Felix Nieberding 2021-11-25
This Repo contains a function and use case for gap-filling Eddy Covariance data using Random Forests algorithm (Breiman, 2001).
The function in _RF_impute_missing_fluxes.R is basically a convenient
wrapper using caret::train() to perform gap-filling of input variables
using randomForest::randomForest(). It is based on the code provided
by Yeonuk Kim based
on his recent publication in Global Change Biology (Kim et al., 2020).
The script gap_filling.R imports the data and sets the processing options (e.g. mtry, n_trees, pre processing steps etc.) Then _RF_impute_missing_fluxes.R estimates the train parameters for every (flux) variable to be filled on an annual basis. The script is performed in parallel to increase computation time and can easily be adapted to the number of cores available.
The other two scripts (check_output.R and get_mtry.R) can be used to inspect model accuracies and results for mtry determination.
The function produces 4 different outputs in 3 different folders.
./RF_models/mtry_[…].csv: Error metrics of mtry estimation, gets only printed if train_mtry = TRUE
./RF_models/mod_RF_[…].rds: The RF model itself.
./RF_results/pred_RF_[…].csv: The results of gap-filling, i.e. the original values, the gap-filling results, the residuals and a column idicating if the values have been used for training only.
./RF_plots/RF_[…].png: Plots showing the results of the gap-filling and accuracy metrics.
[…]: several suffixes getting concatenated indicating the variable which gets filled (_Flux), the year (_Year), the pre-processing method (_impute), the number of trees grown (_N_trees) and a freely choosable suffix (_suffix).
Breiman, L.: Random Forests, Machine Learning, 45, 5–32, doi:10.1023/A:1010933404324, 2001.
Kim, Y., Johnson, M. S., Knox, S. H., Black, T. A., Dalmagro, H. J., Kang, M., Kim, J., and Baldocchi, D.: Gap-filling approaches for eddy covariance methane fluxes: A comparison of three machine learning algorithms and a traditional method with principal component analysis, Global Change Biol, 26, 1499–1518, doi:10.1111/gcb.14845, 2020.