(in no particular order)
Currently, we use the MOD44W product to provide a "reference" water layer. This is removed from the detected surface water, to identify flood. However, there are areas where the MOD44W is not accurate due to seasonal or other issues, or is perhaps simply out of date. We will build a custom reference water layer based on our own algorithm's output. This may, at least for certain regions, be seasonally dependent, so that seasonally normal high water is not reported as flood.
We are working on implementing a cloud shadow mask, which will help us avoid mis-identification of water (and flood) in persistently partially cloudy areas, and also allow us to evaluate products from a single image. Currently, we require multiple (at least 2) water observations of a pixel for it to be marked as water. With an effective cloud shadow mask, we may be able to produce reasonable surface water and flood products from single MODIS observations.
The near real-time processed MODIS imagery has occasional mis-registration issues that may affect product output, particularly along the perimeters of detected water features. Fixes in our pipeline appear to have resolved this issue.
Currently, water is often erroneously detected in areas of terrain shadow, because shadows are often spectrally similar to water. This appears as a scattering of flood pixels in mountainous regions. We have implemented a first-cut mask for this, which removes 70-90% of such false-positives, and are working on an additional fix using a HAND-based mask (Height Above Nearest Drainage).
We have considered reporting detected flood area per tile, so that areas of current concern can be quickly identified. This information would be incorporated into the flood map tile display on the website homepage. It could also be automatically emailed to interested parties, or published on an atom or other feed.