Comparison Of The Models For Short-term Daily Reference Evapotranspiration Forecasting Based On The Public Weather Forecast

The agricultural irrigation water accounts for over 90%of the irrigation water use.Reasonable irrigation decision is an effective way for saving irrigation water,and accurate estimation of crop water requirement(ET)is the foundation of the irrigation scheduling.ET can be calculated by reference crop evapotranspiration(ET0),and accurate forecast of ET0 can not only reduces the risk of crop for suffering drought,but also avoids the waste of irrigation water.The ET0 is mainly influenced by meteorological conditions,and the public weather forecast message are free for the public and accurate enough,thus the near-future daily ET0 can be forecasted based on public weather forecast dataset.By now,privous studies on ET0 forecasting were mainly adopted one kind of model for a single(or more)region forecasting,or certain kinds of models in fewer sites,therefore,it still caused confusion about how to choose the proper model to forecast ET0 under different regions or climates.In this study,61 weather stations were selected for a wide range of China,which also be divided into 4 climate zones.Also,6 ET0 models were employed to forecast daily ET0 based on public weather forecast and the forecast performance by these models were assessed.Then,the performance of each model were cross compared in order to find out the best one for each climate and location.The main conclusions were as followed:(1)The relationship between the accuracy of weather variable and forecast lead time was detected,and results showed that the accuracy of each variable decreased with increasing lead time.Then,the precise of forecasted weather variables in decreasing order of minimum temperature(Tmin),maximum temperature(Tmax),weather type and wind scale.In addition to the Tmin,the reasons for the poor accuracy of Tmax,weather type and wind scale forecasts were further analyzed.The reason why the accuracy of Tmax forecast was higher than that of Tmin was that the the diurnal variation of Tmax is greater than Tmin,making the former more difficult to be accurately predicted.For weather type and wind scale forecasts,since both of them were qualitative data that cannot be directly substituted into model,they need to be converted into quantitative value,and therefore a large error occurred during the transformation process.In general,the accuracy of temperature forecast and weather type forecast met the requirement of ET0 calculation,thus they can be used for ET0 forecasting.However,the wind scale forecast accuracy was relatively lower with large error,so it is not applicable for ET0 forecasting;by comparing four types of common used wind speed data,the best accuracy was found by wind speed with constant value of 2 m/s,which was selected as the model input.(2)The forecasted ET0 provided by 6 models were all declined as the lead time increasing,the reason is that:the accuracy of temperature forecasts and weather type forecast decreased with the increasing lead time,thus the forecasted ET0 played the similar pattern as the variation change of input weather variables.This result also indicated that the ET0 forecast result was directly influenced by the precise of public weather forecast.(3)The forecast performance of each model were compared for each climate.Forecast performance of each model were compared for all climates.For Cwa,the PMF performed best followed by HS,TH,while BC,MC and RPM were not applicable for this climate.In HG climate,the first choice would be PMF,followed with HS and BC,while MC,RPM and TH would not be recommended.For Dwa climate,the PMF appeared as the most accurate one again and the second choice would be HS and BC,whereas TH,RPM and MC were not valid for this climate.For the Dfc,the greatest performance of models ranked as HS,PMF,BC,TH,RPM and MC.(4)As the sole combination model,the PMF provided the greatest average performance in Cwa,HG and Dwa climates,also,the second option for Dfc.Though weather type(SDun)forecast performance was worse than that of temperature,results strongly indicated that adding this factor into model computation improved the forecast performance and confirmed the importance of sunshine duration for ET0 forecasting.For the five temperature-based models,temperature forecast is the only input factor for them,thus the final performance of ET0 reflects the goodness of model structure.The HS performed better than the other temperature-based approaches for most climate regions,and the reason might be due to its robust model structure.Although the rest of four temperature-based models were local calibrated,the performance of them were still remained unsatisfactory for most climates.The BC model was the third choice for HG,Dwa and Dfc climates and can be an alternative to PMF and HS for the mentioned regions.In general,the TH,RPM and MC models provided poor average performance in all climatic regions and would not be recommended.Further,it's essential to local calibrated the temperature-based model for the sake of promoting their accuracy,whereas it's still difficult to obtain reliable prediction for some approaches,such as BC,TH,RPM and MC,whose model structure need to be further optimization.(5)As a whole,the model structure for both PMF and HS were better than the other models and they provided higher accuracy of ET0 forecasts under multiple climates,also,they were the first choice for most climate regions.The overall results could be a practical proposal for the selection of proper model for each climate region and provided some enlightenment about the applicability of each model under different climates.