The transition to geographical data in 3D is facing big challenges regarding the creation of common standards for a functioning global spatial data infrastructure. The need for such data in planning processes has recently increased. Creating 3D-buildings in LOD1 is rather uncomplicated but creating buildings with a higher level of detail, that provide a better illustration of reality, are expensive and time consuming. The resources and efforts to establish geographical data in 3D differ in Sweden's municipalities, where some rural areas often do not have the same opportunities as urban areas. Therefore, a cost-effective and time-efficient method of creating 3D models for, e.g. municipal planning and civil dialogue would be well received. In this thesis project, the method of creating 3D buildings in ArcGIS Pro using the ExtractRoofForm tool is evaluated. Different data sources used as inputs were analyzed by comparing the outcome of successfully created buildings. In ArcGIS Pro, the calculation of root mean square error was used as measure to distinguish successful and unsuccessful buildings. The result of the ArcGIS Pro process was compared to 3D buildings in LOD2 produced by stereo mapping as reference data. The comparison was done by volume calculation, where the result indicates the significance of different inputs to the extraction of 3D buildings, and how well the evaluation of the process works. It could be shown that input data quality largely affects the process. Data with the highest quality gave the best results in the extraction process. The result of the volume calculation shows that successful buildings in ArcGIS Pro also had a high accuracy. Volume calculation and calculation of root mean square error show no correlation. A reason for this could be that there may be some problems evaluating the process of extracted 3D buildings resulting in more time-consuming work with editing after the extraction process. Still, the process of creating 3D buildings shows great potential for a successful outcome. / Övergången till geografisk information i 3D innebär ett stort arbete med att skapa gemensamma standarder för en fungerande GIS-samverkan världen över. Efterfrågan på presentation av geografisk information i 3D som en del i planeringsprocesser eller att utöva medborgardialog har på senare tid ökat. Att skapa 3D-byggnader i LOD1 är relativt enkelt men att kartera byggnader med en högre detaljnivå som ger en mer realistisk bild av verkligheten kan vara både dyrt och tidskrävande. Resurserna varierar mellan olika områden runt om i landet och vissa glesbygdsområden inte har samma möjlighet att erbjuda utbud av t.ex. medborgardialog som i mer urbana områden. Därför skulle en kostnad- och tidseffektiv metod att skapa 3D-modeller för exempelvis kommunal planeringsprocess och medborgardialog vara välkommen. I detta examensarbete utvärderas metoden att skapa 3D-byggnader i ArcGIS Pro med verktyget ”ExtractRoofForm”. Olika datakällor som används som indata analyserades genom att jämföra utfallet av lyckade byggnader. I ArcGIS Pro användes beräkning av medelfel som utvärdering för att skilja lyckade och mindre lyckade byggnader åt. Resultatet av processen i ArcGIS Pro jämfördes med 3D-byggnader i LOD2 framställda av stereokartering som referensdata. Jämförelsen gjordes med volymberäkning där resultatet avser att ge en fingervisning om betydelsen av indatakvalitet vid skapandet av 3D-byggnader samt om hur väl utvärderingen efter processen fungerar. Kvaliteten på indata visade sig efter analys ha betydelse där data med högst kvalitet också gav bäst utfall av lyckade 3D-byggnader. Resultatet av volymberäkning visade att byggnader med lyckat utfall i extraheringsprocessen också uppvisade en hög noggrannhet. Resultaten från volym-beräkning och beräkning av medelfelet visar ingen korrelation vilket tyder på att det kan finnas vissa problem med att utvärdera processen av skapade 3D-byggnader. Detta medför mer tidskrävande efterarbete än nödvändigt. Dock visar processen att skapa 3D-byggnader stora möjligheter till ett gott resultat.

This thesis is dedicated to contributing to the field of semi-automatic extraction of 3D buildings in level of detail 2 (LOD2). The data used in the thesis is open data available in Sweden, such as building footprints from local municipalities and national light detection and ranging (LiDAR) data. The study site is a part of a central street of the city Gävle, Sweden. To perform the automatic extraction of 3D buildings the software ArcGIS PRO and its solution for 3D buildings was used. The method consists of experimenting with the segmentation of the building footprints and then comparing the resulting models RMSE as a whole and for two specific buildings of different size (one small and one large building). There are some improvements that can be made in the quality assessment of the 3D models as the method used in this thesis bases the RMSE of the 3D models the obtained digital surface model (DSM). One possible way to improve the quality assessment would have been to perform a field survey to gather reference points with lower uncertainties than the DSM e.g. employing a total station. The conclusions of the thesis identify key settings that affects the 3D modelling. The best model, lowest RMSE, achieved an RMSE of 2.36 meters with tuning the setting of spatial detailwhile the rest of the settings were kept at their default values. Furthermore, the thesis identifies that the regularization tolerance setting, when segmenting the building footprints, affects the 3D models of large buildings negatively, as seen in an increase of RMSE when it is set to 0.5 meters and lower. However, further studies are required on a larger scale to better validate the conclusions.