1. Statistical Applications of Physical Hydrologic Models and Satellite Snow Cover Observations to Seasonal Water Supply Forecasts
Eric Rosenberg1, Qiuhong Tang1, Andrew W. Wood2, Anne C. Steinemann1, and Dennis P. Lettenmaier1
1Department of Civil and Environmental Engineering, University of Washington, Seattle, WA Tier Incorporated, Seattle, WA
Project Overview
Regression Analysis
Satellite Snow Cover Observations
Despite research demonstrating the value of physically based hydrologic models and satellite observations of snow covered area (SCA) for water supply forecasts, these tools remain underutilized by operational agencies. Part of the reason for this disparity lies in the contrast between experimental forecasting techniques, which tend to employ methods such as ensemble streamflow prediction, and operational forecasting systems that rely on simpler methods like statistical regression to link surface observations of snow water equivalent (SWE) to seasonal runoff volumes. We explore methods that bridge this gap via a hybrid approach which uses model-simulated snow states and raw satellite data as predictors in regression models that are adapted to the operational environment.
In addition to SWE, other predictors included in DWR’s runoff forecasts are precipitation and prior water year runoff. Our model-based regression forecasts were compared with DWR’s ground-based regression forecasts, both trained on data from The skill of each method is compared by plotting the 10th and 90th percentiles of the resulting residuals (black for ground-based; blue, green, or red for model-based) in the so called “funnel plots” below. Mean April-July flows at the yellow forecast points to the left are provided in parentheses.
The Moderate Resolution Imaging Spectroradiometer (MODIS) is a multiband satellite sensor whose 500 m spatial resolution and daily temporal resolution provide near-ideal conditions for hydrologic applications. Our second approach will base regression models on SWE and SCA output from a VIC model that has been updated with snow cover data from MODIS. Our third approach will employ MODIS SCA data directly as predictors in the regression models.
Two challenges inherent in using MODIS data involve its limited record (2000 to the present) and gaps in coverage caused by cloud cover. To address the first problem, we will attempt to extend the MODIS record with data from the 1 km Advanced Very High Resolution Radiometer, the 1 km SCA product from the National Operational Hydrologic Remote Sensing Center, and retrospective VIC simulations. The second challenge will be addressed using the MODSCAG product (Dozier et al. 2008), which also provides advantages offered by fractional snow cover data.
Upper Sacramento
Feather
Yuba
American
Cosumnes
Mokelumne
Stanislaus
Tuolomne
San Joaquin
Merced
Kings
Kaweah
Tule
Kern
U. Sacramento
(2494 taf)
Feather
(1781 taf)
Yuba
(1005 taf)
American
(1240 taf)
+100
The 14
watersheds
of the
Sacramento
(blue), San
Joaquin (green),
and Tulare Lake
(red) hydrologic
regions, which form
our study areas for the
project. Together, the 3
regions are responsible
for roughly 60% of the
state’s runoff. Yellow circles
represent runoff forecast points for
California’s Department of Water Resources
(DWR). Watersheds with both light and dark
colors are divided by DWR into areas of high and
low elevation for purposes of snow measurement.
-100 F M J A F M J A F M J A F M J A
Residual (% of mean annual flow)
Cosumnes
(126 taf)
Mokelumne
(460 taf)
Stanislaus
(702 taf)
+100
-100 F M A J F M A J F M A J
MODIS
image
from April 7,
Snow is
shown in white
and clouds in
cyan. MODSCAG
(Dozier et al. 2008) will be
used to estimate fractional
snow cover and eliminate data gaps due to clouds.
Tuolomne
(1220 taf)
Merced
(632 taf)
San Joaquin
(1254 taf)
+100
The Terra satellite (top), the first to carry the MODIS sensor, was launched in Dec The Aqua satellite (middle), the second to carry the MODIS sensor, was launched in May 2002.
Model-Simulated Snow Water Equivalent
DWR’s forecasts for April-July runoff rely on manual measurements of SWE at various snow courses throughout each watershed (left, for the Feather). In our first approach, we develop new regression equations based on SWE simulated by the Variable Infiltration Capacity (VIC) hydrologic model at points spaced 1/8° apart (right).
-100 F M A J F M A J F M A J
Kaweah
(286 taf)
Tule
(64 taf)
Kern
(461 taf)
Acknowledgements
Residual (% of mean annual flow)
Kings
(1224 taf)
We are grateful to the Division of Flood Management at DWR, and in particular, Adam Schneider and David Rizzardo, for their invaluable assistance. Funding has been provided by NOAA and NASA.
+100
References
-100
Dozier J, Painter TH, Rittger K, and Frew JE Time-space continuity of daily maps of fractional snow cover and albedo from MODIS. Advances in Water Resources. 31(11) F M J A F M J A F M J A F M J A