The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XL-1/W5
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-1/W5, 197–202, 2015
https://doi.org/10.5194/isprsarchives-XL-1-W5-197-2015
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-1/W5, 197–202, 2015
https://doi.org/10.5194/isprsarchives-XL-1-W5-197-2015

  11 Dec 2015

11 Dec 2015

PREDICTION OF WIND SPEEDS BASED ON DIGITAL ELEVATION MODELS USING BOOSTED REGRESSION TREES

P. Fischer1, C. Etienne2, J. Tian1, and T. Krauß1 P. Fischer et al.
  • 1Remote Sensing Technology Institute, German Aerospace Center (DLR), Münchener Str. 20, 82234 Wessling, Germany
  • 2Secquaero Advisors Ltd., 8001 Z¨urich, Switzerland

Keywords: Digital Elevation Model, Wind Speed, Non-Parametric Regression

Abstract. In this paper a new approach is presented to predict maximum wind speeds using Gradient Boosted Regression Trees (GBRT). GBRT are a non-parametric regression technique used in various applications, suitable to make predictions without having an in-depth a-priori knowledge about the functional dependancies between the predictors and the response variables. Our aim is to predict maximum wind speeds based on predictors, which are derived from a digital elevation model (DEM). The predictors describe the orography of the Area-of-Interest (AoI) by various means like first and second order derivatives of the DEM, but also higher sophisticated classifications describing exposure and shelterness of the terrain to wind flux. In order to take the different scales into account which probably influence the streams and turbulences of wind flow over complex terrain, the predictors are computed on different spatial resolutions ranging from 30 m up to 2000 m. The geographic area used for examination of the approach is Switzerland, a mountainious region in the heart of europe, dominated by the alps, but also covering large valleys. The full workflow is described in this paper, which consists of data preparation using image processing techniques, model training using a state-of-the-art machine learning algorithm, in-depth analysis of the trained model, validation of the model and application of the model to generate a wind speed map.