Estimation of canopy chlorophyll in potato based on UAV hyperspectral images

Objectives Chlorophyll content reflects the health status and photosynthesis capacity of crops. The technical parameters that are used to invert canopy images to chlorophyll content quickly and accurately are important in real time monitoring of crop growth condition.

Methods A 3-year potato field experiment was carried out in the main potato producing areas at the northern edge of Yinshan Mountain in Inner Mongolia from 2018 to 2020. Four N rate treatments (198, 202, 229, 287 kg/hm²) were set up, and each treatment had four replications in 2020. During tuber expansion and starch accumulation stage, an unmanned aerial vehicle (UAV) with S185 imaging spectrometer was used to obtain the hyperspectral images of the potato test area, and the spectral reflectance of potato canopy was extracted from them, and the value was converted into chlorophyll content through linear regression. 125 sample points obtained from the study were randomly divided into the training set and validation set in 80% and 20% proportions. Based on the data from the training set, the correlation and estimation models of eight published ratios, normalized spectral indices and optimized spectral indices calculated by the band optimization algorithm with potato chlorophyll content at key growth stages were established, and the accuracy of the models was verified by the data from the validation set. Finally, the distribution map of potato chlorophyll content was made using the estimation model.

Results According to the data from the training set, the distribution of potato chlorophyll content ranged from 10.58 to 23.14 mg/g, with an average chlorophyll content of 19.80 mg/g and a variation coefficient of 14.9%. According to the validation set data, the distribution of potato chlorophyll content ranged from 12.80 to 23.73 mg/g, with an average of 19.59 mg/g and a coefficient of variation of 17.0%. The spectral indices CI_green and ND550 based on green light band had higher coefficient of determination with potato chlorophyll content (R² = 0.48 and 0.61), but the influence of crop type and growth period reduced the accuracy of estimation. The optimized ratio spectral index (RSI) and optimized normalized spectral index (NDSI) calculated by optimizing the bands of 586, 462 nm and 586, 498 nm significantly improved the accuracy of the model and confered a good linear fitting effect. The determination coefficient R² increased from 0.48 and 0.61 to 0.82 and 0.83. After verification, the predicted value of the estimation model was close to 1:1 line with the measured value, the determination coefficient R² was 0.77 and 0.79, and the root mean square error RMSE was low. Through the inversion of potato chlorophyll content distribution, it could be seen that the optimal spectral index (NDSI) model had a good inversion effect, and the distribution range of chlorophyll content varied from 18 to 21 mg/g, which was consistent with the measured value.

Conclusions The best sensitivity bands of RSI and NDSI are 586, 462 and 586, 498 nm, respectively, in which the RSI and NDSI are optimally correlated with chlorophyll content of potato at key fertility stages. The determination coefficients are 0.82 and 0.83, and the validation effect is good. Two spectral indices are applied to the hyperspectral images of the study area for chlorophyll inversion estimation to generate chlorophyll content distribution maps of potato in the field, among which the NDSI estimation is the best and provides theoretical support for spectral index estimation of chlorophyll content of potato.