Wildfire Risk/Danger model

Build a model to estimate the probability of wildfires occuring, using remote sensed data (data from satellite sensors).

Training data

Historical wildfire record - (FireCCI51: MODIS Fire_cci Burned Area Pixel product, version 5.1)

Environmental Variables

Previous month

• CHIRPS Daily: Climate Hazards Group InfraRed Precipitation With Station Data (Version 2.0 Final) - via Google Earth Engine
  • Reduced to monthly sum per pixel (Monthly accumulated precipitation).
  • Recommendation: Explore alternatives, the product usually takes 20 days to process but lately is taking more than a month, so it will be late to produce a risk map with the precipitation from the previous month.
• Potencial Evapotranspiration - MOD16A2 Product
  • Monthly average per pixel
  • Spatial resolution (Ground sampling distance): 1 km
• Evapotranspiration - MOD16A2 Product
  • Monthly
  • Spatial resolution (Ground sampling distance): 1 km
• Temperature MODIS - MOD11A1 Product
  • Monthly average
  • Spatial resolution (Ground sampling distance): 1 km
• NDVI MODIS - MOD13Q1 Product (NDVI Anomalies calculated from this product can be used as a proxy to drought).
  • Monthly average
  • Spatial resolution (Ground sampling distance): 250 m
• Soil Moisture - NASA USDA HSL
  • Deprecated - replaced by Enhanced SMAP
  • Monthly average
  • Spatial resolution (Ground sampling distance) 0.25 º ~ 25 km (Enhanced SMAP ~ 10 km)

Previous year (constant throughout the year)

• Distance to previous fires (Km) - calculated from FireCCI51: MODIS Fire_cci Burned Area Pixel Product, Version 5.1 . We used Dinamica Ego to calculate this variable using the following model. In our opinion, it should be replaced by a script to avoid needing to use another software.

Training data preparation

Data obtained from scripts 0 to 4 can be found here gs://geo-2021/digeo/Incendios/datos_scripts.zip (please download and don't modify the folder structure)

• 0_prepara_datos.R - downloads the product FireCCI51 from Google Earth Engine (GEE).

• 1_construir_muestra.R Using the FIRECCI51 product creates a sample of burned and not burned pixels

• 2_distancia_incendios_anios_previos.egoml - calculates distance to previous fires, using DinamicaEgo

• 3_extracion.R - extract data from the environmental variables (TODO: Check if it really does)

• 4_exploratorio_modelo.R - Creates a model using Random Forests

Inference

• To do the inference you have to download the explicative variables of the previous month using the Jupyter Notebook - 5_Datos_para_inferencia.ipynb, it's recommend running using Google Colab. These data is exported to the incendios_2021 folder in Google Drive, these layers need to be downloaded and copied to the data/correspondingmonth folder.

### Calculates variables for April, which will we used to the inference for the month of May  
iniDate = '2021-04-01'
endDate = '2021-04-30'

• 6_inferencia.R - to conduct the inference simply define the month for which the risk estimation will be calculated

mes_eval <- 5 # Month number

Note

In the last month (May 2021) precipitation (CHIRPS) and soil moisture (SMOS) are delayed in its ingestion into GEE so we instead used an average value obtained using GEE.

var geometry = ee.Geometry.Rectangle([-105.70671, 18.91105, -101.45436, 22.75942])
var m = 5  //##
var chirps = ee.ImageCollection('UCSB-CHG/CHIRPS/DAILY')
  .select('precipitation')
  .filter(ee.Filter.calendarRange(m, m, 'month'))
  .mean()

var smos = ee.ImageCollection('NASA_USDA/HSL/soil_moisture')
.select('ssm', 'susm')
.filter(ee.Filter.calendarRange(m, m, 'month'))
.mean()

Export.image.toDrive({
  image: chirps,
  description: 'precipitacion',
  scale: 250,
  crs: 'EPSG:32613',
  folder: 'incendios_2021' , 
  region: geometry
});

Export.image.toDrive({
  image: smos,
  description: 'smos',
  scale: 250,
  crs: 'EPSG:32613',
  folder: 'incendios_2021' , 
  region: geometry
});