Free surface evaporation is an important process in regional water cycles and energy balance. The accurate calculation of free surface evaporation is of great significance for evaluating and managing water resources. In order to improve the accuracy of estimating reservoir evaporation in data-scarce arid regions, the applicability of the energy balance method was assessed to calculate water surface evaporation based on the evaporator and reservoir evaporation experiment. A correlation analysis was used to assess the major meteorological factors that affect water surface temperature to obtain the critical parameters of the machine learning models. The water surface temperature was simulated using five machine learning algorithms, and the accuracy of results was evaluated using the root mean square error (RMSE), correlation coefficient (r), mean absolute error (MAE), and Nash efficiency coefficient (NSE) between observed value and calculated value. The results showed that the correlation coefficient between the evaporation capacity of the evaporator, calculated using the energy balance method and the observed evaporation capacity, was 0.946, and the RMSE was 0.279. The r value between the calculated value of the reservoir evaporation capacity and the observed value was 0.889, and the RMSE was 0.241. The meteorological factors related to the change in water surface temperature were air temperature, air pressure, relative humidity, net radiation and wind speed. The correlation coefficients were 0.554, −0.548, −0.315, −0.227, and 0.141, respectively. The RMSE and MAE values of five models were: RF (0.464 and 0.336), LSSVM (0.468 and 0.340), LSTM (1.567 and 1.186), GA-BP (0.709 and 0.558), and CNN (1.113 and 0.962). In summary, the energy balance method could accurately calculate the evaporation of evaporators and reservoirs in hyper-arid areas. As an important calculation parameter, the water surface temperature is most affected by air temperature, and the RF algorithm was superior to the other algorithms in predicting water surface temperature, and it could be used to predict the missing data. The energy balance model and random forest algorithm can be used to accurately calculate and predict the evaporation from reservoirs in hyper-arid areas, so as to make the rational allocation of reservoir water resources.

Agricultural intensification has boosted land productivity, but it has also created new sustainability issues. As one of the most important human habitations and agricultural farming areas in arid areas, the Cele Oasis has a very developed agricultural system. This paper studies the long-term effects of different types of agricultural intensification strategies on groundwater level fluctuations in the Cele Oasis. A soil water flow (HYDRUS-1D) and aquifer simulation (MODFLOW) coupling model were used to construct the geometric structures of the vadose zone and saturated zone in the Cele Oasis and to analyze the recharge and discharge mechanism of the oasis. The results showed that HYDRUS-1D accurately simulated soil moisture transport in the Cele Oasis, providing reliable data for calibration of the MODFLOW model. The groundwater level simulated by MODFLOW was in good agreement with the observed value. The results of the R², RMSE, and NSE were ranges of 0.77–0.90, 0.45–0.74 m, and 0.76–0.87, respectively. The errors were acceptable limits. The coupling model predicted the responses of different agricultural types and cropping scenarios to groundwater. Predictions showed that the groundwater level in the Cele Oasis remained stable under the current cropping scenario (100% cropping intensity), and that the groundwater level decreased slightly under the cropping scenario (110% cropping intensity and 130% cropping intensity). When the cropping scenario was at 170% cropping intensity, the groundwater level decreased rapidly, and the maximum drawdown value was 7 m. Therefore, the maximum cropping intensity of the Cele Oasis in the future should be 130% of the current cropping intensity.