Interaction between geospatial monitoring of environmental data and soil carbon from fire

• Grant number: 23/15402-9
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: January 01, 2024
• End date: December 31, 2026
• Field of knowledge: Biological Sciences - Ecology - Applied Ecology
• Agreement: NERC, UKRI
• Principal Investigator: Plínio Barbosa de Camargo
• Grantee: João Arthur Pompeu Pavanelli
• Host Institution: Centro de Energia Nuclear na Agricultura (CENA). Universidade de São Paulo (USP). Piracicaba , SP, Brazil
• Associated research grant: 21/00976-4 - Amazon PyroCarbon: quantifying soil carbon responses to fire, AP.TEM

Abstract

To create the geo-spatial database these project will first compile the data on SOC and its fractions produced by the project and from the literature review. As climatic, topographic, fire regime, biomass, land cover and soil characteristics vary across Amazonian regions and this variation can be reflected in the region's SOC patterns, the project will additionally compile continuous spatially explicit information on these components. This information will be used i to understand the relationships between the drivers and SOC in each sampling location and will be the basis for the generation of SOC maps proposed. A set of regression algorithms will be considered in this study to evaluate de relationships between SOC and drivers, encompassing three main approaches: (i) linear models, (ii) kernel-based models, and (iii) decision tree-based models. All algorithms are implemented in the R package caret. We will use a modelling framework composed of two main steps: (1) selection of the most relevant metrics; and (2) validation of the selected models considering the SOC uncertainty related to field sampling data. For feature selection, we will apply the RFE algorithm (rfe routine of the caret package), using the SOC fractions of each plot as the response variable. The RFE will be used to assess the effect of the number of input features over the model performance. The performance will be evaluated by the Root Mean Squared Error, quantified using a cross-validation scheme, repeated 10 times. Model parameters will be optimized by using an internal cross-validation and selecting the parameters with the lowest RMSE. Following O2 analysis we will use the best set of variables and model for analyzing and produce wall-to-wall maps depicting the spatial variability of SOC. The novel spatial database of SOC for Amazonia, shall help to inform the process of producing the National Inventory of GHG Emissions and as potential indicator of sustainability, (e.g., achievement of the SDGs, or refining the Forest Emissions Reference Level and the REDD+). (AU)