Scripts from paper "Past fire use in deforested land limits secondary forest carbon gains in Amazonia with implications for emissions reduction goals"

Document purpose This repository contains R scripts used to develop the article "Past fire use in deforested land limits secondary forest carbon gains in Amazonia with implications for emissions reduction goals", submitted to Carbon Balance and Management. It includes complete workflows for data processing, statistical modelling, and figure generation, enabling full computational reproducibility. The release of this material complies with the Data Management Plan of FAPESP (grant nº 16581-1/2022). Author correspondence: Isadora HaddadE-mail: [e-mail] For more information, contact us. Data sources and versions Table 1. Data souces used in the research development Dataset Version Temporal coverage Purpose in study Access link Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) v2.0 1985-2020 Monthly precipitation input; MCWD calculation; variables for cluster analysis CHIRPS data TerraClimate - 1985-2020 Minimum and Maximum temperature; variables for cluster analysis TERRACLIMATE data TerraClimate - 1985-2020 Shortwave Radiation; variable for cluster analysis TERRACLIMATE data TerraClimate - 1985-2020 Monthly evapotranspiration; MCWD calculation; variable for cluster analysis TERRACLIMATE data Zuquim et al. (2017) - Soil Cation Concentration (SCC) As published 2017 Soil fertility; variable for cluster analysis https://doi.org/10.1594/PANGAEA.879542 Maximum Cumulative Water Deficit (MCWD) Derived product 1985-2020 Drought intensity indicator; variable for cluster analysis https://doi.org/10.5281/zenodo.4903340 European Space Agency CCI Biomass v.5 2020 Aboveground biomass (AGB) https://gee-community-catalog.org/projects/cci_agb/ TerraClass 2020 Land use and land cover; Secondary Forest Formation (class "2"); Primary Forest Formation (class "1") TERRACLASS data MapBiomas Collection 6 1985-2020 Land use and land cover; caracterization of land use history; Forest Formation (class "3"); input for computer forest age MAPBIOMAS data MapBiomas Collection 4 1985-2020 Fire; bruned area anlysis MAPBIOMAS data Silva-Junior et al. (2020) Secondary forest age Derived product 1985-2020 Forest regeneration age https://github.com/celsohlsj/gee_brazil_sv Note: Biomass values were converted to Carbon (AGC) following the methodology by Martin et al. (2018) Workflow description 1) Edaphoclimatic region (cluster_analysis.R) Performs multivariate analysis (clustering) to define edaphoclimatic regions based on soil and climate variables (grid cell 5x5km). Input: CHIRPS, Zuquim et al. (2017), TerraClimate, MCWD Output: Clustered regions for spatial analysis 2) Land Use History and Carbon Stock CharacterisationThis stage involves data integration from Google Earth Engine (GEE) and spatial refinement in QGIS. The workflow is divided into five main steps: a) Forest Masking and Reference Layers (GEE) Secondary Forest (SF): Defined using the TerraClass "Vegetação Florestal Secundária" (Class 2) as a reference mask. Mature Forest (MF): Defined using "Formação Florestal Primária" (Class 1). To ensure data integrity, a 300m internal buffer was applied to remove edge effects, and pixels affected by fire (1985–2020) were excluded using MapBiomas annual_burned data. b) Age and Biomass Derivation Vegetation Age: Derived for SF areas following Silva-Junior et al. (2020) based on MapBiomas (30m) land cover series. Carbon Stock: Aboveground Biomass (AGB) data from ESA-CCI were extracted for both SF and MF masks and converted to Aboveground Carbon (AGC) using the conversion factors from Martin et al. (2018). c) Grid Integration and Data SystematisationA 100x100m cellular grid was created to integrate all variables. Each grid cell contains: Percentage cover of SF and MF (TerraClass). Percentage cover of each forest age (Silva-Junior et al., 2020). Fire recurrence and time since last disturbance (MapBiomas). Land use history (anthropic class, duration, and prior use) (MapBiomas). Corresponding AGC value (ESA-CCI). d) Spatial Filtering and Characterisation (QGIS) Consistency Filter: Only cells with 100% convergence between the integrated information layers were retained. Stratification: The final dataset was categorised by edaphoclimatic region and prior land-use class. Table 2. Description of previous land use classes Previous land use classes Description Undisturbed Secondary Forest (SFs) SFs without previous anthropogenic land use and fire (i.e., secondary growth started immediately after deforestation) Anthropogenic land use and fire SFs Anthropogenic land use and fire recurrence before the onset of regrowth Anthropogenic land use SFs Anthropogenic land use and the absence of fire before the onset of regrowth Fire SFs SFs are characterised by the forest cover class and fire recurrence before the onset of regrowth Mature Forest (MF) MF areas without fire recurrence are used as the asymptote in the growth models 3) Growth model and statistics (growth-model-estatistical.R)This script performs the core statistical modelling and hypothesis testing of the study. It is organised into three main functional steps: input data: samples of previous land use classes, secondary forest, and mature forest a) Data pre-processing and outlier filtering: Quartile filtering: implements a data-cleaning step that retains AGC only values between the 25th and 75th percentiles for each forest age. Sensitivity analysis: includes a comparative test between mean and median AGC values approaches to determine the most robust input metric for the non-linear model. b) Growth curve modelling (Chapman-Richards): Model fitting: utilises the Chapman-Richards function via the nlsLM algorithm. Edaphoclimatic asymptotes: asymptotic AGC values are fixed based on observed mature forest medians for each edaphoclimatic region. Uncertainty quantification (boostrap): employs a boostrap approach (1,000 interations). c) Statistical testing: Carbon gain rates: calculates the annual carbon accumulation rate, focusing in the <20 years (TIER 1 - IPCC). Non-parametric comparisons: uses Kruskal-Wallis and Dunn's post-hoc tests, to identify significant differences in carbon stocks between regions and land use classes SFs. Distribution testing (Monte Carlo): executes 10,000 Monte Carlo permutations to statistically compare the differences in carbon gain rates between distinct edaphoclimatic regions and previous land use classes. 4) Data visualisation (figures_manuscript.R) Contains the R scripts utilised to generate all figures presented in the manuscript. 5) Future projections (projections-20-35.r) Contains the scripts for future projections spanning the period from 2020 to 2035. Modelling: these projections account for various age-protection scenarios and previous land use classes by edaphoclimatic regions.