Loading...

Downscaling Tehran’s Temperature Field Using Machine Learning Algorithms and Geospatial Interpolation

Jahangir, Mohammad Sina | 2019

782 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 51666 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Moghim, Sanaz
  7. Abstract:
  8. Due to climate change and the increase in the emission of greenhouse gasses, the temperature of the large cities is increasing. Tehran, which is the capital of Iran and the most populated city of it, is no exception. One of the significant tools for characterizing heat in the cities is having access to the temperature field of the region. Different tools can be used for achieving the temperature field. Two methods for doing so are remote sensing and numerical models. Each one of the mentioned methods has their own strength and weaknesses. In this research, the WRF-ARW model (version 3.7) is used for deriving meteorological fields for the city of Tehran. One of the many merits of using numerical methods such as WRF is that they provide high-resolution temporal results which makes them preferable over remote sensing. WRF-ARW is a mesoscale model, and for evaluating phenomenon such as Urban Heat Island (UHI) the output temperature field should be downscaled to a fine resolution. In the present study, Artificial Neural Network (ANN), Bayesian Regression (BR) and Genetic Programming (GP) are applied combined with IDW interpolation for downscaling the WRF-ARW air temperature field (at two meter) to a finer spatial resolution (e.g., 50-m). The results indicate that all of the methods have increased the accuracy of prediction compared to the IDW method. Furthermore, it has been demonstrated that ANN has outperformed the other two methods in all of the defined error criteria (such as RMSE, MSE, and MAE). Also, the results confirm that all of the methods introduced in this study have increased the accuracy of results compared to WRF-ARW outputs in almost all months of the evaluation process (the year 2012). Hence, it can be concluded that the methods used in this study not only can be used for downscaling the temperature field but also they can increase the accuracy of the model output itself. Using BR, Tehran’s Temperature field was downscaled from a 2-km grid spacing to a 50-m grid spacing. Using the first-order reliability methods (MVFOSM and FORM), the formation probability of UHI was assessed. The results indicate that the formation probability is higher during the night and also they are more distributed during the cold months. Districts 12 and 16 were flagged as they were associated with the highest UHI formation probability in warm and cold months, respectively. Furthermore, it was shown that the UHI formation is profoundly affected by the wind field and the land-use of the region. Districts that have dense plant cover (district 22) also have a lower probability to be exposed to UHI. UHIs mostly are evaluated combined with heat waves and cold spells. In the present study, heat wave and the cold spell were evaluated in the city of Tehran during 1995-2016. By reviewing the previous studies, it was deduced that a serious gap exists in the evaluation criteria of these type of events and no standard indicator exists for their assessment. By defining new set of indicators and using their probability distribution instead of their deterministic value, the associated risk of these events was calculated for the city of Tehran. To do so, a reliability framework was used. Also, to compare the obtained results for Tehran with another city with a different climate, the city of Vancouver was also evaluated using the same method. The results show that cold spells are a more considerable threat in both cities and the city of Tehran is at a higher risk of exposure to heat waves in comparison with Vancouver. Although this study has tried to evaluate Tehran’s thermal characteristics as much as possible, it is for sure that more work can be done to complete this research in the future. It is hoped that the output of this study will be useful for both researchers and decision makers
  9. Keywords:
  10. Downscaling Temperature Field ; Weather Research and Forecasting (WRF)Modeling System ; Urban Heat Islands ; Heat Wave and Cold Spells ; System Reliability ; Machine Learning ; Tehran City

 Digital Object List

 Bookmark

No TOC