Study On Predicting Yield Loss In Winter Wheat After Late Frost Stress Using Narrow-broad Waveband Spectral Indices

Crop disaster caused by low temperature is a worldwide agrometeorological disaster.Late frost damage is one of subzero low temperature disasters usually happened after winter wheat entered key jointing stage in which young spikes are taking shape and sensitive to temperature.During this stage,frost damages young spikes firstly,harming their normal development and causing yield reduction.The yield reduction rate of winter wheat can better characterize the degree of frost damage of crops under low temperature.Hyperspectral technology provides a feasible method to identify the degree of frost damage.However,there are few reports on the evaluation and prediction of frost damage from the spatial scale based on satellite multispectral technology.Therefore,proposing early prediction of yield reduction rate of winter wheat after frost is of great practical significance for disaster monitoring and production management decision-making.In this study,Shangqiu City in Henan Province,was selected as the research area.Based on the ASD hyperspectral data of winter wheat canopy,potted plant yield reduction data,sentinel-2 image data and field winter wheat yield reduction data after frost,combined with spectral resampling,spectral index construction,linear regression and neural network modeling methods and technical means,the spectral characteristics of winter wheat canopy after frost were studied,and the spectral indexes responding to the degree of frost damage and the degree of yield loss were evaluated.Based on the candidate broad-band spectral indices,the prediction model of winter wheat yield reduction rate on field scale was established to evaluate the accuracy of broad-band spectral indices and models in predicting yield reduction rate under natural frost.The main conclusions are summarized as follows:(1)In winter wheat at the early and late stages of drug barrier differentiation,the standardized reflectance of the canopy in the near-infrared platform high reflectance area(760?1200 nm)has a tendency to decrease with temperature;in the short-wave infrared band(1300?2500 nm),the normalized reflectivity has a tendency to increase with decreasing temperature.The red-side band(690?780 nm)is the spectral region sensitive to yield reduction,and the sensitive wavelength is near 710 nm.(2)Among the existing spectral indexes calculated based on the ASD hyperspectral narrow-band data after late frost stress,the index MCARI has the best accuracy(correction set R²=0.695,verification set R²=0.635);the newly-built narrow-band spectral index R₉₆₀/R₈₃₆ has the best accuracy in predicting production reduction rate(correction set R²=0.660,validation set R²=0.554).Among the existing spectral indices calculated based on ASD hyperspectral data simulating Sentinel-2 wideband data,the linear regression model of index WBI and yield reduction has the best accuracy(correction set R²=0.707,RMSE=14.850,verification set R²=0.658,RMSE=15.647);B5-B4 has the best accuracy in the new index(correction set R²=0.730,RMSE=14.253,verification set R²=0.661,RMSE=15.545).The results show that it is feasible to screen the spectral index sensitive to winter wheat yield reduction based on Sentinel-2 simulated broadband data,and the best broad-band spectral index predicts the yield reduction accuracy better than the best narrow-band spectral index.(3)Candidate spectral indices calculated based on Sentinel-2 images after natural frost in 2018performed well overall,and 8 indices were significantly related to the ground yield reduction rate.The linear regression model for predicting winter wheat yield reduction rate is the new broadband spectral index B9/B8a with the best accuracy,the modeling set R²=0.445,RMSE=7.759;the verification set R²=0.337,RMSE=7.677;followed by the new spectral index B9/B8a,(B9-B8)/(B9+B8).Comparing the linear regression accuracy of the optimization index,the accuracy of the BP neural network nonlinear model is better than that of the linear regression model on the whole.The linear regression model of the best broadband spectral index B9/B8a predicts that the spatial distribution of winter wheat yield reduction rate is consistent with the ground-measured results and shows a consistent trend with the minimum grass surface temperature of meteorological observation stations.It can be formulated for different regions after the jointing of winter wheat.The best countermeasures against frost stress provide a reference.