Abstract
Comprehensively projecting global fertilizer consumption is essential for providing critical datasets in related fields such as earth system simulation, the fertilizer industry, and agricultural sciences. However, since previous studies have not fully considered the socioeconomic factors affecting fertilizer consumption, huge uncertainties may remain in fertilizer consumption projections. Here, an approach ensembled six machine learning algorithms was proposed in this study to predict global fertilizer consumption from 2020 to 2100 by considering the impact of socioeconomic factors under shared socioeconomic pathway (SSP) scenarios. It indicates that the proposed approach provides a rational and reliable framework for fertilizer consumption prediction that stably outperforms the single algorithms with relatively high accuracy (Nash-Sutcliffe efficiency of 0.93, Kling-Gupta efficiency of 0.89, and mean absolute percentage error of 10.97 %). We found that global N and P fertilizer consumption may decrease from 2020 to 2100, while K fertilizer may buck the trend. N fertilizer consumption showed a declining trend of −1 %, −17.13 %, and −3.43 % under the SSP1, SSP2, and SSP3 scenarios in 2100, respectively. For P fertilizer, those were −0.68 %, −9.68 %, and −2.03 %. In contrast, global K fertilizer consumption may increase by 18.03 %, 9.18 %, and 6.74 %, respectively. On average, N, P, and K fertilizer consumption is highest in China, and the lowest is in Kazakhstan. However, the hotspots of N fertilizer consumption may shift from China to Latin America and the Caribbean. This study highlighted the ensemble machine learning approach could potentially be a robust method for predicting future fertilizer consumption. Our prediction product will not only contribute to a better understanding of global fertilizer consumption trends and dynamics but also provide flexible and accurate key data/parameters for related research. The Projected Global Fertilizers Consumption Datasets are available at doi:https://doi.org/10.5281/zenodo.8195593 (Gao et al., 2023).
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have