Advanced

Modelling near ground wind speed in urban environments using high-resolution digital surface models and statistical methods

Johansson, Lars LU (2012) In Student thesis series INES NGEM01 20112
Dept of Physical Geography and Ecosystem Science
Abstract
Wind is a complex phenomenon and a critical factor in assessing climatic conditions and pedestrian comfort within cities. This master’s thesis attempts to quantify and model the relationship between near ground wind speed and urban geometry using two-dimensional raster data and variable selection methods based on Multiple Linear Regression.

A simple regression model can be implemented in a Geographic Information System (GIS) to assess spatial distribution of wind speed at the street scale in complex urban environments. Wind speed data two meters above ground is obtained from simulations by computer fluid mechanics modelling (CFD) and used as a response variable. Utilizing a shadow-casting raster algorithm, four measures of urban... (More)
Wind is a complex phenomenon and a critical factor in assessing climatic conditions and pedestrian comfort within cities. This master’s thesis attempts to quantify and model the relationship between near ground wind speed and urban geometry using two-dimensional raster data and variable selection methods based on Multiple Linear Regression.

A simple regression model can be implemented in a Geographic Information System (GIS) to assess spatial distribution of wind speed at the street scale in complex urban environments. Wind speed data two meters above ground is obtained from simulations by computer fluid mechanics modelling (CFD) and used as a response variable. Utilizing a shadow-casting raster algorithm, four measures of urban geometry are derived from high-resolution surface models (DSM): Sky View Factor (SVF), Fetch, Frontal Area Index (FAI) and Angular Frontal Area Index (αFAI). To compute Fetch, FAI and αFAI, the shadow-casting algorithm needs a search angle and search distance as input parameters. To evaluate the effect of these parameters, a number of settings were tested resulting in a total of 53 different predictors. A sequential variable selection algorithm followed by all-possible subset regression was used to select predictors and candidate models for further evaluation.

Models including SVF and Fetch are found to explain general spatial wind speed pattern characteristics, but the prediction errors are large, especially so in areas with high wind speeds. However, all selected models did explain 90 % of the wind speed variability (R² ≈ 0.90). The RMSE of predicted wind speed for city models ranges from between 0.01 (dense building geometry) to 0.22 (wide street aligned in the wind direction with air flow channelling). The differences between the selected candidate models are less than 0.02, and no model consistently performs “best” over all city models. Models that included FAI and αFAI did not improve on these results.

Predictors adding information on width and height ratio and alignment of street canyons with respect to wind direction are possible developments to improve model performance. To assess the applicability of any derived model, the results of the CFD model should be thoroughly evaluated against field measurements. (Less)
Abstract (Swedish)
Populärvetenskaplig sammanfattning
Den ökande urbaniseringen värden över och de hälsorisker som identifierats i samband med värmeböljor i städer, har medfört att det urbana klimatet blivit ett allt viktigare forskningsområde. Vindhastighet är en av de viktigaste parametrarna när det gäller att beräkna lokala klimatvariationer och bedöma eventuella hälsorisker. En enkel beräkningsmetod för vind kan därför vara ett användbart verktyg inom både forskning och till exempel inom stadsplanering.

Vind i den urbana miljön styrs främst av bebyggelsens geometri, det vill säga byggnadernas form och inbördes placering. I föreliggande studie har det förhållandet utnyttjas genom att olika mått på bebyggelsegeometri beräknats och deras statistiska... (More)
Populärvetenskaplig sammanfattning
Den ökande urbaniseringen värden över och de hälsorisker som identifierats i samband med värmeböljor i städer, har medfört att det urbana klimatet blivit ett allt viktigare forskningsområde. Vindhastighet är en av de viktigaste parametrarna när det gäller att beräkna lokala klimatvariationer och bedöma eventuella hälsorisker. En enkel beräkningsmetod för vind kan därför vara ett användbart verktyg inom både forskning och till exempel inom stadsplanering.

Vind i den urbana miljön styrs främst av bebyggelsens geometri, det vill säga byggnadernas form och inbördes placering. I föreliggande studie har det förhållandet utnyttjas genom att olika mått på bebyggelsegeometri beräknats och deras statistiska samband till vindhastighet analyserats med linjär multipelregression. En sådan metod resulterar i en enkel matematisk ekvation som kan utnyttjas i ett geografiskt informationssystem (GIS) för att beräkna den rumsliga variationen av vindhastighet.

Digitala höjdmodeller för fyra områden i centrala Göteborg har använts för att beräkna fyra olika mått på bebyggelsegeometrins rumsliga variation: Sky View Factor (SVF), Fetch, Frontal Area Index (FAI) och Angular Frontal Area Index (αFAI). Dessutom har ett antal varianter av dessa beräknats, vilket totalt gett 53 olika mått på bebyggelsegeometrin. Det kan antas att många av dessa inte är nödvändiga för att förklara variationen i vindhastighet, varför en viktig del av den statistiska analysen handlar om att identifiera de viktigaste variablerna, de med den största förklaringsgraden, och att bygga modeller med hjälp av dem. Responsvariabeln vindhastighet modellerades med ENVI-met, som är en tredimensionell datormodell som bygger på computer fluid dynamics (CFD).

Med två olika urvalsmetoder identifierades fyra modeller med två till fem varianter av SVF och Fetch. Förklaringsgraden för FAI och αFAI var inte tillräckligt hög för att dessa skulle inkluderas i någon av de valda modellerna. Resultaten visar att SVF och Fetch kan förklara en betydande del, upp till 90 %, av vindhastighetens rumsliga variation. Däremot varierar felen kraftigt mellan olika platser i förhållande till bebyggelsen. Framför allt lokala höga vindhastigheter till följd av luftströmmar runt hörn ger stora fel i alla fyra modeller. Den summerande slutsatsen är att de viktigaste variablerna, SVF och Fetch, har identifierats och att fler än två variabler inte höjer modellens förklaringsgrad. (Less)
Please use this url to cite or link to this publication:
author
Johansson, Lars LU
supervisor
organization
alternative title
GIS-modellering av marknära vind i städer
course
NGEM01 20112
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Geographic Information System, spatial variability, GIS, digital surface model, wind speed, urban geometry, geomatics, geography, urban climate, physical geography, geografi, naturgeografi, geomatik, stadsklimat, bebyggelsegeometri, vindhastighet, digital höjdmodell, rumslig variation, geografiska informationssystem
publication/series
Student thesis series INES
report number
234
language
English
additional info
External supervisor: Fredrik Lindberg, Department of Earth Sciences, University of Gothenburg
id
2534839
date added to LUP
2012-05-08 16:09:01
date last changed
2012-08-22 12:13:18
@misc{2534839,
  abstract     = {Wind is a complex phenomenon and a critical factor in assessing climatic conditions and pedestrian comfort within cities. This master’s thesis attempts to quantify and model the relationship between near ground wind speed and urban geometry using two-dimensional raster data and variable selection methods based on Multiple Linear Regression.

A simple regression model can be implemented in a Geographic Information System (GIS) to assess spatial distribution of wind speed at the street scale in complex urban environments. Wind speed data two meters above ground is obtained from simulations by computer fluid mechanics modelling (CFD) and used as a response variable. Utilizing a shadow-casting raster algorithm, four measures of urban geometry are derived from high-resolution surface models (DSM): Sky View Factor (SVF), Fetch, Frontal Area Index (FAI) and Angular Frontal Area Index (αFAI). To compute Fetch, FAI and αFAI, the shadow-casting algorithm needs a search angle and search distance as input parameters. To evaluate the effect of these parameters, a number of settings were tested resulting in a total of 53 different predictors. A sequential variable selection algorithm followed by all-possible subset regression was used to select predictors and candidate models for further evaluation. 

Models including SVF and Fetch are found to explain general spatial wind speed pattern characteristics, but the prediction errors are large, especially so in areas with high wind speeds. However, all selected models did explain 90 % of the wind speed variability (R² ≈ 0.90). The RMSE of predicted wind speed for city models ranges from between 0.01 (dense building geometry) to 0.22 (wide street aligned in the wind direction with air flow channelling). The differences between the selected candidate models are less than 0.02, and no model consistently performs “best” over all city models. Models that included FAI and αFAI did not improve on these results.

Predictors adding information on width and height ratio and alignment of street canyons with respect to wind direction are possible developments to improve model performance. To assess the applicability of any derived model, the results of the CFD model should be thoroughly evaluated against field measurements.},
  author       = {Johansson, Lars},
  keyword      = {Geographic Information System,spatial variability,GIS,digital surface model,wind speed,urban geometry,geomatics,geography,urban climate,physical geography,geografi,naturgeografi,geomatik,stadsklimat,bebyggelsegeometri,vindhastighet,digital höjdmodell,rumslig variation,geografiska informationssystem},
  language     = {eng},
  note         = {Student Paper},
  series       = {Student thesis series INES},
  title        = {Modelling near ground wind speed in urban environments using high-resolution digital surface models and statistical methods},
  year         = {2012},
}