Quantifying Urban Morphology-Induced Uncertainty in Urban Meteorology and Heat Stress Simulations in Southern California

This repository provides Jupyter notebooks and supporting datasets for uncertainty quantification (UQ) analysis of WRF-SLUCM simulations, focusing on the impacts of Urban Morphology Parameters (UMPs). The workflow integrates Polynomial Chaos Expansion (PCE), Monte Carlo (MC) sampling, and Sobol sensitivity analysis to evaluate model performance, quantify uncertainty, and assess parameter sensitivities for meteorological variables and heat-stress indicators (HSIs). The repository supports three main objectives:(1) evaluation of WRF-SLUCM outputs against AQS observations,(2) analysis of spatial and temporal impacts of UMP perturbations, and(3) uncertainty and sensitivity analysis for meteorological variables and HSIs. Repository structure and contents: Model_EvaluationThis folder contains notebooks for evaluating WRF-SLUCM simulations against AQS observations for a summer case. Separate notebooks evaluate 2-m air temperature, relative humidity, and 10-m wind speed. Impact_of_UMPsThis folder quantifies the impacts of UMP perturbations on WRF-SLUCM outputs. The notebooks compare spatial bias patterns, station-based performance at AQS locations, and diurnal or time-series responses across different UMP configurations. Uncertainty_R2_MapThis folder focuses on spatial uncertainty mapping and PCE model performance. The main notebook computes spatial distributions of output uncertainty and PCE goodness-of-fit (R²) for meteorological variables and HSIs. A zipped dataset is provided as required input and must be unzipped before running the analysis. SD_UMP_and_Physical_schemeThis folder analyzes standard deviations induced by UMP uncertainty and by different physical parameterization schemes. The notebooks compute PCE-based standard deviations over urban and non-urban areas. Output CSV files include UMP-induced standard deviations and arithmetic standard deviations across 24 physical schemes. MC_Output_DistributionThis folder contains Monte Carlo–based uncertainty analyses. Urban-average, time-aggregated outputs from MC sampling are summarized in a CSV file and used to generate histogram and distribution plots of UMP-induced uncertainty. Meteo_Hourly_UncertaintyThis folder provides hourly (0–23) PCE modeling for meteorological variables. Hourly urban-average meteorological data are provided in a zipped file. The analysis notebook builds hourly PCE models and produces plots of both absolute values and uncertainty by hour. Meteo_Sobol_IndexThis folder contains notebooks for plotting first-order Sobol sensitivity indices for meteorological variables. The analysis uses PCE-derived statistics generated in earlier steps. HSI_Hourly_UncertaintyThis folder focuses on hourly (0–23) PCE modeling for heat-stress indicators. A zipped dataset provides hourly urban-average HSI inputs. The notebook builds hourly PCE models and visualizes both HSI values and uncertainty. HSI_Sobol_IndexThis folder contains Sobol sensitivity analysis for heat-stress indicators. PCE-derived statistics are used to generate first-order Sobol index plots for four HSIs. Sampling design: The root directory contains a CSV file defining 100 Urban Morphology Parameter samples generated using Latin Hypercube Sampling (LHS). These samples serve as the input parameter space for the PCE and MC analyses. Data requirements and notes: Several notebooks require zipped input datasets to be unzipped before execution. File paths and filenames may need to be updated within the notebooks depending on the local directory structure and preprocessing outputs. Suggested run order: A typical workflow begins by unzipping all required datasets. The PCE-based standard deviation analysis should be run first to generate UMP-related statistics used in later Sobol analyses. Model evaluation and UMP impact analyses can then be performed, followed by spatial uncertainty mapping, hourly uncertainty analyses, Sobol sensitivity plotting, and optional Monte Carlo distribution analysis. Computational environment: The notebooks require a standard scientific Python environment, including numpy, pandas, matplotlib, scipy, xarray, netCDF4, and scikit-learn, along with any project-specific utilities used in the workflow.