Analyzing the Impact of Environmental Factors on Drought Intensity Using Machine Learning

Author: Aditya Aiyer
Affiliation: James Logan High School, Union City, CA

​

Abstract:

​

A drought is a period of drier-than-normal conditions. These dry conditions can reduce the quality and quantity of water, raise illnesses and diseases, and, in turn, mortality rates. Drought is a global issue and understanding how different environmental aspects affect drought can allow for preparation and prevention in a time of crisis. The US Drought Monitor categorizes drought on a scale. First, several data visualizations were performed using a real-world environmental dataset to understand how individual factors affected drought intensity. The features were then organized into a graph from highest to lowest contribution to drought prediction. The three top and the three lowest features were plotted on violin plots to analyze how each feature varied across drought categorizations. These results highlight which environmental features should be closely monitored during drought prediction. Five different models, including a neural network, were compared to determine which would produce the highest accuracy results, and the RandomForestClassifier yielded the highest accuracy. A model was then developed that would take multiple environmental factors and predict drought intensity. A time-series split was run on the RandomForestClassifier to compare it to a DummyClassifier, which is a simple model used as a performance baseline. The results revealed that the top 3 drought predictors were surface pressure, temperature range at 2 meters, and the maximum temperature at 2 meters. In addition, the RandomForestClassifier outperformed the baseline model after 10000 lines. This study provides a comprehensive review of important environmental features and effective models for early drought warning systems.

​​

Keywords: environment, drought, permutation, time-series split, cross-validation​

​​

ISSN 3069-8200