DischargeFromSpace background

Technical Background

Overview

Passive microwave satellite sensors provide near-daily, global coverage of the Earth’s land surface. At certain microwave frequencies, these instruments can penetrate cloud cover, making them particularly well suited for river monitoring. Extending earlier efforts based on wide-area optical observations (e.g., Brakenridge et al., 2005; Van Dijk et al., 2016), data from passive microwave missions such as AMSR-E, AMSR-2, TRMM, and GPM can be used to detect and track variations in river discharge.

As river discharge increases, the water-covered surface area within a river reach expands. Since open water emits less microwave radiation than surrounding land, the brightness temperature of a pixel containing both land and water decreases (Brakenridge et al., 2007, 2012). The DFO - Flood Observatory exploits this relationship by examining individual pixels of approximately 10 km × 10 km and tracking their microwave emissions over time. An increasing fraction of water within a pixel leads to a reduction in observed microwave radiation, producing a signal that responds directly to changes in inundated area and indirectly to variations in discharge. These microwave-derived signals are converted to discharge values through calibration against long-term (~10-year) records from in situ gauging stations or outputs from the Water Balance Model (WBM).

Calibration Approach

Like stage-only measurements from ground stations, the surface-water-area-sensitive microwave signal is useful even without conversion to units of discharge. It allows detection of flood peaks, low-flow duration, and relative changes in flow state. However, transforming the signal into m³/s adds substantial value.

Most DFO discharge estimates use WBM output as an independent reference. For this work, the model provides daily discharge values from 1998–2011 at a 10 km global resolution, matching the microwave pixel scale. By comparing modelled discharge with the remote sensing signal, a statistical rating curve can be generated. This curve (linear or low-order polynomial) converts each day’s signal into a discharge estimate.

For calibration, we use monthly maximum, minimum, and mean daily values from both sources (n = 504). Prior work (Brakenridge et al., 2012) showed this method reliably captures low, medium, and high flows. Some scatter is expected, because errors in both the model and the remote sensing data contribute. In some locations, the model may mis-time peak flows, even though the satellite signal closely tracks real discharge changes. Nonetheless, for most pixels, the correlation between WBM discharge and the satellite signal is strong because both independently detect the underlying hydrologic dynamics.

Processing Workflow

We use information from the 36.5–37 GHz channels of AMSR-E, AMSR-2, TRMM, and GPM. The discharge indicator is computed as a ratio:

M - emissivity of the measurement pixel directly over the river and floodplain
C - the 95_th-percentile emissivity from a 9×9 pixel matrix surrounding the site, representing the day’s “driest” land signal

Using the 95_th percentile avoids extreme outliers while providing a stable land-reference value. The ratio M/C is highly sensitive to variations in surface water area within the measurement pixel. This spatial-ratio method also removes regional influences that affect all pixels alike, such as seasonal changes in land surface temperature.

Daily global brightness-temperature grids are produced at the Joint Research Centre (JRC) using near-real-time swath data. The resulting grid has ~10 km resolution near the equator (4000 × 2000 pixels). At lower latitudes, some sensors do not provide complete daily coverage, so we apply a 7-day weighted mean to mitigate gaps. If multiple overpasses occur on the same day, the latest observation is used.

This technique is distinctive because it repurposes microwave channels originally added to these missions to support precipitation retrieval. The ~37 GHz ground-sensing frequency, normally used as a baseline for atmospheric observations, is here leveraged to monitor river discharge.

Defining Satellite based discharge stations

The measurement pixel (M) must be positioned and sized so that:

1. It does not become fully water-covered during major floods (to avoid saturation), and
2. It excludes nearby lakes, reservoirs, or floodplain water bodies whose fluctuations do not reflect changes in river discharge.

Local morphology strongly influences performance, similar to traditional gauging stations.

• In straight or leveed channels, water level (stage) may be a better proxy for discharge than flow area.
• In meandering, braided, or anastomosing rivers, changes in flow area often outperform stage as an indicator. Even relatively small meandering rivers can produce high-quality microwave discharge signals.

Accuracy Considerations

Accuracy is influenced by multiple factors, including river morphology, long-term channel changes, and the intended application. In many operational settings, long and consistent discharge records are more valuable than short but extremely precise datasets. One advantage of this satellite-based approach is that it effectively monitors an 10 x 10 km area, which may reduce sensitivity to local channel-cross-section changes.

Comparisons With Ground Stations

Ongoing evaluations compare satellite-derived discharge with co-located gauge measurements. These tests highlight the strong role of local conditions in determining accuracy. Although the WBM model simulation calibration dataset shows some systematic biases, these biases have limited effect on relative metrics such as:

• flood peak magnitude,
• hydrograph timing and shape,
• recurrence interval estimates,
• low-flow thresholds, and
• flood/low-flow duration and seasonality.

Potential additional Error Sources

Several known issues affect the satellite signal:

1. Irrigation — In agricultural regions, irrigation can alter the emissivity of both the measurement and comparison pixels, reducing interpretability. Such cases are flagged on the SGR interface.
2. Instrument noise — Occasional abrupt, one-day drops in M/C can create false positive discharge spikes. Work is ongoing to filter these artifacts.
3. High-latitude ice — Further development is underway to refine ice-cover detection in cold-region rivers.

Data Usage Notes

The discharge data generated by this system are experimental and provided without warranty. The processing pipeline relies on signal inputs from the Global Flood Detection System (GFDS) at the European Commission’s Joint Research Centre (GDACS project). We are in the process of better place some of the locations of the stations to better curate a set of satellite gauging sites that are more reliable.