基于缓冲区重采样的LSTM林火预测模型*

为提高林火风险预测精度,挖掘地图上隐含的空间信息、时间序列上隐含的长期趋势和循环波动,提出1种基于缓冲区重采样的长短期记忆（LSTM）林火预测模型,选取15个与林火相关的影响因素,以方差膨胀因子为评价指标对其进行多重共线性检验,方差膨胀因子大于10的因素具有共线性,并采用信息增益率验证筛选结果的合理性。考虑到火灾的空间聚集特性,采用缓冲区分析与过采样相结合方法减少样本不均衡现象的影响,最终得到176 732条样本。对12个影响因素和研究时间段的火点建立LSTM预测模型,对森林火灾发生风险进行预测。研究结果表明:基于缓冲区重采样的LSTM林火预测模型有效考虑时空上隐含的信息,预测模型准确率为87.06%,特异性为97.99%,敏感度为76.12%,阳性预测率为97.43%,阴性预测率为80.41%,ROC曲线与AUC值均优于随机森林（RF）和支持向量机（SVM）这2种基准算法。维尔克松秩和检验发现,本文提出的模型与基准算法结果具有显著性差异。研究结果可为提高林火风险预测精度提供参考。     

基于变量重要性评分-随机森林的溶解氧预测模型——以深圳湾为例

运用Pearson相关性分析,变量重要性评分和随机森林方法构建了溶解氧（DO）实时预测模型,并以深圳湾为例采用浮标资料预测1,3,6和12h的溶解氧.模型预测结果表明,模型最优的输入条件为pH值,水温,叶绿素a,氧化还原电位和蓝绿藻5个水质指标,1h预报的相关系数在0.9以上,6h预报结果一定程度上可以满足工程要求,但对低溶解氧事件的预报必须在3h以内.     

机器学习在材料服役性能预测中的应用

针对材料服役性能预测存在误差大、计算复杂、适用性差等问题,提出了基于数据挖掘的机器学习预测方法。首先阐述了机器学习的应用流程,并总结了常用模型原理及其在材料性能预测中的应用。然后采用多种机器学习模型对RPV钢的辐照性能进行预测,并通过Stacking集成方法提高了模型的预测精度。结果表明,机器学习可用于材料服役性能预测,具有较高的预测精度和可靠性。根据材料服役数据的不同特征选择合适的学习模型,同时进行模型融合和参数优化,可有效提高模型的预测精度及运算速度。     

Water Management Applications for Satellite Precipitation Products: Synthesis and Recommendations

This article is an assessment of the current state of the art and relative utility of satellite precipitation products (SPPs) for hydrologic applications to support water management decisions. We present a review of SPPs, their accuracy in diverse settings including the influence of geography, topography, and weather systems, as well as the pros and cons of their use for different water management applications. At the end of this broad synthesizing effort, recommendations are proposed for: (1) SPP developers to improve the quality, usability, and relevance of precipitation products; and (2) SPP users to improve the reliability of their predictions and hydrologic applications to better support water management.     

Employing the shared socioeconomic pathways to predict CO2 emissions

Predicting CO₂ emissions is of significant interest to policymakers and scholars alike. The following article contributes to earlier work by using the recently released “shared socioeconomic pathways” (SSPs) to empirically model CO₂ emissions in the future. To this end, I employ in-sample and out-of-sample techniques to assess the prediction accuracy of the underlying model, before forecasting countries’ emission rates until 2100. This article makes three central contributions to the literature. First, as one of the first studies, I improve upon the Representative Concentration Pathways (RCPs) by incorporating the SSPs, which did not exist when the RCPs have been released. Second, I calculate predictions and forecasts for a global sample in 1960–2100, which circumvents issues of limited time periods and sample selection bias in previous research. Third, I thoroughly assess the prediction accuracy of the model, which contributes to providing a guideline for prediction exercises in general using in-sample and out-of-sample approaches. This research presents findings that crucially inform scholars and policymakers, especially in light of the prominent 2 °C goal: none of the five SSP scenarios is likely to be linked to emission patterns that would suggest achieving the 2 °C goal is realistic.     

坡度图在滑坡研究中的应用

本文阐述了坡度图的概念和作用;详细介绍了坡度尺、坡度图的作法、以及坡度图在兰州地区的应用结果;简要分析了斜坡坡度与滑坡的关系。     

资源型城市接续主导产业的选择研究--以白银市为例

本文论述了资源型城市接续主导产业选择的重要性和选择原则，并以白银市为例，利用综合评价值法广告牌了白银市的主导产业和优势产业，然后用特尔斐法对白银市的接续主导产业进行了选择，选择结果比较符合白银市的实际情况。     

Three-dimensional hydrodynamic and water quality model for TMDL development of Lake Fuxian, China

Lake Fuxian is the largest deep freshwater lake in China. Although its average water quality meets Class I of the China National Water Quality Standard (CNWQS), i.e., GB3838-2002, monitoring data indicate that the water quality approaches the Class II threshold in some areas. Thus it is urgent to reduce the watershed load through the total maximum daily load (TMDL) program. A three-dimensional hydrodynamic and water quality model was developed for Lake Fuxian, simulating flow circulation and pollutant fate and transport. The model development process consists of several steps, including grid generation, initial and boundary condition configurations, and model calibration processes. The model accurately reproduced the observed water surface elevation, spatiotemporal variations in temperature, and total nitrogen (TN), total phosphorus (TP), and chemical oxygen demand (COD) concentrations, suggesting a reasonable numerical representation of the prototype system for further TMDL analyses. The TMDL was calculated using two interpretations of the water quality standards for Class I of the CNWQS based on the maximum instantaneous surface and annual average surface water concentrations. Analysis of the first scenario indicated that the TN, TP and COD loads should be reduced by 66%, 68% and 57%, respectively. Water quality was the highest priority; however, local economic development and cost feasibility for load reduction can pose significant issues. In the second interpretation, the model results showed that, under the existing conditions, the average water quality meets the Class I standard and therefore load reduction is unnecessary. Future studies are needed to conduct risk and cost assessments for realistic decision-making.     

道路交通噪声不同预测模型之比较

目前,我国在进行道路交通噪声环境影响评价工作中主要使用的预测模型有2009声导则模型、2006规范模型,此外,也有部分噪声评价工作采用了德国的CadnaA软件。因各种模型在使用条件和参数选取等方面存在不同,导致预测结果存在差异。如何选取合适的预测模型,一直是国内学者不断研究想解决的问题。本文对已有道路交通噪声进行现场实测,通过设计不同的预测模式进行模型验证,将预测结果与实测值进行比较,结果表明这五种噪声预测模式中,模式一、模式四、CadnaA软件的噪声预测值与实际情况最相符,模式一绝对预测误差昼间在3.3-6.0 dB(A)之间,夜间在-3.4-0.2 dB(A)之间。使用2009声导则模型、2006规范模型相结合的预测模式更为准确。     

香溪河库湾水质特征与非回水区水华响应关系

为探究香溪河在受倒灌影响较弱的河段水华暴发征兆及机理,于水华高发期5~8月对香溪河进行监测,分析电导率、水温、叶绿素a（Chl-a）以及流速.结果表明,在6~8月,香溪河倒灌现象于XX05点（峡口镇）处基本结束,XX06~XX09受倒灌来水影响较弱;香溪河于7月暴发水华,各个监测点位上Chl-a含量均值达到100μg/L以上,XX05~XX06与XX07-XX09点位上暴发不同种水华;在水华暴发前后,水体温度无显著变化,且并无明显分层现象,说明水温分层是水华暴发的主要原因这一理论并不能很好的适用于非回水区;通过对电导率数值的研究发现,数值在垂向上出现显著拐点,而拐点出现在临界层与光补偿层之间,同时与叶绿素a含量分布呈现显著负相关性.香溪河总氮（TN）、总磷（TP）平均值为1.849mg/L和0.157mg/L,均超过富营养化的阈值,水体氮磷含量与Chl-a浓度无显著相关性,水体中除N、P营养盐外的其它离子对香溪河水华的暴发起着重要作用.在水华消退后,电导率数值又逐渐恢复表层高底层低的垂向线性分布特性,与水华的暴发、消退有着明显的响应.