INTEGRATION OF REMOTE SENSING AND SNOW HYDROLOGIC MODELLING USING DATA ASSIMILATION

1
INTEGRATION OF REMOTE SENSING AND SNOW HYDROLOGIC
MODELLING USING DATA ASSIMILATION
Kostas Andreadis
2
Outline

• Motivation
• Remote Sensing/Data Sources
• Data Assimilation Methods
• Problems
• Research Directions
• Discussion

3
Motivation

• Study Area Columbia river basin
• Snowmelt flooding potential
• 70 of streamflow comes from snowmelt
• 62 forested areas

4
Snow Remote Sensing

• Large scale observations
• High albedo of snow
• Cloud-Snow discrimination difficulties

Hall et al. 2002
5
Science Question

• Provide a computationally efficient and
  physically consistent scheme of combining remote
  sensing data with hydrologic modelling for
  snowmelt runoff prediction

6
Variable Infiltration Capacity Snow Model
7
MODIS Snow Products

• Available from March 2000 (Version 4.0)
• Extensive Quality Assurance information
• Improved spatial resolution
• Better snow/cloud discrimination
• Efficient snow mapping in forested areas

8
Snow Data Availability

• MODIS - 0.5 km Daily
  
• GOES - 1 km Hourly
• AVHRR - 1 km - 2 to 3 days
• SNOTEL 660 Stations (U.S.) - Daily

9
Data Assimilation

• Provides time-dependent and spatially distributed
  state estimates that can be updated whenever new
  data become available
• Data-oriented technique
• Uses model to constrain observations

10
General Data Assimilation formulation
DATA
STATE ESTIMATE
ANALYSIS
ERROR ESTIMATE
MODEL
11
Data Assimilation methods

• Optimal Interpolation
• Newtonian Nudging
• 3-D/4-D Variational Assimilation
• Kalman Filters (standard, extended, ensemble)

12
4-D Variational Assimilation
13
Extended Kalman Filter
Compute Kalman gain
Update estimate with measurement
Update the error covariance
14
Problems

• Physical consistency
• Computational effort
• Temperature bias causes divergence
• Need for a multiple product dataset to attain
  maximum potential information
• Issues regarding difference in scale

15
Physical Plausibility

• Dissapearing Layers and states
• Arbitrary redistribution of snow mass between
  layers
• Issue of algorithm stability for binary variables
  (Snow cover1 or 0)
• Snow cover lacks information about snow
  density/depth

16
Computational Effort

• Slight increase in resolution could lead to much
  higher computing cost
• Dimensionality reduction
• Faster optimization algorithms
• Multiple Assimilation window
• Parallel computing algorithms

17
Temperature Bias

• Synthetic experiments showed large-scale errors
  from small temperature bias

Houser et al. 2002
18
Utilizing Multiple Datasets

• Can we combine ground and satellite-derived data
  to produce a snow cover product ?
• Account for areas with thick cloud cover, by
  using station or passive microwave data

19
Differences in scale

• Define the measurement operator in a way that the
  observation value is optimally interpolated
  across the larger/smaller scale

20
Implementation

• Operational application with MODIS for the
  Columbia basin
• Retrospective analysis based on AVHRR data
  (longer record)
• Data Assimilation techniques intercomparison and
  performance assessment

21
Questions?