Search

• A novel maximum volume sampling model is proposed. • A new maximum volume optimization method is developed. • The uniform sampling strategy and Givens transformation are adopted. • The validity is proved by testing both numerical and engineering examples. In this study, a maximum volume sampling model is proposed to improve the accuracy and efficiency of reliability computation. An ellipsoid is constructed with the maximum volume approach in a safe domain, and a new maximum volume optimization method is proposed. The sampling model only computes the samples outside the ellipsoid, which considerably enhances computational efficiency. Furthermore, the uniform sampling strategy and Givens transformation are adopted to efficiently solve the maximum volume optimization model. A series system example, a three-dimensional rock slope example, and an arch bridge example are tested to verify the validity of the proposed maximum volume sampling model. The results indicate that the maximum volume sampling model displays high accuracy and efficiency. [ABSTRACT FROM AUTHOR]

For hyperspectral imagery (HSI) classification, most of the studies focus on how to improve the classification accuracy, while the influence of sampling strategy for classification performance attracts little attention. For now, random sampling (RS) is the most adopted strategy. That is, for a hyperspectral image, a certain number of labeled samples are randomly selected as the training set, and the remaining labeled samples are taken as the test set. However, the RS strategy will produce over optimistic results when used for performance evaluation because of the overlap between training set and test set. Though spectral-spatial classification methods benefit most from the RS strategy, the pixel-wise classification methods can also benefit from it because of the high spectral correlation between training and test samples. However, in practical applications, the RS strategy is not feasible. Because the training and test samples are often collected from different locations. In this situation, the correlation between training and test samples will decrease dramatically and the performance of HSI classification methods will be affected. In this article, a nonoverlapped sampling method is adopted to reduce the correlation between training and test samples and different classic classification methods are evaluated. Experimental results show that the classification performance of all methods drops a lot when nonoverlapped sampling strategy is adopted. After the analysis of some important factors for HSI classification, we also propose a cotraining-based classification method to relief the influence of sampling strategy and obtains much better performance compared with those classic spectral-spatial classification methods. [ABSTRACT FROM AUTHOR]

The core collection construction strategies are critical for the conservation and utilization of germplasm resources. The combined data of phenotypic characteristics and SRAP markers in 780 samples were employed to construct a core collection for Pinus yunnanensis Franch. Based on geographical distribution, a total of four P. yunnanensis germplasm subsets were configured by improving a least distance stepwise sampling method at four sampling proportions. The representativeness of germplasm subsets for original germplasms were evaluated by quantitative and qualitative trait independence tests. For the quantitative traits test, the germplasm subset at a 30% sampling proportion was the best of four germplasm subsets by the mean t-test, the variance F-test, the frequency distribution χ2-test, the Shannon diversity index, and five effective evaluation parameters. For the qualitative traits test, there were significant differences between the original germplasm and germplasm subset at a 10% sampling proportion in the mean t-test of genetic diversity indicators. The genetic diversity within the population was significantly different between the original germplasm and germplasm subsets at 20% and 10% sampling proportions, respectively, in the variance F-test. The results of quantitative and qualitative traits tests revealed that the 30% sampling proportion was appropriate for the construction of the core collection. [ABSTRACT FROM AUTHOR]

In many processes, such as in chemical and process industries, the observations of a quality characteristic to be monitored may be correlated, if sampling intervals are short. Correlation can be modelled by considering the process mean as a random variable wandering according to an autoregressive model and the observations from the process modelled as the mean plus a random error due to short-term variability or measurement error. The sensitivity of the Shewhart control chart in the detection of a special cause is negatively affected by presence of correlation among observations. To overcome this problem, a new sampling strategy, denoted as ESSI (Equally Spaced Samples Items), is proposed to implement the Shewhart control chart as opposed to the traditional rational subgrouping approach. The ESSI sampling strategy allows observations belonging to the same sample to be collected from the process at equally spaced time intervals between two successive inspections. A numerical analysis shows that the implementation of the ESSI strategy in presence of a process wandering mean significantly improves the statistical performance of the Shewhart control chart vs. rational subgrouping for different levels of autocorrelation. Furthermore, by implementing the ESSI sampling strategy, the selection of the width of control limits for the control chart is independent of the correlation. An illustrative example shows the implementation of the proposed strategy. [ABSTRACT FROM AUTHOR]

Probe sampling plans for aflatoxin in corn attempt to reliably estimate concentrations in bulk corn given complications like skewed contamination distribution and hotspots. To evaluate and improve sampling plans, three sampling strategies (simple random sampling, stratified random sampling, systematic sampling with U.S. GIPSA sampling schemes), three numbers of probes (5, 10, 100, the last a proxy for autosampling), four clustering levels (1, 10, 100, 1,000 kernels/cluster source), and six aflatoxin concentrations (5, 10, 20, 40, 80, 100 ppb) were assessed by Monte-Carlo simulation. Aflatoxin distribution was approximated by PERT and Gamma distributions of experimental aflatoxin data for uncontaminated and naturally contaminated single kernels. The model was validated against published data repeatedly sampling 18 grain lots contaminated with 5.8-680 ppb aflatoxin. All empirical acceptance probabilities fell within the range of simulated acceptance probabilities. Sensitivity analysis with partial rank correlation coefficients found acceptance probability more sensitive to aflatoxin concentration (-0.87) and clustering level (0.28) than number of probes (-0.09) and sampling strategy (0.04). Comparison of operating characteristic curves indicate all sampling strategies have similar average performance at the 20 ppb threshold (0.8-3.5% absolute marginal change), but systematic sampling has larger variability at clustering levels above 100. Taking extra probes improves detection (1.8% increase in absolute marginal change) when aflatoxin is spatially clustered at 1,000 kernels/cluster, but not when contaminated grains are homogenously distributed. Therefore, taking many small samples, for example, autosampling, may increase sampling plan reliability. The simulation is provided as an R Shiny web app for stakeholder use evaluating grain sampling plans. [ABSTRACT FROM AUTHOR]

Non-Cartesian magnetic resonance imaging trajectories at golden angle increments have the advantage of allowing motion correction and gating using intermediate real-time reconstructions. However, when the acquired data are cardiac binned for cine imaging, trajectories can cluster together at certain heart rates (HR) causing image artifacts. Here, we demonstrate an approach to reduce clustering by inserting additional angular increments within the trajectory, and optimizing them while still allowing for intermediate reconstructions. Three acquisition models were simulated under constant and variable HR: golden angle (M trd), random additional angles (M rnd), and optimized additional angles (M opt). The standard deviations of trajectory angular differences (STAD) were compared through their interquartile ranges (IQR) and the Kolmogorov-Smirnov test (significance level: p = 0.05). Agreement between an image reconstructed with uniform sampling and images from M trd , M rnd , and M opt was analyzed using the structural similarity index measure (SSIM). M trd and M opt were compared in three adults at high, low, and no HR variability. STADs from M trd were significantly different (p < 0.05) from M opt and M rnd. STAD (IQR × 10−2 rad) showed that M opt (0.5) and M rnd (0.5) reduced clustering relative to M trd (1.9) at constant HR. For variable HR, M opt (0.5) and M rnd (0.5) outperformed M trd (0.9). The SSIM (IQR) showed that M opt (0.011) produced the best image quality, followed by M rnd (0.014), and M trd (0.030). M opt outperformed M trd at reduced HR variability in in-vivo studies. At high HR variability, both models performed well. This approach reduces clustering in k -space and improves image quality. [Display omitted] [ABSTRACT FROM AUTHOR]

Subtropical regions have clay-rich, weathered soils, and long dry periods followed by intense rainfall that produces large fluctuations in soil water content (SWC) and hydrological behavior. This complicates predictions of spatio-temporal dynamics, as datasets are typically collected at too coarse resolution and observations often represent a duration that is too short to capture temporal stability. The aim of the present study was to gain further insights into the role of temporal sampling scale on the observed temporal stability features of SWC order to aid the design of optimal SWC sampling strategies. This focused on both sampling frequency and total monitoring duration, as previous analyses have not considered both of these sampling aspects simultaneously. We collected relatively high resolution data of SWC (fortnightly over 3.5 years) for various soil depths and under contrasting crops (peanuts and citrus) at the red soil region of southeast China. The dataset was split into a three-year training period and a six-month evaluation period. Altogether 13 sampling frequencies (intervals ranging from 15 to 240 days) and eight monitoring duration periods (between three and 36 months) were derived from the training period to identify temporal stability features and the most time stable location (MTSL). The prediction accuracies of these MTSLs were tested using the independent evaluation data. Results showed that vegetation type did affect the spatio-temporal patterns of SWC, whereby the citrus site exhibited a stronger temporal variation and weaker temporal stability than the peanut site. However, the effects of both sampling frequency and observation duration were more pronounced, irrespective of the role of vegetation type or soil depth. With increasing sampling interval or decreasing monitoring duration, temporal stability of SWC was generally overestimated; by less than 10% when sampling interval increased from every 15 to 240 days and by up to 40% with duration decreasing from 36 to 3 months. Our results suggest that sampling strategies and trade-offs between sampling interval and duration should focus on capturing the main variability in hydro-climatological conditions. For subtropical conditions, we found that sampling once every 45 days over 24 months to be the minimum sampling strategy to ensure errors in SWC temporal stability of less than 10%.

Sampling plan is an essential part for measuring profile of turbine blade with coordinates measuring machine(CMM), and the detection of feature parameters is a key component of turbine blade inspection. However, the influence of the sampling strategy on the blade feature parameters evaluation has been rarely studied. In order to understand the correlation between the sampling strategy and the accuracy of blade profile feature parameters, firstly, the extraction methods of turbine blade feature parameters were proposed, and these methods were compiled into an executable program TBGeoInspect. Secondly, based on the level principle, the unified mathematical representations of uniform sampling, curvature-based sampling, chord deviation sampling, weighted curvature-based sampling, and curvature-arc length sampling were given. The effect of these five sampling algorithms on the accuracy of turbine blade profile feature parameters was gained through simulation and experiment results of blade sampling. The results validate that the uncertainty of turbine blade feature parameters and sectional curve fitting error are lowest with curvature-arc length sampling method under the same number of sampling points.

Marine environmental DNA (eDNA) is an important tool for biodiversity research and monitoring but challenges remain in scaling surveys over large spatial areas, and increasing the frequency of sampling in remote locations at reasonable cost. Here we demonstrate the feasibility of sampling from commercial vessels (Mediterranean ferries) while underway, as a strategy to facilitate replicable, systematic marine eDNA surveys in locations that would normally be challenging and expensive for researchers to access. Sixteen eDNA samples were collected from four fixed sampling stations, and in response to four cetacean sightings, across three cruises undertaken along the 300 km ferry route between Livorno (Tuscany) and Golfo Aranci (Sardinia) in the Ligurian/Tyrrhenian Seas, June-July 2018. Using 12SrDNA and 16SrDNA metabarcoding markers, we recovered diverse marine vertebrate Molecular Operational Taxonomic Units (MOTUs) from teleost fish, elasmobranchs, and cetaceans. We detected sample heterogeneity consistent with previously known variation in species occurrences, including putative species spawning peaks associated with specific sea surface temperature ranges, and increased night time abundance of bathypelagic species known to undertake diel migrations through the water column. We suggest commercial vessel based marine eDNA sampling using the global shipping network has potential to facilitate broad-scale biodiversity monitoring in the world’s oceans.

Abstract A central goal in camera‐trapping (CT) studies is to maximize detection probability and precision of occupancy estimates while minimizing the number of CTs to reduce equipment and labor costs. Few studies, however, have examined the effect of CT number on detection probability. Moreover, historically, most studies focused on a specific species and the design could be tailored toward maximizing detection of this target species. Increasingly, however, such studies use data for all captured, non‐target, species (by‐catch data) for animal community‐level analyses. It remains unclear if, and how, the targeting of CTs toward one species affects the detection of non‐target species. We paired CTs from a permanent camera‐trapping grid (with 38 CTs) targeted at monitoring Eurasian lynx (Lynx lynx) in Innlandet County, Norway, with additional randomly placed CTs at two spatial scales (38 CTs within the same habitat patch and 38 CTs within the same 50‐km2 grid cell as the lynx‐targeted CTs) for three months. We combined multi‐scale occupancy models that enable the separation of large‐scale occupancy, CT‐scale site use, and detection probability with single‐scale occupancy models. This allowed us to study the effects of targeted placement and CT number on the detection probability of the target species (lynx) and seven non‐target mammal species (four carnivores, three herbivores, and one rodent). We found that all species, except moose (Alces alces), had the highest detection probability at lynx‐targeted CTs. Moose had equal detection probabilities at all three placement types. Adding extra CTs generally increased detection probabilities. Consequently, for all species, combining a lynx‐targeted CT with one or more randomly placed CTs, increased the accuracy and precision of occupancy estimates for 50‐km2 grid cells compared to single CT estimates. The placement of single CTs underestimated grid‐cell occupancy compared to known minimum occupancy and were similar to site‐use probability estimates of multi‐scale models. It is, however, uncertain to which spatial extent these site‐use probabilities refer. We therefore recommend the use of multiple (targeted) CTs to estimate occupancy in large grid cells and to interpret occupancy estimates from single CTs as site use of an, as of yet undefined, area surrounding the CT.

Agroforestry plays a pivotal role for Sahelian communities by allowing simultaneous improvement of food security and conservation of natural ecosystems and their biodiversity. However, agroforestry systems (AFSs) are particularly heterogeneous in sub-Saharan Africa due to small to very small fields, a large variety of agricultural practices and a diversity of parkland compositions and configurations. This makes spatial sampling processes very important but problematic in terms of representativeness of the landscape heterogeneity to allow an effective study of Sahelian AFSs. In this paper, we proposed, tested and assessed a methodological approach for landscape sampling, mapping and characterization while considering the different types of spatial heterogeneity in complex landscapes, such as Sahelian AFSs. Several complementary methods were combined on the basis of a priori knowledge of agroforestry landscape functioning using multisource data, remote sensing methods, and statistical and spatial analyses applied to landscape ecology. First, the landscape heterogeneity was stratified and used to design two weighted, stratified sampling plans for field surveys of tree species and land use/land cover types. Then, with multisource satellite images together with collected field data, the agroforestry systems were mapped, with a satisfactory accuracy of 85.12% and a Kappa index of 0.81. Finally, we used landscape metrics and diversity indices derived from AFS mapping and the tree species inventory to analyze the diversity of the studied AFS located in the Senegalese Peanut Basin. The results of the analysis evidenced the compositional, configurational and functional heterogeneity found in the study area. This allowed us to demonstrate the ability of the sampling strategy proposed in this paper to capture the various types of heterogeneity in agricultural landscapes. We also showed by implementing the method that it can be used for (i) tree biodiversity analysis, (ii) mapping and (iii) characterization of a complex AFS in sub-Saharan Africa.

Estimating coefficient of variation (CV) and semi-variogram from sample data requires an appropriate sampling spacing. Although some recommendations on the optimal spacings exist, they focused on the simulated random fields. We collected surface soil (0–20 cm) samples from 689 sites based on a regular grid of 100 × 100 m in the Liudaogou watershed (6.89 km2) on the Loess Plateau, China. We grouped the samples by re-sampling at six sampling spacings (100, 200, 300, 400, 500, and 600 m) to determine the characteristic changes in the spatial variability. The mean and CV values varied little among the different spacings (p < 0.05). The sampling spacing can be increased to 300 m if we want to know the possible relationships between OC, TN, and TP with the environmental factors. OC and TN were fitted best by the spherical model whereas TP was fitted best with an exponential model. The optimal model type unchanged as the spacing increased. The apparent parameters of semi-variogram did not differ from their true values when the spacing was less than 200 m for most cases (p < 0.05). Therefore, the spacing should not be greater than 200 m in order to clarify their spatial structures. [ABSTRACT FROM AUTHOR]

A central goal in camera‐trapping (CT) studies is to maximize detection probability and precision of occupancy estimates while minimizing the number of CTs to reduce equipment and labor costs. Few studies, however, have examined the effect of CT number on detection probability. Moreover, historically, most studies focused on a specific species and the design could be tailored toward maximizing detection of this target species. Increasingly, however, such studies use data for all captured, non‐target, species (by‐catch data) for animal community‐level analyses. It remains unclear if, and how, the targeting of CTs toward one species affects the detection of non‐target species. We paired CTs from a permanent camera‐trapping grid (with 38 CTs) targeted at monitoring Eurasian lynx (Lynx lynx) in Innlandet County, Norway, with additional randomly placed CTs at two spatial scales (38 CTs within the same habitat patch and 38 CTs within the same 50‐km2 grid cell as the lynx‐targeted CTs) for three months. We combined multi‐scale occupancy models that enable the separation of large‐scale occupancy, CT‐scale site use, and detection probability with single‐scale occupancy models. This allowed us to study the effects of targeted placement and CT number on the detection probability of the target species (lynx) and seven non‐target mammal species (four carnivores, three herbivores, and one rodent). We found that all species, except moose (Alces alces), had the highest detection probability at lynx‐targeted CTs. Moose had equal detection probabilities at all three placement types. Adding extra CTs generally increased detection probabilities. Consequently, for all species, combining a lynx‐targeted CT with one or more randomly placed CTs, increased the accuracy and precision of occupancy estimates for 50‐km2 grid cells compared to single CT estimates. The placement of single CTs underestimated grid‐cell occupancy compared to known minimum occupancy and were similar to site‐use probability estimates of multi‐scale models. It is, however, uncertain to which spatial extent these site‐use probabilities refer. We therefore recommend the use of multiple (targeted) CTs to estimate occupancy in large grid cells and to interpret occupancy estimates from single CTs as site use of an, as of yet undefined, area surrounding the CT. [ABSTRACT FROM AUTHOR]

Caves and other subterranean habitats fulfill the requirements of experimental model systems to address general questions in ecology and evolution. Yet, the harsh working conditions of these environments and the uniqueness of the subterranean organisms have challenged most attempts to pursuit standardized research.Two main obstacles have synergistically hampered previous attempts. First, there is a habitat impediment related to the objective difficulties of exploring subterranean habitats and our inability to access the network of fissures that represents the elective habitat for the so‐called "cave species." Second, there is a biological impediment illustrated by the rarity of most subterranean species and their low physiological tolerance, often limiting sample size and complicating laboratory experiments.We explore the advantages and disadvantages of four general experimental setups (in situ, quasi in situ, ex situ, and in silico) in the light of habitat and biological impediments. We also discuss the potential of indirect approaches to research. Furthermore, using bibliometric data, we provide a quantitative overview of the model organisms that scientists have exploited in the study of subterranean life.Our over‐arching goal is to promote caves as model systems where one can perform standardized scientific research. This is important not only to achieve an in‐depth understanding of the functioning of subterranean ecosystems but also to fully exploit their long‐discussed potential in addressing general scientific questions with implications beyond the boundaries of this discipline. [ABSTRACT FROM AUTHOR]

The quality of debris flow susceptibility mapping varies with sampling strategies. This paper aims at comparing three sampling strategies and determining the optimal one to sample the debris flow watersheds. The three sampling strategies studied were the centroid of the scarp area (COSA), the centroid of the flowing area (COFA), and the centroid of the accumulation area (COAA) of debris flow watersheds. An inventory consisting of 150 debris flow watersheds and 12 conditioning factors were prepared for research. Firstly, the information gain ratio (IGR) method was used to analyze the predictive ability of the conditioning factors. Subsequently, 12 conditioning factors were involved in the modeling of artificial neural network (ANN), random forest (RF) and support vector machine (SVM). Then, the receiver operating characteristic curves (ROC) and the area under curves (AUC) were used to evaluate the model performance. Finally, a scoring system was used to score the quality of the debris flow susceptibility maps. Samples obtained from the accumulation area have the strongest predictive ability and can make the models achieve the best performance. The AUC values corresponding to the best model performance on the validation dataset were 0.861, 0.804 and 0.856 for SVM, ANN and RF respectively. The sampling strategy of the centroid of the scarp area is optimal with the highest quality of debris flow susceptibility maps having scores of 373470, 393241 and 362485 for SVM, ANN and RF respectively. [ABSTRACT FROM AUTHOR]

Introduction: Pragmatic trials in comparative effectiveness research assess the effects of different treatment, therapeutic, or healthcare options in clinical practice. They are characterized by broad eligibility criteria and large sample sizes, which can lead to an unmanageable number of participants, increasing the risk of bias and affecting the integrity of the trial. We describe the development of a sampling strategy tool and its use in the PREPARE trial to circumvent the challenge of unmanageable work flow. Methods: Given the broad eligibility criteria and high fracture volume at participating clinical sites in the PREPARE trial, a pragmatic sampling strategy was needed. Using data from PREPARE, descriptive statistics were used to describe the use of the sampling strategy across clinical sites. A Chi-square test was performed to explore whether use of the sampling strategy was associated with a reduction in the number of missed eligible patients. Results: 7 of 20 clinical sites (35%) elected to adopt a sampling strategy. There were 1539 patients excluded due to the use of the sampling strategy, which represents 30% of all excluded patients and 20% of all patients screened for participation. Use of the sampling strategy was associated with lower odds of missed eligible patients (297/4545 (6.5%) versus 341/3200 (10.7%) p

A significant challenge with reliability-based design optimization (RBDO) is the high computational cost associated with the double-loop structure that entails a large number of function calls for both the optimization process and reliability analysis. Several decoupling methods have been developed to improve the efficiency of RBDO. In addition, surrogate models have been used to replace the original time-consuming models and improve the computational efficiency. This paper proposes a novel quantile-based sequential RBDO method using Kriging surrogate models for problems with independent constraint functions. An error-controlled adaptive Kriging scheme is integrated to derive accuracy information of surrogate models and develop a strategy that facilitates independent training of the models for the performance function. The proposed independent training avoids unnecessary performance function evaluations while ensuring the accuracy of reliability estimates. Moreover, a new sampling approach is proposed that allows refinement of surrogate models for both deterministic and probabilistic constraints. Five numerical examples are carried out to demonstrate the performance of the proposed method. It is observed that the proposed method is able to converge to the optimum design with significantly fewer function evaluations than the state-of-the-art methods based on surrogate models given the constraint functions are independent. [ABSTRACT FROM AUTHOR]

Low frequency (once a month) but long-term (ca. 6 years) sampling including snow-melt periods in a mountainous stream, the Okura River (Sendai, Japan), revealed that loadings of 5 parameters (COD, TN, TP, TOC and DSiO2) could be expressed exponentially using discharge (Q), while the coefficients for the 5 loadings were all about 1. Here, mathematically, the periodically averaged Q leads to approximation of that of load (L). We analyzed the bias of the spot Q to that of the periodical (30, 14 and 8 days) means. The results ensured the utilization of the spot Q instead of the periodical mean Q for estimating L because of the high correlation factors (0.872, 0.914 and 0.923 on 30-, 14-, 8-day mean Q analyses, respectively) and suggested the validity of the usage of the observed regression slopes of 1.06, 1.22, and 1.22 over 30, 14, 8 days for quantitative correction of L because the fact that the slopes are larger than 1 indicate that the usage of the spot Q instead of the mean Q leads to the overestimation of L. Both changing correlation factors and the regression slopes realized small improvements via shortening the periods from 14 to 8 days. The protocol proposed here is quite original and is applicable to designing sampling strategies at target sites based on quantification of the limitations and/or reliability of L estimations. [ABSTRACT FROM AUTHOR]

Image classification is a key task in image understanding, and multi-label image classification has become a popular topic in recent years. However, the success of multi-label image classification is closely related to the way of constructing a training set. As active learning aims to construct an effective training set through iteratively selecting the most informative examples to query labels from annotators, it was introduced into multi-label image classification. Accordingly, multi-label active learning is becoming an important research direction. In this work, we first review existing multi-label active learning algorithms for image classification. These algorithms can be categorized into two top groups fromtwo aspects respectively: sampling and annotation. The most important component of multi-label active learning is to design an effective sampling strategy that actively selects the examples with the highest informativeness from an unlabeled data pool, according to various information measures. Thus, different informativeness measures are emphasized in this survey. Furthermore, this work also makes a deep investigation on existing challenging issues and future promises in multi-label active learning with a focus on four core aspects: example dimension, label dimension, annotation, and application extension. [ABSTRACT FROM AUTHOR]

Accurate characterization and prediction of soil moisture distribution and solute transport in vadose zone require detailed knowledge of the spatial distribution of soil hydraulic properties. Since the direct measurements of these unknown properties are challenging, many studies invert the soil hydraulic parameters by incorporating observation data (e.g., soil moisture and pressure head) at selected point sampling locations into soil moisture flow models. However, a cost-effective sampling strategy for where and when to collect the data, which is vital for saving the costs for monitoring and data interpretation, is relatively rare compared to the direct parameter retrieving efforts. Here, an optimal spatial–temporal sampling strategy was proposed based on cross-correlation analysis between observed state variables and soil hydraulic parameters. Besides, the effects of meteorological condition, observation type, bottom boundary condition, and correlation scale of soil hydraulic parameters are also demonstrated. The proposed sampling strategy was assessed by both synthetic numerical experiments and a real-world case study. Results suggest the retrieval accuracy of heterogeneous soil is acceptable if the spatial/temporal sampling interval is set to be one spatial/temporal correlation length of soil moisture. Besides, surface observation contains the most plentiful information which could be used to derive root-zone soil moisture/parameters, but this ability depends on the correlation scale of soil hydraulic parameters. Besides, the temporal value of soil moisture depends on meteorological condition. It is not necessary to sample repeatedly during dry periods, but more attention should be paid to the observations after rainfall events. [ABSTRACT FROM AUTHOR]

Accurate inference in phylogeography requires appropriate sampling strategies. Complex questions demand a large sample size at both the population and genetic levels to obtain precise reconstructions. This is the case of the phylogeographic history of Cistus monspeliensis, a plant that displays low plastid (cpDNA) diversity in the Mediterranean Basin but high diversity in the Canary Islands. Here, we aimed to identify Mediterranean refugial areas and to accurately quantify inter‐island colonization events in the Canaries. Using a previous study as starting point, we increased sample size in two ways: (i) additional sampling of plastid genetic markers (from 1041 to 1899 bp); and (ii) additional sampling of populations (from 47 to 69) in long‐term persistence areas suggested by species distribution modeling (SDM). The synergy between SDM and extended population sampling helped find higher genetic diversity. Our deeper phylogeographic sampling of C. monspeliensis revealed the following: (i) potential refugia in long‐term persistence areas with high cpDNA diversity in western Europe and the Canary Islands; and (ii) a significant increase (from 7 to 12) in the number of inferred inter‐island colonization events across the archipelago. Our results stress the usefulness of SDM to identify the genetic signature associated with potential refugial areas. We herein propose a field sampling approach based on SDM that, in combination with a larger cpDNA sampling, can help answer a wide array of phylogeographic questions, such as the location of Quaternary refugia and number of colonizations across archipelagos. [ABSTRACT FROM AUTHOR]

The arcing-involved pulsating coagulation mode with both active and blank periods is essential for modern electrosurgery. This paper begins with a comprehensive introduction to such a pulsating mode, followed by its implementation challenges. Then, an industrial-scale low-speed microcontroller unit (MCU), TMS320F28379D, is utilized to exemplify the proposed output sampling and data-transferring strategy on a gallium nitride (GaN)-based high-frequency inverter that enables coagulation mode with interweaved active periods and blank periods. The inverter prototype fills the active period with 390 kHz sinusoids of amplitude ranging from hundreds to thousands of Volts, while maintaining null outputs during blank periods. The strategy of sampling the above-mentioned sinusoidal outputs, coupled with their data transfer facilitated by direct memory access (DMA), is also articulated for subsequential power computation. Besides that, a novel nonfixed duty cycle approach, featuring an alterable number of sinusoids as the active period, is proposed and integrated into the GaN-based inverter to enhance mode safety. Finally, the power tracking performance of the mode is evaluated initially on resistive load, secondarily on resistive plus capacitive load (R-C), and thirdly on fresh biotissue with the appearance of electrical arcing. The existing necessity of the null blank periods is examined at the end of the paper.

Environmental surveillance can be used for monitoring enteric disease in a population by detecting pathogens, shed by infected people, in sewage. Detection of pathogens depends on many factors: infection rates and shedding in the population, pathogen fate in the sewerage network, and also sampling sites, sample size, and assay sensitivity. This complexity makes the design of sampling strategies challenging, which creates a need for mathematical modeling to guide decision making.In the present study, a model was developed to simulate pathogen shedding, pathogen transport and fate in the sewerage network, sewage sampling, and detection of the pathogen. The simulation study used Salmonella enterica serovar Typhi (S. Typhi) as the target pathogen and two wards in Kolkata, India as the study area. Five different sampling strategies were evaluated for their sensitivity of detecting S. Typhi, by sampling unit: sewage pumping station, shared toilet, adjacent multiple shared toilets (primary sampling unit), pumping station + shared toilets, pumping station + primary sampling units. Sampling strategies were studied in eight scenarios with different geographic clustering of risk, pathogen loss (decay, leakage), and sensitivity of detection assays. A novel adaptive sampling site allocation method was designed, that updates the locations of sampling sites based on their performance. We then demonstrated how the simulation model can be used to predict the performance of environmental surveillance and how it is improved by optimizing the allocation of sampling sites.The results are summarized as a decision tree to guide the sampling strategy based on disease incidence, geographic distribution of risk, pathogen loss, and the sensitivity of the detection assay. The adaptive sampling site allocation method consistently outperformed alternatives with fixed site locations in most scenarios. In some cases, the optimum allocation method increased the median sensitivity from 45% to 90% within 20 updates.

Abstract Background Studies elucidating health-related information and special needs of Syrian migrants living in Germany are urgently required. However, data is scarce and finding appropriate sampling strategies to obtain representative results is challenging. In order to increase survey response in hard-to-reach populations, new methods were developed. One of them is respondent-driven sampling (RDS), a network sampling technique. We aimed to assess if respondent-driven sampling is a better approach to recruit Syrian migrants for health research than classical random sampling via the population registry. Methods A cross-sectional study was conducted in Munich between April and June 2017 inviting adults (18+ years) born in Syria to answer an online questionnaire asking for sociodemographic and health-related information. Recruitment of participants was done using a) random sampling via the population registry (PR) and b) RDS. The two study populations recruited via respondent-driven sampling and the population registry were compared to a sample drawn from the population registry with respect to gender and citizenship. In addition, the two study populations were compared to each other regarding self-reported health status, healthcare utilisation, lifestyle factors, social network size, and acculturation. Results Of 374 persons randomly drawn from the population registry, 49 individuals answered the questionnaire completely (response: 13.1%) while via RDS 195 participants were recruited by 16 seeds. More persons possessed German citizenship in the total sample (20.5, 95% CI: 16.6 to 24.8%) and in the PR study population (28.6, 95% CI: 16.6 to 43.3%) than in the study population (0.5, 95% CI: 0.1 to 1.5%). Participants recruited via the population registry were older, smoked less, reported more often to hold a university degree, and indicated a higher prevalence of chronic diseases, more frequent healthcare utilisation, higher scores of acculturation as well as a larger social network compared to the study population obtained via RDS. Conclusions Response was very low in the PR sample. The number of participants recruited via RDS was larger and led to a study population with substantially different characteristics. Our study thus indicates that RDS is a useful way to gain access to specific subgroups that are hard to reach via traditional random sampling.

As an important unsupervised learning method, clustering can find the hidden structures in data effectively. With the amount of data grows larger, clustering of large data sets is a challenging task. Many clustering algorithms have been developed to deal with small data sets, but they are often inefficient when the data sets are large. Meanwhile, most clustering algorithms require some extra parameters as input, which may not be easy to obtain in practical applications. This paper proposed a new clustering algorithm called hybrid and parameter-free clustering method (HPFCM). HPFCM is able to rapidly perform clustering on large data sets without knowing the number of clusters in advance. HPFCM is based on sampling on large data sets (MMRS* sampling), assessing the clustering tendency on samples (eVAT), determining the number of clusters (EPB), forming different partitions (MST tree cutting), and extending the results to the rest of the data sets. We compare HPFCM with the other three methods, which are popular in clustering large data sets. Several numerical and real-world experiments have been conducted to verify our algorithm. The results show the great potential and effectiveness of HPFCM for clustering large data sets.

Ex situ conservation is widely used to protect wild plant species from extinction. However, it remains unclear how genetic variation of ex situ plant collections reflects diversity of wild source populations. We conducted a global meta‐analysis of the genetic representativeness of ex situ populations by comparing genetic diversity (i.e., AR, allelic richness; He, expected heterozygosity; PPB, percent polymorphic bands; and SWI, Shannon–Winner index), inbreeding coefficient (FIS), and genetic differentiation between ex situ plant collections and their wild source populations. Genetic diversity (i.e., He, PPB, and SWI) was significantly lower in ex situ populations than their wild source populations, whereas genetic differentiation between ex situ and wild populations (ex‐situ‐wild FST), but not that among ex situ populations, was significantly higher than among wild populations. Outcrossing species, but not those with mixed mating system, had significantly lower genetic diversity in ex situ populations and significantly higher ex‐situ‐wild FST. When the collection size for ex situ conservation was ≥30 or 50, PPB, He, and ex‐situ‐wild FST were not significantly different between ex situ and wild populations, indicating a relatively high genetic representativeness. Collecting from the entire natural distribution range and mixing collections from different sources could significantly increase the genetic representativeness of ex situ populations. Type of ex situ conservation (i.e., planting or seed bank) had no effect on genetic representativeness. The effect size of He decreased and the effect size of ex‐situ‐wild FST increased as the duration of ex situ conservation increased. Our results suggest that current ex situ plant collections do not effectively capture the genetic variation of wild populations. Low genetic representativeness of ex situ populations was caused by both initial incomplete sampling from wild populations and genetic erosion during ex situ conservation. We emphasize that it is necessary to employ more thorough sampling strategies in future collecting efforts and to add new individuals where needed. Article impact statement: Low genetic representativeness of living plant collections is a worldwide problem in ex situ conservation. [ABSTRACT FROM AUTHOR]

Copyright of Forest Research is the property of Forest Research Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Field‐based sampling of terrestrial habitats at continental scales is required to build ecosystem observation networks. A key challenge for detecting change in ecosystem composition, structure, and function within these observatories is to obtain a representative sample of habitats. Representative sampling across a continent contributes to ecological validity when analyzing spatially distributed data. However, field resources are limited, and actual representativeness may differ markedly from theoretical expectations. Here, we report a post hoc evaluation of the coverage of environmental gradients as a surrogate for ecological representativeness by a continental‐scale survey undertaken by the Australian Terrestrial Ecosystem Research Network (TERN). TERN's surveillance program maintains a network of ecosystem observation plots initially established in the rangelands through a stratification method (clustering of bioregions by environment) and application of the Ausplots survey methodology. Subsequent site selection comprised gap‐filling and opportunistic sampling. We confirmed that environmental coverage was a good surrogate for ecological representativeness. The cumulative sampling of environments and plant species composition over time were strongly correlated (based on mean multivariate dispersion; r = 0.93). We compared environmental sampling of Ausplots to 100,000 background points and a set of retrospective (virtual) sampling schemes: systematic grid, simple random, stratified random, and generalized random‐tessellation stratified (GRTS). Differences were assessed according to sampling densities along environmental gradients, and multivariate dispersion. Ausplots outperformed systematic grid, simple random, and GRTS in coverage of environmental space (Tukey HSD of mean dispersion, P < 0.001). GRTS site selection obtained similar coverage to Ausplots when employing the same bioregional stratification. Stratification by climatic zones generated the highest environmental coverage (P < 0.001), although resulting sampling densities over‐represented mesic coastal habitats. The Ausplots bioregional stratification implemented under practical constraints represented complex environments well, compared to statistically oriented or spatially even samples. Potential statistical power also depends on replication, unbiased site selection, and accuracy of field measurements relative to the magnitude of change. Consistent with previous studies, our stocktake analysis confirmed that environmental, rather than spatial, stratification is required to maximize ecological coverage across continental ecosystem observation networks, and the approach to establishing TERN Ausplots was robust. We recommend targeted gap‐filling to complete sampling. [ABSTRACT FROM AUTHOR]

Facial expression recognition (FER) has a wide variety of applications ranging from human–computer interaction, robotics to health care. Although FER has made significant progress with the success of Convolutional Neural Network (CNN), it is still challenging especially for the video-based FER due to the dynamic changes in facial actions. Since the specific divergences exists among different expressions, we introduce a metric learning framework with a siamese cascaded structure that learns a fine-grained distinction for different expressions in video-based task. We also develop a pairwise sampling strategy for such metric learning framework. Furthermore, we propose a novel action-units attention mechanism tailored to FER task to extract spatial contexts from the emotion regions. This mechanism works as a sparse self-attention fashion to enable a single feature from any position to perceive features of the action-units (AUs) parts (eyebrows, eyes, nose, and mouth). Besides, an attentive pooling module is designed to select informative items over the video sequences by capturing the temporal importance. We conduct the experiments on four widely used datasets (CK+, Oulu-CASIA, MMI, and AffectNet), and also do experiment on the wild dataset AFEW to further investigate the robustness of our proposed method. Results demonstrate that our approach outperforms existing state-of-the-art methods. More in details, we give the ablation study of each component. [ABSTRACT FROM AUTHOR]

Synthetic chemicals are frequently detected in water bodies, and their concentrations vary over time. Water monitoring programs typically employ either a sequence of grab samples or continuous sampling, followed by chemical analysis. Continuous time‐proportional sampling yields the time‐weighted average concentration, which is taken as proxy for the real, time‐variable exposure. However, we do not know how much the toxicity of the average concentration differs from the toxicity of the corresponding fluctuating exposure profile. We used toxicokinetic–toxicodynamic models (invertebrates, fish) and population growth models (algae, duckweed) to calculate the margin of safety in moving time windows across measured aquatic concentration time series (7 pesticides) in 5 streams. A longer sampling period (14 d) for time‐proportional sampling leads to more deviations from the real chemical stress than shorter sampling durations (3 d). The associated error is a factor of 4 or less in the margin of safety value toward underestimating and an error of factor 9 toward overestimating chemical stress in the most toxic time windows. Under‐ and overestimations occur with approximate equal frequency and are very small compared with the overall variation, which ranged from 0.027 to 2.4 × 1010 (margin of safety values). We conclude that continuous, time‐proportional sampling for a period of 3 and 14 d for acute and chronic assessment, respectively, yields sufficiently accurate average concentrations to assess ecotoxicological effects. Environ Toxicol Chem 2020;39:2158–2168. © 2020 The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC. [ABSTRACT FROM AUTHOR]

In this study, a new sampling strategy for networked control systems, called regular quantification with hysteresis (RQH), is proposed. This alternative presents some benefits with respect to symmetric‐send‐on‐delta sampling, which is one of the most used strategies in the event‐based proportional–integral–derivative (PID) control loops. The behaviour of the RQH is defined by two parameters, the signal quantification and hysteresis, whose effect on the overall system performance is studied and guidelines about its choice are given in terms of noise measurement and steady‐state error. The limit cycle oscillations that could be induced by this sampling strategy are studied and new robustness measures to avoid them are proposed based on the describing function approach. The suitability of some tuning rules for continuous PI when applied to control systems with a RQH sampling is evaluated using the proposed measures. The results show that these tuning rules can be applied under certain conditions. [ABSTRACT FROM AUTHOR]

The chemical analysis of an urolith is often interpreted as "stone's composition". However, it must be taken into consideration, that in most cases, only a fragment of the stone has been sent to the laboratory. In some recurrent patients, stone compositions either vary considerably between episodes or the analytical result obtained from the stone fragment does not fit with the data of e.g. current 24 h-urinalysis or urinary pH-records. The question arises, whether this outcome may be the result of an improper stone sampling scheme. On a simple layered 2D-stone model composed of two mineral phases it is shown, how the choice of a stone fragment process may influence the result of "stone composition". Depending on the initial position of fragment within the whole stone, the respective calculated analyses can relevantly differ from the whole stone composition as well as strongly between two fragments. Even under the simplified conditions of a 2D-2-component-model "grown" under defined conditions, the differences between the analyses of the different specimens taken from a stone are in part remarkable. The more it can be argued that these differences increase if a real 3D-urolith is investigated. Further sampling biases may evolve and increase the problem of proper sampling:, e.g., if an urolith's more resistant parts remain intact while ESWL or laser-based stone fragmentation ("dusting"), the weak parts became fully disintegrated and removed from the body as fine-grained sludge—the stone's fine fraction is lost although its composition may carry important information on the stone's pathogenesis. Consequently, a "stone analysis" only obtained from the harder remains reveals an incomplete result, a fact that in principle limits its clinical interpretation. Choice of stone fragment is crucial. The extent of the uncertainty of an analysis resulting from potential selection biases should not be underestimated. Thus, sampling should be considered as an important part of the processes of quality assurance and management. Errors made at this early stage of diagnosis finding will affect the analytical result and thus influence the clarification of the underlying pathomechanism. This can lead to an improper metaphylactic strategy potentially causing recurrent stone formation which otherwise would have been prevented. A decision scheme for analysis of urinary stones removed using endoscopic methods is suggested. [ABSTRACT FROM AUTHOR]

To reopen the economy safely during the COVID-19 pandemic, governments need the capability to proactively identify new and often asymptomatic infections, as well as contact tracing. Policymakers and public health professionals need a sampling-testing method that can achieve broad population coverage without overwhelming medical workers. We observe that COVID-19 high-risk groups are located in the hubs and cliques of our geo-social network, formed by the close encounters of people during daily life. These individuals are the de facto “canary in a coal mine”. We propose that nations offer free and anonymous testing service to them. With open-source computer algorithms and datasets, only a small fraction of the population selected for COVID-19 testing can cover the majority of high-exposure-risk individuals. A 0.3% sampled testing for a megacity covers 3/4 of its entire population. A 3% sampled testing for a rural town covers 3/4 of its entire population. With government oversight and public consent, this approach can serve each province/state or city/township for decentralized daily testing planning. However, to protect privacy, we recommend constructing the geo-social network of anonymized cellphones, not named individuals. This infrastructure should be dismantled once the pandemic is largely over. This can be achieved by policymakers, health workers, and engineers together in solidarity. [ABSTRACT FROM AUTHOR]

Seed is an important way to store germplasm resources but its genetic integrity will decrease during long-term preservation. So, it's essential to update seeds according to the aging level of different species. Pearl millet [Cenchrus americanus (L.) Morrone syn., Pennisetum glaucum (L.) R. Br.] is a crucial forage grass, biofuel plant and important crops in the world bringing huge economic and ecological benefits. However, there is no report about the impact of aging on genetic integrity of its seeds. In this study, four genetic diversity indexes (the percentage of polymorphic bands, PPB; the effective number of alleles, Ne; the Nei's gene diversity index, H; the Shannon's information index, I) and 20 pairs of genomic-SSR primers were used to certify the optimal sample volume of pearl millet for molecular study and found that the best sample volume was 60. After the artificial aging test, the germination rate and four genetic diversity parameters (the number of alleles, Na; Ne; H; I) were used to evaluate the change of genetic integrity at different aging levels. The results showed that the germination rate and these four genetic diversity parameters declined with the increase of aging levels. Furthermore, when the germination rate of pearl millet seeds went down to 68.23%, a significant difference in genetic integrity was observed with unaged seeds. In conclusion, the optimal sample size of pearl millet was 60 and the critical point of germination rate to renew germplasm resources was 68.23% and these finds might contribute to the scientific study and the safe conservation of pearl millet. [ABSTRACT FROM AUTHOR]

The French government is planning the construction of offshore wind farms (OWF) in the next decade (around 2900 MW). Following the European Environmental Impact Assessment Directive 85/337/EEC, several studies have been undertaken to identify the environmental conditions and ecosystem functioning at selected sites prior to OWF construction. However, these studies are generally focused on the conservation of some species and there is no holistic approach for analysing the effects arising from OWF construction and operation. The objective of this article is to promote a sampling strategy to collect data on the different ecosystem compartments of the future Dieppe-Le Tréport (DLT) wind farm site, adopting an ecosystem approach, which could be applied to other OWFs for the implementation of a trophic network analysis. For that purpose, an Ecopath model is used here to derive indices from Ecological Network Analysis (ENA) to investigate the ecosystem structure and functioning. The results show that the ecosystem is most likely detritus-based, associated with a biomass dominated by bivalves, which could act as a dead end for a classic trophic food web since their consumption by top predators is low in comparison to their biomass. The systemic approach developed for DLT OWF site should be applied for other French and European installations of Offshore Wind Farm. [ABSTRACT FROM AUTHOR]

Key message: We developed and identified a new set of polymorphic EST-SSR markers in Sorbus pohuashanensis (Hance) Hedl. and generated an optimal sampling strategy. Sorbus pohuashanensis (Hance) Hedl. is indigenous to northern China and prized for its graceful appearance and seasonal variations in foliage color. However, only a few molecular biology studies have been conducted on this species. Therefore, its development and increased utilization are limited. The objective of the present study was to develop expressed sequence tag-simple sequence repeat (EST-SSR) markers based on the transcriptome data of this species. A total of 167 primer pairs were designed and 24 were randomly selected from the 55 initially screened for validation. Their polymorphism was assessed using samples of the Mount Kunyu in Shandong Province (SDKYS) population. Fifteen EST-SSR markers were identified as polymorphic, with a mean number of alleles (Na) and a polymorphism information content (PIC) of 8.2 and 0.6907, respectively. Thirty EST-SSRs (including another 15 EST-SSRs being previously developed) derived from S. pohuashanensis had high transferability (90%) to another 14 Sorbus species, and there were nine EST-SSRs applicable to Sorbus species tested. These Sorbus species were divided into two main clusters by an unweighted pair-group method with the arithmetic mean (UPGMA) clustering tree based on the nine EST-SSRs and was validated by principal coordinate analysis (PCoA). The sampling strategy showed that 25 randomly selected individuals from a population of S. pohuashanensis could represent 95% of its genetic diversity level. In all, these findings would provide a powerful theoretical basis for future studies on the genetic diversity and germplasm resources of S. pohuashanensis and its related species. [ABSTRACT FROM AUTHOR]

Surrogate models are popular tool to approximate the functional relationship of expensive simulation models in multiple scientific and engineering disciplines. Successful use of surrogate models can provide significant savings of computational cost. However, with a variety of surrogate model approaches available in literature, it is a difficult task to select an appropriate one at hand. In this paper, we present an overview of surrogate model approaches with an emphasis of their application for variance-based global sensitivity analysis, including polynomial regression model, high-dimensional model representation, state-dependent parameter, polynomial chaos expansion, Kriging/Gaussian Process, support vector regression, radial basis function, and low rank tensor approximation. The accuracy and efficiency of these approaches are compared with several benchmark examples. The strengths and weaknesses of these surrogate models are discussed, and the recommendations are provided for different types of applications. For ease of implementations, the packages, as well as toolboxes, of surrogate model techniques and their applications for global sensitivity analysis are collected. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) have exhibited remarkable success in fraud detection. However, detecting fraud in datasets with scattered graphic densities and multiple relations remains challenging. This is a prevalent issue in fraud detection as fraudsters often employ diverse relationship types to camouflage their activities. Moreover, the constrained interconnectivity between nodes contributes to a scarcity of informative data, thereby intensifying the influence of raw features and further compounding the difficulties in fraud detection processes. To address these challenges, we present SCN_GNN (S trongly C onnected N odes- G raph N eural N etwork), a novel algorithm for fraud detection, that proposes two node sampling strategies based on the fusion of strong node information and graph topology information. Among them, the structured similarity-aware module (SSAM) performs up-sampling to add useful nodes to the sparse graph, while the strong node module (SNM) performs down-sampling based on strong node information and original features. Furthermore, we also reconfigure the RSRL (R ecursive and S calable R einforcement L earning framework) module to improve fraud detection performance by increasing inter-class distances and decreasing intra-class distances, resulting in a refined decision boundary for optimized algorithmic efficacy. We use three metrics (AUC, recall, and G_Mean) to evaluate the performance of SCN_GNN. The experimental results compared with the state-of-the-art models on two real-world datasets demonstrate the superiority of the proposed SCN_GNN. [ABSTRACT FROM AUTHOR]

Detailed urban landuse information plays a fundamental role in smart city management. A sufficient sample size has been identified as a very crucial pre-request in machine learning algorithms for urban landuse classification. However, it is often difficult to recognize and label landuse categories from remote sensing images alone. Alternatively, field investigation is time-consuming with a high demand in human resources and monetary cost. Therefore, previous studies on urban landuse classification have often relied on a small size of labeled samples with very uneven spatial distribution. This study aims to explore the effectiveness of a semi-supervised classification framework with multi-source data for detailed urban landuse classification with a few labeled samples. A disagreement-based semi-supervised learning approach, the Co-Forest, was employed and compared with traditional supervised methods (e.g., random forest and XGBoost). Multi-source geospatial data were utilized including optical and nighttime light remote sensing and geospatial big data, which present the physical and socio-economic features of landuse categories. Taking urban landuse classification in Shenzhen City as a case, results show that the classification accuracy of the semi-supervised method are generally on par with that of traditional supervised methods, and less labeled samples are needed to achieve a comparable result under different training set ratios. Given a small sample size, the accuracy tends to be stable with training samples no less than 5% in total. Our results also indicate that the classification accuracy by using multi-source data is significantly higher than that with any single data source being applied. Among these data, map POI and high-resolution optical remote sensing data make larger contributions on the classification, followed by mobile data and nighttime light remote sensing data.