Machine Learning Predicts Nutrient Concentrations in Tropical Rivers
―Implications for coastal ecosystem conservation efforts―

JIRCAS has created a new model to accurately predict the concentration of nutrients (nitrogen, phosphorus, and silicon) in river water on Ishigaki Island, Okinawa Prefecture, using a machine learning method.

This model predicts riverine nutrient concentrations based on the watershed characteristics such as land use and surface geology using the Random Forest machine learning algorithm. It can also identify the watershed characteristics that have significant impacts on each nutrient. While most of the conventional simulation models target nitrogen and phosphorus, this model can predict the concentration of silicon as well, making it possible to assess the risk of coral reef decline due to excessive nitrogen and phosphorus loads and that of harmful algal blooms such as dinoflagellates due to reduced silicon. 

While estimation and prediction of terrestrial nutrient loadings usually rely on the efforts and expertise in conducting hydrologic and water quality monitoring and operating mathematical simulation models, this model makes it possible to predict nutrient concentrations with relatively simple operations provided that data on land use, surface geology and population density in the target watersheds are available.

As well as the changes in marine environment due to climate change, there are growing concerns about the adverse effects on coral reefs and other coastal ecosystems around tropical islands caused by excessive nutrient inputs from lands. The results of this study are expected to aid in developing policies to conserve healthy coastal ecosystems by appropriately controlling terrestrial nutrient loadings.

The results of this study were published in the electronic edition of Environmental Pollution on November 5, 2022 (JST).

Authors

T Kikuchi, T Anzai, T Ouchi, K Okamoto and Y Terajima

Title

Assessing the impact of watershed characteristics and management on nutrient concentrations in tropical rivers using a machine learning method

Journal

Environmental Pollution
DOI : https://doi.org/10.1016/j.envpol.2022.120599

JIRCAS President　KOYAMA Osamu

Program Director：

HAYASHI Keiichi (Environment Program)

Principal Investigators：

KIKUCHI Tetsuro (Crop, Livestock and Environment Division)　
ANZAI Toshihiko (Tropical Agriculture Research Front)

Press Coordinator：

OMORI Keisuke (Head, Information and Public Relations Office)
Press e-mail：koho-jircas@ml.affrc.go.jp