The University of Montana
Department of Mathematical Sciences

Technical report #6/2010

Prediction of river hydraulic conditions via satellite multi-spectral imagery and statistical learners

Brian M. Steele, Univ. of Montana
Department of Mathematical Sciences, University of Montana, Missoula MT, 59812, USA

Mark S. Lorang
Flathead Lake Biological Station, University of Montana, 32125 Bio Station Lane Polson,
MT 59860-9659 MT, USA


The hydrologic regimes and ecology of many rivers have been altered through human action. Understanding the natural processes and the extent of alteration is a crucial component of management and restoration of these systems. River and floodplain systems are often highly dynamic, requiring data sets that cover a large spatial extent to fully assess ecological structure and function. However, in situ data collection is extremely challenging in these environments, particularly in localities of high river velocity. This article discusses an approach that exploits high resolution remotely sensed images for expanding data coverage, and thereby enhancing the quantification of important hydraulic parameters, specifically water depth, flow velocity and energetic state through Froude number. The primary elements of this approach are (i) acoustic Doppler profilers coupled with GPS for in situ data collection of water depth and velocity (ii) satellite multispectral imagery, (iii) modern distribution-free statistical learners, and (iv) error analysis. Herein, we provide motivation and background for the study of riverine landscapes via remote sensing, though the article concentrates on the third and fourth elements of the approach. An analysis of a 12 km reach of the Tagliamento River located in Northern Italy provides an example of our approach.

Keywords: random forest, Quickbird

Download Technical Report: Pdf (1413 KB)