M. Thyer et al., Probabilistic optimization for conceptual rainfall-runoff models: A comparison of the shuffled complex evolution and simulated annealing algorithms, WATER RES R, 35(3), 1999, pp. 767-773
Automatic optimization algorithms are used routinely to calibrate conceptua
l rainfall-runoff (CRR) models. The goal of calibration is to estimate a fe
asible and unique (global) set of parameter estimates that best fit the obs
erved runoff data. Most if not all optimization algorithms have difficulty
in locating the global optimum because of response surfaces that contain mu
ltiple local optima with regions of attraction of differing size, discontin
uities, and long ridges and valleys. Extensive research has been undertaken
to develop efficient and robust global optimization algorithms over the la
st 10 years. This study compares the performance of two probabilistic globa
l optimization methods: the shuffled complex evolution algorithm SCE-UA, an
d the three-phase simulated annealing algorithm SA-SX. Both algorithms are
used to calibrate two parameter sets of a modified version of Boughton's [1
984] SFB model using data from two Australian catchments that have low and
high runoff yields. For the reduced, well-identified parameter set the algo
rithms have a similar efficiency for the low-yielding catchment, but SCE-UA
is almost twice as robust. Although the robustness of the algorithms is si
milar for the high-yielding catchment, SCE-UA is six times more efficient t
han SA-SX. When fitting the full parameter set the performance of SA-SX det
eriorated markedly for both catchments. These results indicated that SCE-UA
's use of multiple complexes and shuffling provided a more effective search
of the parameter space than SA-SX's single simplex with stochastic step ac
ceptance criterion, especially when the level of parameterization is increa
sed. Examination of the response surface for the low-yielding catchment rev
ealed some reasons why SCE-UA outperformed SA-SX and why probabilistic opti
mization algorithms can experience difficulty in locating the global optimu
m.