Reply to the Discussion of “Evapotranspiration modelling using support vector machines” by R. J. Abrahart et al.

The authors would like to express their gratitude for the interest shown by the discussers (Abrahart et al., 2010) in the paper and for their comments on the subject. We have tried to clarify all the points raised by them in this closure.

In Kisi & Cimen (2009), we gave a general description of the data sets that were used in the Case Study section of our study. We used 10 years of data (from 10 January 1998 to 10 January 2007) for three automated weather stations located at Windsor, Oakville and Santa Rosa in California, because the data of some of these stations had too many missing values before 10 January 1998 (more than 100 missing values in a year).

We did not feel an indication of data reliability was necessary in our study because everybody can easily access these data from the CIMIS web site and can see the data reliability there. These data (CIMIS data) have also been used by other researchers before without indicating data reliability (e.g. Kumar et al., 2002, 2008).

The data in Kisi & Cimen (2009) were downloaded from the CIMIS web server. There may have been a copy–paste error in the presentation of the CIMIS web-site address. The correct address is provided on page 922.

Units of data were overlooked in the tables. The unit for each weather data can be seen within the text in Kisi & Cimen (2009) (for example, see the text after equation (9)). The same units were used for the support vector regression (SVR), artificial neural networks (ANN) and FAO-56 PM methods.

It was clearly stated in the Abstract, Introduction, Application and Results, and Concluding Remarks sections that the ET₀ values calculated using the FAO-56 PM method were those used for the derivation of the SVR and ANN models. We think that this is sufficient clarification. There is no need to give any identifier e.g. F-ET₀. The authors followed the same manner as that used in the related literature (Trajkovic et al., 2003; Trajkovic, 2005; Jain et al., 2008; Kumar et al., 2008; Kim & Kim, 2008).

Kisi & Cimen (2009) obtained daily mean temperature (relative humidity) records by calculating the average of the daily maximum and minimum temperature (relative humidity) records. A linear variation between the daily minimum and maximum temperatures (relative humidity) was assumed in the study. This is the simplest way to obtain daily mean values as surmised by the discussers. Therefore, the authors did not feel an indication of this pre-processing operation necessary in their study.

As previously stated, Kisi & Cimen (2009) downloaded the data from the CIMIS web server. They removed the dates including any missing data from the whole data set. The statistical properties of each data set given in Table 1 (Kisi & Cimen, 2009, p. 921) were calculated using MS Office Excel 2003. The main reason for the difference between the statistics given in Table 1 of Kisi & Cimen (2009) and the discussers’ statistics might have been due to the fact that the authors discarded the whole data set including missing values. For example, if there was a missing relative humidity value in a data set (the date is the same for a data set), the wind speed, temperature and solar radiation values were also discarded from the whole data set. The discussers may have calculated the statistics given in Tables 3–5 considering each variable separately. In addition, the ET₀ statistics given in Table 1 of Kisi & Cimen (2009) were calculated by FAO-56 PM method (not CIMIS Penman), because it is a sole standard method, as reported in Kisi & Cimen (2009). The discussers may have taken these values as CIMIS Penman ET₀ statistics.

The authors believe that the CIMIS Penman is a good method. It can be seen from Table 2 in Kisi & Cimen (2009) that the CIMIS Penman generally performed better than the other empirical models which were employed in the study. Vaughan et al. (2007) compared daily CIMIS ET₀ predictions with daily lysimeter values and found that CIMIS predictions were in good agreement with daily lysimeter data.