Sentinel-2 for Crop Yield Estimation: A Systematic Review

Abstract

Accurate and timely crop yield estimation is critical for global food security, agricultural policy, and farm management. The Copernicus Sentinel-2 satellite constellation, with high spatial, temporal, and spectral resolution, has transformed agricultural monitoring by enabling field- and sub-field-scale analysis. This review synthesizes recent advances in Sentinel-2-based crop yield estimation. A key trend is the shift from regional models to high-resolution field-level assessments driven by three main approaches: (i) empirical models using vegetation indices combined with machine and deep learning methods such as Random Forest and Convolutional Neural Networks; (ii) integration of process-based crop growth models (e.g., WOFOST, SAFY) via data assimilation of Sentinel-2-derived variables like Leaf Area Index (LAI); and (iii) data fusion techniques combining Sentinel-2 optical data with Sentinel-1 SAR to mitigate cloud-related limitations. The review shows that machine learning, deep learning, and hybrid modeling frameworks can explain substantial within-field yield variability across crops and regions. However, performance remains constrained by limited ground-truth data, cloud-induced gaps, and challenges in model transferability across years and locations. Future directions include tighter integration of multi-modal data and improved in-season observations to support robust, operational decision-making in precision agriculture and sustainable intensification.

Figures (5)

• Figure 1: Global trends in Sentinel-2 crop yield estimation research. (A) Geographic distribution of publications by country, showing the number of studies conducted in each nation from 2015 to present. Countries are color-coded by publication count. (B) Temporal trends in crop study frequency, displaying the evolution of research focus across major crop types over time. The stacked bars show the relative contribution of different crops to the literature each year. (C) Continental distribution of publications over time, illustrating the geographic expansion of Sentinel-2 yield estimation research across different regions. In panels B and C, the first four years (2015--2018) are merged into a single bin for consistency with the remaining year ranges.
• Figure 2: Satellite platform comparison for crop monitoring. (A) Spatial resolution comparison showing the same agricultural area near Scott City, Kansas (38.51838°N, -100.91645°W) as captured by Landsat 9 (30 m), Sentinel-2 (10 m), and PlanetScope (3 m), demonstrating the trade-offs between spatial detail and coverage. (B) Spectral band comparison illustrating the wavelength coverage and bandwidth of each platform overlaid on a typical healthy vegetation reflectance curve (order matches panels A and C). (C) Temporal resolution comparison showing the acquisition frequency of each platform over a year, emphasizing the different revisit capabilities for continuous crop monitoring.
• Figure 3: Overview of modeling approaches for yield estimation. (A) Hierarchical bar chart showing the frequency of different modeling approaches organized by category (Machine Learning, Statistical/Regression, Deep Learning, Process-Based/Data Assimilation, and Hybrid/Meta-learning) across the reviewed studies. (B) Category distribution showing the relative prevalence of each modeling approach family in Sentinel-2 yield estimation research. (C) Co-occurrence matrix displaying the frequency of joint usage of the top 10 modeling approaches, revealing common model combinations and co-occurrence patterns in the literature.
• Figure 4: Summary of challenges and solutions in Sentinel-2 yield estimation. A comprehensive table outlining major challenges (e.g., data availability, model transferability), their impact on model performance, and examples of solutions proposed in the literature (e.g., sensor fusion, hybrid modeling).
• Figure 5: Evolution and future outlook of Sentinel-2 yield estimation methods. A schematic timeline illustrating the progression from early methods (c. 2016) like simple VI-regressions to current advanced approaches (c. 2020--2024) using ML/DL and data fusion, and projecting future trends (c. 2025+) such as operational SIF integration and hybrid CGM-ML systems.