ForensX - Space Apps Challenge

ForensX| Automated Detection of Hazards

Awards & Nominations

ForensX has received the following awards and nominations. Way to go!

Global Nominee

Automated Detection of Hazards

Countless phenomena such as floods, fires, and algae blooms routinely impact ecosystems, economies, and human safety. Your challenge is to use satellite data to create a machine learning model that detects a specific phenomenon and build an interface that not only displays the detected phenomenon, but also layers it alongside ancillary data to help researchers and decision-makers better understand its impacts and scope.

Bloom Early Detection (BED) Model

Summary

Natural phenomena occur on a global scale and can have major lasting impacts on local and global ecosystems and economies. Harmful algal blooms (HABs) have a large economic impact on the United States. Approximately $900 million are lost each year just on health and illness costs [1]. Similarly, revenue suffers in tourism, fishing, shipping, dining, property values, and more. HABs are getting worse every year. In the last three decades, HABs have gotten more severe and more frequent across the United States [2].

How We Addressed This Challenge

Being able to forecast and act on HAB development prior to major growth is a cornerstone in an effective algal bloom mitigation response. We developed the Bloom Early Detection model and dashboard to enable early projection and detection of HAB growth. Our model is capable of predicting a HAB outbreak in advance using advanced machine learning modeling and visualizes predictions on an easy to use online dashboard. We hope government organizations like NASA and NOAA can use our tool to drive efforts in combating HAB growth.

How We Developed This Project

What inspired your team to choose this challenge?

We were inspired by NOAA's efforts in predicting forecasts of HAB growth but were dismayed by the limited accessibility and transparency for the general public. We wanted to develop a dashboard for identifying HAB growth with a focus on ease of use and accessibility.

What was your approach to developing this project?

Our project looks to forecast HAB growth via Chlorophyll-a concentrations, as seen from space.

To do so, we had to procure data from various NASA portals. Data was acquired from NASA's MODIS Aqua satellite and processed to extract Chlorophyll-A concentration and Sea Surface Temperature. Dates collected ranged from January 2019 to August 2020 at a temporal resolution of 8 days and a spatial resolution of $4km$. We focused on the Great Lakes to validate a proof of concept. This allows us to gain a better understanding of the Great Lakes' highly variable Chlorophyll-a concentration dynamics. Later, using the GeoPandas Python library (open source), the points were selected using a shapefile of each lake provided by the USGS via a technique called a point-in-polygon test. As we used MODIS Aqua low resolution recordings, temporal dynamics were sparse. To combat this, we performed a non-linear cubic interpolation to increase our temporal resolution.

To enable HAB forecasting via Chlorophyll-a metrics, we employed the Prophet time series model. Prophet is an open-source and scalable time series model from Facebook. Prophet can handle multiple variables robustly. We leveraged this powerful multivariate function by feeding the model location-specific Chlorophyll-A data and some ancillary data. For our proof of concept, we utilized Sea Surface Temperature data. This ancillary data can be easily scaled to study additional relevant environmental and societal dynamics.

To power our easy to use and interactive dashboard, we opted to use the React framework for web development. We used the Deck.gl platform for big data geospatial visualizations. Deck.gl can handle millions of points on the browser in real time, making it an excellent choice for its scalability. Each location is plotted on an interactive map, showing the risk and other relevant statistics for HAB growth.

What problems and achievements did your team have?

Our team achieved writing a scalable, multivariate forecasting model for predicting Chlorophyll-a concentrations for a given location (ie. Great Lakes). We were able to utilize the Deck.gl library to power big data visualization, enabling us to very easily add additional locations, and, in the future, potentially every waterbody in the United States.

Our team faced many challenges in the data procurement and processing tasks. Specifically, MODIS Aqua provides extremely high resolution data, producing almost 100gb of data and making it impossible for us to analyze our data in a realistic time frame. This was especially pronounced when performing point-in-polygon tests. In the future, this can be circumvented using GPU technologies to parallelize the point-in-polygon problem.

How We Used Space Agency Data in This Project

From NASA oceancolor.gsfc.gov:

MODIS-Aqua Sea Surface Temperature
MODIS-Aqua Photosynthetically Available Radiation
MODIS-Aqua Chlorophyll-A concentration

From NASA Earth Observations (NEO):

To model HAB growth, we procured Chlorophyll-A concentration and Sea Surface Temperature data from Aqua MODIS for our desired waterbodies. This data was then used as an input in our machine learning model.

Our main challenge during the hackathon was procuring data — our initial plan involved using the raw 0.0125 degree spatial high resolution of the MODIS Aqua satellite. However, this yielded too many data points to be processed and refined, estimated to take around 7 days. Because of this, we chose to use the data offered by the NASA Earth Observations (NEO) portal that had preprocessed the data into manageable chunks at a slightly lower spatial and temporal resolution. This influenced our design, in that it shaped how many waterbodies we chose to study and the timescale over which this study was feasible.