A. H. Gandomi, Tafresh University, Tafresh, IranDescriptions given in Section 2 and Figures 2, 3 and 4 of the paper studied by the authors clearly indicate that the method utilised for estimating the sediment rate is a tree-based genetic programming (TGP) approach. TGP was introduced by Koza (1992) as an extension of the genetic algorithms (GAs), in which programs are represented as tree structures and expressed in the functional programming language, LISP (Koza, 1992).
According to Section 5 of the paper, the software package Discipulus, which was developed by Conrads et al. (1998), was applied to the sediment rate forecasting problem.
Utilisation of Discipulus software indicates that the approach employed by the authors is machine-code-based, linear genetic programming (LGP) (Deschaine and Francone, 2002; Francone and Deschaine, 2004; Francone et al., 2005; Langdon and Banzhaf, 2005) not TGP. LGP (Brameier and Banzhaf, 2007) is a subset of GP that has recently emerged. Comparing LGP with the traditional Koza's TGP, there are some main differences. LGPs have graph-based functional structures and evolve in an imperative programming language (such as C/C++) (Brameier et al., 1998) and machine code (Nordin, 1994) rather than in expressions of a functional programming language such as LISP (see Figure 10). Unlike TGP, structurally non-effective codes coexist with effective codes in LGPs (Brameier and Banzhaf, 2007). Owing to the imperative program structure in LGP, the non-effective instructions can be identified efficiently. As only effective programs are executed, evaluation can be accelerated significantly. In addition, the machine-code-based, LGP approach searches for the computer program and the constants at the same time (Nordin, 1994). Further information on LGP can be found in Brameier and Banzhaf (2007).
In the paper discussed here, another important task on LGP application is not considered. A popular modularisation concept in LGP is the evolution of program teams (Brameier and Banzhaf, 2001). A team solution is formed by an uneven number of programs, of which every program has one vote. Whereas in general a team solution performs better than a single solution (Brameier and Banzhaf, 2007; Conrads et al., 1998; Gandomi et al., 2009), the prediction qualities of team solutions were not investigated by the authors.
Considering the above arguments, the scientific value of the paper would have increased had the authors investigated the detailed aspects of the utilised method.
AUTHORS' REPLY
Our objective in describing the GP that way was to introduce it as an extension of the familiar genetic algorithm in the most simple terms, although the code used might have applied GP in a different manner.
Considering the scope of our paper it is felt that subsequent investigators can take the lead from our work further and show how prediction quality could be improved by analysing team solutions, etc.
