Check response curve data for common issues — check_response_curve

Checks to make sure an exdf object representing multiple response curves meets basic expectations.

Usage

check_response_curve_data(
    exdf_obj,
    identifier_columns,
    expected_npts = 0,
    driving_column = NULL,
    driving_column_tolerance = 1.0,
    col_to_ignore_for_inf = 'gmc',
    constant_col = list(),
    error_on_failure = TRUE,
    print_information = TRUE
  )

Arguments

exdf_obj: An exdf object representing multiple response curves.
identifier_columns: A vector or list of strings representing the names of columns in exdf_obj that, taken together, uniquely identify each curve. This often includes names like plot, event, replicate, etc.
expected_npts: A numeric vector of length 1 or 2 specifying conditions for the number of points in each curve. If expected_npts is set to a negative number, then this check will be skipped. See below for more details.
driving_column: The name of a column that is systematically varied to produce each curve; for example, in a light response curve, this would typically by Qin. If driving_column is NULL, then this check will be skipped.
driving_column_tolerance: An absolute tolerance for the deviation of each value of driving_column away from its mean across all the curves; the driving_column_tolerance can be set to Inf to disable this check.
col_to_ignore_for_inf: Any columns to ignore while checking for infinite values. Mesophyll conductance (gmc) is often set to infinity intentionally so should be ignored when performing this check. To completely disable this check, set col_to_ignore_for_inf to NULL.
constant_col: A list of named numeric elements, where the name indicates a column of exdf_obj that should be constant, and the value indicates whether the column's values must be identical or whether they must lie within a specified numeric range. If constant_col is an empty list, then this check will be skipped. See below for more details.
error_on_failure: A logical value indicating whether to send an error message when an issue is detected. See details below.
print_information: A logical value indicating whether to print additional information to the R terminal when an issue is detected. See details below.

Details

Basic Behavior:

This function makes a few basic checks to ensure that the response curve data includes the expected information and does not include any mistakes. If no problems are detected, this function will be silent with no return value. If a problem is detected, then the user will be notified in one or more ways:

If error_on_failure is TRUE, then this function will throw an error with a short message. If print_information is also TRUE, then additional information will be printed to the R terminal.
If error_on_failure is FALSE and print_information is also FALSE, then this function will throw a warning with a short message.
If error_on_failure is FALSE and print_information is true, information about the problem will be printed to the R terminal.

This function will (optionally) perform several checks:

Checking for infinite values: If col_to_ignore_for_inf is not NULL, no numeric columns in exdf_obj should have infinite values, with the exception of columns designated in col_to_ignore_for_inf.
Checking required columns: All elements of identifier_columns should be present as columns in exdf_obj. If driving_column is not NULL, it should also be present as a column in exdf_obj. If constant_col is not empty, then these columns must also be present in exdf_obj.
Checking the number of points in each curve: The general idea is to ensure that each curve has the expected number of points. Several options can be specified via the value of expected_npts, as discussed below.
Checking the driving column: If driving_column is not NULL, then each curve should have the same sequence of values in this column. To allow for small variations, a nonzero driving_column_tolerance can be specified.
Checking the constant columns: If constant_col is not empty, then each specified column should either be constant, or only vary by a specified amount. See details below.

By default, most of these are not performed (except the simplest ones like checking for infinite values or checking that key columns are present). This enables an "opt-in" use style, where users can specify just the checks they wish to make.

More Details:

There are several options for checking the number of points in each curve:

If expected_npts is a single negative number, no check will be performed.
If expected_npts is 0, then each curve is expected to have the same number of points.
If expected_npts is a single positive number, then each curve is expected to have that many points. For example, if expected_npts is 7, then each curve must have 7 points.
If expected_npts is a pair of positive numbers, then each curve is expected to have a number of points lying within the range defined by expected_npts. For example, if expected_npts is c(6, 8), then each curve must have no fewer than 6 points and no more than 8 points.
If expected_npts is a pair of numbers, one of which is zero and one of which is positive, then the positive number specifies a range; each curve must differ from the average number of points by less than the range. For example, if expected_npts is c(0, 3), then every curve must have a number of points within 3 of the average number of points.

There are two options for checking columns that should be constant:

A value of NA indicates that all values of that column must be exactly identical; this check applies for numeric and character columns.
A numeric value indicates that the range of values of that column must be smaller than the specified range; this range applies for numeric columns only.

For example, setting constant_col = list(species = NA, Qin = 10) means that each curve must have only a single value of the species column, and that the value of the Qin column cannot vary by more than 10 across each curve.

Use Cases:

Using check_response_curve_data is not strictly necessary, but it can be helpful both to you and to anyone else reading your analysis code. Here are a few typical use cases:

Average response curves: It is common to calculate and plot average response curves, either manually or by using xyplot_avg_rc. But, it only makes sense to do this if each curve followed the same sequence of the driving variable. In this case, check_response_curve_data can be used to confirm that each curve used the same values of CO2_r_sp (for an A-Ci curve) or Qin (for an A-Q curve).
Removing recovery points: It is common to wish to remove one or more recovery points from a set of curves. The safest way to do this is to confirm that all the curves use the same sequence of setpoints; then you can be sure that, for example, points 9 and 10 are the recovery points in every curve.
Making a statement of expectations: If you measured a set of A-Ci curves where each curve has 16 points and used the same sequence of CO2_r setpoints, you could record this somewhere in your notes. But it would be even more meaningful to use check_response_curve_data in your script with expected_npts set to 16. If this check passes, then it means not only that your claim is correct, but also that the identifier columns are being interpreted properly.
Checking identifiers: If the data set includes some identifying metadata, such as a species or location, it may be helpful to confirm that each curve has a single value of these "identifier" columns. Otherwise, the data set may be difficult to interpret.
Checking measurement conditions: If the response curves are expected to be measured under constant temperature, humidity, light, or other environmental variables, it may be helpful to confirm that these variables do not vary too much across each individual curve. Otherwise, parameter values estimated from the curves may not be meaningful.

Sometimes the response curves in a large data set were not all measured with the same sequence of setpoints. If only a few different sequences were used, it is possible to split them into groups and separately run check_response_curve_data on each group. This scenario is discussed in the Frequently Asked Questions vignette.

Even if none of the above situations are relevant to you, it may still be helpful to run run check_response_curve_data but with expected_npts set to 0 and error_on_failure set to FALSE. With these settings, if there are curves with different numbers of points, the function will print the number of points in each curve to the R terminal, but won't stop the rest of the script from running. This can be useful for detecting problems with the curve_identifier column. For example, if the longest curves in the set are known to have 17 points, but check_response_curve_data identifies a curve with 34 points, it is clear that the same identifier was accidentally used for two different curves.

Value

The check_response_curve_data function does not return anything.

Examples

# Read an example Licor file included in the PhotoGEA package and check it.
# This file includes several 7-point light-response curves that can be uniquely
# identified by the values of its 'species' and 'plot' columns. Since these are
# light-response curves, each one follows a pre-set sequence of `Qin` values.
licor_file <- read_gasex_file(
  PhotoGEA_example_file_path('ball_berry_1.xlsx')
)

# Make sure there are no infinite values and that all curves have the same
# number of points
check_response_curve_data(licor_file, c('species', 'plot'))

# Make sure there are no inifinite values and that all curves have 7 points
check_response_curve_data(licor_file, c('species', 'plot'), expected_npts = 7)

# Make sure there are no infinite values, that all curves have 7 points, and
# that the values of the `Qin` column follow the same sequence in all curves
# (to within 1.0 micromol / m^2 / s)
check_response_curve_data(
  licor_file,
  c('species', 'plot'),
  expected_npts = 7,
  driving_column = 'Qin',
  driving_column_tolerance = 1.0
)

# Make sure that there are no infinite values and that all curves have between
# 8 and 10 points; this will intentionally fail
check_response_curve_data(
  licor_file,
  c('species', 'plot'),
  expected_npts = c(8, 10),
  error_on_failure = FALSE
)
#>   species plot npts
#> 1 soybean   1a    7
#> 2 soybean   1b    7
#> 3 tobacco    2    7
#> 4 soybean    5    7
#> 
#> The following curves have fewer than 8 points or greater than 10 points: soybean.1a, soybean.1b, tobacco.2, soybean.5

# Split the data set by `species` and make sure all curves have similar numbers
# of points (within 3 of the mean value); this will intentionally fail
check_response_curve_data(
  licor_file,
  'species',
  expected_npts = c(0, 3),
  error_on_failure = FALSE
)
#>   species npts
#> 1 soybean   21
#> 2 tobacco    7
#> 
#> The following curves have atypical numbers of points (more than 3 away from the mean value of 14): soybean, tobacco

# Split the data set by `species` and make sure all curves have a constant value
# of `plot` and a limited range of `TLeafCnd`; this will intentionally fail
check_response_curve_data(
  licor_file,
  'species',
  constant_col = list(plot = NA, TleafCnd = 0.001),
  error_on_failure = FALSE
)
#>   species npts
#> 1 soybean   21
#> 2 tobacco    7
#> 
#> Not all curves have the same number of points
#> 
#> Curve ID: species = soybean:
#>     The `plot` column takes 3 values: `1a`, `5`, `1b`
#>     The `TleafCnd` column range (4.0666) exceeds the limit (0.001)
#> Curve ID: species = tobacco:
#>     The `TleafCnd` column range (1.8677) exceeds the limit (0.001)
#> 
#> 
#> One or more curves has non-constant values of a column that should be constant.