Assimilation of Remote Sensing Data in a Hydrologic Model to Improve - - PowerPoint PPT Presentation

Assimilation of Remote Sensing Data in a Hydrologic Model to Improve Estimates

• f Spatially Distributed Soil Moisture and

Surface Runoff

• Dr. Muhammad Rehan Anis

Global Change Impact Studies Centre (GCISC) National Centre for Physics Complex (NCP), Near Quaid-i-Azam University Campus, Shahdra Valley Road, Islamabad, Pakistan. rehananis@gmail.com

• Ph. +92 (51) 2077300, Ext: 492; Fax. +92 (51) 2077385

WaSiM

(Water balance Simulation Model)

Multi Layer Landuse (Seasonal) Multi Horizon Soil Table Silting up model and Surface Routing-module Irrigation model Lake model Snow Model Dynamic Glacier model Static glaciers, dynamic glaciers Dynamic phenology Permafrost Model Heat Transfer Model http://wasim.ch/en/index.html

High resolution rainfall-runoff modeling

Schäfertal catchment

Located in Lower Harz Mountains Area = 1.44 km2 Elevation = 392m - 470m

• Avg. Ann. Temp= 6.8 oC

January= -1.8 oC July = 15.5 oC Annual Rainfall 650mm Spatial Resolution = 10 m2 Temporal Resolution = 10 minute

Land use Type

Pine Forest Winter Barley Summer Barley Flax Meadow Tracks Stream

Soil Type

Gleysol Fluvisol Gleyic Cambisol Luvisol Cambisol

Validation 10-min WaSiM-ETH simulation

Distribution of soil moisture in Schäfertal catchment (1st October 2007 at 23 hrs)

IVF: 1.01 NSE: 0.86

Satellite Data Acquisition

• TerraSAR-X and TanDEM-X data collection
• Rapid Eye Data collection

Airborne Campaigns

• Airborne and the associated ground measurements campaigns
• Conduction of the airborne campaigns (May/June)
• Processing of the ground and airborne data
• Model inversion to soil moisture (polarimetric SAR Data)

Polarimetric Decompositions for Soil Moisture Retrieval from Vegetated Soils in TERENO Observatories

Observatory Optic SAR In Situ Bode Catchment Hyperspectral F-SAR & L-band Soil Moisture, Vegetation, Roughness

Observerd soil moisture in Schäfertal

…based on 100 „strategic“ sampling points… …after identification of similar areas based on topography using „fuzzy c-means“… …this similarity was finally used to project the measurements to a continuous map. In this case soil properties are mainly governed by topography.

Validation scheme for SMOS (temperature brightness-Tb) and soil moisture products.

An Overview of Data Assimilation

• J. A. Vrugt et al., 2013. “Hydrologic data assimilation using particle Markov chain Monte Carlo simulation:

Theory, concepts and applications”. Advances in Water Resources 51, 457–478.

Particle filter method is more accurate and faster than the Extended Kalman filter (EKF) or unscented Kalman filter (UKF) Discharge Time (hrs) Excerpt at Time t

? X t+1

• Particle-DREAM

→ Posterior state estimation using particle MCMC

• P-DREAM(VP) → State Augmentation or

Time-variant Parameters

• P-DREAM(IP) → Outside DREAM Loop
• r Time-invariant Parameters
• SODA: Simultaneous Optimization and

Data Assimilation method (EnKF)

Particle - DIFFERENTIAL EVOLUTION ADAPTIVE METROPOLIS (P-DREAM)

Model Catchment

Summer Runoff IVF R2 NSE Simulated 19/06 - 14/07/2007 0.92 0.29 0.29 Assimilated 19/06 - 14/07/2007 0.86 0.85 0.81

Data Assimilation using P-DREAM and WaSiM Hydrological Modeling

Estimated soil moisture generated using remote sensing technique and used in P-DREAM data assimilation algorithm to improve the peak summer runoff. (a) Peak summer discharge at 12:20hrs on 21/06/07 (b) Peak summer discharge at 18:00hrs on 25/06/07 (c) Peak summer discharge at 09:50hrs on 02/07/07.

Conclusion

• Data assimilation of surface soil moisture improves model

predictions especially reducing errors in surface runoff and soil moisture estimation.

• Soil moisture update through data assimilation can be a

supplementary way to overcome the errors created by inaccurate rainfall.

• Soil moisture estimations are useful for improved flood predictions

during transition periods from wet to dry in early spring and from dry to wet in early autumn.

Thank You & Questions

Soil moisture seasonality derived from the combination of six active and passive sensors over the period 1979 to 2010 Credit: Vienna University of Technology – TU Wien (TUW)