NEWS.md

In RFplus: Machine Learning for Merging Satellite and Ground Precipitation Data

Version 1.5-2 (CRAN)

Expected Release Date: 2025-04-01

New Features

• We have added three new categorical metrics: Heidke Skill Score (HSS); Hanssen-Kuipers Discriminant (HK); Equal Threat Score (ETS) or Gilbert Skill Score.

• Aggregate metrics Mean Absolute Error (MAE) within metrics of goodness of fit.

• Since rainfall has a non-parametric behavior, we have modified the correlation coefficient from Pearson to Spearman. Now, 'cc' represents the Spearman correlation coefficient.

• We update the “exdata” data to adapt it to the new RFplus changes.

• We have optimized the core of the RFplus algorithm to reduce the computation time.

• When the parameter 'save_model' is set to TRUE the message 'Saving model. Please wait'. To improve readability we have modified this message to: 'Model saved successfully' to indicate that the model save was successful.

Bug Fixed

• Fixed a bug in the altitude difference calculation, which instead of calculating altitude differences returned a fixed value.
• An error has been corrected in the learning module where, instead of extracting the altitude of the station, the altitude of the centroid of the pixel in which the insitu station was located was extracted.
• Fixed a bug in the calculation of the altitude difference between the in situ stations and the grid, which was incorrectly performed using terra::extract(). This method assumed that the altitude of each insitu station was the altitude of each grid. Now, the difference is correctly calculated between the altitude of each station and all grid cells.
• Fixed a bug in the examples and in the documentation where the “Rain_threshold” parameter was passed as a single value and not as a list, as required by RFplus.

Version 1.4-0 (CRAN)

Expected Release Date: 2025-03-15

New Features

• The “Description” file has been updated to include all authors who have contributed to the RFplus package.
• We have made changes to the documentation to ensure better understanding.
• The evaluation_metrics function has been updated to allow classification of precipitation data into various intensity categories (e.g., light, moderate, heavy rain). It now accepts a list of custom thresholds to define these categories and calculates specific performance metrics for each, such as Critical Success Index (CSI), Probability of Detection (POD) and False Alarm Rate (FAR). This facilitates a more detailed evaluation of model performance at different rainfall intensities. In addition, the function retains traditional fitting metrics, such as RMSE and KGE, providing a complete evaluation tailored to scenarios with rainfall variability.
• Implemented a validation check to identify dates with completely missing data in BD_insitu. This feature allows users to detect and visualize dates where all recorded values are NA, preventing the model from processing them. If such dates are found, the system will trigger a warning, ensuring data completeness before running the Random Forest predictions.
• Two additional categorical metrics have been added when “training” has a value other than 1. The added metrics are: success ratio (SR), Hit BIAS (HB).
• An update of the vignettes was made to address the improvements introduced in the previous versions.

Version 1.3-0 (CRAN)

New Features

• Removed dependency on 'dplyr' and migrated all code to 'data.table' to ensure efficiency and speed for large data sets.

• Added functionality to apply point-to-pixel validation. The metrics analyzed are: Pearson correlation coefficient (CC), root mean square error (RMSE), modified Kling-Gupta efficiency (KGE), relative bias (PBIAS), probability of detection (POD), false alarm rate (FAR), critical success index (CSI).

• Removed dependencies on external libraries for FAR, POD, CSI calculations. Calculations are now performed using R base functions.

• A complete refactoring of the code has been carried out to improve its efficiency and ease of maintenance.

Bug Fixed

• Fixed a bug in 'wet.day' when set to False, rounding was still performed.
• Fixed 'pboptions' slash bug that caused a new slash to be created at each iteration when setting char = “=”.

Version 1.2-2 (release-CRAN)

Bug Fixed

• We modified the description of the package to meet the corrections suggested by CRAN.

• Replaced by \dontrun with \donttest. due to the time of execution of the example. (> 5 seconds)

Version 1.2-1 (release-CRAN)

Bug Fixed

• The word quantile mapping was changed to Quantile Mapping due to CRAN's comment of “ Words possibly misspelled in DESCRIPTION”.

• A test optimization was performed to address the problem of “the error Executing ‘testthat.R’ [421s/114s] the execution of the R code in ‘testthat.R’ had a CPU time 3.7 times higher than the elapsed time” reported by CRAN.

Version 1.2-0 (release)

New Features

• The versioning used has been modified. The semantic versioning has been migrated to the versioning used by CRAN.
• Changed the format for compatibility with the S3 method.
• The input data verification methods were refactored.
• Updated the data for the examples and internal tests

Version 1.1.3

New Features

• Added support for adjusting the simulated distribution by quantile mapping and nonparametric quantile mapping.
• We have changed the ranger library to Random Forest to make it compatible with terra predict.
• Improvements have been made to the code to improve interpretation.

Bug Fixed

• Fixed an ID mess error when transforming the .csv with coordinates to a vector.

RFplus documentation built on April 3, 2025, 5:44 p.m.