vignettes/switchingChambers.md

In bgctw/RespChamberProc: Processing Data from Respiration Chambers

author: "Thomas Wutzler" date: "2024-06-13" output: rmarkdown::html_vignette: keep_md: true vignette: > %\VignetteEngine{knitr::knitr} %\VignetteIndexEntry{Processing several measurement cycles of different setup} %\usepackage[UTF-8]{inputenc}

Processing several measurement cycles of different setup

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.

library(dplyr)
# fit chambers in parallel inside calcClosedChamberFluxForChunkSpecs
library(furrr)
plan(multisession, workers = 4) 

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")

## # A tibble: 6 × 17
##   TIMESTAMP           RECORD Chamber Collar AirTemp AirPres   PAR BodyTemp SurTemp SoilTemp SoilMoist CO2_LI840 H2O_LI840 T_LI840 P_LI840 PTemp  Batt
##   <dttm>               <int>   <int>  <int>   <dbl>   <dbl> <dbl>    <dbl>   <dbl>    <dbl>     <dbl>     <dbl>     <dbl>   <dbl>   <dbl> <dbl> <dbl>
## 1 2015-03-26 06:18:28 261827       1      0    5.14    988.  5.30     6.46    5.10     6.15      18.1      465.      7.30    51.5    98.8  7.53  12.7
## 2 2015-03-26 06:18:29 261828       1      0    5.14    988.  5.30     6.46    5.11     6.18      18.1      465.      7.29    51.5    98.8  7.53  12.7
## 3 2015-03-26 06:18:30 261829       1      0    5.14    988.  5.30     6.46    5.12     6.21      18.1      465.      7.30    51.5    98.8  7.53  12.7
## 4 2015-03-26 06:18:31 261830       1      0    5.13    988.  5.30     6.45    5.10     6.25      18.1      465.      7.29    51.5    98.8  7.53  12.7
## 5 2015-03-26 06:18:32 261831       1      0    5.13    988.  5.30     6.45    5.10     6.29      18.1      466.      7.28    51.6    98.8  7.53  12.6
## 6 2015-03-26 06:18:33 261832       1      0    5.13    988.  5.30     6.45    5.10     6.32      18.1      465.      7.28    51.5    98.8  7.53  12.6

The dataset contains several measurement cycles of light and dark chambers with increasing or decreasing concentrations 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)

## # A tibble: 6 × 23
##   iChunk TIMESTAMP           RECORD Chamber Collar AirTemp AirPres   PAR BodyTemp SurTemp SoilTemp SoilMoist CO2_LI840 H2O_LI840 T_LI840 P_LI840 PTemp  Batt     Pa
##   <fct>  <dttm>               <int>   <int>  <int>   <dbl>   <dbl> <dbl>    <dbl>   <dbl>    <dbl>     <dbl>     <dbl>     <dbl>   <dbl>   <dbl> <dbl> <dbl>  <dbl>
## 1 4      2015-03-26 06:19:20 261864       1      1    4.99    988.  3.31     6.40    5.29     7.57      17.9      468.      7.24    51.6    98.8  7.53  12.6 98774.
## 2 4      2015-03-26 06:19:21 261865       1      1    4.99    988.  3.97     6.39    5.29     7.59      17.9      468.      7.25    51.6    98.8  7.53  12.6 98780.
## 3 4      2015-03-26 06:19:22 261866       1      1    4.98    988.  5.30     6.39    5.3      7.61      18.0      468.      7.25    51.6    98.8  7.53  12.6 98774.
## 4 4      2015-03-26 06:19:23 261867       1      1    4.98    988.  5.30     6.39    5.3      7.62      18.0      468.      7.26    51.6    98.8  7.53  12.6 98780.
## 5 4      2015-03-26 06:19:24 261868       1      1    4.98    988.  5.30     6.39    5.3      7.64      18.0      468.      7.26    51.6    98.8  7.53  12.6 98780.
## 6 4      2015-03-26 06:19:25 261869       1      1    4.98    988.  5.30     6.39    5.30     7.66      17.9      468.      7.26    51.6    98.8  7.53  12.6 98780.
## # ℹ 4 more variables: CO2_dry <dbl>, H2O_dry <dbl>, VPD <dbl>, collar <int>

The modified dataset contains a new variable, iChunk, which reports 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)")

Associating Chamber information to chunks

Different collars may have different depth leading to different volume, or are connected by tubing of different length to the sensor.

The user needs to provide - a mapping of iChunk to collar in column collar - dimensional information about the collar in a data.frame

Function subsetContiguous creates a default for the collar mapping by assigning the values of the index column.

dsChunk %>% group_by(iChunk) %>% summarise(collar = first(collar)) %>%  head()

## # A tibble: 6 × 2
##   iChunk collar
##   <fct>   <int>
## 1 4           1
## 2 5           1
## 3 6           2
## 4 7           2
## 5 8          31
## 6 9          31

DataFrame collar_spec then needs to specify for each collar id in column collar, the colums area (m2) and volume (m3), as well a tlag (s), the lag time between start of the cycle , i.e. the start of the chunk (usually chamber closing time), and the time when the gas reaches the sensor.

In this example, we specify the same surface area and the same tlag for each collar but simulate removing the litter and then measuring the depth of each collar to recompute the volume. The depth, here, are random numbers around 3 cm. If the lagtime is set to missing (NA) then it is estimated in each chunk by a breakpoint detection.

chamberVol = 0.6*0.6*0.6        # chamber was a cube of 0.6m length
surfaceArea = 0.6*0.6
collar_spec <- tibble(
  collar = unique(dsChunk$collar), 
  depth = pmax(0,rnorm(length(collar), mean = 0.03, sd = 0.015)),
  area = surfaceArea,
  volume = chamberVol + surfaceArea * depth,
  tlag = NA)
head(collar_spec)

## # A tibble: 6 × 5
##   collar   depth  area volume tlag 
##    <int>   <dbl> <dbl>  <dbl> <lgl>
## 1      1 0.00862  0.36  0.219 NA   
## 2      2 0.0117   0.36  0.220 NA   
## 3     31 0.0282   0.36  0.226 NA   
## 4     32 0.0517   0.36  0.235 NA   
## 5     28 0.0155   0.36  0.222 NA   
## 6     27 0.0129   0.36  0.221 NA

Problems with association setups to the data can be checked by function checkCollarSpec, which returns FALSE and attribute msg in case of problems.

checkCollarSpec(dsChunk, collar_spec)

Computing the flux

Function calcClosedChamberFluxForChunkSpecs applies function calcClosedChamberFlux to each subset.

# for demonstration use only the first 20 cycles
dsChunk20 <- subset(dsChunk, as.integer(iChunk) <= 20) 
resChunks1 <- calcClosedChamberFluxForChunkSpecs(
  dsChunk20, collar_spec
  , colTemp = "AirTemp"
  # linear and saturating shape
  , fRegress = c(lin = regressFluxLinear, tanh = regressFluxTanh)   
  , debugInfo = list(omitEstimateLeverage = TRUE)   # faster
)
head(resChunks1)

## # A tibble: 6 × 17
##   iChunk collar  flux fluxMedian sdFlux  tLag lagIndex autoCorr   AIC sdFluxRegression sdFluxLeverage iFRegress sdResid iqrResid    r2 times       model 
##   <fct>   <int> <dbl>      <dbl>  <dbl> <dbl>    <int>    <dbl> <dbl>            <dbl>          <dbl>     <dbl>   <dbl>    <dbl> <dbl> <list>      <list>
## 1 4           1  2.58         NA 0.0360     0        1    0.570 -136.           0.0360             NA         2   0.197    0.249 0.998 <dbl [180]> <gnls>
## 2 5           1  2.69         NA 0.0276     0        1    0.486 -156.           0.0276             NA         2   0.175    0.223 0.999 <dbl [181]> <gnls>
## 3 6           2  2.38         NA 0.0303     6        7    0.538 -168.           0.0303             NA         2   0.173    0.228 0.999 <dbl [175]> <gnls>
## 4 7           2  2.59         NA 0.0242     0        1    0.419 -160.           0.0242             NA         2   0.167    0.244 0.999 <dbl [181]> <gnls>
## 5 8          31  1.35         NA 0.0277    11       12    0.447 -177.           0.0277             NA         2   0.157    0.198 0.996 <dbl [171]> <gnls>
## 6 9          31  1.91         NA 0.0111     3        4    0.498 -148.           0.0111             NA         1   0.172    0.197 0.998 <dbl [183]> <gls>

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

Plotting faceted data and fits

We recommend to plot the results together with the concentration data 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 peculiarities 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)

## 
##  0  3  5  6  8 11 12 14 16 
##  9  1  1  2  2  1  1  1  2

The plots do not indicate problems, and the longest estimated lagtime varies between 0 and 16 seconds. Since slightly overestimating the lagtime does not change the flux but only might slightly increase the uncertainty, we infer that for this campaign a lag-time of about 16 seconds is appropriate.

One can save processing time and avoid failures in the non-robust breakpoint-detection by specifying a fixed lag-time (may differ across collars) with the collar specification.

collar_spec2 <- mutate(collar_spec, tlag = 16)

resChunks2 <- calcClosedChamberFluxForChunkSpecs(
  dsChunk20, collar_spec2
  , colTemp = "AirTemp"
  # linear and saturating shape
  , fRegress = c(lin = regressFluxLinear, tanh = regressFluxTanh)   
  , debugInfo = list(omitEstimateLeverage = TRUE)   # faster
)
head(resChunks2)

## # A tibble: 6 × 17
##   iChunk collar  flux fluxMedian sdFlux  tLag lagIndex autoCorr   AIC sdFluxRegression sdFluxLeverage iFRegress sdResid iqrResid    r2 times       model 
##   <fct>   <int> <dbl>      <dbl>  <dbl> <dbl>    <int>    <dbl> <dbl>            <dbl>          <dbl>     <dbl>   <dbl>    <dbl> <dbl> <list>      <list>
## 1 4           1  2.51         NA 0.0348    16       17    0.499 -126.           0.0348             NA         2   0.185    0.250 0.998 <dbl [164]> <gnls>
## 2 5           1  2.62         NA 0.0130    16       17    0.497 -118.           0.0130             NA         1   0.178    0.231 0.999 <dbl [165]> <gls> 
## 3 6           2  2.36         NA 0.0140    16       17    0.557 -137.           0.0140             NA         1   0.175    0.219 0.998 <dbl [165]> <gls> 
## 4 7           2  2.58         NA 0.0293    16       17    0.449 -144.           0.0293             NA         2   0.170    0.260 0.999 <dbl [165]> <gnls>
## 5 8          31  1.36         NA 0.0291    16       17    0.444 -169.           0.0291             NA         2   0.158    0.187 0.995 <dbl [166]> <gnls>
## 6 9          31  1.91         NA 0.0127    16       17    0.497 -127.           0.0127             NA         1   0.176    0.205 0.997 <dbl [170]> <gls>

bgctw/RespChamberProc documentation built on June 11, 2025, 3:52 p.m.