Background

A phytoplankton bloom or harmful algal blooms (HABs) are defined as a "high concentration of phytoplankton in an area, caused by increased reproduction"; [this] often produces discoloration of the water" (Garrison, 2005). They are a predictable seasonal occurrence, linked to heavy rainfall, or to increased nutrient loading from human activities. Rising summer surface temperatures of lakes will likely increase algal blooming by 20% over the next century (O'Reiley).

Our Solution

We propose Lost Short-Term Memory (LSTM) model, a state of the art deep learning model for time series forecasting. Our model predicts the occurrence of HABs with given confidence. Our service enables policy makers to preempt algae’s projected growth of 20% over the coming years.

Our presentation can be found here: https://drive.google.com/file/d/1WTVUQFKNMDg3HyreaO8B3GUoJ-V9_qWN/view?usp=sharing

In the presentation, you can look at our proposed LSTM model. Unlike other existing predictive model available in the market, we are simultaneously using the current environmental changes parameters (water temperature, wind speed etc.) and the past algae concentration data extracted from NASA's chlorophyll data satellite images to predict the likelihood of algae's growth. The idea here is that current environmental changes parameter account for all the causes of HABs we are currently aware of, and past algae data fills up the gap for the parameters which are unknown to human as they don't affect HABs production directly (mystery behind HABs growth). When these features are concatenated together and fed into our LSTM model, the backpropagation algorithm minimized the difference between our prediction and actual HABs data (minimizing L2 Loss), thus learning the complex mapping between our input and output features (encoding solves the mystery!).
A similar solution has been shown to work for stock market analysis but no prior research exists for prediction of HABs most likely due to relatively less popularity of the given topic.

Demo

We have made a demo of how this deep learning based program would work. In the video shown below, the client wants to see how likely it is for an algal bloom to grow in Hawaii. After the client types Hawaii and clicks search, a map appears and zooms in to Hawaii meanwhile it calculates the water temperature and wind speed. These are the two parameters that mainly affect the growth of algal bloom. In Hawaii the water temperature is too high and the wind power is too strong. Therefore the program predicts that it is unlikely for algal bloom to grow.

https://drive.google.com/open?id=1ihhjrS7HrD4ujbeg...

The code for the demo can be found here: https://github.com/gogeorge/al_bloom.

The above code shows the front-end part of our service. Currently, the front-end part is not connected with our back-end deep learning model because the proposed LSTM model requires an intensive amount of training not otherwise possible in a local machine without any GPU support. We aim to integrate real-time prediction in future with the availability of proper resources and funding.

The Future Product

The final model will predict where algal blooms will occur using a deep learning model trained on NASA's data, as mentioned previously. Currently, we are considering the factors of water temperature and wind speed in addition to the chlorophyll data to train the DL model. In future iterations, we aim to have the parameters be adjusted according to the client's needs, since each case is unique.

Furthermore, more useful features will be added to the program as well as a freemium plan where the basic plan will be free for the clients but the premium plans will be aimed toward larger organizations and municipalities with local algal blooms.

References

[1] Anderson, D. M., Hoagland, P., Kaoru, Y., & White, A. W. (2000). Estimated annual economic impacts from harmful algal blooms (HABs) in the United States (No. WHOI-2000-11). National Oceanic and Atmospheric Administration Norman OK National Severe Storms Lab.

[2] US Department of Commerce, & National Oceanic and Atmospheric Administration. (2014, August 1). Why do harmful algal blooms occur? Retrieved October 19, 2019, from https://oceanservice.noaa.gov/facts/why_habs.html.

[3] O'Reilly, C. M., Sharma, S., Gray, D. K., Hampton, S. E., Read, J. S., Rowley, R. J., ... & Weyhenmeyer, G. A. (2015). Rapid and highly variable warming of lake surface waters around the globe. Geophysical Research Letters, 42(24), 10-773.

[4] Lee, S., & Lee, D. (2018). Improved prediction of harmful algal blooms in four Major South Korea’s Rivers using deep learning models. International journal of environmental research and public health, 15(7), 1322.