Landslide susceptibility mapping (LSM) is a commonly used approach to reduce landslide risk. However, conventional LSM methods generally only consider the influence of each single conditioning factor on landslide occurrence or absence, which neglects the interactions of different conditioning factors and may lead to biased LSM results. Therefore, this study aims to use a new machine learning model—attentional factorization machines (AFM)—to explicitly consider the influence of feature interactions in LSM to improve and obtain more reliable LSM results. The Anhua County in China is chosen as the study area. The area under the receiver operating characteristic curve (AUC) and statistical indicators are used to evaluate the performance of LSM models. For comparison, the common LSM models such as the logistic regression (LR) and random forest (RF) models are also used to conduct the LSM. The results show that the performance of AFM is a little better than RF in the AUC metric, whereas the LR model has the worst performance. Compared with general LSM models, AFM considers feature interactions by introducing an attention mechanism to learn the weight of different feature combinations, which not only ensures the model interpretability but also improves the model performance.

The visible and near-infrared (VNIR) spectroscopy prediction model is an effective tool for the prediction of soil organic matter (SOM) content. The predictive accuracy of the VNIR model is highly dependent on the selection of the calibration set. However, conventional methods for selecting the calibration set for constructing the VNIR prediction model merely consider either the gradients of SOM or the soil VNIR spectra and neglect the influence of environmental variables. However, soil samples generally present a strong spatial variability, and, thus, the relationship between the SOM content and VNIR spectra may vary with respect to locations and surrounding environments. Hence, VNIR prediction models based on conventional calibration set selection methods would be biased, especially for estimating highly spatially variable soil content (e.g., SOM). To equip the calibration set selection method with the ability to consider SOM spatial variation and environmental influence, this paper proposes an improved method for selecting the calibration set. The proposed method combines the improved multi-variable association relationship clustering mining (MVARC) method and the Rank–Kennard–Stone (Rank-KS) method in order to synthetically consider the SOM gradient, spectral information, and environmental variables. In the proposed MVARC-R-KS method, MVARC integrates the Apriori algorithm, a density-based clustering algorithm, and the Delaunay triangulation. The MVARC method is first utilized to adaptively mine clustering distribution zones in which environmental variables exert a similar influence on soil samples. The feasibility of the MVARC method is proven by conducting an experiment on a simulated dataset. The calibration set is evenly selected from the clustering zones and the remaining zone by using the Rank-KS algorithm in order to avoid a single property in the selected calibration set. The proposed MVARC-R-KS approach is applied to select a calibration set in order to construct a VNIR prediction model of SOM content in the riparian areas of the Jianghan Plain in China. Results indicate that the calibration set selected using the MVARC-R-KS method is representative of the component concentration, spectral information, and environmental variables. The MVARC-R-KS method can also select the calibration set for constructing a VNIR model of SOM content with a relatively higher-fitting degree and accuracy by comparing it to classical calibration set selection methods.

Under the influence of complex environmental conditions, the spatial heterogeneity of soil organic matter (SOM) is inevitable, and the relationship between SOM and visible and near-infrared (VNIR) spectra has the potential to be nonlinear. However, conventional VNIR-based methods for soil organic matter estimation cannot simultaneously consider the potential nonlinear relationship between the explanatory variables and predictors and the spatial heterogeneity of the relationship. Thus, the regional application of existing VNIR spectra-based SOM estimation methods is limited. This study combines the proposed partial least squares–based multivariate adaptive regression spline (PLS–MARS) method and a regional multi-variable associate rule mining and Rank–Kennard-Stone method (MVARC-R-KS) to construct a nonlinear prediction model to realize local optimality considering spatial heterogeneity. First, the MVARC-R-KS method is utilized to select representative samples and alleviate the sample global underrepresentation caused by spatial heterogeneity. Second, the PLS–MARS method is proposed to construct a nonlinear VNIR spectra-based estimation model with local optimization based on selected representative samples. PLS–MARS combined with the MVARC-R-KS method is illustrated and validated through a case study of Jianghan Plain in Hubei Province, China. Results showed that the proposed method far outweighs some available methods in terms of accuracy and robustness, suggesting the reliability of the proposed prediction model.

Flows are usually represented as vector lines from origins to destinations and can reflect the movements of individuals or groups in space and time. Revealing and analyzing the spatiotemporal flow patterns are conducive to understanding information underlying movements. This paper proposes a new method called the OD – EOF (Origin – Destination – Empirical Orthogonal Function) to discover important spatiotemporal flow patterns on the premise of maintaining the pairwise connections between origins and destinations. We first construct a spatiotemporal flow matrix that contains connection information between origins and destinations and temporal flow information by adding a temporal dimension to the OD map. Then, we decompose the spatiotemporal flow matrix into spatial modes and corresponding time coefficients by EOF decomposition. The decomposition results depict the prominent spatial distribution of and temporal variation in flows, with most of the spatiotemporal characteristics highly concentrated into the first few spatial modes. The method is evaluated by five synthetic datasets and a user study and subsequently applied to analyze the impact of the COVID-19 pandemic on the spatiotemporal patterns of human mobility in China during the Spring Festival travel rush in 2020 and 2021. The results show the prominent spatiotemporal patterns of human mobility during these periods under the influence of the COVID-19 pandemic outbreak and the normalization of pandemic prevention and control.