Estimation Of Cotton Growth Information Using Two Ground-based Visible Imaging Sensors

[Object] The objectives of this paper were to monitor main growth information using ground-based visible imaging sensors in different growth stages, and to determine the spectral and color parameters for estimating canopy and leaf nitrogen nutrition status, and to develop quantitative model for estimating LAI and fIPAR and fAPAR with image transmittance and image fIPAR. Consequently, spectral and color parameters in visible region and image fractions index (transmittance and fIPAR) may provide theoretical basis and technical support of precise monitoring technique.[Methods] In this study, we used various kinds methods to odserve data in different growth stages from 2006 to 2010 in cotton field. (1) A systematic analysis was undertaken on quantitative relationships of nitrogen nutrition indices to color parameters, sensitive wavebands and major spectral indices, such as the ratio index (RI), normalized difference index (ND) and difference index (DI), which composed of any two wavelengths with original reflectance. Futhermore, the quantitative models were developed and the optimum models werw selected which with the maximum determination coefficients (R2) and the minimum RMSE. (2) Analysis of the diurnal pattern of transmittance and fIPAR of the cotton canopy showed that the best time for observe data was around solar noon, because at this time the solar elevation angle is high and remains relatively constant during measurements. Additional, by analyzing the relationships among various transmittance or fIPAR, we determined that Timag and fIPARimag could be used to estimate light attenuation and light interception in the cotton canopy. Hence, the estimated models of LAI and fIPAR and fAPAR were established using Timag and fIPARimag parameters. Overall, the ability of all estimated models in this paper were validated using an independent dataset, accepted statistics indices included determination coefficients (R2), RMSE and RRMSE.[Results] The main results of this paper as follows:(1) The results indicated that the maximum sensitivity of reflectance to variation in chlorophyll, nitrogen contents and SPAD readings was found in the far-red wavelength region at 710 nm and in the red wavelength region (R) for two sensors, respectively. Furthermore, spectral indices could improve the prediction ability obviously, and difference indices (DI and R-B) of different sensors composed of blue and red wavelengths gave a better prediction performance. The models to retrieve chlorophyll, nitrogen contents and SPAD readings using DI were the most feasible models with the maximum determination coefficients (R2) and the minimum RMSE, especially, DI(R440, R710), DI(R440, R710), DI(R420, R710), DI(R420, R720) and DI(R49o, R710) were the optimum indices for the models of chlorophyll a+b, chlorophyll a, chlorophyll b and N, and SPAD readings, respectively. R-B was the optimum index of digital camera but its prediction performances were lower than these of DI. Additional, b* (CIE 1976 L*a*b* color model) and S (HSI color model) were the optimum color parameters, and the prediction ability of b* was lower than that of DI. However, the prediction performance of S was relative weak with the highest RRMSE and the lowest R2.(2) The results showed that canopy spectral reflectance decreased with increasing leaf nitrogen content, and the bands sensitive to leaf nitrogen content occurred the green and red regions mainly. Furthermore, the models to retrieve canopy leaf nitrogen contents using DI(R580, R680) and G-R were most feasible with the maximum determination coefficients (R2) and the minimum standard error (SE) for two visible sensors, respectively. Additional, b* (CIE 1976 L*a*b* color model) and H (HSI color model) were the optimum color parameters. On the whole, for the fitting effects, the spectral index was superior to color parameters for the same sensor, and MSI200 superior to digital camera. Then, the prediction performances of the spectral indices of digital camera were validated by using independent dataset. We found that difference indices DI(Rs8o, R680) and G-R were the optimum indicators of canopy leaf nitrogen content with the highest predictive precision (0.8131 and 0.7636, respectively)and accuracy (1.0149 and 0.9661) and the lowest RMSE (2.3313 and 2.7406 mg g-1, approximately 6.52% and 8.24% of the mean).(3) Analysis of the diurnal pattern of transmittance of the cotton canopy showed that the best time for measuring Timag was around solar noon, because at this time the solar elevation angle is high and remains relatively constant during measurements. Around solar noon, Timag was in good agreement with Tquan (transmittance measured with a linear quantum sensor). By analyzing the relationships among Timag, Tquan, and diffuse non-interceptance (DIFN), we determined that Timag could be used to estimate light attenuation in the cotton canopy at different stages, except for the boll opening stage. In addition, Timag was saturated at LAI>5. We analyzed the relationship between LAIdest (LAI measured destructively) and Timag using data from 2009 and 2010. The R2 and SE of the calibration model were 0.8438 and 0.5605, respectively. The ability of Timag to predict LAI was validated using an independent dataset (2007 data). The determination coefficient and RMSE of the validation model were 0.8767 and 0.4305, respectively. However, the model underestimated LAI as the LAI exceeded 5. The Timag saturation, which was largely because of errors in image recognition and segmentation, resulted in underestimation of LAI. Intercomparisons of LAI estimates showed that there were small discrepancies and significant correlations among data obtained from digital images, the LAI-2000, and destructive sampling methods. Data from the LAI-2000 was highly consistent with that obtained by destructive sampling. (4) The results indicated that the fIPARimag which calculated with canopy images taken around solar noon can model the fIPAR and fAPAR of cotton canopy. Then, regression analysis were made on the relationships between fCover and fIPARimag and flPAR and fAPAR, with the determination of coefficient (R2) exceeded 0.93. Tests with independent dataset (2009 data) showed that the prediction performance of fIPARimag is superior to fCover.[Conclusions] Hence, spectral and color parameters in visible region may provide an effective and feasible means of estimating canopy and individual leaf nitrogen nutrition status quantitatively in cotton field. Moreover, Timag and fIPARimag which derived from four image fractions can provide a new and accurate means of estimating LAI and fAPAR and fIPAR. Consequently, the digital camera could be mounted on a tractor or farm vehicle for real-time, non-destructive monitoring of LAI to support field management.