Localization And Calibration Of Model Parameters Based On PCSE Crop Model And Sensitivity Analysis

As a big country with a long history of agriculture,China's agricultural development is an important guarantee for the prosperity of the motherland.The stable production of crops is related to the stable maintenance of the domestic society,the strategic control of the military,and the development of new energy science and technology.With the advancement of science and technology,the application of computer technology has gradually spread to the agricultural growth simulation system,and the simulation research of crop growth process has become a research hot spot of current technology application.PCSE is a commonly used simulation model of plant growth process,which has a large number of parameter configurations,involving many biological and physical evolution processes of the Earth's ecology.There are four parameters for PCSE,namely climate,soil,crop and management parameters.Among them,crop parameters are particularly important.Due to the uncertainty of internal structure,the diversity of parameter extraction,and the uncertainty of driving factors,the growth model accumulates error during daily growth prediction.The final prediction results in a large error.Therefore,how to establish a parameter adjustment strategy has a significant impact on the error reduction of crop growth model prediction.Therefore,the adjustment strategy of crop parameters is the key direction of current crop growth simulation and crop model application.This paper makes a detailed study on crop parameters in the PCSE growth model.Firstly,the regional range of each crop parameter was determined,and the EX-SMOTE method was used to generate crop parameter samples in the whole parameter sample space.The effective sample parameters of 5000 groups in each parameter range were obtained.Then,using EFAST and GBDT to verify thesensitivity of 5000 sets of sample parameters in different climates,the four sensitive parameters were selected by the analysis results,the average sensitivity index was0.216.SPAN(leaf life of the leaf area at 35?),SLATB078 with an average sensitivity index of 0.096(specific leaf area at a growth period of 0.78),TBASE with an average sensitivity index of 0.288(low temperature threshold for leaf age),The average sensitivity index is 0.08 CVO(the assimilation efficiency of the storage organ).Secondly,combined with the variance of historical data and simulated data,the average sensitivity value calculated in the second step,and the calculated variance value of the four-year simulated value and the true value,the minimum value is selected as the initial iteration point for the sensitive gradient descent iterative calculation.,obtain the PCSE growth model localization optimal parameter combination.Finally,the growth model was used to simulate the growth process of maize in Changchun City,and different experimental methods were used to predict the predicted values.The experimental results are: the average error value without calibration is 1609.3 kg/ha;the average error with the baseline type value as the initial calibration value is 1353.6 kg/ha;the sample selected by the principle of minimum mean square error under different climatic conditions The average error of the point as the initial calibration point is reduced to 857.4 kg/ha.In this paper,the experiment proves that this error can be reasonably localized and calibrated through the test procedure,so that the valid parameter set after calibration can be obtained,and the regional crop yield prediction error can be narrowed.