FarmDynR

Developed by Hugo Scherer and Marc Müller at Wageningen Economic Research.

The goal of FarmDynR is to give the user the ability to aggregate FADN data to create representative farms for any grouping available, generate descriptive statistics, and to run FarmDyn from R.

Additionally, it includes useful functions to analyze the results of FarmDyn simulations.

Requirements

This package requires that you have GAMS and FarmDyn installed. Additionally, the package imports from:

  • gdxrrw (remember to load GAMS with igdx() with the path to your version of GAMS)

  • readr

  • dplyr

  • tidyr

  • readxl

  • data.table

It also suggests:

  • stringr

Installation

You can install the development version of FarmDynR like so:

 install.packages('https://gitlab.iiasa.ac.at/mind-step/FarmDynR')

Workflow

The workflow for this package is as follows:

  1. Load the individual farm FADN data into R (with desired countries and years)

  2. Run fadn2fd() with the FADN data, farmbranch desired, the mapping and the option to save GDX files based on the mapping - Yields will be calculated and the FADN data will be prepared - Outliers are removed

    • The mapping is a list of vectors, where each vector contains the names of the columns in the FADN data that correspond to the desired grouping. For example, if you want to group by NUTS0 and organic status, or NUTS0 or NUTS2, the mapping would be list(c("NUTS0", "misc%OrganicCode"), "NUTS0", "NUTS2"). The first vector is the grouping you want to use, and the following vectors are the groupings you want to aggregate to. Currently, depending on the farm branch selected, either dairy or arable, different subset conditions apply. In the case of dairy, more than 5 cows and those farms that have the top and bottom 10 % of milk yields are removed, also farm typology (TF14) of 45 is selected. In the case of arable, the farms are selected whose land is covered with 80% FarmDyn crops, also the Farm typologies 15, 16, 20 are selected.

    • From this data, the variable names are selected and/or computed for use in FarmDyn. At the same time, yield calculations per region are happening. If there is a region that does not have any yield it is imputed with the higher level region, country or European average. Once this is all done, the data is mapped using the provided list of mappings, so that different files can be generated for p_farmdata. If save_gdx is set to TRUE, the resulting files are in gdx format with a parameter p_farmdata and names depending on the provided mappings.

    • Once the gdx files are made, they have to be placed in the “dat” folder in FarmDyn

  1. Write batch file with writeBatch() and run FarmDyn with runFarmDynfromBatch() using the created batch file - Optional: Create descriptive statistics for reporting with fd_desc() with the farmbranch.
    • This will copy the segment towards the end of runInc.gms containing the settings to run FarmDyn from batch files and replace the scenario description with the mapping, the farmData file with the format “farmData_{ mapping }.gdx”, as is created in the FarmDynR package using fadn2fd(), and will include all the farmIds from this gdx file
    • This only works for one mapping, however, a list of mappings will not work like it does for the aggregation. However, a single mapping made of different categories or characteristics can be accepted (for example c("NUTS0", "misc%OrganicCode")). The mappings have to match the ones created with fadn2fd()
    • Once the batch file is created, runFarmDynfromBatch() will run FarmDyn by giving the location of the batch file, the ini and the xml
    • The descriptive statistics are calculated using the weighted mean with the (summed up) SYS02 weights

Once the simulations have been run, the next step is to analyze the results. In FarmDynR there are many ways to do this. You can load parameters, scalars, and marginals from dump files, or access to results, automatically setting a column sims as the name of your simulation run. scen_analysis() will load the results of the different scenarios (p_res) found in res_{scenario name}_until_2010.gdx by merging them together. load_dumps() does the same, but with dump files with the pattern dump_{scenario name}_{farmId}.gdx. From dump files, many other objects can be accessed, like variables (vars_dump()), scalars (load_dump_scalar()), and marginals (load_dump_marg()), in case they are needed.

Taking advantage of the scen_diff() and abs_diff() functions, the relative and absolute differences compared to the reference scenario can be easily calculated. Technically, these latter functions can be used for any data, not just FarmDyn scenario results.

Example 

library(FarmDynR)

# Read in FADN data, for example using:
FADN <- setDT(
  ldply(
    FADNFiles, fread,
  )
)

# Create mapping
mapping <- list(c("NUTS0", "misc%OrganicCode"), "NUTS0", "NUTS2")

# Create FarmDyn data
fd_data <- fadn2fd(fadn, "Dairy", mapping, save_gdx = FALSE)

# Create descriptive statistics
fd_desc(fd_data, farmbranch = "dairy", csv = FALSE, dir = NULL)

# Write batch file
writeBatch("path/to/FarmDyn", mapping, farmIds = unique(fd_data[[mapping]]$farmIds))

# Run FarmDyn
runFarmDynfromBatch("path/to/batch/file",
  IniFile = "path/to/Ini.ini", XMLFile = "path/to/xml.xml",
)

# Analyze the scenarios
scenarios <- scen_analysis(res_fold=file.path("FarmDyn", "sample", "results"), scen_name = c("reference", "scenario1", "scenario2"))

# Calculate differences from reference
selected_cols <- c("colname", "othercolnames")
other_cols <- c("colname1", "colname2")

scenarios <- abs_diff(scenarios, vars_to_diff = selected_cols)
scenarios <- scen_diff(scenarios, vars_to_diff = other_cols)