APPLICATION OF THE WEATHER GENERATOR IN PROBABILISTIC
CROP YIELD FORECASTING
Martin Dubrovský (1), Zdeněk Žalud (2), Mirek Trnka (2), Jan Haberle (3)
Petr Pešice (1)
(1) Institute of Atmospheric Physics, Prague, Czech Republic
(2) Mendel University of Agriculture and Forestry, Brno, Czech Republic
(3) Research Institute of Crop Production, Prague, Czech Republic
project QC1316 of NAZV(National Agency for Agricultural Research, Czech Republic)
PERUN = system for crop model simulations under various meteorological conditions
• development of the software started last year
• tasks to be solved by PERUN: crop yield forecasting (main task) climate change/sensitivity impact analysis (task for future)
PERUN - components:
1) WOFOST crop model (v. 7.1.1.; provided by Alterra Wageningen)
new: Makkink formula for evapotranspiration was implemented
(motivation: Makkink does not need WIND and HUMIDITY data)
2) Met&Roll weather generator (Dubrovský, 1997)
some modifications had to be made (will be discussed later)
3) user interface- input for WOFOST (crop, soil and water, start/end of simulation,
production levels, fertilisers, ...)
- launching the process (weather generation, crop model simulation)
- statistical and graphical processing of the simulation output
seasonal crop yield forecasting1. construction of weather series
seasonal crop yield forecasting2. running the crop model
Modifications of previous version of the4-variate Met&Roll generator
(1) 4-variate 6-variate: To generate all six daily weather characteristics required by WOFOST (PREC, SRAD, TMAX, TMIN, VAP, WIND), the separate module adds values of VAP and WIND to the previously generated four weather characteristics (SRAD, TMAX, TMIN, PREC) using the nearest neighbours resampling from the observed data.
(2) The generator may produce series which consistently follow with the observed data at any day of the year.
(3) The second additional module allows to modify the synthetic weather series so that it fits the weather forecast
A) 4-variate 6-variate:
4-variate series:@DATE SRAD TMAX TMIN RAIN
...
99001 1.9 -2.7 -6.3 0.3
99002 2.1 -3.6 -3.7 0.7
99003 1.5 0.1 -1.3 2.4
99004 2.4 0.3 -2.7 0.6
99005 1.4 -1.4 -5.1 0.1
...
...
learning sample:@DATE SRAD TMAX TMIN RAIN VAPO WIND
...
xx001 1.6 1.3 -1.5 3.3 0.63 1.0
xx002 1.6 -0.8 -3.8 0.3 0.53 1.7
xx003 3.9 -2.3 -9.9 0.0 0.23 2.0
xx004 4.5 -2.3 -11.4 0.0 0.38 1.0
xx005 1.6 -6.1 -12.9 0.0 0.33 1.3
xx006 1.6 -1.8 -12.4 1.1 0.23 3.3
xx007 3.8 1.2 -2.3 0.0 0.52 4.7
xx008 1.7 -0.1 -4.3 0.0 0.39 1.3
xx009 1.7 -1.8 -6.7 0.4 0.42 4.0
xx010 1.7 -3.8 -8.0 1.0 0.36 2.0
xx011 1.7 0.0 -3.9 8.3 0.46 2.0
xx012 2.9 3.7 -0.3 2.8 0.57 1.7
xx013 1.8 2.6 -0.8 1.0 0.62 2.0
xx014 4.0 2.9 -3.3 0.0 0.45 2.7
xx015 4.0 2.4 -5.9 0.0 0.37 1.3
...
6-variate series:@DATE SRAD TMAX TMIN RAIN VAPO WIND
...
...
99001 1.9 -2.7 -6.3 0.3 0.34 3.099002 2.1 -3.6 -3.7 0.7 0.28 3.0
99003 1.5 0.1 -1.3 2.4 0.61 3.099004 2.4 0.3 -2.7 0.6 0.57 3.0
99005 1.4 -1.4 -5.1 0.1 0.47 3.0
B) series which consistently follow with the observed data
(1) generator working in normal-mode (1st-order generator assumed):
X(t) = f [X(t-1), e] [e = vector of random numbers]
X(0) ~ PDF(X)
(2) series which follows with the observed data at day D0:
(observed weather data are available till D0-1)
X(t) = f [X(t-1), e] for t>D0
X(D0) = f [XOBS(D0 -1), e]
B) series which consistently follow with the observed data
C) modification of the synthetic weather series so that it fits the weather forecast:
weather forecast is given in terms of- expected values valid for the forthcoming days (e.g., first week: 12±2 °C, second week: 7±3 °C, …)
alternative formats of the weather forecasts: (useful in climate change/sensitivity analysis)
- increments with respect to long-term means (first week: temperature = + 2 C above normal;
precipitation = 80% of normal;second week: ….., ….
- increments to existing series
crop yield forecasting at various days of the year
probabilistic forecast <avg±std> is based on 30 simulations
input weather data for each simulation =
[obs. weather till D−1] + [synt. weather since D; without forecast!]
(better to say: forecast = mean climatology)
a) the case of good fit between model and observationsite = Domanínek, Czech Rep.
crop = spring barley
year = 1999
emergency day = 122
maturity day = 225
observed yield = 4739 kg/ha
model yield = 4580 kg/ha
(simulated with
obs. weather series) enlarge >>>
crop yield forecasting at various days of the year
a) the case of good fit between model and observation
Crop yield forecasting at various days of the year
b) the case of poor fit between model and observation
site = Domanínek, Czech Republic
crop = spring barley
year = 1996
emergency day = 124
maturity day = 232
observed yield = 3956 kg/ha
model yield = 5739 kg/ha
(simulated with
obs. weather series)
enlarge >>>
crop yield forecasting at various days of the year
b) the case of poor fit between model and observation
crop yield forecasting at various days of the year
b) the case of poor fit between model and observation
task for future research: find indicators of the crop growth/development (measurable during the growing period) which could be used to correct the simulated characteristics, thereby allowing more precise crop yield forecast
indicators
PERUN - user’s interface
E N D
Thank you for your attention!
PERUN = system for crop model simulations under various meteorological conditions (development of the software started last year)
• tasks to be solved by PERUN: crop yield forecasting climate change/sensitivity impact analysis
• comprises all parts of the process: preparing input parameters for the crop model
(with a stress on the weather data) launching the crop model simulation statistical and graphical analysis of the crop
model output
input data to crop modela) non-weather data: info about crop, soil and hydrology;
start/end of simulation, nutrients, ... (as in WCC)
b) weather data:
- station
- day D0 (observed data till D0 −1, synthetic since D0;
- weather forecast table (weather series are postprocessed to fit the weather forecast)
- weather series = observed till D0−1, synthetic since D0
- formula for evapotranspiration: Makkink or Penman
c) N = number of re-runs (new weather data are generated for each simulation; other input data are always the same)
a) weather forecast is given in terms of the absolute values
* weather forecast random component
METHOD = 1
...averages... ..std. deviation..
@JD-from JD-to TMAX TMIN PREC TMAX TMIN PREC
99121 99130 17 6 30 2 2 10
99131 99140 14 4 60 3 3 20
99141 99150 21 10 10 4 4 10 @
b) weather forecast is given in terms of increments to existing series
* weather forecast random component
METHOD = 2
@
c) increments with respect to the long-term means
* weather forecast random component
METHOD = 3
...averages... ..std. deviation..
@JD-from JD-to TMAX TMIN PREC TMAX TMIN PREC
99121 99130 1 1 1.2 2 2 0.1
99131 99140 0 0 1.0 2 2 0.1
99141 99150 -1 -1 0.9 2 2 0.1
@