River Watch 2
Satellite River Discharge and Runoff Measurements: Technical Summary
G. Robert Brakenridge
CSDMS/INSTAAR, University of Colorado
Tom De Groeve
Joint Research Centre of the European Commission, Ispra, Italy
Surface Dynamics Modeling Laboratory, Dept. of Geography, University of Alabama
Son V. Nghiem
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA
February 20, 2013
Satellite microwave sensors provide global coverage of the Earth’s land surface on a near-daily basis and, at certain wavelengths, without significant interference from cloud cover. Using a strategy first developed for wide-area optical sensors (see for example: Brakenridge and others, 2005), such data can be used to measure river discharge changes. As rivers rise and discharge increases, floodplain water surface area increases. Microwave emission over river measurement sites, observed from space, can monitor such changes.
Transformation of the remote sensing signal to actual discharge values is accomplished via use of a rating equation (as the case for most stage-based gauging station discharge measurements on the ground). For River Watch 2, the calibrating discharge values are obtained by Sagy Cohen via runs of a global runoff model (WBM). Five years (2003-2007) provide abundant data for calibration. The model produces daily discharge values for these years, at each measurement site (the global grid resolution is 10 km). As shown on the site displays (click "obtain data" on each site), a signal-discharge rating curve results. The rating curve equation then transforms all daily signal data, including new data, to discharge.
In most cases, the rating curve uses monthly maximum, minimum, and mean daily discharges for both model output and the remote sensing ( n=180). We expect much scatter (both model and remote sensing error are included). However, in many cases, along well-chosen river reaches, there is strong correlation of modeled discharge values and the remote sensing signal. In such cases, it is clear that: 1) the remote sensing signal is indeed tracking discharge, and 2) the modeling is producing a realistic time series of discharge variation. The degree of correlation does not, however, verifiy the discharge magnitudes (there could be model bias, thus making uniformly too high or too low discharge output values). Testing is therefore underway, within the U.S. and elsewhere, to compare River Watch discharge output with ground-based gauging station data, which are normally calibrated via in-situ current meter measurements. See the online U.S, examples in this regard as well, as well as these examples.
River Watch 2 uses the NASA/Japanese Space Agency "Advanced Scanning Microwave Radiometer (AMSR-E)" band at 36.5 GHz (descending orbit only, horizontal polarization) and also the NASA/Japanese Space Agency TRMM 37 GHz channel. The discharge estimator is a ratio of a daily calibrating values ("C") that represent that day's background emissivity near the site, and "M", the emissivity from a measurement pixel centered over the river and its floodplain. C/M is very sensitive to changing surface water area within the M pixel. The original River Watch used a simple ratio of two pixels. For River Watch 2, and using a technique developed by T. De Groeve, "C" is the 95th percentile value of the brightest (driest) calibration pixel brightness temperature for a 7x7 pixel array of calibration pixels centered on M, the measurement target. This provides sensitive and less noisy background ground surface microwave information, to which the "M" pixel values can be compared. Even small changes due to water/land proportions within the "M" pixel are thereby observed.
At mid-latitudes, AMSR-E and TRMM pixel dimensions are approximately 10 km. It is important that the M pixel be large enough to avoid saturation (complete filling of the measurement pixel by water) during flood events. Because a river reach, ~10 km in length (one pixel) is used to sense discharge changes, it is also important that the pixel include a relatively uniform stretch of river without major tributary junctions, or nearby streams or variable water bodies. Local site characteristics in fact strongly affect the sensitvity and signal/noise ratio of this method. Each site display thus provides a link to visualize the measurement pixel (river and floodplain reach) being used as the monitoring site. Careful quality control must be provided at each site, beginning with visual examination of the site characteristics: we are still performing such work, so that some sites may be deleted in the future. However, there are thousands of more suitable river measurement sites to be added.
A little more on how this works: Due to low emission from water surfaces compared to land surfaces, and depending on channel and floodplain morphology, the C/M ratio responds strongly to river discharge changes. Modeling and empirical studies demonstrate that the ratio is, in contrast, relatively unaffected by soil moisture, vegetation, or other changes affecting both land parcels. The initiation and removal of river ice cover can also be detected: ice breakup immediately affects the C/M ratio as low-emission water replaces ice within the pixel. See Brakenridge and others (2007) for the initial research paper describing this approach, Brakenridge and others (2012) for updated information, and another technical report by Kugler and De Groeve (2007). A list of relevant references is also available.
The approach is novel in that microwave sensors used previously to monitor the atmosphere and precipitation are here employed to directly measure river discharge changes and watershed runoff on the ground. This is possible because the sensors included multiple microwave bands designed for different purposes in. Thus, to examine atmospheric conditions, such as precipitation, ground-sensing channels were included in order to provide the background component of upwelling microwave radiation. River Watch 2 uses just these channels to monitor surface water changes. River Watch 2 replaces an earlier version that used only AMSR-E and, as noted, somewhat different signal algorithms. It is now running forward in time using only TRMM information (AMSR-E has ceased operation). River Watch 2 will add AMSR-2 information when it becomes available and is ingested into the GDACs-GFDS system that provides our source remote sensing signal data (see link to that data source on each site page)..
Currently, all data are processed as forward-running 4 day means. This is because, especially for lower latitudes and AMSR-E, sensor coverage at a given measurement pixel sometimes skips a day or two. Daily variation is, however, still captured, and although smoothed by the averaging process.
Discharge data are fundamental for observation of surface runoff (commonly expressed as discharged water volume/watershed area, in mm). Daily runoff maps showing runoff and runoff anomaly values for each watershed associated with a discharge measurement site are presently online. Soon 7-day accumulated runoff and runoff anomaly displays will also be provided.
This is a cooperative project between the University of Colorado, Boulder, CO, USA and GDACS-GFDS (Global Disaster Alert Coordination System, Global Flood Detection System), European Commission Joint Research Centre, Ispra, Italy, the University of Alabama and the Jet Propulsion Laboratory. Drs Kettner and Syvitski also provide processing and analysis support and input at the University of Colorado.
The Dartmouth Flood Observatory at the University of Colorado is supported in part by grants from NASA. We wish to thank Dr. Bob Adler, University of Maryland, for initial suggestion that the TRMM sensor data could be used in addition to that from AMSR-E.
Brief listing of known errors: 1) In agricultural areas, irrigation of the measurement pixel can affect the signal, even when the surrounding landscape is also irrigated. 2. The rating curves include both model and measurement errors, and, as is the case for many stage/discharge rating curves, the relation between the two data sets may vary with different flow regime (but not yet be expressed by the single-equation rating curves now being used). 3. Intermittent sensor noise (commonly marked in the remote sensing data files by occasional very negative values) occasionally produce intermittent positive spikes in discharge. We are working to filter such instrument noise. 4. Comparison to ground station (stage-based) daily data sometimes indicates significant (1-3 day) lags between the stage peak and the water area (River Watch) flood peak: this introduces error into the scatter plots used to develop rating equations, which can be removed, in some cases, by adjusting the two time series incorporating the observed lag.
Flood Observatory Homepage