Research On Optimization And Evaluation Of Simulation Methods For Greenhouse Cucumber Evapotranspiration In Southern Jiangsu

Accurate estimation of crop evapotranspiration（ET_c）is the basic scientific study of water-saving theory,which is of great significance for guiding water-saving irrigation decision-making,deciding crop drought degree and improving irrigation utilization efficiency.Due to the difference of micrometeorological environment between greenhouse and open field,the transpiration mechanism of greenhouse crops was different from that of open field environment.Penman-Monteith（P-M）model commonly used on open field crops and its key parameters should be improved based on the micrometeorological environment of greenhouse and crop growth structure before its application in greenhouse.With the development of the current temperature measuring technology,it is of high practical value to estimate greenhouse crop ET_c based on T_c-T_a.Meanwhile,due to the demand of intelligent and automatic development of greenhouse agriculture,the application of artificial neural network model on the estimation of greenhouse crops ET_c will become a new trend of water-saving irrigation decision-making in greenhouse.Therefore,the simulation of canopy resistance parameters（r_c）in the P-M model was optimized in this study.The response mechanism of r_s and various environmental factors was systematically analyzed by observing water consumption rule,leaf stomatal resistance（r_s）,micrometeorological factors and plant growth status of cucumber in greenhouse.The factors influencing r_s to r_c scale conversion were quantified,and the improved Jarvis canopy resistance model was constructed and verified.At the same time,this study discusses the application value of T_c-T_a in greenhouse irrigation decision-making:By comprehensively analyzing the change rule of T_c-T_a and the main influencing factors,a greenhouse crop water stress index（CWSI）model based on T_c-T_a was established,and the influences of CWSI and soil water content（SMC）,as quantitative indexes of greenhouse irrigation time,on greenhouse irrigation decision-making were compared.The response relationship between T_c-T_a and crop ET_c was clarified.In addition,based on modern information technology,an artificial neural network BP model was constructed.The Penman-Monteith model（P-M-J）based on Jarvis canopy resistance model,the improved P-M-J model,the T_c-T_a based Jackson empirical model and the artificial neural network BP model were respectively determined.The simulation accuracy of different models for cucumber ET_c in greenhouse was verified based on the measured values of the lysimeter,and the applicability of the models in greenhouse was evaluated.The main research conclusions were as follows:（1）The main factors which affected the simulation accuracy of Jarvis model on crop canopy resistance r_c were the differences of leaf size,spatial structure distribution and photosynthetic rate.The effective leaf area index function（LAI_e）in Jarvis model was optimized by multi-layer analysis method.The simulation of r_c in the optimized Jarvis model was better than that in the original model.The correlation coefficient between the former model and the measured value was 0.68,which was higher than that of the latter（R²=0.64）,and the root mean square error was 1.63 s cm^-1.Lower than the latter(RMSE=1.75 s cm^-1).Compared with the original Jarvis model,the LAI_e in the improved Jarvis model better reflected the actual physiological growth conditions of cucumber,and the improved Jarvis model had better simulation results for cucumber r_c and higher applicability in greenhouse.（2）The diurnal variations of T_c-T_a of greenhouse cucumbers were obvious,and the T_c-T_a of greenhouse cucumbers at different cucumber growing stages showed unimodal variation,and the peak value of T_c-T_a appeared at 13:00～14:00.The T_c-T_a of greenhouse cucumbers at 14:00 had the highest correlation with ET_c.T_c-T_a was negatively correlated with R_n,T_a and VPD,and positively correlated with RH at different growth stages.The correlations between T_c-T_a and meteorological factors were ranked as T_a>R_n>VPD>RH.Water stress index（CWSI）was negatively correlated with T_c-T_a and RH,and positively correlated with R_n and VPD.The correlation order was as follows:VPD>R_n>T_c-T_a>RH,CWSI was used as a quantitative index of greenhouse irrigation time,which can accurately reflect the actual water demand of crops with the change of meteorological factors and physiological conditions of crop growth.（3）Compared with the original P-M-J model,the simulated values of greenhouse cucumber ET_c by the improved P-M-J model were closer to the measured values(R²=0.88,MAE=0.08mm h^-1,RMSE=0.10 mm h^-1),and the deviation was smaller.With the increase of LAI for cucumber,the ET_c simulation accuracy of the original P-M-J model and the improved model was improved.The optimal period of ET_c estimation performance for both models was 14:00～16:00,and the worst period was 10:00～14:00.The Jackson empirical model based on T_c-T_a fitted the simulation results of ET_c well with the measured values(R²=0.97,MAE=0.59mm d^-1,RMSE=0.74 mm d^-1),and the coefficient of determination was slightly higher than that of the original P-M-J model(R²=0.96,MAE=1.21 mm d^-1,RMSE=1.51 mm d^-1),the estimation deviation was smaller.The simulation results of cucumber ET_c using artificial neural network BP model with different combinations of meteorological parameters were all better(R²>0.96,MAE<0.80 mm d^-1,RMSE<0.70 mm d^-1),and the fitting degree between the estimated and measured values of the model ET_c were higher than that of the original P-M-J model.Moreover,the accuracy of BP model was improved with the increase of input parameters.Based on the results of this study,the improved P-M-J model was based on the principle of energy balance,and the model was more widely applied to different types of greenhouse and crop species,with higher stability.With fewer types of meteorological data,the artificial neural network model still had a good performance,but this kind of model cannot reflect the physical and physiological processes of crop evapotranspiration,and requires a large amount of data,so it is more suitable to simulate the ET_c of planting crops under stable micrometeorological conditions.