Martin Dubrovsky
Institute of Atmospheric Physics, Hradec Kralove, Czech Republic
*** presented at 78th Annual Meeting of the American Meteorological Society, Phoenix, 11-16 January 1998 ***
To study impacts of climate variations on crop production, the crop growth models are employed (Mearns et al., 1997). These models require series of daily weather characteristics to simulate crop growth in daily steps. The weather series representing current and changed climate conditions may be synthesised by weather generator (WG) whose parameters are derived from observed series and modified in accordance with climate change scenario.
The contribution reviews results of experiments made with weather generator Met&Roll and crop growth model CERES-Maize (Dubrovsky et al., 1998).
Met&Roll (Dubrovsky, 1997) is a 4-variate (daily sum of global solar radiation SRAD, daily precipitation amount RAIN, extreme daily temperatures TMAX and TMIN) WG designed for use with crop growth model CERES. Precipitation occurrence is modelled by the first-order Markov chain, precipitation amount by Gamma distribution and radiation and extreme temperatures by trivariate first-order autoregressive (AR) model (Wilks, 1992). The set of parameters of Met&Roll consists of:
(a) means and standard deviations of SRAD, TMAX and TMIN, defined separately for wet and dry days (annual cycle is expressed in daily steps),
(b) two parameters of Gamma distribution and two parameters of Markov chain model (annual cycle is expressed in monthly steps),
(c) two 3x3 matrices of AR model (annual cycle is not considered).
The stochastic structure of synthetic series generated by WG should be the same as the structure of observed series. The validation tests have, however, revealed some discrepancies:
(i) the frequency of occurrence of long dry spells, extreme values of daily precipitation amount and variability of monthly means are underestimated by MET&ROLL;
(ii) solar radiation and extreme temperatures do not follow normal distribution assumed by AR model;
(iii) correlations and lag-1 correlations of weather characteristics have significant annual cycle not assumed by AR model.
(iv) The model generally performs better in summer months.
The question stands how the imperfections in reproducing the stochastic structure of weather series affect the model yield characteristics. To give an answer, the distributions of grain yields simulated with use of observed vs. synthetic (generated by Met&Roll) weather series for 16 Czech locations were compared. No statistically significant difference was detected and it is thus accepted that the synthetic weather series generated by MET&ROLL are applicable to crop growth simulations (at least in Central European climate conditions).
The problem arises how to project climate change scenario into parameters of WG. For example, the scenario may prescribe to increase monthly precipitation by 10%. However, it does not specify if the change is due to increasing number of wet days, or increasing mean daily precipitation amount, or both. To gain a notion on possible errors resulting from ambiguities in projecting climate change scenario into parameters of WG, the sensitivity of model grain yields to selected characteristics of the weather series was examined. The set of characteristics considered includes those which are not explicitly treated by typical climate change scenario but they are controllable by WG. Selected results are displayed in Fig 1.
(i) The weather generator fails to reproduce some features of stochastic structure of daily weather series.
(ii) However, the differences between distributions of model grain yields simulated with synthetic and observed daily weather series were found statistically insignificant. This indicates that the imperfections involved in generator's model need not have a pronounced effect on reliability of yields simulated by the crop growth model.
(iii) The sensitivity analysis (Fig. ) indicates that the characteristics not included in typical climate change scenario may affect the yields. For example, increasing persistence of the series (scenarios D and I) or increasing variability of extreme temperatures and solar radiation (C) decrease model grain yields.
This study was supported by the Grant Agency of the Czech Republic under project 205/96/1669.
Figure 1. Variability of model grain yields
simulated with modified characteristics of daily weather series.
The horizontal bars demarcate quantiles of sets of the yields
from the 99-year crop growth simulation experiments. The numbers
on the right to each bar are values of the standardised Wilcoxon
statistic for testing the hypothesis that the distribution of the
grain yields does not differ from the reference "no-change"
distribution related to unmodified parameters of WG. The values
of the statistic beyond <-1.96, 1.96> indicate
statistically significant (level of significance = 5%) difference
of the two distributions.
A) temperature means are additively modified by dT = -1, +1, +2 C, daily temperature amplitudes are preserved (means of TMAX and TMIN are changed by the same increment);
B) daily temperature amplitudes are changed, daily temperature means are preserved;
C) standard deviations of TMAX, TMIN and SRAD are multiplicatively modified (the means are preserved);
D) persistence of daily series of SRAD, TMAX and TMIN is modified ("no persist": lag-1 correlations are set to zero, "high persist": lag-1 correlations are increased to about 90% of lag-0 correlations);
E) monthly sums of precipitation are changed by multiplying Gamma distribution parameters (mean daily precipitation amounts are changed, percentages of wet days are preserved);
F) monthly sums of precipitation are changed by modifying probabilities of wet day occurrence (monthly means of temperature and solar radiation are preserved);
G) shape of Gamma distribution is changed (mean daily precipitation sums are preserved);
H) daily precipitation sums and percentages of wet days are changed (monthly sums and means of RAIN, TMAX, TMIN and SRAD are preserved);
I) persistence of the wet day occurrence series is changed by modifying transition probabilities of the Markov chain model.