pdm

A reproducible framework for monthly spatial prioritization of deforestation risk using data-driven modeling in R

---

Overview

pdm is an R package designed to support environmental decision-making by identifying and prioritizing areas at risk of deforestation.

The package operationalizes a modeling pipeline that:

• Generates spatial predictors from environmental datasets
• Trains predictive models for land-use change and fire occurrence
• Produces prioritization maps to support monitoring and intervention strategies
• Evaluates model performance through temporal validation and error metrics

This framework was developed with a focus on Amazon biome applications, but it is fully adaptable to other regions.

---

Motivation

Deforestation and fire dynamics are complex, multi-scale processes influenced by climate, accessibility, and land-use pressures.

While many models exist, there is a gap between research prototypes and operational tools that:

• Run consistently over time
• Produce standardized outputs
• Support decision-making workflows

pdm bridges this gap by providing a reproducible, modular, and extensible pipeline.

---

Installation

# install.packages("devtools")
devtools::install_github("albhasan/prioritizedeforestationhotspots")# Necessary in this first package version
devtools::install_github("LucasUfmg/pdm")

---

Workflow

The package follows a structured pipeline:

1. Data preparation

   • Spatial covariates (e.g., land cover, climate, accessibility)
   • Temporal aggregation (monthly/annual)

2. Feature engineering

   • Generation of derived variables
   • Standardization and filtering

3. Model fitting

   • Statistical or machine learning models
   • Configurable training periods

4. Prediction and prioritization

   • Spatially explicit predictions
   • Ranking of high-risk areas

5. Evaluation

   • Commission error
   • Variable importance
   • Temporal performance diagnostics

---

Minimal Example

library(pdm)
library(prioritizedeforestationhotspots)
# Run model with default parameters for September 2022
pdm::run_pipeline(folder = "your_folder_path",
                  mes_inicial = 9,mes_final = 9, ano_inicial = 2022,
                  run_prep = T,run_prio = T)

# Outputs
names(result)
#> "priority_map" "commission_error" "variable_importance" "monthly_error"

# Plot prioritization map
plot(result$priority_map)

---

Outputs

The model generates:

• Prioritization maps: spatial ranking of risk areas
• Commission error metrics: evaluation of false positives
• Variable importance: model interpretability
• Temporal diagnostics: monthly performance trends

These outputs are designed to support:

• Environmental monitoring programs
• Policy targeting and enforcement
• Research on land-use dynamics

---

Reproducibility

The package was built with reproducibility as a core principle:

• Fully scriptable workflow
• Deterministic outputs given fixed inputs
• Modular functions for each processing step

---

Technical Stack

• R (>= 4.x)
• Spatial analysis: sf, terra
• Data manipulation: dplyr
• Modeling: customizable (user-defined or built-in methods)

---

Author

Lucas Santos Data Scientist | Environmental Modeling | Spatial Analysis

• Experience in Amazon deforestation and fire modeling
• Background in academic and applied research environments
• Former PhD researcher (University of California, Irvine)

---

Potential Applications

• Monitoring deforestation frontiers
• Fire risk prediction
• Conservation planning
• Environmental policy support
• Early warning systems

---

Future Work

• Integration with real-time data pipelines
• Expansion to other biomes and regions
• Incorporation of advanced ML models
• Deployment as a decision-support tool

---

License

MIT License

---

Contact

For collaborations or academic inquiries, feel free to reach out.