Multivariate evaluation of a physically based distributed model in
snow-fed river basins of Hindu Kush Himalayan Region
Dr. Sangam Shrestha Assistant Professor, WEM, AIT
Pallav Kumar Shrestha Research Associate, WEM, AIT
[PRESENTER]
Background, Study Area INTRODUCTION
OUTLINE
Bias correction, Modeling approach, Evaluation approach
METHODOLOGY
Sensitivity Analysis, Model evaluation – hydrologic and snow response, Equifinality
RESULTS
Findings, Limitations CONCLUSION
BACKGROUND
- Recent evidences - Low latitude mountainous cyrosphere
undergoing change
- HKH lies in this range
- Distinct pattern – glaciers east of Karakoram experiencing negative mass balances
- Hindu-Kush-Karakoram-Himalaya (HKH) - Third pole, Water Towers of Asia
- Highest points on Earth – data scarcity
- Snow Dominant hydrology…
- Snow as a second variable to validate!
- Multivariate Model Evaluation - Q and Snow Cover (MODIS)
- Snow cover – spatial distribution
- Distributed modeling
- Precipitation coverage - Precipitation forcing at higher
altitudes
- APHRODITE estimates
- Panday et al (2013) - Tamor
- Hydrology evaluation - Drawbacks of NSE and R2
- Tailoring of balanced set of criteria
INTRODUCTION
STUDY AREA
INTRODUCTION
China
India
Nepal
Koshi Basin
Tamor Basin
Koshi Basin
8385 masl
235 masl (outlet) CA : 5884 km2
Tamor DEM
DATA . PRECIPITATION FORCING . BIAS CORRECTION
METHODOLOGY
Data Source
Topography ASTER – METI (Japan) and NASA
Hydrometeorology DHM, Nepal
Precipitation Estimate APHRODITE, Japan
Land use Survey Department, Nepal
Soil SOTER – ISRIC
Snow Cover MODIS (processed by ICIMOD)
- Hybrid precipitation input
- Ground Stations Data for lower elevations (9 rain gauges)
- APHRODITE estimates for higher elevation (6 fabricated stations)
- Power Transformation Method
- To remove bias in APHRODITE precipitation estimate
𝑷𝒄𝒐𝒓𝒓 = 𝒂.𝑷𝒖𝒏𝒄𝒐𝒓𝒓𝒃
- FORTRAN code utilizing secant root finding algorithm developed
Multivariate Modeling . SWAT . Snow Cover
METHODOLOGY
HYDROLOGY
- Calibration – 2000 to 2003 (4 yrs.); Validation – 2004 to 2006 (3 yrs.)
SNOW COVER
- Data for validation – RS Snow Cover (MODIS) – spatial data
- SWAT gives tabulated output of SWE (Snow Water Equivalent) in each elevation band of each HRU
- Tabulated Spatial – ArcGIS model builder
- Pixel by pixel comparison using Snow Pixel Efficiency (Seff) 𝑆𝑒𝑓𝑓 =
𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑒𝑙𝑙𝑠 𝑚𝑎𝑡𝑐𝑖𝑛𝑔
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑒𝑙𝑙𝑠 𝑐𝑜𝑚𝑝𝑎𝑟𝑒𝑑𝑋 100 %
MODIS SWAT
Evaluation Indices for Hydrology
METHODOLOGY
- R2, NSE, PBIAS – popular criteria for hydrology simulation
- Drawback of R2 - Biased to the pattern of the hydrograph
- Under/over-estimation - Unchecked
- weighted R2 incorporated
- Limitation of NSE - Biased to simulation of higher flows
- Relative NSE incorporated
𝑁𝑆𝐸𝑟𝑒𝑙 = 1 −
𝑦𝑜𝑏𝑠− 𝑦𝑠𝑖𝑚
𝒚𝒐𝒃𝒔
2
𝑦𝑜𝑏𝑠− 𝑦 𝑜𝑏𝑠
𝒚 𝒐𝒃𝒔
2
R2 = 1 wR2 = 0.7
𝑤𝑅2
= 𝑏 . 𝑅2 𝑓𝑜𝑟 𝑏 ≤ 1
𝑏 −1. 𝑅2 𝑓𝑜𝑟 𝑏 > 1
2011 2012
Sensitivity Analysis . Bias Correction
RESULTS
SWAT Cup
Alpha_Bnk
Ch_N2
Ch_K2
SNO50COV
TIMP
1
2
3
5
6
Power Transformation – APHRODITE Rainfall
Uncorrected
MBE
Corrected
RMSE
Multivariate Evaluation – Discharge
RESULTS
NSE 0.73 NSErel 0.89
NSE 0.78 NSErel 0.87
Multivariate Evaluation – Snow Cover Extent
RESULTS
- Hundreds of maps representing daily Snow Cover from SWAT throughout 2000 - 2007
- Overall average Snow Pixel Efficiency (Seff) throughout model years : 78.7 %
- Median – 76.2%
- Maximum – 89.4%
- Seff utilized by Pelliciotti et al (2012) Average of
Model Period
Seff : 78.7 %
Multivariate Evaluation – Equifinality
RESULTS
Indicators
Without Snow Calibration
Optimizing SNO50COV
Calibration Validation Calibration Validation
R2 0.79 0.79 0.78 0.79
wR2 0.53 0.66 0.57 0.70
NSE 0.69 0.76 0.71 0.77
NSErel 0.88 0.90 0.88 0.89
PBIAS -14.5 -6.9 -13.6 -6.4
Seff 0.71 0.79
Seff 0.71 Seff 0.79
CONCLUSIONS
- Power Transformation Method - a successful Bias correction method
- Precipitation forcing – Hybrid approach : gridded data + ground data – tackle data scarcity
- Distributed modelling – possibility of Multivariate evaluation
- SWAT model – Decent performance in both hydrology and snow extent
- Hydrology evaluation indices – balanced set of evaluation criteria with NSErel and weighted R2
- Multivariate evaluation – dealing with Equifinality
CONCLUSIONS
- Bias correction of temperature - APHRODITE – daily single value
- SWAT – daily extremes (2 values)
- Temporal variability of Lapse rates - SWAT – single value for all seasons
- Temporal stationarity
- Glacial hydrology - 10.6 % of Tamor is Glaciers (GLIMS database)
- Next step – models with glacier/ ice module
- SWAT Holes - Why?? - Spatial coverage incomplete due to thresholds
in HRU definition step
- ~22% blank in Tamor SWAT model
- Next step – fully distributed model
LIMITATIONS
Gratitude
Thank you for your attention!