Remote Sensing Data-Based Sugarcane Acreage Estimation and Yield Forecasting

This study employs three distinct methods: Stepwise Multiple Linear Regression (SMLR),
Ridge Regression, and XG-Boost was used to forecast the sugarcane yield of the Navsari
district of Gujarat, India. The integration of meteorological data and remote sensing-derived
Vegetation Indices (VIs) is a key component in these approaches. A Sentinel-2 satellite image
from May 2023 was utilized for accurate acreage estimation, revealing measuring 8.53%
error compared to government data. Ridge regression emerges as the most accurate model for
yield forecasting, demonstrating consistency across validation years. The combination of
remote sensing data, meteorological data, and machine learning algorithms proves effective
in predicting sugarcane yield, offering a cost-effective, time-efficient, and error-free
alternative. This approach not only enhances the accuracy of crop yield forecasts but also
addresses the challenges associated with traditional methods, such as human error, expense,
and time consumption. Overall, this study underscores the effectiveness of remote sensing in
conjunction with meteorological data and machine learning for precise and efficient
sugarcane yield forecasting, it may provide valuable insights for stakeholders such as
policymakers, crop insurance companies, and agro-processing entities. A constraint of this
study lies in the presence of cloudy images, especially during the months from June to
September. The presence of cloudy conditions introduces contamination, thereby presenting a
specific challenge to accurately forecast the yield, particularly for Kharif crops. Another limitation of this study is the low temporal resolution of Landsat satellite imagery, making it
challenging to obtain real-time data on crop conditions within very short intervals.