NASA Logo, National Aeronautics and Space Administration

NRT Global MODIS Flood Mapping

Product Description

Basic System

Onboard NASA's Terra and Aqua satellites, the MODIS instrument provides twice daily near-global coverage at 250 m resolution in two optical bands; these are the key data source for our flood products. The LANCE processing system at NASA Goddard provides the products that we ingest typically within a few hours of satellite overpass. The Terra equatorial overpass is at ~10:30 AM local solar time, and Aqua at 1:30 PM. Although other instruments provide higher resolution data, none provide global daily coverage. We are working to bring targetted radar data into the system to allow us to "see through" clouds, and so provide more timely products.

Water Detection Algorithm

We detect water using an algorithm developed by Bob Brakenridge of the Dartmouth Flood Observatory. This algorithm uses a ratio of MODIS Band 1 and Band 2, and a threshold on Band 7 to provisionally identify pixels as water.


We then composite the water detections over the product window (2 days, typically). If a pixel is identified as water over several (2 or more) observations during the product window, it is then definitively marked as water, and output in the MSW, or "MODIS Surface Water" product. 2 (or more) observations are required because cloud shadow can appear quite spectrally similar to water. In cases where cloud shadow occurs in the same spot in 2 observations, the product may incorrectly flag such areas as water. A 3 observation requirement helps further, but also increases latency of the product. At the moment, we are using the 2 observation requirement as a balance between accuracy and timeliness.

Flood Identification

Finally, the detected water is compared to a reference water layer that shows "normal" water extent, and any pixels found outside the normal water extent are marked as flood, and output in the MFW ("MODIS Flood Water") products.

Known Issues

Resolution: The MODIS data we use is at 250 m resolution, so flooded features smaller than this size will not be reliably detected.

Geo-Referencing: An initial examination of the geo-referencing of the water identification algorithm for this product has indicated an inconsistency with other data, including Landsat, and also the increased latency MODIS science products. This is an important issue which is under investigation.

Reference Water: We currently use the MOD44W product as a fixed global reference water layer. This is also derived from MODIS data, but shows consistent differences in places from what our algorithm is detecting as water in non-flood conditions. This issue may be related to the geo-referencing issue cited above. As a part of this project, we will build our own reference water layer using the MSW product, and allow seasonal variations in the normal water extent.

Clouds: MODIS cannot see through clouds, so we are unable to determine surface water extent when an area is cloudy. Our standard products are built using 2 (or more) days of data, so there are several (2 per day) chances to observe an area. The graphic map products (MFM, MODIS Flood Map) show areas where there was insufficient clear imagery for water to be detected.

Cloud Shadows: These are often spectrally similar to water, and thus can be flagged as water with our algorithm. Currently, we avoid cloud shadows by requiring a water identification on multiple (at least 2) observations. This solution fails when partly cloudy conditions prevail allowing cloud shadows to cover the same ground over one or two days. It also limits the time-sensitivity of the products, since at least 2 water observations are required to mark pixels as water (and potentially flood).

As of July 2012, we have implemented an initial version of terrain shadow masking to reduce the number of false water detections under terrain shadow.