Data-Driven Framework Integrating Database Analysis and Bayesian Segmented Quantile Regression to Determine Critical Nutrient Levels in Soil and Leaves of Peach Trees

Abstract

Fertilization recommendations are frequently proposed by a limited number of calibration experiments conducted in a few regions, with few cultivars. Thus, recommendations concerning nutrients’ critical levels (CLs) and sufficiency ranges (SRs) are not always the most suitable ones. In addition, it is not known whether CL and SR differ when it comes to peach yield and quality variables, such as pulp firmness or total soluble solids (TSS) concentration. CL and SR can be estimated by using Bayesian segmented quantile regression (BSQR) in combination with databases. The aim is to propose nutrients’ CL and SR in soil and in leaves of peach trees grown under a subtropical climate, by combining databases and BSQR models. The dataset comprised 208 observations of Prunus persica cultivars ‘Maciel’ and ‘Chimarrita’, grown in southern Brazil. Models were developed through plateau regression based on BSQR models to measure the association between dependent variables (fruit yield and quality) and nutrient concentrations in the soil and leaves. CL and SR of N and K in leaves recorded values for variables related to fruit quality, pulp firmness, and TSS lower than values recorded for variables related to yield. We show for the first time that CL and SR in leaves for N and K related to yield are different between cultivars. A similar outcome was recorded for CL and SR of N in leaves between growing regions/sites presenting different soil types and climatic variables. CL and SR of N in leaves related to yield were higher in peach trees grown in sandy-soil sites with low organic matter content. This study presents critical advances for the Southern Brazilian peach market, but with implications for the approach presented in the present study (combination of database and BSQR modeling) relevant for application in orchards around the world.