超大城市土地覆盖与热环境的随机森林回归模型研究

目前对于超大城市土地覆盖和热环境定量模型研究报道不足，这主要是因为大城市地表温度和地表生物物理组分之间存在复杂的潜在非线性关系，这使得准确评估城市热环境情况遇到了严峻的技术挑战。研究选取中外6个典型超大城市（北京、上海、广州、伦敦、纽约和东京）为研究对象，以Landsat遥感影像为主要数据源，利用单通道算法反演各城市地表温度，采用随机森林回归模型（RFR）建立土地覆盖类型与城市热环境定量关系模型（LCT），综合分析城市土地覆盖因子与热环境间的多维定量关系。土地覆盖与地表温度的定量关系显示，城市地表热场的空间结构在很大程度上被下垫面用地类型所左右，不透水面会导致高温热场的聚集，而植被和水体则有降温作用。6个超大城市地表覆盖结构变化产生的升温/降温效应有所差异，北京、上海、纽约和东京等城市区域的植被和水体降温效应较广州和伦敦显著。基于随机森林回归方法建立了NDVI、MNDWI和NDISI等3种土地覆盖类型与城市热环境的综合定量关系模型（LCT），模型得到的精度高于基于多元线性回归方法建立的模型。LCT_RF模型的R²值在0.623~0.826之间，比LCT_MLR模型高0.021~0.074；RMSE比LCT_MLR模型低0.07℃~0.35℃。研究超大城市土地覆盖与城市热环境的互动作用机理，能为未来生态城市建设提供宝贵建议。

Study on the Random Forest Regression Model of Land Cover and Thermal Environment in Megacities

Rapid urbanization has led to rapid change in land cover and the landsurface heat balance in megacities. Due to the complex potential nonlinear relationship between land surface temperature and surface biophysical components in megacities， the quantitative models and the response mechanism between land cover and thermal environment in megacities is not yet clear. Takesix Chinese and foreign megacities （Beijing， Shanghai， Guangzhou， London， New York and Tokyo） as the study area， Landsat images were used to comprehensively analyze the quantitative relationship between urban land cover factors and thermal environment. The single-channel algorithm was used to retrieve the land surface temperature of thesix megacities.The random forest regression model was used to establish the quantitative relationship （LCT） model between land cover types and urban thermal environment （LST）. The quantitative relationship between land cover type and LST showed that the LST was closely related to urban land surface types. The spatial pattern of the urban thermal field depends to a great extent on the spatial distribution pattern of the urban land surface types. The impervious surface will lead to the accumulation of high LST fields， while vegetation and water had a significant cooling effect. The land cover compositionin six megacities had different heating/cooling effects. In urban areas， such as Beijing， Shanghai， New York， and Tokyo， the cooling effects of vegetation and water were more pronounced than thosein Guangzhou and London. The established LCTmodel between the three land cover types， NDVI， MNDWI， and NDISI， and the urban thermal environment showed that the LCT model had higher precisionthan that was based on the multiple linear regression method. The R² value of the LCT_RF model is 0.021~0.074， which is higher than that of the LCT_MLR model. The RMSE is 0.07℃~0.35℃， which is lower than that of the LCT_MLR model.It will be helpful for future construction of eco-cities by studying the interaction mechanism between the land cover and the urban thermal environment in megacities.