Wildfire_Prediction_Model

For more information on the project and method procedures, refer to the ArcGIS StoryMap website: https://storymaps.arcgis.com/stories/f53d780418ad4cc6bb2386db05165ed4

Repository Information

This repository contains four Jupyter notebooks:

1. Combining_Wildfire_Climate_Data.ipynb: This notebook process and merges wildfire and climate data. The climate data is a daily summary of the average temperature (TAVG), average wind speed (AWND), and precipitation (PRCP) within each county.
2. Decision Tree Modeling.ipynb: This notebook creates a decision tree model of different conditions (one-variable, two-variable, etc.) and comapares it with a random classification model. A hypothesis test is applied to see if all the decision tree models have a significantly higher accuracy versus the random classification model.
3. Random Forest Modeling.ipynb: This notebook creates a random forest model and compares it with a random classification model (dummy classifier). A hypothesis test is applied to see if the random forest model has a significantly higher accuracy versus the random classification model.
4. Wildfire_Dataset_Variable_Methods.ipynb: Contains data processing and visualization instructions regarding adding variables to our wildfire dataset.

This repository contains the following visualization attachments:

1. V4_WDF_OFFICIAL.csv: This CSV file is the cleaned and processed database already provided. If you want to skip to the decision tree and random forest modeling, ignore the Combining_Wildfire_Climate_Data notebook.
2. ML_Model_Tables.pdf: Contains two readable tables. One table is the scikt-learn parameters applied in our code. The other table is the accuracy and p-values for each tested model.
3. dt_all.png: Graphviz image of a decision tree that factors in all our variables (AWND, PRCP, TAVG, DURING_A_DROUGHT, WATERSHD, POP_BY_COUNTY). Max depth of tree is 6.
4. dt_awnd.png: Graphviz image of a decision tree that factors in only AWND. Max depth of tree is 3.
5. dt_climate_all.png: Graphviz image of a decision tree that factors in our climate variables (AWND, PRCP, TAVG). Max depth of tree is 1.
6. dt_pop_by_county.png: Graphviz image of a decision tree that factors in only POP_BY_COUNTY. Max depth of tree is 13.

---

Datasets

Dataset #1: Spatial Wildfire Occurrence Data for the United States, 1992-2015 [FPA_FOD_20170508] (4th Edition)

This data publication contains a spatial database of wildfires that occurred in the United States from 1992 to 2015.
Project Goal: Extracted Southern California wildfires from 2000-2015.
Source: USDA

Dataset #2: CAL FIRE Incident Archive

CAL FIRE contains an archive of wildfire incidents from 2013 to present.
Project Goal: Extracted Southern California wildfires from 2016-September 2020.
Source: CAL FIRE

Dataset #3: Climate Data Online Search (Daily Summaries)

Contains climate summaries for the average temperature (TAVG), average wind speed (AWNG), and precipitation (PRCP).
Project Goal: Extracted daily climate summaries from 2000-September 2020.
Source: NOAA

Dataset #4: Population By County

Population by counties were collected from the census.
Project Goal: Acquired the population data and appended it to our wildfire database.
Source: U.S. Census Bureau

Dataset #5: California High Hazard Zones (Tier 2)

Tier Two High Hazard Zones are based on watersheds and represent areas for forest health restoration and fire planning.
Project Goal: Extracted watershed data in Southern California using ArcGIS Pro's intersection tool.
Source: CAL FIRE

Dataset #6: Weeks In Drought

The Consecutive Drought Dataset for California was collected from the National Drought Mitigation Center in the form of a CSV file.
Project Goal: Extracted the counties in our study area and confirmed which wildfires occurred during a drought.
Source: NDMC

Note: Watershed, drought, and population data was appended to our merged wildfire and climate database using Microsoft Access. Refer to Wildfire_Dataset_Variable_Methods.ipynb for further instructions on this data processing step.