AUTHOR=Ding Haoxi , Hu Wei , Zhu Hongfen , Bi Rutian TITLE=Spatial Scaling Effects to Enhance the Prediction for the Temporal Changes of Soil Nitrogen Density From 2007 to 2017 in Different Climatic Basins JOURNAL=Frontiers in Ecology and Evolution VOLUME=10 YEAR=2022 URL=https://www.frontiersin.org/journals/ecology-and-evolution/articles/10.3389/fevo.2022.848865 DOI=10.3389/fevo.2022.848865 ISSN=2296-701X ABSTRACT=

Soil nitrogen density (SND), which is influenced by environmental factors operating at different spatial scales and intensities, is critical for agricultural production and soil quality. Although the spatiotemporal distribution of top-layer SND has been well explored, the scale effects of environmental factors on the temporal changes of SND (SNDT) are poorly studied, which might promote the predictive accuracy of SNDT. Thus, SNDT during a certain period was calculated to explore the multiscale effects of environmental factors on it. In the study, three sampling transects under the basins of warm-temperate, mid-temperate, and warm-temperate zones were established with 200 km long and 1 km intervals to explore the spatial variation of SNDT, examine the multiscale effect of environmental factors on it, construct the predicting models based on its scale-specific relations with environmental factors, and validate the models in each basin or in other climate-zone basins. The results indicated that the increment of SND during a certain period was the greatest in the mid-temperate basin, and the variation of SNDT was ranked as cool-temperate > mid-temperate > warm-temperate basins. Under different soil types, the spatial characteristics of SNDT were different in different climate-zone basins, but the average SNDT under cropland was the greatest in each basin. Considering the influencing factors (climatic, topographic, and vegetation factors), they had controls on SNDT operating at different spatial scales. In regard to the prediction of SNDT, the method of partial least square regression (PLSR) combined with a multiscale analysis was found to be more preferable for dependent SNDT prediction than the traditional method of stepwise multiple linear regression but could not be validated for the independent validation data in other basins. Thus, the spatial multiscale relations of SNDT with environmental factors could provide more information for each basin, and the integration of the extra information decomposed by wavelet transform into the method of PLSR could enhance the SNDT prediction for dependent datasets. These findings are of great significance for future studies in the spatial modeling of SND temporal dynamics under the influence of environmental changes.