USING EARTH OBSERVATION TO MONITOR NO-TILL PRACTICES OVER AGRICULTURAL CROPS IN EASTERN ONTARIO AND PRINCE EDWARD ISLAND, CANADA

摘要

The benefits of reduced tillage and no-till practices over agriculture fields have been well documented and Canadian growers are increasingly implementing such practices. Detecting and quantifying crop residue cover over agricultural fields is thus critical in identifying conservation tillage practices which is an important objective within the Agricultural Policy Framework of Agriculture and Agri-Food Canada (AAFC). Earth observation can play an important role in crop residue monitoring and thus in delivering residue information products to AAFC policy and programs. Several methods are being tested at AAFC to quantify percent residue cover, with spectral unmixing analysis (SMA) showing the greatest potential. Several multispectral (Landsat and Spot) images were acquired during the 2006 spring tillage season, over two agricultural sites in Canada, namely Eastern Ontario and Prince Edward Island. Simultaneous ground data were collected to characterize various crop residue types (corn, soybean, small grain, potato, forage). The main objectives of this study are: (1) to propose a standard methodology for endmember selection and extraction for input in the SMA algorithm; (2) to validate the SMA approach for percent residue cover mapping; and (3) to determine if the SMA approach can be utilized for no-till mapping. Results comparing the four different endmember extraction techniques did not provide a clear solution as the use of individual residue endmembers performed only slightly better than the other techniques evaluated. Further testing is thus required for defining a standard method for endmember extraction. As for the crop residue fraction validation, root mean square errors (RMSEs) produced mixed results fluctuating between 20 to 30% with the residue corn fields achieving the best results (RMSE ≈ 20-24%), with the small grains producing the poorest results (RMSE ≈ 35-40%). The analysis also concluded that SMA, in combination with Landsat and Spot image data, can be used to identify no-till (60-100%) fields to accuracies that exceed 85%.