Initial BioFertilizer Experiment Data Analysis

This R Publication is intended to show some initial data analysis from the first three test BioFertilizer treatments. Because there are no controls or replicates available in the data, this analysis will have to be limited only to asking whether we see significantly different responses from the three treatments.

First, we do some data processing to prepare for visualization and analysis.

Plot the raw data

To the naked eye, the responses of the three treatments appear to be quite similar. But there is a lot of noise in the data. Let’s dig deeper.

Break out the plots by individual sensor, with some fit lines

It looks to me like the response from the Water-only treatment is parabolic, where the other two might respond quickly and then plateau. Let’s remove some of the noise from the signal:

Look at average signals by treatment type to visually look for overall trends

This makes it look like the two BioFertilizer treatments are more similar to one another, whereas the water-only treatment looks like it may be following different trends.

Look at some shorter decomposition windows

In general, our approach for decomposition sensor data analysis has been to extract individual sensor slopes and then look to correlate these slopes to other parameters of interest, for example the enzymatic activity within each instrumented plot. This initial experiment does not include enough replicates or control conditions to have a good sense of what windows should be chosen, so some arbitrary breakpoints will be considered, breaking the timeseries into three windows:

#Define breakpoints for starts and ends of windows
bp1 <- rEnd
bp2 <- as.POSIXct("2024-07-15 00:00:00")
bp3 <- as.POSIXct("2024-07-17 15:00:00")
bp4 <- endDate

#Define the factor "window" based on breakpoints
aveData$window <- 0
aveData$window[aveData$dateTime > bp1 & aveData$dateTime < bp2] <- 1
aveData$window[aveData$dateTime > bp2 & aveData$dateTime < bp3] <- 2
aveData$window[aveData$dateTime > bp3 & aveData$dateTime < bp4] <- 3
aveData$window <- as.factor(aveData$window)

Without more data, it’s very hard to say how different these responses really are. However, it’s clear that both the initial slope and the slope further into the experiment differs across treatments, especially between the water-only signal and the other two signals.

To put some numbers to this, we can look at the slopes within each of the windows:

## # A tibble: 12 × 3
## # Groups:   treatment [3]
##    treatment                 window    slope
##    <fct>                     <fct>     <dbl>
##  1 BioFertilizer Frontloaded 0       0.0999 
##  2 BioFertilizer Over Time   0       0.100  
##  3 Water                     0       0.120  
##  4 BioFertilizer Frontloaded 1       0.00588
##  5 BioFertilizer Over Time   1       0.00812
##  6 Water                     1       0.0149 
##  7 BioFertilizer Frontloaded 2       0.0196 
##  8 BioFertilizer Over Time   2       0.0162 
##  9 Water                     2       0.0105 
## 10 BioFertilizer Frontloaded 3      -0.00100
## 11 BioFertilizer Over Time   3      -0.00148
## 12 Water                     3      -0.00807