可解释性长短期记忆模型用于预测湖泊总磷浓度变化

对湖泊总磷的变化预测和来源识别对水资源调度和流域生态治理有着重要的意义,然而复杂的生化反应和水动力条件导致的非平稳性给湖泊总磷浓度的准确预测带来极大的困难。为克服这一挑战,本文引入了基于加权回归的季节趋势分解(seasonal and trend decomposition using Loess,STL)技术和夏普利加法(SHapley additive exPlanations,SHAP)结合长短期记忆网络(long short-term memory neural network,LSTM)和门控循环单元(gated recurrent unit,GRU)构建了一个可解释的预测框架,以增强对湖泊总磷浓度演变的预测并提高其可解释性。研究表明:(1)在骆马湖总磷浓度的预测中,该框架拥有较好的预报精度(R²=0.878),优于LSTM和卷积长短期记忆模型(convolutional neural networks and long short term memory network,CNN-LSTM)。当预测时间步长增加到8 h时,该框架有效提高了总磷浓度的预测精度,平均相对误差和均方根误差分别降低了47.1%和33.3%。从预测趋势来看,骆马湖在汛期的总磷平均浓度为0.158 mg/L,相较于非汛期的平均浓度,增加了202.1%。(2)运河来水是骆马湖总磷浓度最重要的影响因素,贡献权重为60.0%,并且不同断面(三湾、三场)的污染源受水动力、气象等因素的影响存在显著的时空差异。本文凸显了神经网络模型在预警水体污染方面的可实施性,并且为提高传统神经网络的学习能力和可解释性的开发与验证提供了重要方向。     

A deep-sea in-situ suspended sediment sampling system: Comment

A recent article by Beer et al. (1974) in Marine Geology describes an in-situ suspended sediment sampling system which utilizes a pump/motor unit, filter holders, and a lead-acid battery power source. Two aspects of their article may be misleading: the use of certain filter material as X-ray diffraction mounts, and the technique by which an oil-filled battery case may be operated.     

Improved Petrological Modal Analyses from X-ray Powder Diffraction Data by use of the Rietveld Method I. Selected Igneous, Volcanic, and Metamorphic Rocks

Mineral modes have been determined for specimens of eight rocktypes from CuK X-ray powder diffraction data using the Rietveldmethod. The samples include two granites, a granodiorite, adamellite,gabbro, basalt, trachyte, and two granulite-facies metamorphicrocks. Up to eight individual mineral components have been measuredin each sample (no glassy phases were observed), with a detectionlimit of {small tilde}1 wt.%, depending on the mineral assemblage.Marked variations in grain size (i.e., granite vs. trachyte)provide no difficulties for the X-ray method. The X-ray resultscompare very favourably with (1) optical modes determined forthe medium–coarse-grained samples by point counting, (2)normative calculations obtained using locally enhanced catanormand mesonorm software, and (3) corresponding Rietveld modesdetermined, for two samples, from neutron powder data. Wheredifferences occur, these are discussed in relation to the limitationsof each of the methods. The improved accuracy of the X-ray method is due primarily tothe incorporation of the full diffraction profile in the Rietveldanalysis calculations, and the elimination of preferred orientationby collecting the data from samples packed in glass capillaries(i.e., Debye–Scherrer mode). The good agreement of theX-ray and neutron modes shows that the usual problems encounteredwith microabsorption, extinction, and sampling are of littleconcern in these rocks. The results highlight one of the majoradvantages provided by Rietveld modal analysis over the moretraditional ‘reference intensity’ X-ray methods,namely, that the crystal chemistry (and thus the calibrationconstants) of the individual phases can be adjusted dynamicallyduring each individual analysis. This not only provides moreaccurate phase abundances, but also gives important supplementaryinformation about the mineralogy of the major components.     

A case study of warm air advection over a melting snow surface

When relatively warm, moist air moves over a snow surface, sensible heat and moisture are extracted from its lower layers and used to melt the snow. The depth of the cooled layer depends on horizontal wind speeds and the presence of high vertical wind shear. The mechanism for air mass modification appears to be turbulent mixing.     

Authigenic iron and titanium oxides in triassic red beds: (St. Bees Sandstone), Cumbria,Northern England

Authigenic iron and titanium oxides occur in three main textural varieties in the St. Bees Sandstone, a Triassic red bed succession in Cumbria. These are: (a) overgrowths of haematite, titaniferous haematite, anatase, and rutile which generally occur on detrital host grains of similar compositions. The overgrowths may occur as syntaxial rim cements or rhombohedral and prismatic projections and are always in optical continuity, (b) discrete crystals of anatase (including octahedrite) and haematite 10-40 μm in size occur in pore spaces, and (c) fine grained ( < 1 μm) pigmentary haematite occurs as grain coatings around detrital grains and in the interstitial matrix. The euhedral nature and pore-filling habit clearly indicate the authigenic origin of these iron- and titanium oxides. All three textural varieties are believed to have been precipitated from ground waters containing dissolved ferrous and titanium ions. These ions were derived from the intrastratal solution of detrital iron and titanium bearing grains including ferromagnesian silicates and iron-titanium oxides. The precipitation of iron oxides from migrating ground waters in the manner described here goes some way to explaining how thick, uniformly red successions can be formed.     

On the eigenvalues of kinematic α-effect dynamos

The kinematic α-effect dynamo problem is investigated in the case of an exterior perfect conductor. It is shown that certain approximate symmetries discovered in the numerical analysis of ROBERTS (1972) are exact for this case. As an illustration, an exact solution is given in a cylindrical geometry, where the equations can be written in terms of one variable. The implications for the earth's dynamo are discussed.     

Fe3+ in shocked olivine crystals of the ALHA 77005 meteorite

ALHA 77005, a shocked achondrite of the shergottite group, is unique in containing cumulus olivine crystals which display a brownish color in thin section. Mössbauer, EPR, and optical spectroscopic studies show the presence of Fe³⁺ in the olivine. Approximately 4.5 wt.% of the total iron in the olivine (24 wt.% as FeO) is in the trivalent state. The Fe³⁺ ions preferentially occupy M2 lattice positions. Charge transfer between Fe³⁺ and Fe²⁺ as well as between Fe³⁺ and oxygen ligands causes a broad absorption band in the optical spectrum at high energies which is the reason for the olivine color. It is concluded that the color of the ALHA 77005 olivine is preterrestrial and may be the result of shock-induced oxidation of the olivine crystals on the meteorite parent body.     

TRACING BY GRAVITY OF A NARROW BURIED GRABEN STRUCTURE,DETECTED BY SEISMIC REFRACTION,FOR GROUND-WATER INVESTIGATIONS IN NORTH CHILE*

In the Andes mountains of North Chile seismic refraction measurements revealed the existence of a buried graben structure. The special geological environment permitted the deduction, by detailed analyses of diffraction patterns and delayed arrival times, of a low velocity layer underneath a high velocity ignimbritic sheet. The gravity method was chosen as an economic secondary aid to trace the course of the buried structure. This unusual combination of detailed gravity measurements following a general seismic survey was not only successful in detecting and tracing a buried structure, but also provided complementary data about the deeper subsurface conditions.     

Saturn's electrostatic discharges: Could lightning be the cause?

It is proposed that Saturn's electrostatic discharges (SED) might be generated in the planet's equatorial atmosphere, perhaps as lightning from a storm system. The 10^h10^m periodicity of the signal envelope duplicates that of Saturn's equatorial jet. The rings shield the atmosphere from solar EUV photons, and thereby substantially reduce the local ionospheric cutoff frequency to allow low-frequency SED to leak out. Many of the unusual properties of SED could be explained in terms of changes in the storm system, the relative spacecraft position in the beaming pattern of the source, local refraction of the signal by the highly disturbed ionosphere, and the influence of the ring particles on the highest frequency component of SED. A comparison of SED with planetary lightning on other planets shows that the two are similar in general character and some time behavior; the power output of SED may be higher than most planetary lightnings but that is unclear because of uncertainties in the measurements and variations in the signal's spectrum. Our simple discussion suggests that lightning could be a viable source for SED and that exotic ring mechanisms are not necessarily required.