Estimating Evaporation Using ANFIS

Abstract

Water resources engineering assessment requires simple but effective evaporation estimation procedures, especially from readily measurable meteorological factors. Unfortunately, such approaches are rather scarce in the literature. In this paper, an adaptive neural-based fuzzy inference system (ANFIS) was applied to daily meteorology data from the Lake Eğirdir region in the southwestern part of Turkey. Daily evaporation, solar radiation, air and water temperatures, and relative humidity measurements were used to develop the ANFIS method, which helps to assess possible contributions that each input variable has on the evaporation estimates. Such an assessment is not possible by any conventional procedure including the Penman method. However, the Penman method daily evaporation estimations were used as output data for the verification of the ANFIS approach. Classical evaporation estimation models treat the data individually. However, ANFIS models process past data collectively and then adaptively provide estimates as new sets of data become available. In the ANFIS architecture as developed in this paper, there are four measured input variables and one output variable to estimate evaporation. The estimation results from the ANFIS model had a high coefficient of determination of 0.98 when compared with the Penman method results and a low average performance error of 4.6% among other alternatives. The average performance error is less than the practically acceptable limit of 10%.