Estimating Cultivated Land Quality in Sugarcane Planting Areas Using Remote Sensing and Machine Learning in Guangxi, China Zhihe Hu1,Xuqing li1,2*,Ruiyin Tang1,2,Guohong li1,2,Yancang Wang1,2,Zekun zhang1
1North China Institute of Aerospace Engineering, Langfang 065000, China
2Hebei Province Collaborative Innovation Center for Space Remote Sensing Information Processing and Application, Langfang 065000, China
Abstract
Guangxi is a pivotal sugarcane region in China, where cultivated land quality underpins high yields and sustainable production. Traditional assessments relying on labor-intensive field sampling and manual interpretation are time-consuming, spatially constrained, and ill-suited to rapid regional evaluations. To overcome these limitations, we develop an integrated pipeline that fuses high-resolution remote sensing, deep learning-based semantic segmentation, and machine learning to estimate a comprehensive Cultivated Land Quality Index (CLQI). Using GF-6 multispectral imagery, we generated RGB composites from the visible bands and curated a segmentation dataset from authoritative sugarcane distribution records and expert-annotated samples to train OCRNet. The trained model accurately delineated sugarcane parcels in Nanning and Chongzuo, yielding a fine-grained distribution map and an estimated sugarcane area of 3,785.48 km^2. Representative sample points were selected, and the comprehensive Cultivated Land Quality Index (CLQI) was computed in accordance with national standards, integrating historical soil data. On Google Earth Engine, spectral indices (e.g., NDVI, EVI, VARI, NDWI) were derived- Spearman rank correlation then screened predictors, retaining NIR, DGSI, DRSI, and DVI as most informative. Two estimators were trained: Random Forest (RF) achieved R2_train=0.798, R2_test=0.687, RMSE=0.016- XGBoost achieved R2_train=0.833, R2_test=0.707, RMSE=0.019. Thus, XGBoost achieved a higher test-set R2 (0.707) with a comparable RMSE, indicating stronger predictive power and generalization for CLQI than RF. The framework provides operational evidence and a technical reference for precision land management and broader agricultural land quality assessments across Guangxi.
Keywords: Cultivated Land Quality- Remote Sensing- Sugarcane fields- Semantic Segmentation
