Unmixing corn and soybean using Multiple Endmember Spectral Mixture Analysis (MESMA) based on Google Earth Engine

Authors: Ke Li*, SUNY-Buffalo, Le Wang, SUNY-Buffalo
Topics: Remote Sensing, Agricultural Geography
Keywords: MESMA, corn, soybean, Google Earth Engine
Session Type: Guided Poster
Day: 4/9/2020
Start / End Time: 8:00 AM / 9:15 AM
Room: Agate A/B, Hyatt Regency, Third Floor
Presentation File: No File Uploaded


Mapping the distribution of cropland poses a significant role in evaluating food production and commercial trade. With advantages about spatial, temporal resolution, Landsat data provides a promising way to map crop land type. However, the fraction of each cropland of per-pixel level have been poorly investigated and remain unclear. Multiple Endmember Spectral Mixture Analysis (MESMA), especially in heterogeneous environments, represents a key tool to investigate the fraction of croplands for per-pixel basis. This study evaluated the potential for displaying the distribution of cropland type taking corn and soybean as an example. In this study, three types of endmembers signature were devised, signal date during the growing season and two types of temporal signature that capture cropland phonology characteristics accordingly. Firstly, we automatically choose the best candidate endmembers using Count based endmember selection (Cob), Endmember Average RMSE (EAR) and Minimum Average Spectral Angle (MASA), and then we find the best fit endmembers combination when unmixing. All methods were designed based on Google Earth Engine (GEE). Cropland Date Layer (CDL) was used to compare with the result we calculated. For fraction distribution map of corn and soybean, by assigning pixel with fraction about more than 50% to one class (the pixel higher than 50% corn was assigned to corn), this distribution map could hope to be comparable with CDL and have high similarity.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login