Climate-state-dependent design rainfall and infrastructure reliability in Bangladesh: recurrence compression of post-monsoon extremes under Indian Ocean Dipole and ENSO variability

ZENODO REPOSITORY DESCRIPTION Title: Climate-state-dependent design rainfall and infrastructure reliability in Bangladesh: Data, code, and results Authors: Sadman Arshad (Bangladesh University of Engineering and Technology) --- DESCRIPTION: This repository contains the complete analysis workflow, datasets, and results supporting the manuscript "Climate-state-dependent design rainfall and infrastructure reliability in Bangladesh: recurrence compression of post-monsoon extremes under Indian Ocean Dipole and ENSO variability" submitted to the Journal of Hydrology. The research quantifies how observable ocean-atmosphere variability modulates the reliability of stationary extreme-rainfall assumptions underlying hydraulic infrastructure design in Bangladesh. Using 64 years (1950–2013) of post-monsoon (October–December) rainfall records and climate indices, the study demonstrates that the Indian Ocean Dipole (IOD) and El Niño–Southern Oscillation (ENSO) significantly alter the effective return periods of design rainfall events through a process termed "recurrence compression." KEY FINDINGS: - The stationary 100-year design rainfall event in Central Bangladesh (452 mm) has an effective return period of approximately 37 years under strongly negative IOD states and 190 years under strongly positive states—a five-fold modulation.- ENSO is a persistent covariate across the full record; the IOD effect strengthens in the post-1982 period, with both showing comparable magnitude in recent decades.- Lifetime failure probabilities of stationary designs are state-dependent, ranging from 7–23% under positive-phase persistence to 63–77% under negative-phase persistence, confirming that climate-state temporal clustering creates design risk independent of average-case climate weighting.- The analysis supports a four-component design and risk-management framework comprising a climate surcharge factor (~1.17), seasonal operational triggers, lifetime reliability assessment, and asset-portfolio prioritization. --- REPOSITORY CONTENTS: 1. ANALYSIS CODE: - Master script implementing all nonstationary GEV models (stationary, DMI-conditioned, ENSO-conditioned, dual-covariate) - Leave-one-out sensitivity audit functions - Bootstrap confidence interval construction (2,000 resamples) - Rolling-window temporal stability analysis (eight 30-year windows) - Fixed-panel confirmation on six continuous stations - Goodness-of-fit tests (Kolmogorov–Smirnov, Anderson–Darling) - Engineering metrics computation: return levels, recurrence compression, lifetime failure probabilities - Cross-model sensitivity tables and visualization 2. PROCESSED DATASETS: - Central Bangladesh regional OND annual maximum rainfall (1950–2013, n=64; 1982–2013, n=32) - Northeastern, Coastal, Northwestern regional maxima (1950–2013, n=64) - Standardized climate indices: Dipole Mode Index (DMI) from HadISST; Niño 3.4 from ERSSTv5 - All indices aligned to October–December season and standardized within each analysis period - Fitted model parameters (location, scale, shape) and covariate coefficients for all six model formulations - Residual diagnostics and probability-integral transforms 3. RESULTS TABLES: - Table 1: Model comparison (stationary, DMI, ENSO, dual), 1950–2013 - Table 2: Leave-one-out sensitivity (1988, 1998, 2004, 2007 exclusions) - Table 3: Regional single-covariate results by region - Table 4: Climate-state-dependent return levels (50-year, 100-year) - Table 5: Equivalent return period of the stationary design event across IOD spectrum - Table 6: Lifetime failure probabilities by state assumption and model - Table 7: Cross-model sensitivity of engineering metrics 4. FIGURES: - Figure 1: Station locations and regionalization (Southwestern region excluded) - Figure 2: AIC improvement and regional covariate coefficients - Figure 3: Design rainfall as function of IOD state - Figure 4: Recurrence compression curve (equivalent return period vs. DMI) - Figure 5: Lifetime reliability under climate-state variability - Figure 7: Data diagnostics for DMI-conditioned GEV (scatter, QQ-plot, PP-plot) - Figure 8: Rolling-window temporal stability (coefficient paths, recurrence compression trajectory) - Figure 9: Fixed-panel confirmation (six continuous stations, 1948–2013) 5. SUPPLEMENTARY MATERIALS: - Interaction test results (formal regime-dependence testing) - Covariate correlation matrix (DMI, Niño 3.4, 1950–2013 and 1982–2013 windows) - Monthly rainfall data (quality-control flags, station-level completion rates) - Standardization summary (index means, standard deviations by period) --- TECHNICAL SPECIFICATIONS: Software: Python 3.x, R 4.xDependencies: scipy.optimize, numpy, matplotlib, pandas, bootstrap resampling libraryData format: CSV for rainfall and climate indices; NetCDF for intermediate fitsCode documentation: Inline comments, function docstrings, parameter definitionsReproducibility: All random seeds fixed; multi-start optimization grids specified explicitly --- METHODOLOGY SUMMARY: The analysis employs nonstationary generalized extreme value (GEV) distributions with location-parameter conditioning on the OND DMI and Niño 3.4 indices. Maximum-likelihood estimation used Nelder–Mead optimization initialized from a 625-point grid (5 starting values per coefficient, 4 coefficients in the dual model) to guard against local optima. Model comparison used likelihood-ratio tests (nested) and Akaike information criterion (non-nested). Distributional adequacy verified with Kolmogorov–Smirnov and Anderson–Darling tests; no model rejected at 5%. Robustness established through:- Leave-one-out sensitivity: removal of 1988, 1998, 2004, 2007 (highest-leverage years)- Temporal regime analysis: split at 1982; formal interaction tests; nonparametric bootstrap of regime coefficients- Rolling-window audit: eight overlapping 30-year windows (1950–2013) with equivalent recurrence period recalculated in each- Fixed-panel replication: six stations with continuous 1948–2013 coverage- Cross-model comparison: all metrics reported under DMI-only, ENSO-only, and dual-covariate formulations Engineering metrics derived from audited fits:1. Climate-state return levels: T-year quantile conditional on DMI (or Niño 3.4) state2. Equivalent return period: T_e(s) = 1 / [1 − F_s(x₁₀₀)], where x₁₀₀ is the stationary 100-year depth3. Lifetime failure probability: 1 − ∏_s [F_s(x₁₀₀)]^(L_s) for L-year service life under state-occupation weighting --- DATA SOURCES & AVAILABILITY: Original rainfall data: Bangladesh Meteorological Department (BMD)- 34 stations, daily records, 1948–2014- Quality control: >95% monthly completeness- Available by institutional request to BMD (http://www.bmd.gov.bd)- Not publicly redistributable per BMD institutional arrangement Climate indices (PUBLIC, included in this repository):- Dipole Mode Index (DMI): HadISST monthly sea surface temperature, computed after Saji et al. (1999) Source: UK Met Office Hadley Centre (https://www.metoffice.gov.uk/hadobs/hadisst/)- Niño 3.4 index: NOAA ERSSTv5 monthly sea surface temperature anomaly Source: NOAA Physical Sciences Laboratory (https://origin.cpc.ncei.noaa.gov/products/analysis-monitoring/ensoyears.shtml) Processed rainfall dataset in this repository: Regional OND annual maximum monthly rainfall for five regions (Central, Northeastern, Coastal, Northwestern, Southwestern) and their indexed variants. --- CITATION: If you use this code or data, please cite:Arshad, S. (2026). Climate-state-dependent design rainfall and infrastructure reliability in Bangladesh: Data, code, and results. Zenodo. And the associated manuscript:Arshad, S. (2026). Climate-state-dependent design rainfall and infrastructure reliability in Bangladesh: recurrence compression of post-monsoon extremes under Indian Ocean Dipole and ENSO variability. Journal of Hydrology. --- KEYWORDS: nonstationary extreme-value analysis; generalized extreme value distribution; Indian Ocean Dipole; ENSO; return period; infrastructure reliability; design rainfall; Bangladesh; climate variability; hydrologic design --- FUNDING: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. --- LICENSE: [Select appropriate license, e.g., Creative Commons Attribution 4.0 International or MIT License] --- CONTACT & SUPPORT: For questions regarding code, data processing, or methodology:Sadman ArshadBangladesh University of Engineering and Technology (BUET)Department of Civil EngineeringEmail: [e-mail] For access to original BMD rainfall data, contact:Bangladesh Meteorological Departmenthttp://www.bmd.gov.bd --- ADDITIONAL NOTES: This repository is intended to support full reproducibility of the published analysis. The code can be executed in sequence to regenerate all tables and figures from the processed regional rainfall and climate index datasets. Users should note: 1. Sample sizes are modest (n=32 for the primary 1982–2013 engineering window; n=64 for the full record), so upper-tail interval estimates are unstable; point estimates and cross-model ranges are preferred.2. Monthly block maxima are used in preference to daily maxima to integrate multi-day rainfall episodes typical of post-monsoon extremes; sub-daily urban drainage design requires separate analysis.3. Regional station-mean indices characterize seasonal flood-loading environment, not point design storms; direct application to individual hydraulic structure sizing requires local recalibration.4. The framework applies to Central, Northeastern, and Coastal Bangladesh; the Northwestern region shows no significant climate linkage and was excluded from the engineering framework. The Southwestern region was excluded following a pre-analysis data-quality audit.5. Lifetime reliability metrics condition on rainfall alone; because the post-monsoon season exposes delta infrastructure to compound surge, river-stage, and rainfall loading, the exceedance probabilities reported are lower bounds on true multi-hazard risk. --- VERSION HISTORY: Version 1.0 (2026-06-17): Initial release supporting manuscript submission to Journal of Hydrology