r-geo-viz

Historical Geospatial Forest Fire Visualization: A Narrative & Didactic Guide

This document started with an AI prompt. It is an outline of our workflows, explaining how we progressed from scanned maps and a couple of shapefiles to a set of R scripts that overlays of historic forest wildfire data across Oregon and California.

1. Project Background

The Core Historical Mystery:
- We began with a set of USGS Annual Report forest classification maps from the late 19th and early 20th centuries. These maps depict historic land cover, standing timber volume, and notably, massive “burnt” areas.
- While modern research heavily documents paleofire records, native cultural burns, and the 20th-century fire suppression regimes that paved the way for modern megafires, the exact perimeters and timing of the large burns shown on these scanned maps (which I downloaded from davidrumsey.com) remained a mystery.
- To investigate this, we needed to compare the “burnt” zones marked on the historical scanned maps directly with historical wildfire databases.
The Georeferencing Stage:
- Raw scans of maps like the Big Trees Quadrangle, Placerville Quadrangle, and Port Orford Map were found and georeferenced.
- This is the process of assigning spatial coordinate systems (such as local Polyconic or Albers projections) to the raw image pixels, producing georeferenced raster files (.tif and world files) stored under data/forest_maps/. Soon the robots will be able to do this for us.
The Vector Search Challenge:
- The next crucial step was to find vector shapefiles of historical fire perimeters.
- Initially, automated assistants struggled to download these databases. They got stuck in robot-unfriendly redirect and download loops between several public open-data platforms.
- Ultimately, I paused the agent and downloaded these datasets ‘by hand’:
  - Oregon Historic Fires: Found in data/Historic_OR_Fires/ (covering pre-2000 fires).
  - California FRAP Fire Perimeters: Found in data/California_Historic_Fire_Perimeters_FRAP/ (containing California wildfire histories dating back to the late 1800s).

2. The R scripts

To handle the geospatial projection, coordinate transformation, cropping, and rendering, we (the Agent and I) developed a sequence of R scripts. Below is the step-by-step breakdown of how these scripts run and interact:

run_all.R:
- Purpose: The master driver script.
- Workflow: It sequentially loads and runs each of the specialized scripts below, handling any individual file errors gracefully to ensure consistent reproduction of the entire pipeline.
01_plot_all.R:
- Purpose: Individual map verification and context mapping.
- Workflow:
  - Iterates through each georeferenced raster found in the workspace.
  - Projects global vector layers (US Counties, Oregon Fires, and California FRAP Fires) into the exact native Coordinate Reference System (CRS) of each individual raster.
  - Applies custom southwest-expanded padding to provide geographical context, plotting the raster cleanly on top of county boundaries with a dedicated prompt tracing box in the lower-left corner.
- Outputs:
  - Individual maps such as output/01_map_Plate_CXXVI_v1.1.jpg and output/01_map_Plate_LXXII_v1.1.jpg.
02_plot_southwest.R:
- Purpose: Regional composite overview.
- Workflow:
  - Computes the bounding box of Southwest Oregon in the rasters’ native CRS.
  - Overlays all georeferenced Oregon raster sheets onto a single canvas.
  - Adds county boundaries and all pre-2000 historic fire vectors on top as a unified layer.
- Outputs:
  - output/02_oregon_southwest_composite_map.jpg
03_plot_southwest_1900.R:
- Purpose: Narrowing the historical lens to a specific year.
- Workflow:
  - Replicates the Southwest Oregon composite viewport from Script 02, but filters the historical fire shapefile to display only fires recorded in the key year 1900.
- Outputs:
  - output/03_oregon_southwest_1900_map.jpg
04_plot_western_1900.R:
- Purpose: High-performance multi-sheet seamless stitching.
- Workflow:
  - Focuses on the westernmost sheets (Plates CXXVI & CXXVII).
  - Digitally “trims” the legend collars (cropping 200 pixels + 10% boundary on each edge) to remove overlapping text margins.
  - Generates both a standard-resolution composite and an ultra-high-resolution (8000x7200px) stitched map.
- Outputs:
  - Standard resolution: output/04_oregon_western_1900_map.jpg
  - High-resolution: output/04_oregon_western_1900_map_highres.jpg
05_california_composite.R:
- Purpose: Multi-state decadal synthesis.
- Workflow:
  - Projects all Oregon and California raster sheets into Web Mercator (EPSG:3857).
  - Groups the massive combined fire perimeters into decadal categories (1800s through 2020s).
  - Renders them using a beautiful custom color ramp where red represents the oldest fires and yellow represent the newest, showing how fire patterns have evolved.
  - Displays a dynamic prompt version panel (currently v5.5) in the lower left alongside the legend.
- Outputs:
  - output/05_california_composite_map_v5.5.jpg
06_detect_burnt.R:
- Purpose: The “Holy Grail” of feature extraction. Haven’t gotten this to work yet
- Workflow:
  - Evaluates data/burned.jpg as a training template to extract the specific color signatures (using k-means clustering) used by historical cartographers to denote burnt regions.
  - Analyzes the scanned maps, calculates Red-to-Green density ratios, and extracts the historical hand-drawn burn polygons.
  - Converts identified cells into vector polygons and saves them as GeoJSON.
- Outputs:
  - Raster mask: output/06_burnt_Plate_CXXVI.tif
  - Vector polygons: output/06_burnt_Plate_CXXVI.geojson
  - Diagnostic plot: output/06_burnt_plot_CXXVI.jpg
07_frap_overlay.R:
- Purpose: California-specific wildfire overlay validation.
- Workflow:
  - Directly overlays the high-resolution California FRAP database onto the Placerville, Big Trees, and Port Orford historical maps in their native coordinates.
- Outputs:
  - output/07_frap_overlay_Placerville.jpg
  - output/07_frap_overlay_BigTrees.jpg
  - output/07_frap_overlay_PortOrford.jpg
agents.md:
- Along the way, I edited the agents file a few times so that I could re-use it with other data. That’s where the UCSB maps come from.

3. Visual Gallery of Generated Maps

Below is a selection of the visual outputs generated by our R geospatial pipeline, demonstrating the progression from raw scans to vector overlays.

A. Regional Composite & Overlays

Southwest Oregon Composite Map
- Displays georeferenced raster sheets seamlessly aligned with county borders and historic pre-2000 fire shapefiles.
Southwest Oregon Year 1900 Fires
- Isolates the fire polygons specifically recorded in the year 1900 to match the approximate era of the survey maps.
Western Oregon 1900 Fire Overlays
- Features collar-trimmed, stitched sheets (Plates CXXVI & CXXVII) plotted alongside 1900 fire perimeters.

B. California FRAP Wildfire Overlays

Placerville Quadrangle
- Shows the georeferenced Placerville scan with actual historic FRAP fire perimeters overlaid in transparent red.
Big Trees Quadrangle
- Overlay of historic FRAP fire perimeters on the Big Trees Quadrangle.
Port Orford Quadrangle
- Overlay of historic FRAP fire perimeters on the Port Orford map.

C. Multi-State Decadal Composite

Oregon & California Decadal Composite (Version 5.5)
- Synthesizes all available maps in Web Mercator projection. Fire perimeters are styled dynamically by decade using an inverted heat ramp (red representing older fires, yellow representing newer fires).

D. Automated Feature Extraction

Burnt Area Detection Plot
- Highlights our k-means template-matching detection process, outlining classified burn zones with a cyan border.
- Some day. Most work is being done with remote sensing imagery. Scanned maps, not so much.

E. Infrastructure & Contextual Mapping

UCSB Infrastructure Map
- Demonstrates the flexibility of our vector mapping setup when applied to alternative, non-forest spatial data layers.