GEOSCAN Search Results: Fastlink

GEOSCAN Menu


TitlePredicting building ages from LiDAR data with random forests for building energy modeling
 
AuthorTooke, T R; Coops, N C; Webster, J
SourceEnergy and Buildings vol. 68, pt. A, 2013 p. 603-610, https://doi.org/10.1016/j.enbuild.2013.10.004
Image
Year2013
PublisherElsevier BV
Documentserial
Lang.English
Mediaon-line; digital
File formatPDF; HTML
ProvinceBritish Columbia
AreaVancouver
Lat/Long WENS-123.1167 -123.1000 49.2500 49.2333
SubjectsScience and Technology; machine learning; Classification
Illustrationsgraphs
Released2013 10 02
AbstractIn this study we examine the use of airborne light detection and ranging (LiDAR) data to augment the prediction of building age and energy performance for numerous buildings using a random forests machine learning approach. Analysis is conducted for a residential neighborhood in Vancouver, Canada. Four separate models are developed to represent the increasing sophistication of spatial data available to municipalities. Results indicate that the random forests model using only LiDAR-derived building attributes predicts age with an average error (RMSE) of 16.8 years and 33.5% of the variance explained. When all attributes are combined from the various datasets the predictive capacity of the model is increased by 20% and the average error reduced to 15.8 years. Furthermore, examination of variable importance suggests that while mean building height derived from the LiDAR data is the most important attribute predicting age, the three subsequently ranked variables relate to non-LiDAR based attributes. Discussion is given to observable trends between select predictor variables and building age, and to the implications for building energy modeling and simulation efforts.
Summary(Plain Language Summary, not published)
In this study we examine the use of LiDAR data to predict building age and energy performance of numerous buildings at the neighbourhood scale using a random forests machine learning approach. Analysis is conducted for a residential neighbourhood in Vancouver, Canada. Four separate models are developed to represent the increasing sophistication of spatial data available to municipalities. Building height was found to be the most important variable derived from LiDAR, supported by zoning and building footprint information derived from traditional municipal datasets. The relevance of predictor variables for building energy modelling and simulation efforts is discussed.
GEOSCAN ID298931

 
Date modified: