code_ALCZM_GM

Introduction

This repository is supplementary to the paper "Optimized LCZ Mapping with Automated Machine Learning Re- veals Thermal Disparities during a Heatwave Event."

The objectives of this project are:

• Building AutoML-driven LCZ classifier

• Assesses the impact of resultant LCZ mapping on urban climate simulations

• Examines the associated implications for heat impact analysis.

Scripts and Data

0_data_prepare

This section covers the LCZ, remote sensing, population, and urban surface data used in this project, along with how to process these datasets and spatially integrate them with the LCZ training areas (TAs).

Note: Please define the storage path and data source you require in these scripts.

Num	Scripts	Description
1	Download_Remote_Feature.py	Download features from Google Earth Engine via API.
2	Load_TAs_data_from_WUDAPT.py	Select Manchester, London, Barcelona TAs from WUDAPT database.
3	Feature_process.py	Calculate remote sensing indices using downloaded features.
4	TAs_split.py	Prepare 25 sets of TA bootstrap indices.
5	Feature_link_TA.py	Spatially integrate processed features with the TAs.

1_model_dev

Scripts in this folder employ AutoML to conduct 25 bootstrap experiments, recording accuracy metrics to evaluate the performance of the machine learning classifier. Subsequently, it utilizes all training data and features to train the final LCZ classifier, applying it to the target cities—in this project, London, Barcelona, and Greater Manchester.

Note: Please define the storage path and data source you require in these scripts.

Num	Scripts	Description
1	Manchester_AutoML_bootstrap.py	Take Greater Manchester as an example to evaluate the performance of the proposed LCZ classifier.
2	AutoML_LCZ_classfier.py	Use all training information to build the final LCZ classifier.

2_LCZ_mapping

This section of the script uses the final LCZ classifier to generate the LCZ map for the target city.

Num	Scripts	Description
1	LCZ_map_Generate.py	Load the final LCZ classifier and generate locally optimized LCZ maps with Gaussian filters
2	Results_map_vis4GM.ipynb	Visualize Greater Manchester LCZ maps (GLCZM, ALCZM, WLCZM) with WUDAPT LCZ standard

3_simulation_landsurfdata

This section of the script converts the generated LCZ maps into the surface dataset required for CLMU model simulations. The code for this section originates from @Yuan Sun.

Num	Scripts	Description
1	step0_pct_crop.ipynb	Calculate pct crop in land surface data.
2	step1_process_lcz_map.ipynb	Convert LCZ to land cover percentage.
3	step2_process_land_cover.ipynb	Calculate and assign land cover percentage.
4	step3_process_building_hight.ipynb	Process urban features in land surface data.
5	step4_generate_lcz_surface.ipynb	Generate final land surface data with LCZ maps.

4_resulst_ana

This section of the script evaluates the performance of machine learning classifiers and assesses simulation results under different LCZ map configurations using observational data.

Num	Scripts	Description
1	ml_table_metrics.ipynb	Calculated the OA indicators for three cities. These correspond to Figure C.8.
2	multi_sensor_results_validation.ipynb	Simulation results were validated under three LCZ configurations. These correspond to Tables 1, 2, and G5 in the paper.

5_Figure_plot

This section of scripts is leveraged to render the images appearing in the paper.

Num	Scripts	Description	Figure number in the paper
1	Fig2_a_TA_spatial_distribution.ipynb	Visualize detailed study area (Greater Manchester)	Figure 2(a)
2	Fig2_bcd_TA_plot.ipynb	Visualize TAs in the detailed study area	Figure 2(b)(c)(d)
3	Fig3_a_accuracy.ipynb	Visualize the performance of LCZ classifier	Figure 3(a)
4	Fig3_b_lczmap_distribution.ipynb	Visualize the classified results in Greater Manchester	Figure 3(b)
5	Fig4_Fig5_simulation.ipynb	Visualize the simulation results with different LCZ maps	Figure 4, Figure 5
6	Fig6_abcd_spatial_plot.ipynb	Visualize the spatial patterns of urban air temperature and human heat stress exposure disparities across three simulations	Figure 6(a)(b)(c)(d)
7	Fig6_ef_heat_index.ipynb	Visualize the spatial distribution of cumulative differences in human heat stress exposure between two simulations	Figure 6(e)(f)
8	FigC7_3cities_TA_distribution.ipynb	Visualize the TA distribution across three cities	Figure C.7
9	FigE9_a_Results_map_vis4GM.ipynb	Visualize three LCZ maps across the GM region	Figure E.9 (a)
10	FigE9_bc_GM_TA_distribution.ipynb	Visualize the distribution of building types and differences in sensor layout across three LCZ maps.	Figure E.9 (b)(c)
11	Fig.B_10_SHAP.ipynb	Visualize the feature importance of the developed LCZ classifier	Figure B.10
12	Fig.B_11_gaussian_filter_comparison.ipynb	Visualize the Gaussian filter comparison	Figure B.11
13	Fig.B_12_confusion_matrix.ipynb	Visualize the normalized confusion matrix	Figure B.12
14	Fig.B_13_t-test.ipynb	Visualize the result of t-test	Figure B.13

Acknowledgments

• This work was supported by the Natural Environment Research Council [grant number UKRI1294].
• We gratefully acknowledge NVIDIA Corporation for providing the GPUs used in this research through the NVIDIA Academic Grant Program.

Citation

Place holder