科研之友
微信
新浪微博
Facebook
分享链接
摘要
Rapid and accurate predictions of heavy metal contents in farmland are of great significance to ensure the safety of agricultural products and maintain ecosystem balance. Combining multispectral images and chemometric modeling provides a feasible means of estimating cadmium (Cd) and lead (Pb) contents in farmland. We collected 640 samples from the surface soils of farmland in Xiangzhou District, Hubei Province, China. The Cd, Pb, soil organic matter (SOM), pH, and Fe of the soil samples were measured in a laboratory. In this study, the dimidiate pixel model was used to process remote sensing images. In particular, we used random forest (RF) to screen the best spectral indices for use as input variables. Partial least squares regression (PLSR), backward propagation neural network (BPNN), and RF were used to calibrate the spectral data with Cd and Pb contents, and the optimal model was used for the regional mapping of soil Cd and Pb contents. Additionally, we explored the potential of using spectral estimation mechanisms to estimate Cd and Pb contents. The mechanism for estimating Cd and Pb contents with multispectral images depended mainly on the covariance of Cd and Pb contents with that of SOM. For both Cd and Pb estimations, the double-date image estimation model performed better than the single-date image estimation model, and the unmixed image estimation model was more accurate than the original image estimation model. Overall, the estimation model using the best spectral indices as input variables performed better than the model using full-band data as input variables. The RF model outperformed the PLSR and BPNN models in all cases. Relatively satisfactory estimates of Cd and Pb contents in farmland of the study area (maximum R-val(2) (determination coefficient of the validation set) = 0.60 for Cd and R-val(2) = 0.63 for Pb) were obtained. Our results show that areas of farmland contaminated by Cd and Pb throughout the study area have increased and that contamination levels have worsened since 1990. In addition, the spatial patterns of Cd and Pb contents in farmland throughout the study area were analyzed and validated using field survey results. The results of the study provide a theoretical basis and methodological reference for the rapid prediction of Cd and Pb contents in regional farmland.
-
单位y
