Deriving corn and soybeans fractions with Land Remote-Sensing Satellite (System, Landsat) imagery by accounting for endmember variability on Google Earth Engine

Timely mapping of corn and soybean plays an important role in food security in the USA. A subpixel fraction derived from Land Remote-Sensing Satellite (System, Landsat) imagery during growing seasons is desired in order to help local farmers monitor crop growth and manage them in a timely fashion. However, two obstacles need to be surpassed before such fractional information can be made available: 1) the endmember spectral reflectance of corn and soybean varies with time and location; 2) no methods have been tested for deriving fractional maps of corn and soybean throughout the growing season. Therefore, in this research, we have set aside two objectives: 1) To account for endmember variability of corn and soybean during their growing seasons; 2) To derive multi-temporal fractional maps for corn and soybeans with Landsat and monitor the growing status of corn and soybean. Accordingly, we employed three endmember optimization methods and the state-of-the-art unmixing method Multiple Endmember Spectral Mixture Analysis (MESMA) to acquire corn and soybean fractional maps based on Google Earth Engine (GEE). Applying the method on Landsat 8 images from April to September 2017, we generated multi-temporal fractional maps of corn and soybean in Grundy County and analysed their changes. Up to 94.76% of our study area was successfully explained by the unmixing model. The crop fraction in both corn and soybean fields was about 15.00% during the planting stage, and increased to nearly 80.00% in the peak growing season. The crop fractions remained high during harvest, which could be attributed to crop residues in the field. These findings correspond well with the growth stages provided by the United States Department of Agriculture (USDA). That the corn growing season was earlier than soybeans was also well represented by the fractional change analysis. Moreover, among all the fractional maps, the results in the peak growing time (29 July 2017 in this study) had the highest agreement with classification results, with an overall accuracy of 85.07%. This research shows the great potential of monitoring corn and soybean growth conditions with fractional maps. The methods in this study, implemented in GEE, can be easily transferred to other crops and other locations.