| Crop growth models can simulate the processes of crop growth,development,and yield formation in response to environmental change,which provides an effective tool for crop yield forecast.In this study,we tried to establish a method for maize yield forecast based on the maize growth simulation model CERES-Maize and historical weather record.Two years' experiment data of three sites at Yangling(2014 and 2015),Heyang(2009 and 2011)and Changwu(2010 and 2011)in Shaanxi Province,were used to test the reliability and accuracy of the method established above.The weather data needed for model simulation were divided into two different sections as known and unknown weather data in the future during the whole growth season of maize.Recorded weather data were obtained from local weather stations,while unknown data were supplemented with historical weather data of multiple years in the local experimental sites.Multiple complete climatic data series were then created and used to run the CERES-Maize model to forecast maize yield for a given year.As the advancing of maize growth season,daily weather data were gradually merged into measured weather data in a target year.Consequently,daily maize yield could be forecasted from sowing to harvest.In addition,in order to reduce the number of model runs and the uncertainties in yield forecasts,this study compared the daily meteorological data(including maximum temperature 8)(6,?;minimum temperature 8)4)9),?;rainfall,mm;solar radiation SRAD,MJ 8)-2(9-1)of historical and target years with normal K-NN(K nearest neighbor)algorithm and a modified K-NN algorithm to select several historical analogue years whose weather properties were similar to the target year.Then the weather data of selected analogue years were used to supplement the unknown data in maize growth season to create complete weather series and forecast maize yield.Based on forecast yields with the method above,linear regressions were made for the median values of daily simulated yields two days after sowing.If the slope of linear regression curve kept decreasing for a giving number(e.g.10,20 or 30)days,which means the poor weather conditions of the target year drove yield mainly decreasing,so an irrigation was needed.Some main conclusions have been drawn based on the results as follows.(1)The distributions of predicted yields began to converge and the uncertainties decreased rapidly after tasseling stage(about 45-60 d before harvest),CV and MARE of daily forecasted yields are almost lower than 15%,so before harvest people can forecast the yield of target growth season with high accuracy.(2)Yield forecast accuracy was generally lower than expectation for the method based on climatic analogue years selected with normal K-NN method.However,in the modified K-NN algorithm,instead of daily values,seven-day average values of meteorological variables were compared between historical years and the target year to select the climatic analogue years.Then the weather data of selected analogue years were used for model running and yield forecast.Consequently,yield forecast accuracy was improved and time consumption was reduced due to the decrease of CERES-Maize model runs.(3)Based on the dynamic distribution of daily forecast yields and the meteorological factors,especially rainfall,in a given periods of the target year,a method for irrigation decision was established.This method was able to catch the point at which forecast yields had continuously decreased for a given number of days.Then an irrigation was triggered and a given depth of water was applied.(4)There are usually tens of years with measured weather data available for most weather stations in China.Only a limited number of scenarios of future climate could be generated through the combination of known and historical weather fata in a given growing season,which could not comprehensively and sufficiently represent the possible series of weather data that are in conformity with local meteorological properties. |
PDF Full Text Request