Research On Hyperspectral Characteristics,Quantitative Diagnostic Models,and Topdressing Application For Winter Oilseed Rape Nitrogen Status

Real-time and precision nitrogen?N?management is one of the most promising methodology to match N supply and with crop N demand in both space and time in winter oilseed rape?Brassica napus L.?production.Therefore,accurate and rapid monitoring of oilseed rape N status with the hyperspectral remote sensing technique is always the hot topic in the field of sustainable agriculture and site-specific N management strategy.Then,based on this,winter oilseed rape field experiments were conducted over various consecutive growing seasons?2012-2017?at different sites?Wuhan,Wuxue and Shayang?in Hubei province,China,using different N application rates,planting patterns,cultivars,growing stages and ecological sites.In situ canopy or leaf hyperspectral reflectance data were acquired over a wavelength range from350 to 2500 nm?the visible and middle-infrared region?,and quantitative correlatives between N status and their hyperspectra were determined from five aspects in following logic order:canopy scale?group?,leaf scale?individual?,diagnostic mechanism of N nutrition,diagnostic specificity of N nutrition,and N fertilizer recommendation model based on canopy hyperspectra.The main conclusions from the study were as followed:?1?On the canopy?group?scale.Firstly,the optimal spectral parameters of winter oilseed rape plant N accumulation?PNA?and chlorophyll density?ChD?were determined;and then,a quantitative technique was developed to estimate oilseed rape yield accurately during pod-filling stage;moreover,the monitoring models were built depending on ground-based canopy reflectance spectra for the above N nutrition indicators prediction.For the PNA,the model in which the optimum reflectance ratios?R1259/R492?and first derivative of the reflectance spectra at the given wavelength at742 nm?FD742?as variables would be perfect of estimating PNA of winter oilseed rape using hyperspectral techniques.The model estimation accuracy was high,the r²values were 0.98 and 0.98,respectively,and the mean relative error?MRE?values were 14.42 and 10.31%,respectively.For the ChD,using a training dataset,the best results for assessing ChD status were observed when using the newly-developed red-edge area parameter,which indicated a difference between the double-peak areas based on the position of the main peak?DIDRmid?.Tests conducted on the independent validation dataset showed that DIDRmid can be used to accurately predict ChD in oilseed rape,with a MRE of 7.93,6.06,7.52 and 7.22%at seedling,budding,flowering and the whole growth stages,respectively.For the yield and its components,the best results were obtained with the first-derivative reflectance-partial least square?FDR-PLS?model for the prediction of yield and pod number,with values of r²_valof 0.90 and 0.91,and ratio prediction to deviation(RPD_val)of 3.11 and3.12 using the independent validation dataset,respectively.The variable importance in projection?VIP?scores resulted from the PLS regression analysis were used to determine the effective wavelengths and reduce the dimensionality of the spectral reflectance data.The newly-developed FDR-PLS model using the effective wavelengths?628,753,882,935,1061 and 1224 nm?performed well in yield prediction with r²_valof 0.91,and RPD_valof 2.34;Similar results were also obtained for pod number prediction with r²_valof 0.87,and RPD_valof 2.52 using the effective wavelengths?628,758,935,1063,1457 and 1600 nm?.?2?On the leaf?individual?scale.Field experiments were first conducted to study the effect of N nutrition on the spatial distribution of SPAD value of rapeseed leaves on the main stem,and the optimal leaf position and leaf area was identified.Subsequently,the hyperspectral characteristics for the sensitivity leaf were analyzed again under various N application rates.The correlation analysis between the SPAD value and chlorophyll concentration,leaf N concentration and plant N concentration revealed that the fourth leaf from the top is better than others,recommending as a diagnostic leaf area.Simultaneously,the central part on the fourth leaf is better than others,recommending as a diagnostic leaf position.Additionally,for the hyperspectral study on the special leaf,that LNC and leaf reflectance significantly varied with the levels of N fertilization,and a good correlation was observed for all the spectral methods.Tests with the independent validation dataset showed that the FDR-PLS method could well predict LNC in oilseed rape,with the values of r²_valand RPD_valal being 0.97 and 5.49,respectively.The VIP scores resulting from this PLS regression analysis were also used to determine the effective wavelengths.The newly-developed FDR-PLS model using the effective wavelengths?432,467,519,614,772,912 and1072 nm?performed well in LNC prediction with r²_val=0.88,and RPD_val=2.97.?3?For the diagnostic mechanism of N nutrition with the canopy hyperspectral technique.Seedling-budding stage,field sampling revealed that a vertical distribution pattern of LNC existed,presenting an evident decline from the upper to lower layer.The FDR-PLS model for LNC prediction in different layers yielded a relatively higher accuracy compared to the R-PLS based on the full range hyperspectra.And then,seven?437,565,667,724,993,1084 and 1189 nm?,six?423,570,598,659,725 and877 nm?,and five bands?420,573,597,667 and 718 nm?were identified as effective wavelengths for assessing the vertical LNC distribution in the upper,middle and lower layer,respectively.The newly-developed support vector machine-FDR?SVM-FDR?regression model using the effective wavelengths also performed well for upper(RPD_val=2.36),middle(RPD_val=2.56),and lower(RPD_val=2.03)layer LNC prediction.Budding-flowering stage,the LNC and canopy hyperspectral reflectance both showed a vertical distribution pattern,the second and third layers of winter oilseed canopy were an effective depth to monitor NC using hyperspectral data.A PLS model was developed to estimate the LNC on the first,second,third and fourth leaf layers,with the RPD value was 1.64,1.61,1.88 and 1.05,respectively.The model,which considered the vertical distribution patterns of the LNC and the effective canopy layers,has demonstrated great potential to estimate the LNC status of the middle-lower layer within oilseed rape canopies.?4?For the diagnostic specificity of N nutrition.Firstly,the canopy hyperspectral specificity between LNC and leaf phosphorus concentration?LPC?was clarified based on eight field experiments and 25 farmers' s fields;and then,the effective wavelengths which could be accurately and specifically monitoring winter oilseed rape LNC and LPC status were identified.In the LNC study,canopy reflectance decreased from 400 to 715 nm?visible region?and increased from 715 to 1300 nm?near-infrared region?as LNC increased;however,the canopy spectral characteristics under different P treatments showed inconsistent trends.With increasing LPC,the visible reflectance exhibited no clear regularity,whereas the near-infrared reflectance increased.Moreover,seven wavelengths centred at 445,556,657,764,985,1082,and1194 nm and six wavelengths at 755,832,891,999,1196,and 1267 nm were identified as effective wavelengths for predicting the LNC and LPC values based on the FDR-PLS regression model.The newlydeveloped FDR-PLS models for LNC(r²_val=0.85,RPD_val=2.10)and LPC(r²_val=0.78,RPD_val=1.94)provided accurate estimations based on field experiment validations using the effective wavelengths.The overall results demonstrated the applicability and feasibility of the FDR-PLS model for estimating the N and P status of winter oilseed rape using in situ canopy hyperspectral reflectance data.?5?For the N fertilizer recommendation model based on canopy hyperspectra.N real-time recommendation model for winter oilseed rape was developed based on the“LNC-canopy hyperspectral reflectance”fitting function,and a complicated test plan which integration of various N treatments was proposed and carried out to revalidate the robustness and generalized applicability of the technique.Based on the previous field experiments,the optimal spectral parameter,ratio vegetation index(RVI_FDR?764,657?)was selected to estimate winter oilseed rape LNC status.Then the critical RVI_FDR?764,657?value at six-leaf,eight-leaf,ten/over-wintering,over-wintering,budding and flowering stage were calculated with the“N application rates-relative yield”method:4.340,6.781,4.400,3.792,5.493 and 1.379.A new practical field model was evaluated for N recommendation under different N treatments.Results showed that the six-leaf stage?30 days after transplanting?was the first critical period when using canopy hyperspectral technique to guide N fertilizer topdressing control;and budding stage?120 days after transplanting?was the second critical period.The methodology developed by canopy hyperspectra in our study reveals great potential to diagnose the N status of winter oilseed rape.