Urban air temperature is a crucial variable for many urban issues. However, the availability of urban air temperature is often limited due to the deficiency of meteorological stations, especially in urban areas with heterogeneous land cover. Many studies have developed different methods to estimate urban air temperature. However, meteorological variables and local climate zone (LCZ) have been less used in this topic. Our study developed a new method to estimate urban air temperature in canopy layer during clear sky days by integrating land surface temperature (LST) from MODIS, meteorological variables based on reanalysis data, and LCZ data in Szeged, Hungary. Random forest algorithms were used for developing the estimation model. We focused on four seasons and distinguished between daytime and nighttime situations. The cross-validation results showed that our method can effectively estimate urban air temperature, with average daytime and nighttime root mean square error (RMSE) of 0.5 ℃ (R² = 0.99) and 0.9 ℃ (R² = 0.95), respectively. The results based on a test dataset from 2018 to 2019 indicated that the optimal model selected by cross-validation had the best performance in summer, with time-synchronous RMSE of 2.1 ℃ (R² = 0.6, daytime) and 2.2 ℃ (R² = 0.86, nighttime) and seasonal mean RMSE of 1.5 ℃ (R² = 0.34, daytime) and 1.2 ℃ (R² = 0.74, nighttime). In addition, we found that LCZ was more important at night, while meteorological data contributed more to the model during the daytime, which revealed the temporal mechanisms of the effect of these two variables on air temperature estimation. Our study provides a novel and reliable method and tool to explore the urban thermal environment for urban researchers.

城市气温是许多城市问题的关键变量。然而，由于气象站的缺乏，城市气温的可用性往往受到限制，特别是在土地覆盖不均匀的城市地区。许多研究开发了不同的方法来估计城市气温。然而，气象变量和当地气候区（LCZ）在该主题中较少使用。我们的研究开发了一种新方法，通过整合 MODIS 的地表温度 (LST)、基于再分析数据的气象变量以及匈牙利塞格德的 LCZ 数据来估计晴天期间冠层的城市气温。随机森林算法用于开发估计模型。我们关注四个季节，并区分白天和夜间的情况。交叉验证结果表明，我们的方法可以有效地估计城市气温，平均白天和夜间均方根误差（RMSE）分别为0.5℃（R ²  = 0.99）和0.9℃（R ²  = 0.95）。基于2018-2019年测试数据集的结果表明，交叉验证选择的最优模型在夏季表现最好，时间同步RMSE分别为2.1℃（R ²  = 0.6，白天）和2.2℃（R ²  = 0.86，夜间），季节性平均 RMSE 为 1.5 ℃（R ²  = 0.34，白天）和 1.2 ℃（R ²  = 0.74，夜间）。此外，我们发现LCZ在夜间更为重要，而气象数据在白天对模型的贡献更大，这揭示了这两个变量对气温估计影响的时间机制。我们的研究为城市研究人员探索城市热环境提供了一种新颖可靠的方法和工具。