Embed Size (px)
Citation preview
Using Climate Models for Yield Predictions: Working with Uncertainties
Shrikant Jagtap
Global Climate Technology for Development
University of Florida
US Department of State
0
10
20
30
40
50
1952 1956 1960 1964 1968 1972 1975 1980 1984 1988 1992 1998 2002
Bags/a
cre
Traditional Forecast
Using Climate Models for Yield Predictions: Working with Uncertainties
0
20
40
60
80
100
0 2 4 6 8 10 12 14 16
Yield-98
Climatology
El Niño
Regional 98
Global 98
Using Climate Models for Yield Predictions: Working with Uncertainties
(a)
0
50
100
150
200
250
300
350R
ain
fall- m
m/m
onth
Station
FSU-GCM
FSU-RCM
(b)
0
5
10
15
20
25
30
Wetd
ays- days/m
onth
Station
FSU-GCM
FSU-RCM
(c)
0
5
10
15
20
25
Mar Apr May Jun Jul Aug
Radia
tion- M
J/m
2/d
ay
Station
FSU-GCM
FSU-RCM
(d)
0
5
10
15
20
25
30
35
Mar Apr May Jun Jul Aug
Maxim
um
Tem
p- Deg C
/day
Station
FSU-GCM
FSU-RCM
Using Climate Models for Yield Predictions: Working with Uncertainties
From USER’s Perspectives
1. While it is clearly desirable to improve a forecast model by resolving model shortcomings, we may simply not have the data or understanding to pin down crucial uncertainties in the models.
2. Model may have two types of error: random errors due to the cumulative impact of unknown processes on the known processes, and systematic errors due either to parameters not being adequately constrained by available observations or to the structure of the model being incapable of representing the phenomena of interest.
3. Therefore, a systematic treatment of model error is essential for forecasts to be useful in decision making. The net result of minimizing errors is that
usable forecasts can be made with existing models.
Using Climate Models for Yield Predictions: Working with Uncertainties
Study RegionSE USA
Length of Study1987-1999 Hind-Cast
2000-2006 Forecast
Forecast ProductsMonthly March-August
Weather Generator
Crop management
Irrigation management
Crop yields
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1987 1989 1991 1993 1995 1997 1999Years
Co
rrela
tio
n o
ver
a s
easo
n
600
690
780
870
960
1050
1140
1230
1320
1410
1500
Seaso
nal ra
infa
ll (
mm
)
FSU-RSM MBC k-NN Seasonal rainfall total (mm)
Temporal linear correlation skill comparison among the FSU-RSM, the mean bias correction (MBC), and the k-nearest neighbor method at Tifton, Georgia with the station observed March through August monthly rainfall.
Variable, Monthly Rsq- Raw Rsq- Corrected
Rainfall ~0 0.32
Maximum Temp 0.86 0.98
Minimum Temp 0.84 0.98
Radiation 0.42 0.95
10-Ensembles
30
60
90
120
150
180
1988 1991 1993 1995 1997 1999
On
se
t o
f P
lan
tin
g (
DO
Y)
10-Ensembles
0
2
4
6
8
10
12
14
1988 1991 1993 1995 1997 1999
Ma
ize
yie
ld, T
/ha
Climatology Recorded Forecast
Validation 0-6
~ 154,000 Sq Km
Forecast 2008 2009 2010 2011
Onset ? 66% 79% 81%
End-of-Rain ? 85% 83% 93%
Rainfall ? 76% 68% 95%
Seasonal Rainfall Prediction – Nigeria case study
Using Climate Models for Yield Predictions: Working with Uncertainties
Shrikant Jagtap
1980-2012: 60+ Countries
I’ll be working in India here after
+91 773 8899 302
