Research Of Forage Yield And Nutrient Content In Herdsman Pasture Based On Neural Network And Near Infrared Spectroscopy

In this paper,BP-ANN model and NIRS model were established based on the field measurement data,high-resolution remote sensing data and chemical method measurement.The two models were used to rapidly estimate the grass yield and nutrient output of an herding pasture in Qinghai Lake township,Haiyan County,and the accuracy and applicability of the model were evaluated.In order to improve the accuracy of estimation of annual grass yield of pastureland,estimate the yield and nutrient output of pastureland in cold and warm seasons more accurately,and show its monthly variation visually,evaluate the stocking and grazing situation of pastureland in cold and warm seasons with double indexes,and give specific suggestions on monthly supplementary feeding and rotational grazing.The results are as follows:1.NIRS quantitative model of 5 nutrients was established based on 2435 samples of dry grass in the cold season and 2765 samples of green grass in the warm season,in the area around Qinghai Lake.The prediction fitting degree of DM,CP and Ca contents of a.subtilis was good,among which,the DM model was the best(RSQ=0.960),the content of phosphorus was expected to be improved(RSQ=0.840),and the EE model was not available(RSQ=0.526).The DM,CP,Ca and P content prediction models of green herbage were verified,among which the CP model was the best(RSQ=0.971)and the EE model was not verified(RSQ=0.472).2.The BP neural network was trained with 9 grassland subtypes and 2 plantation subjects as the sample data of 2844 grass yield,97 landscape high score data and grassland type data in the area around qinghai lake.After testing,the grass yield estimation model R =0.743,RMSE=58.531g/m.It is proved that the reliability and high fitting degree of the yield estimation model can satisfy the yield estimation of both cold and warm seasons.3.The BP-ANN model was used to estimate the grass yield of specific pasture in Qinghaihu rural area,and the measured values were compared.R and RMSE were 0.766 and 43.654g/m~2(P < 0.01),confirming the applicability of the model in alpine meadow pasture.The total forage yield in the cold season was only 43.36% that in the warm season,and the maximum difference between months was about three times.Unit output reached a minimum of 40.46g/m~2 in May,and a peak of 370.41 g/m~2 in September,with a maximum monthly difference of 9.15 times.The overall unit output in the warm season was 2.48 times that in the cold season.4.NIRS model was used to determine the content of DM,CP,P and Ca of herbage in cold season and warm season respectively,and the chemical values were used to correct and adjust the model.When analyzing the content of Ca,P,CP and DM of various kinds of herbage in the alpine meadow,the fitting degree was relatively high.The results showed that the herbage quality in this herding area was lower than that in the warm season from June to August.The whole Ca/P of year-round forage is far above the standard of healthy feeding.The outputs of DM,CP,P and Ca in the cold season were only 80%,43%,33% and 66% of the units in the warm season,respectively.The seasonal differences in monthly DM,CP,Ca and P outputs were significant,with differences between the extremes of 5.22 times,6.69 times,5.89 times and 13.1 times,respectively.5.With the predicted values of the above two groups of models,the quantity stocking value and DCP stocking value were taken as the bottom line values of warm season and cold season respectively to calculate the stocking situation.According to the real overloading rate,the consistency of overload prediction reached 87.5%,which has practical reference significance.Only in August and September 2018 and February 2019 did not break the critical overload value,while the other months all showed the overload of grassland in different areas.