In this post we’ll continue our analysis of wildfire activity using spatial analytics techniques. In part 1, we downloaded historical wildfire information for the years 2002-2016 from a USGS ArcGIS Server map service for the United States. Each year was exported to an individual feature class containing wildfires for that particular year only. After downloading the data we appended the annual wildfire information into a single feature class containing all wildfire information for the 2002-2016 time period.
Today’s post will continue to massage the dataset and we’ll also define our study parameters, and perform some basic analysis. Future articles will focus on using various spatial analytics tools and the R programming language to analyze the data in support of our study parameters.
Part of the analysis phase of this project will be to create various hot spot maps. Hot spot maps can be created from point or polygon feature classes. We’ll create hot spot maps from both at various points of the analysis, but before doing so we’ll need to aggregate the points to a polygon layer. In this section of the article we’ll download a US Counties polygon feature class that will be used in the aggregation. We’ll also project the datasets to a projected coordinate system instead of a geographic coordinate system which is necessary for the spatial statistics tools. We’ll also do a little data cleanup.
- When distance is a component of your GIS analysis, which is most often the case when dealing with spatial statistics tools, you need to project your data using a projected coordinate system rather than a geographic coordinate system. Right now the HistoricalWildfires feature class is stored in a geographic coordinate system. Use the Project tool to project the data to a USA_Contiguous_Alberts_Equal_Area_Conic projection with an output feature class name of HistoricalWildfiresPrj and store it in the USGS Wildfire Data geodatabase. In this coordinate system, all areas are proportional to the same areas on earth, and distance is most accurate in the middle latitudes where many of the states in the lower 48 are located.
- During the analysis phase of this project we’re going to use a U.S. Counties feature class for part of our analysis. We’ll spatially join the HistoricalWildfiresPrj feature class to the county feature class. Esri provides a layer containing these counties. Download this layer from Esri, select the lower 48 states, and export them to a new layer called USCounties_Lower48. Project the USCounties_Lower48 feature class to the USA_Contiguous_Alberts_Equal_Area_Conic projection as a layer called USCounties_Lower48Prj and save it in the USGS Wildfire Data geodatabase.
- Right click the USCounties_Lower48Prj layer in the ArcMap table of contents and select Joins and Relates | Join. Spatially join the layer to the HistoricalWildfiresPrj layer to create a new feature class called USCountiesJoined in the USGS Wildfire Data geodatabase. This will create a new field called Count_ in the USCountiesJoined feature class. The Count_ field contains a count of the number of wildfires that occurred in each county. You can see this is the screenshot below.
- Some counties haven’t had any wildfires during the study period. The Count_ field will contain <Null> values for these records. Use the Field Calculator to set the <Null> values to 0.
Defining the Parameters of the Study
For this study we’re going to focus on the analysis of wildfires from the years 2002-2016 in the United States. We’ll examine the data from a national perspective as well as by individual state. In particular, here are some of the questions that we are going to attempt to answer (we made add more in the future):
- Where do wildfires cluster?
- Are human induced wildfires concentrated in particular areas (what about naturally occurring wildfires)?
- What are the temporal variations in wildfires?
- In what areas are the largest wildfires concentrated?
- What areas of the country are similar to others with respect to the occurrence of wildfires?
We’ll examine some of these questions in this article, and others in future articles.
Some Basic Analysis
We’ll start by doing some basic analysis on the dataset. The HistoricalWildfiresPrj feature class is a point feature class containing the location of each wildfire from our study period. The attribute table for this feature class contains a firecause field, which is a one character code indicating the cause of the fire. For now we’re going to create some basic maps and analysis without looking at the underlying cause of the fire. Later we’ll break the fires out into categories based on whether the fire was started by natural or human events.
Color Coded Map of Wildfire by US County
- Using the Count_ field in the USCountiesJoined layer, create a color coded map that is normalized by the SQMI field. Normalization takes the size of the county out of the equation. Otherwise, counties with a larger area will naturally have more fires, all other variables being equal. I’ve provided a couple screenshots below to show you how this is done as well as the result. This reveals some fairly obvious patterns of high wildfire activity in the western United States including California, Oregon, and Idaho along with Oklahoma, the Appalachian region, and Florida. However, ArcGIS includes a number of spatial statistics tools that we can use to get a more accurate picture of wildfire activity.
Hot Spot Map of Wildfires by US County
- In the next step we’re going to create a hot spot map using the same data. One of the fields the Hot Spot Analysis tool needs is an input field. Create a new field called NormCount (Double data type) in the USCountiesJoined feature class. This will hold the results of the normalization we used in the creation of the color coded map in the last step.
- Use the Field Calculator to calculate the values of this field to be the Count_ field divided by the SQMI field.
- Find the Hot Spot Analysis tool in the Spatial Statistics toolbox and open it. Input the parameters shown in the screenshot below. There are many options for determining the Conceptualization of Spatial Relationships parameter. For more information on the Hot Spot Analysis tool and the various input parameters please take a look at my video on this subject.
- Run the tool and you should see the results shown in the screenshot below. You’ll see some distinct concentrations of wildfire hot spots in various locations throughout the country for the years 2002-2016. You may want to re-run the Hot Spot Analysis tool a number of times with different input parameters for Conceptualization of Spatial Relationships to see how this parameter changes the output.
Cluster and Outlier Analysis of Wildfires
In this section we’ll use the Cluster and Outlier Analysis tool to identify counties that have a high incidence of wildfires even though the surrounding counties have a low incidence of wildfires, and vice versa. The Cluster and Outlier Analysis tool should be run anytime you run Hot Spot Analysis. Outliers can be extremely important when examining many types of problems. When examining wildfire activity this tool would enable us to find areas with high wildfire activity that are surrounded by areas of low wildfire activity and vice versa. It attempts to find values that are different from their neighbors and is an underutilized tool.
- Find the Cluster and Outlier Analysis tool in the Spatial Statistics toolbox and open it. Input the parameters shown in the screenshot below.
- Click OK to execute the tool. The results of the analysis should appear as seen in the screenshot below. The counties in dark red and blue are the outliers. The dark blue counties indicate areas of low wildfire activity in close proximity to areas of high wildfire activity. The dark red county indicates an area of high wildfire activity in close proximity to counties of low wildfire activity.
Human Induced Wildfire Hot Spots
In this section we’ll dive a little deeper into the data and examine the distribution of human caused wildfires.
- The HistoricalWildfiresPrj feature class includes a firecause attribute field with a single character to indicate the cause of the fire. A value of ‘H’ indicates a human caused wildfire. Use the Select by Attributes tool to select these features. The selection set includes 15,400 human caused wildfires.
- Export the selected set to a feature class called HumanCausedWildfires in the USGS Wildfire Data geodatabase.
- Spatially join the HumanCausedWildfires feature class to the USCounties_Lower48Prj feature class to create a new feature class called USCountiesJoined_HumanCaused.
- This will create a Count_ field in the USCountiesJoined_HumanCaused feature class. There will be some Null values that you’ll need to recalculate to 0.
- This value needs to be normalized by the population in each county. Create a new field called NormCount as a double data type and populate it by dividing Count_ by the Population field (the Population field is an alias for the POP2015 field.
- Create a color coded map using the NormCount and you should see a map similar to that shown in the screenshot below. What does this map reveal? Not a whole lot, but it does hint at the possibility that human caused wildfires are perhaps higher in Oklahoma, Montana, and Idaho. Let’s run the Hot Spot Analysis tool to see if the patterns are more apparent.
- Run the Hot Spot Analysis tool using the USCountiesJoined_HumanCaused feature class as the input feature class, NormCount as the input field, and Contiguity_Edges_Corners as the Conceptualization of Spatial Relationships parameter. The output should appear as seen in the screenshot below. Now the patterns showing hot spots of human caused wildfires are evident.
In the next article we’ll investigate wildfire activity at the state rather than the national level.
More information on Spatial Analytics: