Repository Mirror for your Cloud Server and Webhosting

Title:

Linear and Nonlinear Regression for Agricultural Data

Version:

0.1.0

Description:

Fit, compare, and visualise linear and nonlinear regression models tailored to field-trial and dose-response agricultural data. Provides S3 classes for mixed-effects models (via 'lme4'), nonlinear growth curves (logistic, 'Gompertz', asymptotic, linear-plateau, quadratic), and four/five-parameter log-logistic dose-response models (via 'drc'). Includes automated starting-value heuristics, goodness-of-fit statistics, residual diagnostics, and 'ggplot2'-based visualisation. Methods are based on Bates and Watts (1988, ISBN:9780471816430), Ritz and others (2015) <doi:10.1371/journal.pone.0146021>, and Bates and others (2015) <doi:10.18637/jss.v067.i01>.

Depends:

R (≥ 4.1.0)

Imports:

lme4 (≥ 1.1-35), drc (≥ 3.0-1), ggplot2 (≥ 3.4.0), patchwork (≥ 1.1.0), stats, utils

Suggests:

broom (≥ 1.0.0), broom.mixed (≥ 0.2.9), emmeans (≥ 1.8.0), nlme (≥ 3.1-0), methods, testthat (≥ 3.0.0), knitr, rmarkdown, lmerTest, car, multcomp

License:

MIT + file LICENSE

Encoding:

UTF-8

Language:

en-US

RoxygenNote:

7.3.3

VignetteBuilder:

knitr

Config/testthat/edition:

NeedsCompilation:

Packaged:

2026-03-27 06:10:56 UTC; acer

Author:

Sadikul Islam

[aut, cre]

Maintainer:

Sadikul Islam <sadikul.islamiasri@gmail.com>

Repository:

CRAN

Date/Publication:

2026-03-31 15:00:23 UTC

agriReg: Linear and Nonlinear Regression for Agricultural Data

Description

agriReg provides a unified interface for fitting, comparing, and visualising regression models commonly used in agricultural research:

Linear & mixed-effects models via fit_linear() (wraps lm / lme4::lmer)
Nonlinear growth curves via fit_nonlinear() — logistic, Gompertz, asymptotic, linear-plateau, quadratic
Dose-response models via fit_dose_response() — 4/5-parameter log-logistic and Weibull (wraps drc)
Model comparison via compare_models() — AIC, BIC, RMSE side by side
Diagnostics via gof_stats(), residual_check(), and S3 plot() methods

Typical workflow

library(agriReg)

# 1. Clean field trial data
trial <- clean_agri_data(my_data, yield_col = "yield_tha")

# 2. Linear mixed model (treatment + block structure)
lm_fit <- fit_linear(trial,
                     formula = "yield_tha ~ nitrogen + phosphorus",
                     random  = "(1|block)")
summary(lm_fit)
plot(lm_fit, type = "residuals")

# 3. Nonlinear growth curve
nl_fit <- fit_nonlinear(trial, x_col = "days", y_col = "biomass",
                        model = "logistic")
plot(nl_fit)

# 4. Dose-response
dr_fit <- fit_dose_response(herb_data,
                            dose_col = "dose_g_ha",
                            resp_col = "weed_control_pct")
ed_estimates(dr_fit, respLev = c(50, 90))

# 5. Compare models
compare_models(linear = lm_fit, nonlinear = nl_fit)

Author(s)

Maintainer: Sadikul Islam sadikul.islamiasri@gmail.com (ORCID)

Clean and validate agricultural field-trial data

Description

Removes rows with missing yield values, coerces the yield column to numeric, optionally flags IQR-based outliers, and returns an agriData object with a concise summary.

Usage

clean_agri_data(
  df,
  yield_col = "yield",
  flag_outliers = TRUE,
  na_action = c("remove", "warn")
)

Arguments

df

A data.frame containing field trial records.

yield_col

Character. Name of the column that holds the primary response variable (e.g. "yield", "biomass"). Default "yield".

flag_outliers

Logical. When TRUE (default), a logical column .outlier is appended: TRUE marks observations that fall more than 1.5 x IQR below Q1 or above Q3.

na_action

One of "remove" (default) or "warn". "remove" drops rows with NA in yield_col; "warn" keeps them but issues a warning.

Value

A data.frame with additional class "agriData". Attributes n_removed (rows dropped) and n_flagged (outliers flagged) are attached.

Examples

df <- data.frame(
  yield = c(3.1, 4.2, NA, 8.9, 3.8, 100),
  block = c("A", "A", "B", "B", "C", "C")
)
clean_agri_data(df, yield_col = "yield")

Compare multiple agriReg models side by side

Description

Computes AIC, BIC, RMSE, MAE, and R² for a collection of agriLM, agriNLS, or agriDRC objects and returns them in a tidy data frame ranked by AIC.

Usage

compare_models(...)

Arguments

...

Named model objects. Names are used as row identifiers. If unnamed, models are labelled model1, model2, …

Value

A data.frame with columns model, engine, n_par, AIC, BIC, RMSE, MAE, R2, and delta_AIC (difference from best).

Examples


dat <- data.frame(
  nitrogen = rep(c(0, 40, 80, 120, 160), each = 8),
  yield    = c(2.1, 2.8, 3.6, 4.1, 4.3,
               2.0, 2.9, 3.5, 4.2, 4.4,
               2.2, 2.7, 3.7, 4.0, 4.5,
               2.1, 3.0, 3.4, 4.3, 4.2,
               2.3, 2.8, 3.6, 4.1, 4.3,
               2.0, 2.9, 3.5, 4.2, 4.4,
               2.2, 2.7, 3.7, 4.0, 4.5,
               2.1, 3.0, 3.4, 4.3, 4.2)
)
m1 <- fit_linear(dat, "yield ~ nitrogen")
m2 <- fit_nonlinear(dat, "nitrogen", "yield", "quadratic")
compare_models(linear = m1, quadratic = m2)

Compute effective dose (ED) estimates from a dose-response model

Description

Returns the dose required to achieve a specified percentage of the maximum response, with confidence intervals via the delta method.

Usage

ed_estimates(
  drc_fit,
  respLev = c(10, 50, 90),
  interval = "delta",
  level = 0.95
)

Arguments

drc_fit

An agriDRC object.

respLev

Numeric vector of response levels (default c(10, 50, 90)). E.g. 50 = ED50 (dose for 50% of maximum effect).

interval

Type of confidence interval: "delta" (default) or "none".

level

Confidence level (default 0.95).

Value

A matrix of ED estimates and confidence bounds (printed and returned invisibly).

Fit a log-logistic or Weibull dose-response model

Description

A wrapper around drc::drm() providing the four model families most commonly used in herbicide/pesticide and nutrient-response studies. Returns an agriDRC object with print, summary, and plot methods.

Usage

fit_dose_response(
  data,
  dose_col,
  resp_col,
  fct = c("LL.4", "LL.5", "W1.4", "W2.4"),
  curveid = NULL,
  ...
)

Arguments

data

A data frame containing dose and response columns.

dose_col

Character. Name of the dose/concentration column.

resp_col

Character. Name of the response column (e.g. percentage inhibition, weed biomass reduction, germination rate).

fct

Character. Model function:

"LL.4": 4-parameter log-logistic (default). Parameters: b, c, d, e.
"LL.5": 5-parameter log-logistic (asymmetry parameter f added).
"W1.4": 4-parameter Weibull type-1.
"W2.4": 4-parameter Weibull type-2.

curveid

Optional column name for grouping curves (e.g. "species" or "treatment"). Passed directly to drc::drm().

...

Additional arguments forwarded to drc::drm().

Value

An object of class "agriDRC" containing:

fit: The underlying drc object.
fct: The model function name.
dose_col, resp_col: Column names used.
data: The data used.
call: The matched call.

Examples


herbicide_trial <- load_example_data("herbicide_trial")
dr <- fit_dose_response(herbicide_trial,
                        dose_col = "dose_g_ha",
                        resp_col = "weed_control_pct")
summary(dr)
ed_estimates(dr, respLev = c(10, 50, 90))
plot(dr)

Fit a linear or mixed-effects model for agricultural trials

Description

A wrapper around stats::lm() and lme4::lmer() that returns an agriLM object with consistent print, summary, plot, coef, fitted, residuals, and predict S3 methods.

Usage

fit_linear(data, formula, random = NULL, weights = NULL, ...)

Arguments

data

A data frame or agriData object.

formula

A model formula as a character string or formula object. Example: "yield ~ treatment + block".

random

Optional character string specifying the random-effects term in lme4 syntax, e.g. "(1|block)" or "(1|block) + (1|location)". When supplied lme4::lmer() is used; otherwise stats::lm().

weights

Optional numeric vector of observation weights passed to lm() or lmer().

...

Additional arguments forwarded to lm() or lmer().

Value

An object of class "agriLM" (a named list) containing:

fit: The underlying lm or lmerMod object.
engine: Character: "lm" or "lmer".
formula: The fixed-effects formula.
random: The random-effects term (or NULL).
data: The data used for fitting.
call: The matched call.

Examples


wheat_trial <- load_example_data("wheat_trial")
# Ordinary least squares
m1 <- fit_linear(wheat_trial, "yield ~ nitrogen + phosphorus")

# Mixed model with block as random effect
m2 <- fit_linear(wheat_trial,
                 "yield ~ nitrogen + phosphorus",
                 random = "(1|block)")
summary(m2)
plot(m2, type = "residuals")

Fit a nonlinear growth or response curve to agricultural data

Description

Provides a unified interface for five model families commonly used in agronomic research. Automatic starting-value heuristics are applied when start is not supplied, making the function usable without prior knowledge of parameter ranges.

Usage

fit_nonlinear(
  data,
  x_col,
  y_col,
  model = c("logistic", "gompertz", "asymptotic", "linear_plateau", "quadratic"),
  start = NULL,
  control = nls.control(maxiter = 500, tol = 1e-06)
)

Arguments

data

A data frame (or agriData) with at least x_col and y_col.

x_col

Character. Name of the independent variable column (e.g. "days", "dose", "nitrogen_kg_ha").

y_col

Character. Name of the response column (e.g. "yield", "biomass", "plant_height").

model

One of:

"logistic": 3-parameter logistic: y = Asym / (1 + exp((xmid - x)/scal))
"gompertz": 3-parameter Gompertz: y = Asym \cdot \exp(-\exp(b2(b3-x)))
"asymptotic": Asymptotic / Mitscherlich: y = Asym(1 - \exp(-rate \cdot x))
"linear_plateau": Linear-plateau (broken-stick): y = a + b \cdot \min(x, cp)
"quadratic": Quadratic polynomial: y = a + b \cdot x + c \cdot x^2

start

Optional named list of starting parameter values. When NULL (default), heuristic starting values are derived from the data.

control

A list passed to stats::nls.control(). Increase maxiter if convergence fails.

Value

An object of class "agriNLS", a named list containing:

fit: The nls object.
model: Character: the model name.
x_col, y_col: Column names used.
data: The data used for fitting.
call: The matched call.

Examples


maize_growth <- load_example_data("maize_growth")
nl <- fit_nonlinear(maize_growth, x_col = "days", y_col = "biomass_g",
                    model = "logistic")
summary(nl)
plot(nl)

Fit a sequence of polynomial linear models and pick the best

Description

Convenience function for testing linear, quadratic, and cubic fits of a single continuous predictor and selecting the best by AIC.

Usage

fit_polynomial(data, x_col, y_col, max_degree = 3)

Arguments

data

A data frame.

x_col

Predictor column name.

y_col

Response column name.

max_degree

Maximum polynomial degree to test (default 3).

Value

An agriLM object for the best-fitting model. A comparison table is printed as a side-effect.

Examples


wheat_trial <- load_example_data("wheat_trial")
best <- fit_polynomial(wheat_trial, x_col = "nitrogen", y_col = "yield")
plot(best)

Compute goodness-of-fit statistics for a model

Description

Extracts R², adjusted-R², RMSE, MAE, AIC, and BIC from an agriLM, agriNLS, or agriDRC object (or any object with fitted(), residuals(), and AIC() methods).

Usage

gof_stats(model)

Arguments

model

An agriLM, agriNLS, agriDRC, or compatible model object.

Value

A named list with elements R2, adj_R2, RMSE, MAE, AIC, and BIC.

Simulated herbicide dose-response dataset

Description

Weed control percentage measured across seven herbicide doses for two weed species.

Format

A data frame with 84 rows and 4 variables:

species: Weed species (Amaranth, Ryegrass).
dose_g_ha: Herbicide dose (g active ingredient per hectare).
weed_control_pct: Weed control percentage (0–100).
rep: Replicate number (1–6).

Source

Simulated for package demonstration.

Load a bundled agriReg example dataset

Description

Reads one of the three example datasets shipped with the package from inst/extdata. Under normal installation the datasets are available directly via data(wheat_trial) etc; this function is a fallback that works even when .rda files have not been compiled.

Usage

load_example_data(name = c("wheat_trial", "maize_growth", "herbicide_trial"))

Arguments

name

One of "wheat_trial", "maize_growth", or "herbicide_trial".

Value

A data.frame.

Examples

wheat <- load_example_data("wheat_trial")
head(wheat)

Simulated maize growth time-series

Description

Daily above-ground biomass measurements from 10 maize plants tracked from emergence to physiological maturity under well-watered conditions.

Format

A data frame with 200 rows and 4 variables:

plant_id: Plant identifier (1–10).
days: Days after emergence.
biomass_g: Dry biomass (g/plant).
treatment: Water treatment (WW = well-watered, DS = drought-stressed).

Source

Simulated for package demonstration.

Comprehensive model summary with fit statistics

Description

Prints a formatted block with coefficient table, goodness-of-fit statistics, and (for agriLM) an ANOVA table.

Usage

model_summary(model, anova = TRUE)

Arguments

model

An agriLM or agriNLS object.

anova

Logical. Include ANOVA table for agriLM models (default TRUE).

Value

Invisibly returns the model object (called for its side effect of printing a formatted summary block containing coefficients, goodness-of-fit statistics, and, for agriLM models fitted with engine = "lm", an ANOVA table).

Find the optimal x value (e.g. economic optimum dose)

Description

For quadratic and linear-plateau models, computes the x value that maximises the predicted response.

Usage

optimum_dose(nls_fit)

Arguments

nls_fit

An agriNLS object fitted with model = "quadratic" or "linear_plateau".

Value

A named numeric vector with x_opt (optimum x) and y_max (predicted maximum y).

Flag outliers using multiple methods

Description

A standalone outlier-detection function supporting IQR, Z-score, and modified Z-score (Iglewicz-Hoaglin) methods.

Usage

outlier_flag(df, col, method = c("iqr", "zscore", "modz"), threshold = NULL)

Arguments

df

A data frame.

col

Column name to test.

method

One of "iqr" (default), "zscore", "modz".

threshold

Numeric threshold: for "iqr" the IQR multiplier (default 1.5); for "zscore" the SD multiplier (default 3); for "modz" the modified Z-score cutoff (default 3.5).

Value

df with a logical column ⁠<col>_outlier⁠.

Plot a fitted dose-response curve

Description

Plot a fitted dose-response curve

Usage

## S3 method for class 'agriDRC'
plot(x, log_dose = TRUE, n_points = 200, ...)

Arguments

x

An agriDRC object.

log_dose

Logical. Plot dose on a log10 scale (default TRUE).

n_points

Integer. Points on the smooth curve (default 200).

...

Ignored.

Value

A ggplot object (printed invisibly).

Plot diagnostics for an agriLM model

Description

Plot diagnostics for an agriLM model

Usage

## S3 method for class 'agriLM'
plot(x, type = c("fit", "residuals", "qq", "scale"), ...)

Arguments

x

An agriLM object.

type

One of "fit" (observed vs fitted), "residuals" (residuals vs fitted), "qq" (normal Q-Q), or "scale" (sqrt(|residuals|) vs fitted). Default "fit".

...

Ignored.

Value

A ggplot object (printed invisibly).

Plot a fitted nonlinear curve over observed data

Description

Plot a fitted nonlinear curve over observed data

Usage

## S3 method for class 'agriNLS'
plot(x, n_points = 300, show_residuals = FALSE, ...)

Arguments

x

An agriNLS object.

n_points

Integer. Number of points on the fitted curve (default 300).

show_residuals

Logical. When TRUE, adds vertical line segments from observed points to the curve. Default FALSE.

...

Ignored.

Value

A ggplot object (printed invisibly).

A convenience wrapper for ggplot2 visualisation of model fit

Description

Dispatches to plot.agriLM, plot.agriNLS, or plot.agriDRC depending on model class. Useful in pipeline contexts.

Usage

plot_fit(model, ...)

Arguments

model

An agriLM, agriNLS, or agriDRC object.

...

Arguments passed to the underlying plot method.

Value

A ggplot object, returned invisibly. Called primarily for the side effect of printing the plot. The exact structure of the returned object depends on the class of model: plot.agriLM, plot.agriNLS, or plot.agriDRC is dispatched accordingly.

Residual diagnostic plot panel

Description

Generates a 2×2 panel of residual diagnostics: residuals vs fitted, normal Q-Q, scale-location, and a histogram of residuals.

Usage

residual_check(model, ncol = 2)

Arguments

model

An agriLM or agriNLS object.

ncol

Integer. Number of plot columns (default 2).

Value

Invisibly returns the list of four ggplot objects.

Simulated wheat nitrogen-response trial

Description

A simulated dataset from a randomised complete block design (RCBD) testing four nitrogen rates across three blocks and two phosphorus levels.

Format

A data frame with 72 rows and 6 variables:

block: Block identifier (A, B, C).
nitrogen: Nitrogen application rate (kg/ha).
phosphorus: Phosphorus level (low, high).
variety: Crop variety (V1, V2).
yield: Grain yield (t/ha).
biomass: Above-ground dry biomass (t/ha).

Source

Simulated for package demonstration.

Normalise yield values within groups

Description

Z-score or min-max normalises a numeric column, optionally within groups (e.g. per trial site or year).

Usage

yield_normalize(
  df,
  yield_col = "yield",
  method = c("zscore", "minmax"),
  group_by = NULL
)

Arguments

df

A data frame.

yield_col

Character. Column to normalise.

method

"zscore" (default) or "minmax".

group_by

Character vector of grouping column names, or NULL for global normalisation.

Value

df with a new column ⁠<yield_col>_norm⁠.

Examples

df <- data.frame(yield = c(3, 4, 5, 6), site = c("A","A","B","B"))
yield_normalize(df, group_by = "site")