is appropriate.

One can save processing time and avoid breakpoint-detection failures by specifying a fixed lag-time during the concentration fitting by parameter useFixedTLag.

resChunks2 <- calcClosedChamberFluxForChunks(
  dsChunk20, colTemp = "T_LI840"
  # linear and saturating shape
  , fRegress = c(lin = regressFluxLinear, tanh = regressFluxTanh)   
  , debugInfo = list(omitEstimateLeverage = TRUE)   # faster
  , volume = chamberVol
  , area = surfaceArea
  , useFixedTLag = 15
)
head(resChunks2)

ds2 <- resChunks2 %>% 
  select(iChunk, flux) %>% 
  rename( fluxFixed = flux )
ds <- left_join(resChunks1, ds2, by = "iChunk")
#ggplot( ds, aes(flux, fluxFixed)) + geom_point()
ggplot( ds, aes(flux, I((fluxFixed - flux)/flux)*100) ) + geom_point()