WRF/UCM Simulations of Urban Heat Island in Guangzhou with an Extracted Land-use Map from the Remote Sensing Data

Guang Chen, Lihua Zhao, Akashi MochidaTELKOMNIKA 2016 · DOI

TL;DR

This paper tackles the challenge of accurately simulating urban heat islands (UHIs) by developing and testing new, high-resolution land-use maps derived from remote sensing data within the WRF/UCM model. The findings demonstrate that these custom land-use maps significantly improve UHI simulation accuracy compared to default datasets, highlighting the critical role of detailed land-use information for urban climate modeling.

What Problem Does This Paper Address?

Urbanization is a major driver of climate change at local scales, leading to phenomena like the Urban Heat Island (UHI) effect, where cities are significantly warmer than surrounding rural areas. Accurately simulating these effects is crucial for urban planning, public health, and understanding future climate impacts. However, existing climate models often rely on outdated or coarse land-use datasets that fail to capture the rapid and complex changes occurring in rapidly urbanizing regions like Guangzhou, China. This mismatch between model inputs and real-world conditions leads to inaccurate simulations of UHI intensity and its diurnal variations. The paper aims to bridge this gap by creating and evaluating more realistic land-use representations for urban climate modeling.

Methods

The researchers employed a multi-pronged approach to create and test new land-use datasets for the WRF/UCM model:

  • Land-Use Extraction (RS_12): They used supervised classification with the maximum likelihood algorithm on Landsat ETM+ imagery from 2012 to create a detailed land-use map of Guangzhou. This map was designed to be more up-to-date than the default WRF land-use data.
  • Urban Land-Use Sub-classification (UCM_12): To further refine the urban representation, they developed a "human settlement index" (HSI). This index combined nighttime light data (DMSP-OLS) and normalized difference vegetation index (NDVI) to classify urban areas into three subcategories: Commercial/Industrial/Transportation (CIT), High-Intensity Residential (HIR), and Low-Intensity Residential (LIR). This allowed for a more nuanced representation of urban structure within the Urban Canopy Model (UCM).
  • Numerical Simulation: The Weather Research and Forecasting (WRF) model, coupled with a single-layer Urban Canopy Model (UCM), was used to simulate the thermal environment. Three experiments were conducted: one using the default MODIS land-use data, one using the extracted RS_12 land-use map, and one using the UCM_12 classified land-use map.
  • Evaluation: Simulation results for 2-m air temperature and 10-m wind velocity were compared against observations from automatic weather stations at both urban and suburban points. The Root-Mean-Square-Error (RMSE) was used as a quantitative metric for temperature comparison.

Datasets

  • Landsat-7 Enhanced Thematic Mapper Plus (ETM+) imagery (2012): Used for extracting the primary land-use map (RS_12).
  • DMSP-OLS nighttime-light data (2012): Used in conjunction with NDVI to derive the human settlement index (HSI).
  • Terra MODIS surface reflectance (MOD09A1) and NDVI (MOD13A2) images (2012): Used for generating NDVI composites and as the default land-use dataset for comparison.
  • National Centers for Environmental Prediction (NCEP) Global Forecast System Final Analyses: Used for initial and lateral boundary conditions for the WRF model.
  • Automatic weather station data (Guangzhou, August 2012): Used for validating the simulation results (2-m temperature and 10-m wind velocity).

The paper introduces two new land-use datasets (RS_12 and UCM_12) specifically tailored for the Guangzhou region in 2012, offering a more up-to-date and detailed representation compared to the commonly used, older MODIS or USGS datasets.

Key Insights and Ideas

  • The Power of Detail: The study strongly suggests that the level of detail and accuracy in land-use data is paramount for simulating urban climate phenomena like UHIs. Even a single urban land-use category derived from recent remote sensing (RS_12) shows improvement over older, generalized datasets.
  • Urban Structure Matters: Sub-classifying urban areas into different densities (UCM_12) provides even greater accuracy. This highlights that the physical characteristics of urban development (e.g., building density, road width) significantly influence local thermal dynamics.
  • Nighttime UHI Capture: The new land-use models (RS_12 and UCM_12) were better at capturing the peak UHI intensity at night, a critical period for human thermal comfort and health.
  • Limitations of Default Data: The default MODIS land-use data, while convenient, is shown to be less effective in capturing the nuances of urban climate, particularly in rapidly developing cities.

Key Results

  • Temperature Simulation: The UCM_12 experiment (with sub-classified urban land use) consistently produced the lowest Root-Mean-Square-Error (RMSE) for 2-m temperature simulations, indicating better accuracy compared to RS_12 and the default MODIS dataset. For urban points, UCM_12 achieved an RMSE of 1.82, significantly lower than MODIS (2.30) and RS_12 (2.00).
  • Diurnal UHI Intensity: Both RS_12 and UCM_12 showed improved reproduction of UHI intensity evolution, particularly during nighttime, compared to the MODIS dataset. UCM_12 demonstrated a slightly better performance overall.
  • Wind Velocity: While all WRF/UCM experiments produced higher wind velocities than observed, the RS_12 and UCM_12 runs showed similar performance to each other, suggesting that while land-use data impacts temperature simulation, its effect on wind velocity simulation might be less pronounced in this setup. However, the paper notes that the model still struggles to accurately reflect the reality of building influences on wind environments.

Limitations and Future Work

  • Negative UHI at Noon: The study acknowledges a failure of all experiments to accurately depict negative UHI intensity at noon, suggesting a potential limitation in the model's ability to capture certain atmospheric conditions or the specific land-use interactions at that time.
  • Underestimation of UHI Intensity: The paper notes that while the new models improve UHI simulation, they generally underestimate the intensity at times other than the peak, and the WRF model performs poorly before daybreak.
  • Wind Velocity Discrepancies: The simulations for wind velocity, even with improved land-use data, did not fully capture observed patterns, indicating a need for further model development or refinement in representing urban wind dynamics.
  • Future Work: The authors suggest future studies to assess how different urban development modes can affect the urban thermal environment, implying a desire to use these refined modeling techniques for urban planning and policy analysis.

Summary updated

Abstract

In order to investigate the performance of weather research and forecasting (WRF) model coupled with an urban canopy model (UCM) with an extracted land use data from Remote Sensing data (RS), three numerical experiments with different geographic models were carried out. Supervised classification with the maximum likelihood was applied to extract the land-use map in Guangzhou 2012 named the geographic model RS_12. Then based on the satellite-measured night time light data and the normalized difference vegetation index, a human settlement index was used to classify the urban land category to three urban land subcategories in the UCM as another geographic model named UCM_12. Both new geographic model simulation results are capable of reasonably modeling the observation result. The UCM_12 reasonably reproduced the best 2-m temperature evolution and the minimum root-mean- -square-error as compared with other experiments. Experiments with new geographic models can capture the strongest UHI time occured at night and gradually decreases in the morning but failure to get negative at noon. A better accuracy when comparing the new geographic models to the observation proved that extract land use data from RS can be used in urban scale simulation under the fast urbanizition in China.
BibTeX
@inproceedings{chen2016wrf,
  author = {Guang Chen and Lihua Zhao and Akashi Mochida},
  title = {{WRF/UCM Simulations of Urban Heat Island in Guangzhou with an Extracted Land-use Map from the Remote Sensing Data}},
  booktitle = {TELKOMNIKA},
  year = {2016},
  doi = {10.12928/telkomnika.v14i2a.4374},
}
Figure