Zelin Li

We constructed base model, dummy variable model, and mixture model with three variables including knot diameter, loose knot length, and sound knot length with three typical coniferous species, Pinus koraiensis, Larix olgensis, and Pinus sylvestris var. mongolica, from the Linkou Forestry Bureau and Mengjiagang forest farm in Heilongjiang Province in 2020. We analyzed the differences in knot properties among different tree species and simplified the modeling work. Firstly, we collected relevant knot property data through the sectioning method based on relevant literature, transformation of the model form and substitution of related variables to conduct a base model. We transformed the species into dummy variables as qualitative factors, and introduced the dummy variable model of the relevant attributes into the base model. We introduced the random effects of sample trees and sample plots when constructing the mixture model. By comparing evaluation indicators, such as Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), the mixture model with the best fitting effect was selected. We selected the optimal universal equation by comparing the fitting accuracy of the base model, dummy variable model and mixture model. The fitting accuracy of the dummy variable model and mixture model was higher than that of the basic model. The evaluation indicators (AIC and BIC) showed that the mixture model had a better fitting effect on knot properties than the dummy variable model. In the model comparison results, R² of mixture models for sound knot length, the loose knot length, and knot diameter increased by 13.2%, 84.8% and 40.3%, respectively. The predictive accuracy of the three base models for different tree species' knot attributes was above 90%, and both the prediction accuracy of the dummy variable model and mixture model were above 94%, indicating that the constructed models could well predict knot-related properties. From the perspective of tree species, the sound knot length, knot diameter, and loose knot length was in order of P. sylvestris var. mongolica > P. koraiensis > L. olgensis. Fitted results of the dummy variable model and the mixture model were superior to the basic model, with higher accuracy.

The l₂-norm minimization is a common means for the 3D inversion of gravity data. The unconstrained l₂-norm inversion will produce a smooth solution, which contains redundant structures and artifacts. Positivity-constrained l₂-norm inversion can eliminate redundant structures and artifacts, resulting in a more reliable solution. However, the positivity constraint restricts the applications of gravity inversion to some extent because the measured gravity data are likely to be caused by both positive and negative sources. To address this issue, we propose a strategy that combines the l_p-norm regularization and fine adjustment of the depth weighting function to refine the unconstrained gravity inversion results. Synthetic tests show that the proposed strategy yields an improved smooth solution compared with the unconstrained l₂-norm inversion method. The proposed strategy is also applied to the inversion of gravity data collected over a Layikeleke iron–copper skarn deposit, Xinjiang, China.

Euler deconvolution is a popular technique used for analyzing potential field data because it requires little prior information. However, the reliability of Euler solutions can be impacted by interference from adjacent field sources, or background fields. In this manuscript, we present an effective Euler deconvolution algorithm that accounts for linear background fields. Our algorithm, called improved finite-difference Euler deconvolution, builds upon the finite-difference method and is less susceptible to interference from nearby sources. We use this algorithm to achieve a joint estimation of the coefficients of the source coordinates, the structure index, and the linear background trend. Compared to Euler deconvolution methods based on differential similarity transformations, which also account for linear background fields, our method is easier to understand and implement programmatically and is faster. We tested our method using both 2D and 3D synthetic data, and the results indicate that our algorithm has better computational accuracy than the finite-difference algorithm and is comparable to the Euler deconvolution algorithm based on differential similarity transformations. In addition, our method was shown to be effective when tested on real data.