Illustration by Susana Ortiz

Words by Jack Riewe

‍

We often say you can use machine learning for anything—you just have to know what kind of questions to ask. The questions aren’t limited to user behavior, but also data on our environment. 

The wildfires in Brazil—specifically the Amazon rainforest, have been under close watch in recent years as the rainforest experiences threats from deforestation and decreasing biodiversity. Because wildfires are unfortunately so common, humans have collected a good amount of data from them and can better understand the patterns they make and the relationships they have with the regions they are most likely to spark in.

Can Machine Learning Predict Wildfires?

Of course. With the right data, and the right machine learning solution, you can predict wildfire damage, but what kind of questions should we be asking this data and how can we apply it to reverse the damage they cause?

This is a really exciting use case for us, because it’s our first on environmental ML. We think this is relevant for those who want a no-code option for better understanding natural disasters. 

This post will be beneficial for:

• Those who want to self-educate or make their own wildfire predictions.
• Land managers who wish to best prepare for wildfires.
• Nonprofits looking to allocate aid to Brazilian states prone to wildfires.

Before we get into the use case, we’d like to say Obviously AI’s thoughts are with those affected by the wildfires in Australia. We are unnerved by the news and hope for a fast recovery. If you'd like to donate to the Red Cross disaster relief and recovery fund, click here.

About the Dataset

We will be using a dataset on wildfires in Brazil from 1998 to 2017. The data was obtained from the official website of the Brazilian government.

With this dataset, we will find patterns in the time of year and state related to the number of wildfires. The columns represent the year the forest fire happened, the Brazilian state, the month the forest fire happened, the number of forest fires reported, and the date they were reported. 

Our Analytic and Prediction Process

If you don’t know by now, we’re a no-code data science tool. We have taken the traditional time-consuming machine learning process and simplified it into 3 steps. Essentially, we’re following the same process as this Kaggle Notebook—only without code.

With this dataset, we can upload into our platform, ask questions, and get an algorithmic output in English.

Comparatively the traditional process of preparing and importing data would look like this:

The Queries We’ll Ask to Predict and Analyze Wildfire Patterns

Since we cut out SQL queries, we can ask our datasets questions in our natural language. We encourage you to try these questions yourself on our platform! 

The questions we’ll ask for this use case:

1. What Are the Predicted Number of Wildfires in Brazil in 2020?
2. What Month Will be the Worst for Wildfires in Brazil?
3. What States Have the Highest Number of Wildfires?
4. Has the Average Number of Wildfires Increased in Brazil Year-Over-Year?

Exploring the Results

What Are the Predicted Number of Wildfires in Brazil in 2020?

According to the data, São Paulo is the most wildfire prone in 2020 with the predicted number to be just above 208 wildfires in 2020. Coming in next are Mato Grasso and Bahia. 

As you might have guessed, the other top drivers by impact on the number of wildfires include the time of year. 

What Month Will be the Worst for Wildfires in Brazil?

According to the data, October will have the highest number of wildfires with all of Spring being a wildfire-prone time of year.

What States Have the Highest Number of Wildfires?

Rio has had the highest number of wildfires. Because the dataset had repeating values, I simply chose “unique occurences” in the drop down menu to the left of the graph in our platform. 

‍

We can even ask our dataset if the number of wildfires has increased over the years.

And get this line graph:

Takeaways

Using these outputs, we can map out where in Brazil wildfires are concentrated.

‍

From the dataset we can also conclude:

• The average number of wildfires over the years has a general upward trend year-over-year. 
• The São Paulo and Rio de Janeiro regions are the most prone to wildfires and generally need the most emergency relief.
• Spring in Brazil is the most wildfire-prone season in Brazil as the temperature rises into the summer months.

If you would like to explore this dataset yourself you can download it here and request a demo to explore Obviously AI’s platform to analyze and make predictions from the data without code. If you want some more ideas of what to explore with data related to wildfire damage, see how the number of wildfires are increasing with temperature or how much emergency relief and fire precautions fire-prone regions are taking to combat these natural disasters.

‍