Document

Report
Satellite rainfall application in the
Narayani basin
CORDEX South Asia science workshop
Kathmandu, Nepal
27-30th August 2013
Mandira Singh Shrestha
International Centre for Integrated Mountain Development
Kathmandu, Nepal
Presentation outline
 Satellite rainfall estimates
 Hydrological modelling using GeoSFM
 Study area
 Input data
 Intercomparison of products
 Calibration and validation
 Results
One third of the disaster are floods
Pakistan floods: 2000 killed, 20 million
affected
Most floods are transboundary which requires
cooperation across borders
Uttarakhand disaster:
>5000 killed, millions affected
Why satellite rainfall estimate?

Inadequate density of
hydrometeorological stations

Delay in data transmission

Not adequate lead time –
limited data sharing across
transboundary borders
Failure to Capture
Significant Rainfall
Application of satellite-based
rainfall estimates
•Objective 1: Assess the accuracy
of satellite-based rainfall estimates
and make intercomparisons
•Objective 2:Applying the satellitebased rainfall estimates for flood
prediction and integration of snow
and glacier into rainfall runoff
modelling
Satellite - based rainfall estimate:
NOAA CPC-RFE2.0

Initial version became operational in January 2001

Originally run over the African continent then expanded to
southern Asia and western Asia / eastern Europe

Product is a combination of surface and satellite precipitation
information

Spatial resolution: 0.1 degree

Temporal resolution: daily

Domain: 5o to 40oN, 60o to 110oE
Methodology for verification of
satellite-based rainfall estimates
Verification of Satellite Based Rainfall
Estimates
Visual
Verification:
(Qualitative)
Maps at
same scale
and colour
Continuous Verification
(Quantitative)
RMSE
Bias
Multiplicative Bias
Percentage error
Correlation Coefficient
Categorical Verification
(Qualitative)
False Alarm Ratio
Probability of Detection
Equitable Threat Score
Study area: Narayani basin
Basin characteristics:

Located in central Nepal

Catchment area = 32, 000 km2

Elevation variation 8167-100 m

More than 70% of rainfall
occurs in monsoon

High spatial and temporal
variation of precipitation
200 mm – 6000 mm
Study area and hydrometeorological
network
Input data
Spatial datasets
 Digital Elevation Model: Hydro 1k DEM
 Soil data (FAO)
 Landcover (USGS)
Dynamic datasets (daily time series)
 Satellite rainfall estimates: NOAA CPC_RFE2.0
rainfall estimates
Period: 2003 – 2004
 Gauge observed rainfall (DHM)
 Discharge data – period of record 1964-2006
(DHM)
Rainfall bias: Narayani
Gauge observed rainfall (mm)
Low: 0
High: 74
Low: 0
Contingency Table
Estimated
Rainfall
High: 129
CPC_RFE2.0 rainfall estimates (mm)
Observed Rainfall
37
2
1
1
Probability of Detection (POD)
0.97
False Alarm Ratio (FAR)
0.05
Mean Absolute Error (MAE)
36.1
Bias
-33.7
Root Mean Square Error (RMSE) 45.2
Correlation Coefficient
0.60
Percentage Error
-62.1
Intercomparison
Accumulated June, July, August and September rainfall for 2003
Comparison of annual rainfall
Annual Total Rainfall 2002 (OBS)
Annual Total Rainfall 2002 (SRE)
Annual Total Rainfall 2003 (OBS)
Annual Total Rainfall 2003 (SRE)
GeoSFM model overview
 GeoSFM simulates the dynamics of runoff processes by
using remotely sensed and widely available global
datasets
 Catchment scale modeling framework
 Semi-distributed hydrologic model
 Inputs aggregated to the catchment level
 GIS based Modeling
 ArcView 3.0 environment
GeoSFM model framework
Semi distributed
hydrologic model
Outputs
Hydrologic simulation
Inputs
Precipitation
PET
Terrain analysis
Basin characteristics
Basin response
Soil water balance
Streamflow generation
Discharge
Soil moisture


Spatial information database
DEM, Soil, Landcover

Semi-distributed, physically based
hydrologic model
Simulates runoff process using remotely
sensed data and global datasets
Graphical user interface within GIS for
model input and visualization (ArcView
version 3.x with the Spatial Analyst
Extension)
GeoSpatial streamflow modelling for
flood risk monitoring
GIS Preprocessing
Satellite
Rainfall
Estimates
Processing and
Analyzing
GDAS PET
Fields
FAO Soil
Data
3
80,000
60,000
40,000
20,000
month/day/year
Sim
Obs
11/23/2004
6/25/2004
1/26/2004
8/28/2003
3/30/2003
10/30/2002
6/1/2002
0
1/1/2002
DEM
100,000
Discharge (m /sec)
Land Use/
Land Cover
Geospatial Stream Flow Model (GeoSFM))
Water Balance
Routing
•1D
Sub-basin 1
•2D
Sub-basin 2
+
Main channel
Sub-basin 3
+
Sub-basin 4

Main channel
Subbasins are the modeling units
for water balance and routing
•Muskingum-Cunge
•Diffusion
+
Outlet
•Lag
GeoSFM rainfall-runoff component has three main
modules: water balance, catchment routing, and
distributed channel routing
Source: Guleid Artan
Model components
 Terrain analysis module
 Parameter estimation module
 Data preprocessing module
 Water balance module
 Flow routing module
 Post-processing module
GeoSFM modelling
 39 subbasins were considered
 Hydro 1K DEM - hydrologic
parameters, such as slope, aspect,
flow direction, and accumulation
were derived
 focused in particular on the months
of June, July, August, and
September (the monsoon season)
 gridded gauge observed rainfall
data for the monsoons of 2003 and
2004 were used in the GeoSFM to
predict floods
Model performance indicator
Nash Sutcliff Coefficient of Efficiency (NSCE)


2


2
 n
  Si  Oi
NSCE  1   i n1

Oi  O

 i 1





where Oi is observed discharge, Si is simulated discharge,
and
is the mean value of the observed discharge.
0.0
14000
20.0
12000
40.0
Discharge (m3/sec)
16000
10000
60.0
8000
80.0
6000
100.0
Rainfall (mm)
Model calibration and validation
4000
120.0
2000
Observed and simulated
streamflow at Devghat using
2003 monsoon gauge observed
rainfall (June to September)
NSCE = 0.84, Correlation =
0.94
140.0
0
152
167
182
197
212
227
242
257
272
Days
Gauge Rain
Observed Discharge
12000
Simulated Discharge (Gauge)
0.0
10.0
10000
8000
30.0
6000
Observed and simulated streamflow at
Devghat using 2004 monsoon gauge4000
observed rainfall (June to September)2000
NSCE =0.77, Correlation 0.94
40.0
50.0
60.0
70.0
0
80.0
152
162
172
182
192
202
212
222
232
242
252
262
Days
Gauge Rain
Observed Discharge
Simulated Discharge (Gauge)
272
Rainfall (mm)
Discharge (m3/sec)
20.0
Comparison of observed and
simulated with CPC_RFE
12000
Discharge (m3/sec)
10000
8000
6000
4000
2000
0
152
172
192
212
232
252
272
Days
Observed Discharge
Simulated Discharge (RFE)
Simulated Discharge (Gauge)
Comparison of observed and simulated daily flows at Devghat using
gauge observed rainfall and RFE data as input rainfall (June to
September 2003)
Why bias-correction?

Regional and country level satellite-based rainfall estimates have
indicated discrepancies between SRE and gauge observed rainfall

The uncertainty involved in Geo-SFM modeling that has been
observed using the SRE

Bring observed and predicted/estimated values as close to each
other as possible

The bias in precipitation was found to vary spatially in a given
domain/basin
Methods of bias-adjustment
 Ratio based or multiplicative bias-adjustment
 Ingestion of local rain gauges into the RFE
algorithm
 Anomaly – based bias adjustment
 Correcting the mean and coefficient of variation
 Many other methods
RFE and improved RFE: Narayani
basin
70.0
Rainfall (mm)
60.0
50.0
40.0
30.0
Basin averaged
RFE
20.0
10.0
272
264
256
248
240
232
224
216
208
200
192
184
176
168
160
152
0.0
Days
RFE
Observed Rain
70.0
50.0
40.0
Basin averaged
Improved RFE
30.0
20.0
10.0
Days
Improved RFE
Observed Rain
256
248
240
232
224
216
208
200
192
184
176
168
160
0.0
152
Rainfall (mm)
60.0
Improved RFE
RFE-the shape of precipitation is given by
the combination of satellite estimates,
magnitude is inferred from GTS station
data, need the maximum availability of the
rain gauge stations - Incorporate more
gauge data for improved rainfall estimates
Rainfall ingestion: Narayani basin
Statistical comparison of
performance of GeoSFM with
unadjusted and adjusted RFE
NSCE
r
Bias
RMSE
RFE (unadjusted)
-1.23
0.75
-2458
2750
0.81
0.19
0.58
RFE adjusted with a seasonal factor
0.27
0.8
-345
1561
0.49
0.51
0.87
RFE adjustment with a monthly factor
0.38
0.81
-378
1471
0.45
0.55
0.87
RFE adjustment with running 7 day average factor
0.22
0.79
106
1654
0.52
0.48
0.85
RFE (gauge-satellite merged)
0.53
0.90
-990
1316
0.57
0.43
0.89
Dataset
RMSEs RMSEu
d
Daily observed and simulated flows
using bias-adjusted CPC_RFE2.0
rainfall
16000
16000
Seasonal adjustment
14000
12000
12000
10000
10000
Discharge ( m3/sec)
Discharge (m3/sec)
14000
Monthly adjustment
8000
6000
4000
2000
8000
6000
4000
2000
0
0
152
162
172
182
192
202
212
222
232
242
252
262
152
272
162
172
182
192
Time (Days)
Observed
Observed
Simulated
232
242
252
262
272
232
242
252
262
272
Simulated
16000
7-Day moving average
14000
12000
Discharge (m3/sec)
Discharge (m3/sec)
16000
14000
202 212 222
Time (Days)
10000
8000
6000
4000
2000
Improved RFE
12000
10000
8000
6000
4000
2000
0
0
152
162
172
182
192
202
212
222
232
242
252
262
272
152
162
172
182
192
202
Time (Days)
Observed
Simulated
212
222
Time (Days)
Observed
Simulated
Shrestha, M.S., Artan, G.A.,Bajracharya, S.R., Gautam, D.K. and Tokar, S.A. (2011) Bias-adjusted
satellite-based rainfall estimates for predicting floods: Narayani Basin. J. Flood Risk Management
Summary

Using new technology and advanced scientific knowledge for
monitoring, assessing, forecasting and communicating information

Data formats are important as data preparation takes a lot of time and
energy

Good quality insitu data is essential for model calibration and validation

Improved understanding of flood forecasting methods and models

More accurate (quantitative) and high resolution data are necessary for
reasonable flood predictions. It is difficult to predict the floods quantitatively
using current satellite based data. We can only give an indication of
probability of occurrence

Intercomparison of satellite based rainfall estimates and models for
flood forecasting needs to be further explored.

Continued training and capacity building in state of the art technology
such as application of satellite-based precipitation in the region is necessary
to enhance flood risk management.
Thank you

similar documents