GPM-PAF: Reproducible Plotting, Sensitivity Analysis and Post‑Processing Code for the Tropical Cyclone Wind Speed Retrieval Paper (v1.0)

This repository provides the complete Python implementation for generating all main and supplementary figures, performing sensitivity analyses, and compiling statistical summaries for the paper on the GPM‑PAF (Passive‑Active Fusion) tropical cyclone (TC) wind speed retrieval algorithm. Important: This code package only includes the plotting, post‑processing, and sensitivity analysis modules. It does not contain the original data extraction (CSV parsing) or the model training (coefficient fitting) steps. Those pre‑processing stages have been completed prior to this work, and their output intermediate CSV files are required as inputs to run this code. With these provided data files, the code fully reproduces all visualisations and diagnostic tests presented in the paper. Key functionalities included: Generation of all main figures (Figures 1–9) and supplementary figures (Figures S1–S5): Global distribution of 165 TC overpasses. Case study for Typhoon BOLAVEN (wind speed map). Training set characteristics: NCEP background wind vs. reference, and fusion term vs. wind correction (with normal/extreme segmentation). Independent hurricane validation: retrieved vs. IBTrACS wind speeds, and RMSE by wind‑speed bins. Time evolution of the symmetry index (S) before rapid intensification (RI). Composite wind maps for four intense cyclones (BOLAVEN, YUTU, IRMA, YASA). Prediction error versus reference wind speed and versus rain rate. Sensitivity of retrieval performance to the scale factor A. Automatic sensitivity computations: Tests the impact of the scale factor A (from 1/14 to 1/6) on the validation RMSE and the regression coefficients B_normal and B_extreme (output: sensitivity_A.csv). Evaluates the sensitivity to segmentation thresholds: varying Tb thresholds (220–240 K) and U10 thresholds (25–45 m/s) (outputs: sensitivity_Tb.csv and sensitivity_U10.csv). Automated statistical summary generation: Produces a comprehensive explain.txt file that lists, for each figure, the key statistics: sample sizes, RMSE, correlation coefficient (R), bias, Spearman’s ρ, Mann‑Whitney U test p‑values, and bin‑wise RMSE values. This document is directly useful for manuscript writing. Input data required (to be provided by the user):Place the following pre‑processed CSV files in the designated directory (DATA_DIR): all_cyclones_features.csv – features of all 165 samples. training_dataset_typhoon_cyclone.csv – training set (73 typhoons/cyclones). validation_dataset_with_predictions.csv – independent hurricane validation set with predictions. model_parameters.csv – retrieved coefficients and overall performance metrics. (Optional) RI_composite.csv – if missing, the RI‑related figure and statistics will be skipped. Outputs generated: All figures in high‑resolution PNG format (600 dpi). Sensitivity tables in CSV format. A detailed explain.txt file with all statistical values.