Sugarcane Productivity Dataset

This dataset consists of processed, plot-level observations developed for sugarcane productivity prediction using multi-source satellite imagery and machine learning approaches. The dataset integrates field-based productivity measurements, remote sensing-derived indicators, and environmental variables, collected from three sugarcane plots situated within a single, homogeneous agroecological zone over four consecutive planting seasons.

Each record represents a plot-level observation aggregated at a specific observation date, capturing the temporal progression of crop development while maintaining spatial consistency. The target variable is sugarcane productivity, measured at the plot level and used as the response variable in predictive modeling. The predictor variables comprise:

Vegetation indices extracted from multispectral imagery acquired by Sentinel-2 and Landsat-9,

Morphological attributes, including planting age and crop development indicators,

Meteorological variables representing local climatic conditions during the growing season,

Derived features engineered to enhance model sensitivity to productivity-related dynamics.

All data sources were temporally synchronized to ensure consistency between satellite acquisitions, field observations, and meteorological records. Preprocessing steps included data cleaning, plot-level aggregation, and temporal alignment to minimize noise, missing values, and temporal mismatch across variables.

The dataset is intended to support reproducible research, methodological benchmarking, and comparative evaluation of machine learning models for sugarcane productivity prediction under controlled agroecological conditions. While spatial variability is intentionally limited, the dataset emphasizes temporal generalization, multi-source data fusion, and predictive robustness, making it particularly suitable for methodological studies rather than large-scale spatial extrapolation