Leveraging Soil Moisture for Early Flood Detection

Abstract:

Publicly available multispectral and radar satellite sensors can map flooded areas at spatial resolutions ranging up to a few hundred meters. However, they often do not observe the maximum inundation extent, either because clouds obscure the view or due to infrequent overpasses. In places where flood water moves rapidly or where clouds are more persistent, this proves to be a significant barrier to successful monitoring of large floods in ungauged watersheds.

Microwave sensors can penetrate clouds and have a high revisit frequency. Preliminary studies have shown soil moisture signals from passive microwave satellites to be useful for flood identification (Wu et al, 2019), but the data is only available at coarse spatial resolution (~10-km ~ 36-km). In this work, we combined microwave-based soil moisture retrievals from SMAP, SMOS, AMSR2, and ASCAT to use soil moisture as a surrogate for flooding at a 10-km resolution with (at most) 3-hourly temporal resolution.

We use high (3-m, 10-m) to medium (250-m)-resolution optical and SAR data to establish the relationship between the coarse resolution soil moisture signal and the surface water extent within a pixel. Using frequent soil moisture observations as a proxy for surface water allows us to know when higher resolution observations occur in relation to the flood peak and recession. Additionally, we assess precipitation-soil moisture-inundation relationships for understanding flood risk.

We present case studies assessing the potential of these data for high-temporal resolution flood monitoring where ground and optical data is not available during large events such as the recent cyclone Amphan in Kolkata, India and 2019 floods in Khartoum, Sudan. By leveraging multi-sensor and multi-resolution datasets, we aim to augment flood maps and fill key information gaps in data-sparse regions when multispectral and radar data is also not available.