GI for Sustainability

GI & Sustainability 7

6708.1 - Where in the Rural? Exploring Human Settlement Patterns in Rural Landscapes

Thursday, July 6
4:10 PM - 4:30 PM
Location: Maryland C

It is estimated that just under half (46% per UN, 2014) the world's population lives in rural settings, yet most global population datasets lack meaningful precision to discern the spatial distribution of these rural populations. They are the underrepresented inconspicuous half as a direct consequence of the difficulties to identify wide-spread residential areas in rural settings in remote sensing imagery. This misrepresentation becomes obvious in comparison to the saliency of urban settings in population distributions which are easy to detect due to development intensity and density of built-up land. The Global Human Settlement Layer (GHSL) is a global dataset at medium resolution (~38 m) which indicates the presence of built-up area within a pixel. Derived from imagery of the Landsat family, it has the potential to make rural populations more conspicuous. However, while GHSL is an improvement to other existing datasets, most residential land containing rural populations remains still faded into the background (Uhl et al, 2016).

Motivated by this chronic underestimation of rural residential land we explored the potential of statistical modeling techniques in identifying places in rural regions that are likely developed. We created logistic regression models to predict the presence of a structure using various independent variables derived from medium resolution data (30 m), including Landsat 8 reflectance values, digital elevation models (DEM), GHSL-derived built-up land, and Open Street Maps (OSM) road data. From Landsat 8 we used the panchromatic, infrared and short wave infrared bands, and derived different normalized difference ratios. From the DEM we derived slope, topographic wetness index (TWI) and mean minimum variance (MMV). From OSM we calculated Euclidean distance to a road, local straightness index, and local road length. As GHSL reliably represents urban and periurban areas, we take advantage of it as a covariate to determine distance to urban areas. In effect, we are modeling built-up land in rural areas and thus carry out an effort to extend and enhance the GHSL data product.

Recognizing that not all regions are equally covered by OSM road datasets, models are presented with and without using OSM roads as a covariate in order to examine the level of predictive power in a statistical model without relying on road proximity or road density. Another important factor in our methods is the training (and validation) dataset. The models are trained using a database of actual building footprints.
We present results of a model trained in Boulder County Colorado, USA and then apply the model in Los Angeles County California, USA and Berkshire County, Massachusetts, USA in order to test for sensitivity and regional variation of drivers and explanatory factors for modeling built-up land. For each of these counties building footprint data is publicly available and can be used to validate the performance of the model.
Initial results for Boulder County are promising with Tjur pseudo R-squared values (a measure of model discrimination; Tjur, 2009) of 0.54 and 0.45 and the area under the ROC curve values (AUC) of 0.93 and 0.91 for models including and not including roads as a covariate, respectively. We found distance to roads and slope to be important variables and strong predictors of a pixel being built-on. When properly transformed, Landsat variables also enhanced the predictive power of the models, particularly the panchromatic (note we resampled the panchromatic to 30 meters). While the model which included road variables has greater discrimination power, the model without road variables still performed well.
The analytical setup described above represents a promising general framework to determine places in rural settings that are likely built-up.

Alexander Stum

University of Colorado

I am a PhD student at the University of Colorado-Boulder. I have researched soil spatial predictive models, generalization of hydrographic features, and the use of NLCD and parcel data for dasymetric mapping approaches. My current research focuses on exploring rural human development patterns and how these data can be used.


Send Email for Alexander Stum

Stefan Leyk

Associate Professor
University of Colorado Boulder

PhD University of Zurich, Department of Geography (Dr. sc. nat.), 2005
M.S. Dresden University of Technology, Department of Forest Sciences, 2002
B.S. Dresden University of Technology, Department of Forest Sciences, 2001
03/2016-present FACULTY ASSOCIATE - Institute of Behavioral Science. University of Colorado Population Center (CUPC)
07/2014-05/2015 VISITING PROFESSOR - Department of Geography, University of Zurich, Switzerland
07/2014-present ASSOCIATE PROFESSOR - Department of Geography, University of Colorado Boulder, U.S.A.
08/2007-06/2014 ASSISTANT PROFESSOR - Department of Geography, University of Colorado Boulder, U.S.A.


Send Email for Stefan Leyk

Michael Govorov

Vancouver Island University

Michael Govorov, professor at Vancouver Island University and instructor/facilitator at University of Toronto, holds a PhD in Technical Science, with a specialization in cartography, from the Siberian State Academy of Geodesy, as well as a Post-doctoral Research Certificate in Cartography and Geographic Information Systems from Wuhan Technical University of Surveying and Mapping. Dr. Govorov has 20 years research and teaching experience in GIS, Cartography and Remote Sensing. He has experience leading and successfully completing several research and educational projects funded internationally by government and private organizations.


Send Email for Michael Govorov


6708.1 - Where in the Rural? Exploring Human Settlement Patterns in Rural Landscapes

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Where in the Rural? Exploring Human Settlement Patterns in Rural Landscapes