How Accurate and Usable is GDEM? A Statistical Assessment of GDEM using LiDAR Data

Geomorphometry 2011Geomorphometry 2011, 07-09 September 2011, Esri, Redlands, California

Tomislav Hengl and Hannes I. Reuter

“A methodological framework for assessing accuracy of the GDEM product is described using four small case studies in areas of variable relief and forect canopy (Booschord in the Netherlands, Calabria in Italy, Fishcamp in USA and Zlatibor in Serbia). Focus is put on evaluating the true accuracy of ASTER GDEM using LiDAR data aggregated to 30 m resolution. Three aspects of accuracy have been evaluated: (a) absolute accuracy of elevations (goodness of fit between true and GDEM elevations), (2) accuracy of stream networks (goodness of fit for buffer distance maps for stream networks), and (3) accuracy of surface roughness parameters (goodness of representation of nugget variation and residual errors). Results show that GDEM seems to be of little use in areas of low relief ( <20 m), as in such areas the difference between the topographic features will be statistically significant. Nugget variation in all cases is 3-8 times lower than in the LiDAR DEMs, which indicates that surface roughness is under-estimated.

Comparison between stream network (case study Calabria) derived using LDEM (left) and GDEM (right) using exactly the same parameters.

Comparison between stream network (case study Calabria) derived using LDEM (left) and GDEM (right) using exactly the same parameters. The Rsquare between the buffer distance to streams is statistically significant (R2=.46), but from a practical point of view there are obvious differences. Notice also that surface roughness in LDEM is at the order of magnitude higher than in the GDEM. The size of the rectangle is 3.95 by 4.075 km.

“These results also suggest that an adjusted R-square of >.995 could be used as the threshold level for a satisfactory fit between LiDAR and GDEM (this R-square corresponds to RMSE of <10 m). For stream networks, an R-square of >.60 seems to be satisfactory. Analysis of the short-range variability allows determination of the effective grid cell size that more closely matches the true surface roughness. These results support previous work that indicates that a more suitable grid cell size for GDEM v1 is about 90 m.”

All presentation materials and reviewed papers from Geomorphometry 2011 are available at

Applying GIS Methods to Public Health Research at Harvard University

Journal of Map & Geography LibrariesJournal of Map & Geography Libraries, Volume 7, Issue 3, 2011

Jeffrey C. Blossom, Julia L. Finkelstein, Weihe Wendy Guan, and Bonnie Burns

“The Center for Geographic Analysis (CGA) at Harvard University supports research and teaching that relies on geographic information. This includes supporting geographic analysis for public health research at Harvard. This article reviews geographic concepts that apply to public health, pertinent data available in geographic format, and GIS analytical techniques. The work-flow methodology the CGA has developed for conducting research with geographic data will be presented, highlighting successful practices to follow and pitfalls to avoid.

Map illustrations as published in The Lancet, 376 (9756), Julio Frenk, Lincoln Chen, Zulfiqar A. Bhutta, et al. Health professionals for a new century: Transforming education to strengthen health systems in an interdependent world. Pp. 1923–1957, ©2010, with permission from Elsevier.

“Applications of this work flow are illustrated through an in-depth discussion of specific case studies in public health research at the university.”