Trade-off between cost and performance of earth observation data in olive trees health assessment: Digital crop mapping approach using machine learning algorithms

Abstract

• Comparison of UAV, Mohammed VI, and Sentinel-2 data for olive tree health assessment. • Random Forest model outperforms Cubist and XGBoost across all datasets. • Sentinel-2 data shows competitive performance despite lower resolution. • Cost-effectiveness analysis reveals trade-offs between data quality and price. • Framework for selecting optimal EO products based on monitoring needs and budget. This study evaluates the efficacy and cost-effectiveness of different Earth Observation (EO) products for assessing olive trees’ health in a 6.2 ha orchard. We developed a digital crop mapping approach using high-resolution multispectral imagery from three sources: unmanned aerial vehicle (UAV) at 2 cm spatial resolution, Mohammed VI satellite at 50 cm, and Sentinel-2 satellite at 10 m. Leaf chlorophyll content, measured for 75 trees using a SPAD meter, served as the dependent variable for predictive modeling and a proxy of trees’ health. Furthermore, three machine learning algorithms, Random Forest (RF), Cubist, and Extreme Gradient Boosting (XGBoost), were employed to predict chlorophyll content. The RF model showed superior performance across all datasets, achieving the lowest RMSE values. While UAV data provided the highest spatial detail, capturing fine-scale health variations, it came at a significantly higher cost. The commercial Mohammed VI satellite data offered a balance between detail and coverage, potentially cost-effective for areas approaching its 100 km 2 minimum purchase requirement. However, the free Sentinel-2 data showed competitive performance, particularly with the RF model (RMSE: 7.773, RPIQ: 1.251), suggesting its viability for large-scale monitoring. Spatial chlorophyll maps revealed distinct capabilities of each EO product in capturing orchard variability. Our findings highlight the trade-offs between EO product costs, resolutions, and predictive accuracy, providing crucial insights for selecting appropriate remote sensing approaches based on specific monitoring requirements and budget constraints. This research underscores the potential of integrating diverse EO products with advanced ML techniques for cost-effective and scalable olive tree health assessment, with implications for precision agriculture and sustainable orchard management. Furthermore, all data and scripts used in this study are provided for reproducibility and transparency.