Prediction of runoff seasonality in ungauged basins
1. Prediction of Runoff Seasonality
in Ungauged Basins
Pierluigi Claps [claps@polito.it]
Francesco Laio
Politecnico di Torino, Italy.
AGU Fall Meeting 2009
Thans to: Gianluca Vezzù, Daniele Ganora, Elisa Bartolini
2. Runoff Seasonality
• (1) A hydrological signature in Catchment Classification
• (2) A starting point for streamflow Prediction in Ungauged
Basins
• (3) A key pattern for assessing effects of global warming in
critical (mountain) areas
AGU Fall Meeting 2009 P Claps - F Laio
3. Runoff seasonality
Mean annual runoff [mm]
North-Western Italy
41 basins
AGU Fall Meeting 2009 P Claps - F Laio
5. (1) Classification (and prediction) based on Distances
Regime curves (treated like patterns) are related to each other by means of
their similarity - dissimilarity
• Dissimilarity = distance, e.g.:
|qi,s1 – qi,s2|
• “complex” basin descriptors can be used,
as, e.g., precipitation regimes
•Regime distances are put in a
distance matrix
•Analogous distance matrices are
created for each descriptor
AGU Fall Meeting 2009 P Claps - F Laio
6. Distance-based approach
e.g. Ganora et al. (WRR 2009) applied to FDC
• A regression model identifies the most significant descriptors
Regime distance Descriptors’ distance
matrix matrices
• Significance of regression coefficients: Mantel test [Mantel and Valand ,
1970], Lichstein [2007] (distance matrices contain dependent values)
• Cluster analysis or nearest neighbors with the selected descriptors
will allow for curve estimation in Ungauged Basins
AGU Fall Meeting 2009 P Claps - F Laio
7. Significant Variables
Centroid latitude, Mean elevation, main orientation angle
Distance between avg. elevation
Distance between regimes
Best regression model: highest R2 with all the covariates being significant
AGU Fall Meeting 2009 P Claps - F Laio
9. (2) Prediction by Parametric (Fourier) approach
e.g. Claps et al. (JHE 2008) applied to Temperatures in Italy
AGU Fall Meeting 2009 P Claps - F Laio
10. Regression Model Selection
based on cross-validation RMSE and MAE
Optimal regressions: common set of descriptors
R2adj=0.68-0.89
AGU Fall Meeting 2009 P Claps - F Laio
11. Regression Model Selection
based on cross-validation RMSE and MAE
Optimal regressions: common set of descriptors
R2adj=0.68-0.89
Best regressions: highest R2 with all the covariates being still significant
R2adj=0.84-0.93
AGU Fall Meeting 2009 P Claps - F Laio
13. (3) Role of rainfall regime and prediction (mountain areas)
Quasi-deterministic model (Snow-affected runoff)
Snow storage effects on the runoff regime;
Detection of unreliable rainfall measurement;
Assessment of precipitation underestimation due to undercatch
regime
Model input
Precipitation and temperature regime
Digital Elevation Model
Model output
Runoff regime
Snow storage and melting
AGU Fall Meeting 2009 P Claps - F Laio
14. Model Features
1. Sub-monthly temperature variability
Logistic distribution with
m=mean temp and
variance to calibrate
1-t
temp < 0
The cumulative probability
snow allows to partition different
temp > 0
physical processes within
melt and
t ET
the same month
2. Snowmelt based on degree-day approach
d: number of days
tmpj : monthly positive tmp
tmpb : threshold tmp
k : melting rate
AGU Fall Meeting 2009 P Claps - F Laio
15. Model Features
1. Sub-monthly temperature variability
Logistic distribution with
m=mean temp and
variance to calibrate
1-t
temp < 0
The cumulative probability
snow allows to partition different
temp > 0
physical processes within
melt and
t ET
the same month
2. Snowmelt based on degree-day approach
d: number of days
tmpj : monthly positive tmp
tmpb : threshold tmp
PARAMETERS TO CALIBRATE:
- within-month variance of T k : melting rate
- Melt factor k
AGU Fall Meeting 2009 P Claps - F Laio
16. Application to 41 basins in North-Western Italy
Mountain areas
AGU Fall Meeting 2009 P Claps - F Laio