Abstract The growing number of densely populated cities has resulted in a decrease in vegetation coverage, which in turn, has resulted in a temperature increase in urban areas. This phenomenon is known as urban heat island (UHI). Several strategies have been proposed to mitigate the effect of urban heat islands in recent years, including usage of green roof coverage. However, there is a need for a rigorous analysis of the relationship between UHI and different urban characteristics using advanced models for urban planners to make policy-decisions to mitigate the UHI effect. In this study, the cooling effect of the green roof strategy in the city of Austin, considering 2D/3D urban characteristic parameters was investigated. To begin with, land surface temperature (LST) was estimated from Landsat 8 TIRS data in July 2016. Also, 3D urban morphology parameters derived from light detection and ranging (LiDAR) data. To simulate the green roof strategy, Sentinel 2A satellite images were used to calculate the normalized difference vegetation index and the normalized difference built-up index because of higher spatial resolution than Landsat 8 OLI. Then a multilayer feed-forward neural network was applied as a nonlinear model to find a relationship between LST and various urban characteristic parameters simultaneously. Furthermore, the importance of the variables in LST modeling was evaluated using sensitivity analysis. After that, some downtown residential and office buildings, which had the potential to become a green roof, were selected to implement the green roof strategy. Finally, by analyzing the relationship between LST reduction due to green roof simulation and urban indicators, the best buildings for green roof implementation were determined. Results showed that the accuracy of the LST modeling was reached to R2 = 0.786 and RMSE = 0.956 °C. In addition, by greening 3.2% of total building roofs, the average of LST decreased by 1.96 °C. Moreover, the results indicated that the building green roofs with (i) heights of 15–25 m, (ii) the highest values of sky view factor and solar radiation, and (iii) the lowest distance to the water body, had the greatest cooling effects on LST. Consequently, these findings indicated that the green roof has a significant effect on temperature reduction, especially by selecting the buildings with the above mentioned characteristics.

中文翻译：

---

使用人工神经网络模拟绿色屋顶及其对城市热岛的潜在缓解作用：德克萨斯州奥斯汀的案例研究

摘要 人口稠密的城市数量不断增加，导致植被覆盖率下降，进而导致城市地区温度升高。这种现象被称为城市热岛 (UHI)。近年来，已经提出了几种策略来减轻城市热岛的影响，包括使用屋顶绿化。但是，需要使用先进的模型对 UHI 与不同城市特征之间的关系进行严格分析，以便城市规划者做出减轻 UHI 影响的政策决策。在这项研究中，考虑到 2D/3D 城市特征参数，研究了奥斯汀市绿色屋顶策略的冷却效果。首先，地表温度 (LST) 是根据 2016 年 7 月的 Landsat 8 TIRS 数据估算的。此外，来自光检测和测距 (LiDAR) 数据的 3D 城市形态参数。为了模拟绿色屋顶策略，由于空间分辨率高于 Landsat 8 OLI，使用 Sentinel 2A 卫星图像计算归一化差异植被指数和归一化差异建成指数。然后应用多层前馈神经网络作为非线性模型，同时寻找 LST 与各种城市特征参数之间的关系。此外，使用敏感性分析评估了 LST 建模中变量的重要性。此后，一些有潜力成为绿色屋顶的市中心住宅和办公楼被选中实施绿色屋顶战略。最后，通过分析绿色屋顶模拟导致的 LST 减少与城市指标之间的关系，确定了实施绿色屋顶的最佳建筑。结果表明，LST 建模的精度达到了 R2 = 0.786 和 RMSE = 0.956 °C。此外，通过绿化 3.2% 的建筑屋顶，平均 LST 下降了 1.96 °C。此外，结果表明，(i) 高度为 15-25 m，(ii) 天空视野因子和太阳辐射值最高，以及 (iii) 到水体的距离最小的建筑绿色屋顶，具有最大的冷却对 LST 的影响。因此，这些发现表明屋顶绿化对降温有显着影响，尤其是通过选择具有上述特征的建筑物。确定了实施绿色屋顶的最佳建筑。结果表明，LST 建模的精度达到了 R2 = 0.786 和 RMSE = 0.956 °C。此外，通过绿化 3.2% 的建筑屋顶，平均 LST 下降了 1.96 °C。此外，结果表明，(i) 高度为 15-25 m，(ii) 天空视野因子和太阳辐射值最高，以及 (iii) 到水体的距离最小的建筑绿色屋顶，具有最大的冷却对 LST 的影响。因此，这些发现表明屋顶绿化对降温有显着影响，尤其是通过选择具有上述特征的建筑物。确定了实施绿色屋顶的最佳建筑。结果表明，LST 建模的精度达到了 R2 = 0.786 和 RMSE = 0.956 °C。此外，通过绿化 3.2% 的建筑屋顶，LST 的平均值下降了 1.96 °C。此外，结果表明，(i) 高度为 15-25 m，(ii) 天空视野因子和太阳辐射值最高，以及 (iii) 到水体的距离最小的建筑绿色屋顶，具有最大的冷却对 LST 的影响。因此，这些发现表明屋顶绿化对降温有显着影响，尤其是通过选择具有上述特征的建筑物。总建筑屋顶的 2%，LST 的平均值下降了 1.96 °C。此外，结果表明，(i) 高度为 15-25 m，(ii) 天空视野因子和太阳辐射值最高，以及 (iii) 到水体的距离最小的建筑绿色屋顶，具有最大的冷却对 LST 的影响。因此，这些发现表明屋顶绿化对降温有显着影响，尤其是通过选择具有上述特征的建筑物。总建筑屋顶的 2%，LST 的平均值下降了 1.96 °C。此外，结果表明，(i) 高度为 15-25 m，(ii) 天空视野因子和太阳辐射值最高，以及 (iii) 到水体的距离最小的建筑绿色屋顶，具有最大的冷却对 LST 的影响。因此，这些发现表明屋顶绿化对降温有显着影响，尤其是通过选择具有上述特征的建筑物。