Environmetrics最新文献

This study provides a comprehensive evaluation of the computational performance of R, MATLAB, Python, and Julia for spatial and spatio-temporal modelling, focusing on high-dimensional datasets typical in geospatial statistical analysis. We benchmark each language across key tasks, including matrix manipulations and transformations, iterative programming routines, and Input/Output processes, all of which are critical in environmetrics. The results demonstrate that MATLAB excels in matrix-based computations, while Julia consistently delivers competitive performance across a wide range of tasks, establishing itself as a robust, open-source alternative. Python, when combined with libraries like NumPy, shows strength in specific numerical operations, offering versatility for general-purpose programming. R, despite its slower default performance in raw computations, proves to be highly adaptable; by linking to optimized libraries like OpenBLAS or MKL and integrating C++ with packages like Rcpp, R achieves significant performance gains, becoming competitive with the other languages. This study also provides practical guidance for researchers to optimize R for geospatial data processing, offering insights to support the selection of the most suitable language for specific modelling requirements.

Recent advances in local models for point processes have highlighted the need for flexible methodologies to account for the spatial heterogeneity of external covariates influencing process intensity. In this work, we introduce tessellated spatial regression, a novel framework that extends segmented regression models to spatial point processes, with the aim of detecting abrupt changes in the effect of external covariates on the process intensity. Our approach consists of two main steps. First, we apply a spatial segmentation algorithm to geographically weighted regression estimates, generating different tessellations that partition the study area into regions where model parameters can be assumed constant. Next, we fit log-linear Poisson models in which covariates interact with the tessellations, enabling region-specific parameter estimation and classical inferential procedures, such as hypothesis testing on regression coefficients. Unlike geographically weighted regression, our approach allows for discrete changes in regression coefficients, making it possible to capture abrupt spatial variations in the effect of real-valued spatial covariates. Furthermore, the method naturally addresses the problem of locating and quantifying the number of detected spatial changes. We validate our methodology through simulation studies and applications to two examples where a model with region-wise parameters seems appropriate and to an environmental dataset of earthquake occurrences in Greece.

Long-term exposure to ambient air pollution has previously been associated with childhood asthma, but endeavors have focused on single and pairwise pollutant models. We introduce a novel framework for selection of effect drivers from an environmental mixture, which is based on an entropy rank agreement measure. We apply the method in a nationwide study, relating prenatal exposure to ambient air pollution to asthma incidence in Danish children aged 0–19 years that are born from 1998 to 2016. Also, we estimate effects through population-wide G-estimation contrasts. We conclude that being exposed to the observed levels of ambient air pollution in contrast to the hypothetical case of the minimum of the observed subject-specific exposure levels and the 2.5% quantile levels is associated with relative risk increases that exceed 30% and absolute risk differences that exceed 2 percentage points across Danish municipalities. For selection we discover that SO$��{\hspace{0.17em}}_4^{�2�-�}�$ and primary organic aerosols appear the most important predictors of asthma amongst the included ambient air pollutants and that these are both associated with a risk increase. The developed methodology is a promising approach to handling an environmental mixture of exposures in statistical analyses, which allows for discovery of important drivers of associations.

Large-scale environmental surveys relying on intensive fieldwork are expensive, but survey sampling methodology offers several options to improve their cost-efficiency. For example, sites selected for field assessments can be arranged in clusters to reduce the time spent moving between the sites, and auxiliary data can be utilized to stratify the survey region and sample less important strata less densely. Geographically balanced and well-spread sampling can yield further improvements since the target variables of environmental surveys tend to be spatially autocorrelated. A combination of these ideas was illustrated and evaluated in the context of a national forest inventory, and alternative methods of spatially balanced sampling were compared. The main findings were that (i) both the local pivotal method and the generalized random-tessellation stratified design guaranteed a clearly better spatial regularity than systematic sampling when applied to fragmented regions resulting from stratification and (ii) they also ensured better global balance in unstratified sampling. In our case study, where stratification and sample allocation were based on high-quality auxiliary data, stratified sampling was clearly more efficient than unstratified for the primary survey target parameter. However, our results also illustrate that highly nonproportional sample allocation can be dangerous in a multi-purpose survey.

Scientists and statisticians often seek to understand the complex relationships that connect two time-varying variables. Recent work on sparse functional historical linear models confirms that they are promising as a tool for obtaining complex and interpretable inferences, but several notable limitations exist. Most importantly, previous works have imposed sparsity on the historical coefficient function, but have not allowed the sparsity, hence lag, to vary with time. We simplify the framework of sparse functional historical linear models by using a rectangular coefficient structure along with Whittaker smoothing, then reduce the assumptions of the previous frameworks by estimating the dynamic time lag from a hierarchical coefficient structure. We motivate our study by aiming to extract the physical rainfall–runoff processes hidden within hydrological data. We show the promise and accuracy of our method using eight simulation studies, further justified by two real sets of hydrological data.

While many studies on climate have focused on location shifts, none have specifically tested whether lower or upper tails of the climate data generating process have structurally changed over time. This manuscript applies a new test that can detect either distributional or tail structural change to various annual and daily U.S. climate measures. Notably, we find both distributional and tail structural change and, quite interestingly, tend to observe greater evidence in one tail versus the other for most climate measures. We also find the presence of multiple breaks. Our results imply that climate modeling, and specifically climate-crop yield modeling, should account for significant and asymmetric changes in climate distributions and not only location shifts.

Inconsistent estimation issues in the Matérn covariance function pose significant challenges to constructing confidence intervals using traditional methods. This paper addresses these challenges by employing the bootstrap method and comparing two straightforward approaches: the percentile bootstrap (PB) and the reverse percentile interval (RPI). We assess their efficacy through coverage rates and interval scores, focusing on accuracy and breadth. Theoretically, we prove that PB outperforms RPI, a claim substantiated by simulation experiments showing its superior coverage accuracy and interval scores. Moreover, the simulation results show strongly interdependent phenomena between parameters. Accordingly, by exploring the micro-ergodic parameter's impact, the study provides insights into these findings' underlying factors, particularly relevant for large spatial datasets. In the empirical study, our approach exhibits greater reliability and effectiveness in confidence interval estimation for large datasets with uniformly and non-uniformly distributed locations, as compared to several other methods. Furthermore, we applied the method to sea surface temperature data, demonstrating its strong applicability for analysis. This study provides theoretical insight and practical guidance for constructing confidence intervals, particularly in mitigating inconsistent estimation issues, especially in the context of the Matérn covariance function.

In this article, we study the problem of forecasting the next year's number of Atlantic hurricanes, which is relevant in many fields of applications such as land-use planning, hazard mitigation, reinsurance and long-term weather derivative market. Considering a set of well-known predictors, we compare the forecasting accuracy of both machine learning and classical statistical models, showing that the latter may be more adequate than the first. Quantile regression models, which are adopted for the first time for forecasting hurricane numbers, provide the best results. Moreover, we construct a new index showing good properties in anticipating the direction of the future number of hurricanes. We consider different evaluation metrics based on both magnitude forecasting errors and directional accuracy.