DavesNotes

MODIS data in real time. 500-m resolution for things like snow, daily coverage

• MODO-9 surface reflectance
• Fraction snow cover, RMS error 10-12% versus landsat.
• Time-space filtering solves for interruptions, noise
• Pixels are different sizes.

Energy Balance Reconstruction

• Day of full melt--> day of snow depth for most recent peak
• Peak snow depth is a good predictor of stream runoff
• Estimate incoming energy from solar radiation etc. Topographic downscaling
• Also air temp, outgoing radiation
• Based on this and day of melt, can estimate how much water was released
• Considered pretty good, but don't know its value until

Real time passive microwave image from a different satellite

• The stuff you download from AMSR is off by a factor of 10.
• In addition to poor calibration, it has low R^2. Sometimes even negative
• Principle: ice is much more transparent in microwave range than ice is
• Microwaves come from soil (top 5 cm), but is scattered/attenuated by ice
• Varies with soil (liquid) moisture.
• More emissions implies less snow. Their algorithm uses two frequencies
• Also, pixel size is pretty big & intensity varies a lot by day
  • Some pixels fluctuate a lot more than others
  • If snow is wet, estimates will be too low
  • If crystals are big, estimates can bee too high.
• Microwave data is daily. Avoid the daytime data, use nighttime instead.
• Problems: heterogeneity within pixels, vegetation, miscalibrated.
• Estimates are too high below 100 mm and too low above.
• Clear sensor saturation

Validation:

• In Sierra-Nevada, we get much better estimates of how much snow actually melted
• Reconstruction based on date of snow melt does pretty well.
• It doesn't work until the snow has melted
• Direct estimates of snowfall from satellites are considered unreliable

Training Data:

• MODIS snow cover % (mean and sd)
• (Other land cover?)
• topography (elevation/slope/aspect/more?)
• daily temperature, radiation
• Daily microwave data

Test data:

• Day of snow melt