Spatial modeling of soil physicochemical properties and fertility in tropical agricultural systems under different structural heterogeneity using multispectral UAV and geostatistics

• Autores: Vega-Herrera, Sergio, Ysuiza-Perez, Alfredo, Perez-Tello, Mónica, Goigochea-Pinchi, Diego, Rios-Rios, Raúl, Dominguez-Yap, Percy, Garcia, Leonela, Barrera-Torres, Cicerón, Oliva-Cruz, Carlos, Santillán-Gonzáles, Manuel, Arratea-Pillco, David, Alejos-Patiño, Italo
• Formato: artículo
• Fecha de Publicación: 2026
• Institución: Universidad Nacional de Trujillo
• Repositorio: Revistas - Universidad Nacional de Trujillo
• Lenguaje: español
• OAI Identifier: oai:ojs.revistas.unitru.edu.pe:article/7531
• Enlace del recurso: https://revistas.unitru.edu.pe/index.php/scientiaagrop/article/view/7531
• Nivel de acceso: acceso abierto
• Materia: agricultura de precisión; interpolación espacial; índices de vegetación; teledetección con UAV; geoestadística; sensores multiespectrales

Spatial variability of soil properties is a key factor that influences productivity, nutrient management, and overall sustainability in tropical agricultural systems. This is especially true where differences in soil types make it difficult to apply site-specific management strategies. In this context, the study sets out to compare the performance of the same analytical workflow based on multispectral UAV imagery, multiple linear regression, and geostatistical interpolation, across two tropical farming systems that differ in their level of structural heterogeneity. One was a station-wide multicrop system, and the other was a rice system under varying planting densities, both located at the El Porvenir Agricultural Experimental Station in San Martín, Peru. The study included 60 soil samples from the multicrop component and 27 from the rice system. All samples were georeferenced and taken at 30 cm depth, then analyzed in the lab for pH, electrical conductivity, nitrogen, phosphorus, potassium, soil organic carbon, and texture. The exact same workflow was applied to both systems: Spearman correlation, stepwise multiple linear regression, and ordinary kriging. Results showed that the rice system gave better predictive accuracy for specific variables like nitrogen and phosphorus. On the other hand, the multicrop component proved more useful for mapping spatial patterns and defining management zones, thanks to its greater heterogeneity. In addition, indices based on NIR and red edge bands had stronger links with the main soil properties. Overall, the performance of the approach clearly depended on the structural heterogeneity of each system: more uniform environments favored point-based predictions, while more variable ones were better suited for operational zoning.