• Created geographic and lag variables in ArcGIS and R and compiled hourly departure data and predictors for each station.

• Trained an OLS model on one week and predicted for the following week. Successfully avoided serial and spatial autocorrelation by adding spatial and time lags. Wrote replicable analyzing instruction in html via Rmarkdown and published video for practical uses in YouTube, both with maps and graphs made from ggplot and ggmap. 

• Collected and created geographic variables from MPO and ArcGIS. Built a hybrid stepwise multivariate linear regression based on cluster analysis that successfully avoided heteroscedasticity. Built binominal regression to predict ridership changes. Built alternative Principle Component Regression.

Built an application to help the user to find the closest available bike stations and the nearest station that needs rebalancing according to users’ real-time location and generate routes for both stations. The closest available station is located based on live station data and users’ visiting purpose. The rebalancing station is located based on two years’ historical data. The application also provides users the availability of current stations and historical daily changes for the whole system.

• Built a transportation 13-intersection network for the study area in VISSIM from scratch. Successfully avoided unrealistic lane changes via balancing traffic counts proportionately and calculating OD matrices for each origin in study area accordingly. Added signal timing and priority rules in the model so that cars will yield for pedestrians. Calibrated model.

• Created simulation videos of before after scenarios and compared travel time changes in different segments to analyze its impact on vehicle travel. Used VISUM to analyze and display travel volume and isochrones changes in neighborhood. 

• Assisted with the 3-year CM2 project to build a travel demand model for the Northeast region in TransCAD. Accomplished tasks include analyzing the national household survey, creating TAZ systems, and creating transit networks.

• Create a user interface in Google Earth Engine where users can inspect the severity and possibility of wildfire of any place across the world (by clicking) at any date (by entering) after 1984. Severity is measured by adjusted NBR and possibility is measured from temperature, slope and land cover type.

• Upon receiving the date from user, this program will choose the most reliable Landsat data, compute and map the NBR changes in recent two week globally. By clicking, two charts will pop up in the widget to show NBR changes over that year in comparison to previous 10-year benchmark, as well as exact score severity and possibility score.

• Wrote python script in ArcGIS that analyzes urban runoff in San Diego. The urban topography is updated by extracting building height and footprint from Lidar and converting to pyramid based on pixels’ distances to its building centroid. The flow accumulation is updated with filter layer, where drainage buffer has calculated CN. 

Built two Models in ArcGIS. The first one score the suitability of each raster pixel according to primary disaster risk (possibility and severity), secondary disaster risk (flooding, landfall, health, electric shock) and vulnerabilities (population) through 70 steps of raster analysis including zonal statistics, cost distance, flow direction, raster calculator and etc. The second model works solely with the suitability raster and selects out three best and distant assembly sites through conditioning and iteration. It created a set of candidate sites, scored them based on neighborhood security, distance to evacuation zones and zones’ areas, and select out the highest sites. Then remove the selected sites from the candidates and add distance to selected sites as an additional criterion and select the next best site and then iterate the process.

• Created an earthquake risk analysis tool in ArcToolbox via model builder. The tool includes 40 steps of conversions, joins, dissolve, buffer, calculators, interpolate, regression etc. that analyze the risk in three aspects: primary & secondary disasters’ possibility and vulnerability. 

• Built OLS regression to predict housing price in Boston