HR: 1340h
AN: H53G-1515 [Abstracts]
TI: Robotic Stream Flow and Solute Mass Balance Measurements Guided by a Non-Stationary Gaussian Process Model
AU: * Singh, A
EM: singh.amarjeeet@gmail.com
AF: Electrical Engineering Department, UCLA, Los Angeles, CA 90095, United States
AU: * Singh, A
EM: singh.amarjeeet@gmail.com
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AU: Fisher, J
EM: jfisher@ucmerced.edu
AF: Environmental Systems Program and Sierra Nevada Research Institute, University of
California, Merced, Merced, CA 95340, United States
AU: Fisher, J
EM: jfisher@ucmerced.edu
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AU: Pai, H
EM: hpai@ucmerced.edu
AF: Environmental Systems Program and Sierra Nevada Research Institute, University of
California, Merced, Merced, CA 95340, United States
AU: Pai, H
EM: hpai@ucmerced.edu
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AU: Villamizar Amaya, S
EM: svillamizar_amaya@ucmerced.edu
AF: Environmental Systems Program and Sierra Nevada Research Institute, University of
California, Merced, Merced, CA 95340, United States
AU: Villamizar Amaya, S
EM: svillamizar_amaya@ucmerced.edu
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AU: Harmon, T C
EM: tharmon@ucmerced.edu
AF: Environmental Systems Program and Sierra Nevada Research Institute, University of
California, Merced, Merced, CA 95340, United States
AU: Harmon, T C
EM: tharmon@ucmerced.edu
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AU: Kaiser, W
EM: kaiser@ee.ucla.edu
AF: Electrical Engineering Department, UCLA, Los Angeles, CA 90095, United States
AU: Kaiser, W
EM: kaiser@ee.ucla.edu
AF: Center for Embedded Networked Sensing (CENS), UCLA, Los Angeles, CA 90095, United
States
AB:
Spatially distributed hydraulic and water quality property characterization is important to understanding a broad
range of river issues including confluence and discharge mixing phenomena, groundwater-surface water
exchanges, and flow and temperature distributions in the context of habitat restoration efforts. Such
characterization efforts often need to be completed rapidly to avoid complications associated with transient
upstream conditions ( e.g., reservoir operational changes, time-variable irrigation drainage). In this work, we test
a non-stationary Gaussian Process (GP) model for increasing sampling efficiency during a robotic deployment of
velocity (ADV) and electrical conductivity (EC) sensors across a river transect. GP modeling is a common
statistical approach for addressing spatially distributed phenomena. We first develop velocity and salinity
observations within the mixing zone of the Merced-San Joaquin River confluence robotically in the form of high
resolution (114 point) raster scans. We train the GP model by dividing the river cross-section into three sub-
regions corresponding to Merced river side (east), mixing zone (center), and San Joaquin River side (west). An
information criterion was selected that assigned each observation location a quantitative value in terms of the
uncertainty about our prediction of the EC value given the measurement made at that location. We then executed
a path-planning algorithm optimizing 16 locations out of the original 114. Using the observations from these 16
locations, and the trained GP model, we predicted the values at the rest of the 98 unobserved locations. EC
distributions are compared for the raster- and GP-based data and suggest that the GP modeling strategy is viable
for enhancing sampling efficiency in the context of spatially distributed river characteristics.
DE: 1856 River channels (0483, 0744)
DE: 1860 Streamflow
DE: 1871 Surface water quality
SC: Hydrology [H]
MN: 2007 Fall Meeting