Marginal agricultural land identification in the Lower Mississippi Alluvial Valley based on remote sensing and machine learning model

Abstract

Marginal agricultural lands are considered unsuitable for conventional crop production but could be utilized for biofuel production, groundwater recharge, afforestation, and other land use purposes. Effective management of these lands requires accurate identification and understanding of their spatial distribution. We used random forest regression (RFR) with remotely sensed vegetation indices, soil properties, and climate data to estimate soybean and corn yield and identify marginal agricultural lands in the Lower Mississippi Alluvial Valley (LMAV). The developed RFR models were utilized to estimate gridded crop yields and identify marginal agricultural lands within the LMAV. Depending on the cost and revenue data used, the total area of marginal agricultural lands ranged from 18,888 to 720,655 ha. Using the 2021 sales price and associated production costs, 18,888 ha (1.03% of total cropland) of marginal lands were identified within the study area. However, using the six-year (2016–2021) average of the inflation-adjusted sales price and production costs, the estimated area of marginal agricultural lands increased to 720,655 ha (36.35% of total cropland). The findings have significant implications for land use planners and farmers, who can use the identified spatial distribution data of marginal lands within the study area to update existing field crops and plan for alternative land uses such as biofuel/bioenergy crop production, afforestation, groundwater recharge, restoring wildlife habitat, and soil restoration. These alternative land uses could result in higher returns and conservation of marginal lands, leading to sustainable and productive land use management.

Keywords

• Remote sensing
• Random forest
• Marginal agricultural land
• Crop yield prediction
• Land use change