Assessment of the sensitivity and prediction uncertainty of evaporation models applied to Nasser Lake, Egypt

Summary This paper examines the sensitivity and uncertainty of evaporation estimates from Nasser Lake, south of Egypt using data collected at the floating station Raft, 2 km upstream of the Aswan High Dam, and 12 evaporation models. Based on their approach the evaporation models were ranked in five classes, respectively the Bowen ratio energy budget (BREB) approach, three energy–aerodynamic based methods, a mass transfer method, two temperature and five solar radiation–temperature approaches. State dependent parameter models were used to estimate the sensitivity indices of the different evaporation estimation methods at daily and monthly frequencies. A drastic reduction in the cost of the sensitivity analysis was obtained when coupled with efficient sampling methods, producing quasi-Monte Carlo sequences (Latin Hypercube samples). Analysis of the sensitivity indices ( S i ) revealed that the energy and advective parameters ( Q x and Q v ) in the Bowen ratio energy budget method most strongly affect evaporation estimates. The most sensitive parameters for the energy–aerodynamic methods (Penman, Priestley–Taylor and the deBruin–Keijman methods) are the change in heat storage ( Q x ) and solar radiation. For the other methods the most sensitive parameters were solar radiation, air temperature and vapour pressure, but with lower sensitivities. After investigation of the first and second-order sensitivity indices, it was found that the first-order sensitivity indices capture more than 90% of the total variance of the evaporation estimates. Interactions between input parameters were found limited with a maximum contribution to variance of less than 3.3% for daily frequency and less than 8.5% for the monthly frequency. Results indicated that many parameters in the 12 evaporation models could be fixed without appreciably affecting the output of these models. This helped to reduce the computational load of these models. The uncertainty results show that the Dalton group is subject to the highest error. The overall uncertainty ranged from 9% (BREB method) to 15% (mass transfer method) of the mean evaporation rate.