Guangdong develops an explainable four-dimensional prediction model for soil heavy metals in space and time

Recently, Wang Qi, a researcher at the Institute of Eco-Environment and Soil Science, Guangdong Academy of Sciences, and others, with the support of the Guangdong Outstanding Youth Fund and the National Natural Science Foundation of China, independently developed a geographically explainable soil heavy metal spatiotemporal four-dimensional prediction model that goes beyond the “black box” of artificial intelligence. Related research papers were published in Hazardous Materials. Wang Qi is the first author of the paper, and Li Fangbai is the corresponding author.

Evaluating the temporal and spatial changes of regional soil heavy metal content and predicting the future trend of soil heavy metal content is an important prerequisite and basis for effective prevention and control of soil pollution. With the continuous maturity of artificial intelligence technology, the integration of artificial intelligence technology and prediction model can help improve the quantitative level of soil heavy metal prediction and management, which can not only improve the scientific decision-making, but also bring new opportunities for the precision and intelligence of soil pollution prevention and control.

Conceptual framework of soil heavy metal spatiotemporal four-dimensional prediction model. Photo courtesy of the research team

Given the paucity of current historical data on soil geochemistry, it is extremely difficult to establish future spatial prediction models of soil heavy metal content by using multi-period time series data of atmospheric and water content. At the same time, a large number of studies have shown that the more accurate the prediction results, the worse the interpretability of the model, and most of the models with higher accuracy have complex and changeable internal structures, which cannot be intuitively understood. Therefore, how to find a balance between the interpretability and accuracy of predictions is the difficulty and focus of AI prediction.

Using the idea of machine learning ensemble learning, the researchers organically combine the advantages of multiple excellent single machine learning and deep learning models in stability and accuracy in series and parallel to form a strong combinatorial prediction model, and construct a spatiotemporal prediction model of soil heavy metal content based on the deep integration of explainable ensemble learning model and soil heavy metal geospatial source-sink mechanism. It can more accurately predict the future spatiotemporal situation and evolution trend of soil heavy metal content, including quantitative estimation of the impact of interaction between drivers on soil heavy metal content and cross-scale collaborative prediction of source-sink relationship from sample scale to regional scale.

The prediction accuracy of the model is 93.8% and the spatial resolution is 1 km, which brings new insights and inspiration for the development of methods for deciphering the complex spatiotemporal mechanism between soil heavy metal source-sink process and predicting the spatiotemporal pattern of soil pollution. (Source: China Science News Zhu Hanbin)

Related paper information:

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button