Over Time Efficiency of Predictive Models Based on Proximal Sensing to Assess the Dynamics of Soil Fertility Attributes

Autor(es)

Figueiredo, Veronica Martins ; Bocoli, Fernanda Almeida ; de Padua, Eduane Jose ; Reis, Renata Andrade ; Mancini, Marcelo ; Teixeira, Anita Fernanda dos Santos ; Carneiro, Marco Aurelio Carbone ; Curi, Nilton ; Silva, Sergio Henrique Godinho

• Tipo de documento: Artigo Científico
• Fonte: JOURNAL OF SOIL SCIENCE AND PLANT NUTRITION; v. N/A, p. 15-pg., 2025-02-03.

Resumo

Portable X-ray fluorescence (pXRF) can be used to estimate many soil attributes to help define appropriate management. However, little is known about the over time efficiency of predictive models trained with pXRF data. Here we evaluated the temporal variation of soil properties and tested whether prediction models trained with pXRF data from 2017 will remain accurate when applied to data collected in 2021. 43 georeferenced soil samples were collected at the same places in 2017 and 2021 under five land uses, including managed and natural areas. Models were trained to predict soil fertility attributes based on pXRF data. Prediction models created in 2017 were applied to samples collected in 2021 to assess the over time efficiency of such models. After soil fertility analysis conducted in 2021, it was noticed that coffee plantation areas, which received annual fertilization, showed better fertility conditions in 2017 and 2021 compared with eucalyptus, Cerrado, and native forest. Soil under eucalyptus, which were fertilized only at planting, exhibited a slight improvement in fertility. Unmanaged soils under native vegetation (Cerrado and forest) had their fertility increased in 2021 due to deposition of nutrients applied on adjacent managed areas. The usefulness of the predictive models was maintained after 4 years for effective CEC, available Ca, pH, and remaining P (R-2>0.60). However, prediction accuracy for exchangeable Al and Mg, base saturation, soil organic matter, and CEC decreased (R-2 <0.52). Prediction models generated with 2017 data can still predict some soil attributes accurately after 4 years. (AU)

• Processo FAPESP: 21/06968-3 - Da semente à xícara: internet das coisas na cadeia produtiva de cafés de qualidade
• Beneficiário: Antonio Chalfun Junior
• Modalidade de apoio: Auxílio à Pesquisa - Temático