Dissolved sulfides in the oxic water column of San Francisco Bay,California

Trace contaminants enter major estuaries such as San Francisco Bay from a variety of point and nonpoint sources and may then be repartitioned between solid and aqueous phases or altered in chemical speciation. Chemical speciation affects the bioavailability of metals as well as organic ligands to planktonic and benthic organisms, and the partitioning of these solutes between phases. Our previous, work in south San Francisco Bay indicated that sulfide complexation with metals may be of particular importance because of the thermodynamic stability of these complexes. Although the water column of the bay is consistently well-oxygenated and typically unstratified with respect to dissolved oxygen, the kinetics of sulfide oxidation could exert at least transient controls on metal speciation. Our initial data on dissolved sulfides in the main channel of both the northern and southern components of the bay consistently indicate submicromolar concenrations (from <1 nM to 162 nM), as one would expect in an oxidizing environment. However, chemical speciation calculations over the range of observed sulfide concentrations indicate that these trace concentrations in the bay water column can markedly affect chemical speciation of ecologically significant trace metals such as cadmium, copper, and zinc.     

Spatial and seasonal patterns in sediment nitrogen remineralization and ammonium concentrations in San Francisco Bay,California

Nitrogen remineralization and extractable ammonium concentrations were measured in sediments from several locations in North and South San Francisco bays. In South Bay, remineralization rates decreased with depth in sediment and were highest in the spring following the seasonal phytoplankton bloom. At the channel stations, peak remineralization lagged peak water-column phytoplankton biomass (as measured by chlorophylla) by a month. Remineralization rates were generally higher in South Bay than North Bay. The lower remineralization rates in North Bay may be a result of anomalously low phytoplankton production and thus reduced deposition to the sediments, as well as low reiverine organic inputs to the upper estuary in recent years. Remineralization rates were positively correlated to carbon and nitrogen content of the sediments. In general, ammonium profiles in South Bay sediments showed no increase in deeper (4–8 cm) sediments. In North Bay, ammonium concentrations were greatest at stations with highest remineralization rates, and, in contrast to South Bay, extractable ammonium increased in deeper sediment. Differences in ammonium pools between North Bay and South Bay may be a result of increased irrigation by deep-dwelling macrofauna, which are more abundant in South Bay.     

Late Holocene Marsh Expansion in Southern San Francisco Bay, California

Currently, the largest tidal wetlands restoration project on the US Pacific Coast is being planned and implemented in southern San Francisco Bay; however, knowledge of baseline conditions of salt marsh extent in the region prior to European settlement is limited. Here, analysis of 24 sediment cores collected from ten intact southern San Francisco Bay tidal marshes were used to reconstruct spatio-temporal patterns of marsh expansion to provide historic context for current restoration efforts. A process-based marsh elevation simulation model was used to identify interactions between sediment supply, sea-level rise, and marsh formation rates. A distinct age gradient was found: expansion of marshes in the central portion of southern San Francisco Bay dated to 500 to 1500 calendar years before present, while expansion of marshes in southernmost San Francisco Bay dated to 200 to 700 calendar years before present. Thus, much of the tidal marsh area mapped by US Coast Survey during the 1853–1857 period were in fact not primeval tidal marshes that had persisted for millennia but were recently formed landscapes. Marsh expansion increased during the Little Ice Age, when freshwater inflow and sediment influx were higher than during the previous millennium, and also during settlement, when land use changes, such as introduction of livestock, increased watershed erosion, and sediment delivery.     

Pesticides associated with suspended sediemnts entering San Francisco Bay following the first major storm of water year 1996

Estuaries receive large quantities of suspended sediments following the first major storm of the water year. The first-flush events transport the majority of suspended sediments in any given year, and because of their relative freshness in the hydrologic system, these sediments may carry a significant amount of the sediment-associated pesticide load transported into estuaries. To characterize sediment-associated pesticides during a first-flush event, water and suspended sediment samples were collected at the head of the San Francisco Bay during the peak in suspended sediment concentration that followed the first major storm of the 1996 hydrologic year. Samples were analyzed for a variety of parameters as well as 19 pesticides and degradation products that span a wide range of hydrophobicity. Tidal mixing at the head of the estuary mixed relatively fresh suspended sediment transported down the rivers with suspended sediments in estuary waters. Segregation of the samples into groups with similar degrees of mixing between river and estuary water revealed that transport of suspended sediments from the Sacramento-San Joaquin drainage basin strongly influenced the concentration and distribution of sediment-associated pesticides entering the San Francisco Bay. The less-mixed suspended sediment contained a different distribution of pesticides than the sediments exposed to greater mixing. Temporal trends were evident in pesticide content after samples were segregated according to mixing history. These results indicate sampling strategies that collect at a low frequency or do not compare samples with similar mixing histories will not elucidate basin processes. Despite the considerable influence of mixing, a large number of pesticides were found associated with the suspended sediments. Few pesticides were found in the concurrent water samples and in concentrations much lower than predicted from equilibrium partitioning between the aqueous and sedimentary phases. The observed sediment-associated pesticide concentrations may reflect disequilibria between sedimentary and aqueous phases resulting from long equilibration times at locations where pesticides were applied, and relatively short transit times over which re-equilibration may occur.     

Lateral migration and bank erosion in a saltmarsh tidal channel in San Francisco Bay, California

Saltmarsh tidal channels have often been recognized as being stable landscape features, despite highly sinous planforms, severely undercut banks, and high rates of bank erosion. In an effort to solve this paradox, a saltmarsh tidal channel in the San Francisco Bay was monitored from March 1995 to March 1996. The short-term rate of bank erosion was measured using erosion pins and found to be 57 ± 10 mm yr^?1 on the outside banks of meander bends. In addition, a long-term maximum lateral migration rate of 23 ± 23 mm yr^?1 was estimated from aerial photos, producing a dimensionless channel migration rate (defined as the rate of migration divided by channel with), of 0.5% yr^?1. The difference in the rates of lateral migration and bank erosion is attributed to the persistence of failed bank material (slump blocks) in the channel. The slump blocks induce sedimentation, protect the banks, and prevent further bank erosion. A published stability analysis method for undercut banks is applied to determine a maximum overhanging width. Using the measured compressive and tensile strengths of rooted bank material, 16.55 ± 1.16 kPa and 2.93 ± 0.71 kPa, respectively, the maximum width of an undercut bank is calculated to be 0.69 m. The average width of slump blocks measured in the field is 0.67 ± 0.25 m. A simple numerical model predicting the rate of lateral migration is derived using the results from the stability analysis and data from sedimentation and erosion pins inserted throughout the channel. This model accurately predicts a rate of 23 ± 3 mm yr^?1.     

Impact of high herbivore densities on introduced smooth cordgrass,Spartina alterniflora,invading San Francisco Bay,California

Spartina alterniflora, smooth cordgrass, invading San Francisco Bay, California (USA), is attacked by high densities of a plant hopper, Prokelisia marginata, and a mirid bug, Trigonotylus uhleri. Both herbivores are sap-feeders. We investigated the impact of these herbivores on S. alterniflora's growth rate, vegetative spread, and seed production by manipulating herbivore densities in the field and in a greenhouse. Herbivore densities in the field peaked in early fall, with P. marginata averaging more than 300 individuals per mature culm of S. alterniflora (about 100,000 per m²) and T. uhleri densities exceeding 10 per culm (about 3,000 per m²). Field reductions of herbivore densities by approximately 70% with insecticidal soap did not result in greater vegetative growth rates or lateral spread of plants; plants grew vigorously with the highest densities of insects. In the greenhouse study, conducted with seedlings, herbivory significantly reduced plant mass and tiller number in some but not all replicate herbivory treatments. In both field and greenhouse, there were significant differences between some clones' growth rates independent of herbivory. Inflorescence production in the field was not affected by reduced-herbivory treatments. Seed set was low under conditions of both natural and reduced herbivory, averaging 0.4%. Despite densities of P. marginata and T. uhleri that are much higher than typically observed in areas where S. alterniflora is native, herbivory by these particular insects appears to have little impact and in unlikely to limit S. alterniflora’s spread through San Francisco Bay. *** DIRECT SUPPORT *** A01BY070 00009     

Stratigraphy of late Quaternary estuarine deposits and amino acid stereochemistry of oyster shells beneath San Francisco Bay,California

The sequence of Quaternary deposits beneath the floor of San Francisco Bay includes four to seven noncontemporaneous estuarine units intercalated with alluvium and dune sand. Units L (0–10,000 B.P.), M (>40,000 B.P., probably ca. 80,000–140,000 B.P.), and N (older than unit M) are distinctly superposed. The dominant molluscan fossil in each of these three units is Ostrea lurida Carpenter, the native oyster along much of the pacific Coast of North America. Despite a lamellar structure that suggests vulnerability to contamination, O. lurida shells generally yield amino acid enantiomeric ratios that are analytically reproducible and stratigraphically consistent. The kinetics of racemization in O. lurida conceivably resembles that of Protothaca and Saxidomus, other bivalves whose kinetics of racemization are relatively well understood. Assuming such a resemblance, enantiomeric ratios in O. lurida imply that (1) unit M is the same approximate age as estuarine terrace deposits bordering San Pablo Bay and Carquinez Strait, providing that the terrace deposits have been at diagenetic temperatures 1°-2°C warmer than unit M; and (2) the age of unit N is about four times greater than that of unit M, providing that both units have been at the same approximate diagenetic temperature.     

Sources of mercury to San Francisco Bay surface sediment as revealed by mercury stable isotopes

Mercury (Hg) concentrations and isotopic compositions were examined in shallow-water surface sediment (0-2 cm) from San Francisco (SF) Bay to determine the extent to which historic Hg mining contributes to current Hg contamination in SF Bay, and to assess the use of Hg isotopes to trace sources of contamination in estuaries. Inter-tidal and wetland sediment had total Hg (Hg_T) concentrations ranging from 161 to 1529 ng/g with no simple gradients of spatial variation. In contrast, inter-tidal and wetland sediment displayed a geographic gradient of δ²⁰²Hg values, ranging from −0.30‰ in the southern-most part of SF Bay (draining the New Almaden Hg District) to −0.99‰ in the northern-most part of SF Bay near the Sacramento-San Joaquin River Delta. Similar to SF Bay inter-tidal sediment, surface sediment from the Alviso Slough channel draining into South SF Bay had a δ²⁰²Hg value of −0.29‰, while surface sediment from the Cosumnes River and Sacramento-San Joaquin River Delta draining into north SF Bay had lower average δ²⁰²Hg values of −0.90‰ and −0.75‰, respectively. This isotopic trend suggests that Hg-contaminated sediment from the New Almaden Hg District mixes with Hg-contaminated sediment from a low δ²⁰²Hg source north of SF Bay. Tailings and thermally decomposed ore (calcine) from the New Idria Hg mine in the California Coast Range had average δ²⁰²Hg values of −0.37 and +0.03‰, respectively, showing that Hg calcination fractionates Hg isotopes resulting in Hg contamination from Hg(II) mine waste products with higher δ²⁰²Hg values than metallic Hg(0) produced from Hg mines. Thus, there is evidence for at least two distinct isotopic signals for Hg contamination in SF Bay: Hg associated with calcine waste materials at Hg mines in the Coast Range, such as New Almaden and New Idria; and Hg(0) produced from these mines and used in placer gold mines and/or in other industrial processes in the Sierra Nevada region and SF Bay area.     

Late Holocene δC and pollen records of paleosalinity from tidal marshes in the San Francisco Bay estuary, California

Records of stable carbon isotopes (δ¹³C) are presented from cores collected from four San Francisco Bay marshes and used as a proxy for changes in estuary salinity. The δ¹³C value of organic marsh sediments are a reflection of the relative proportion of C₃ vs. C₄ plants occupying the surface, and can thus be used as a proxy for vegetation change on the marsh surface. The four marshes included in this study are located along a natural salinity gradient that exists in the San Francisco Bay, and records of vegetation change at all four sites can be used to infer changes in overall estuary paleosalinity. The δ¹³C values complement pollen data from the same marsh sites producing a paleoclimate record for the late Holocene period in the San Francisco Bay estuary. The data indicate that there have been periods of higher-than-average salinity in the Bay estuary (reduced fresh water inflow), including 1600-1300 cal yr B.P., 1000-800 cal yr B.P., 300-200 cal yr B.P., and ca. A.D. 1950 to the present. Periods of lower-than-average salinity (increased fresh water inflow) occurred before 2000 cal yr B.P., from 1300 to 1200 cal yr B.P. and ca. 150 cal yr B.P. to A.D. 1950. A comparison of the timing of these events with records from the California coast, watershed, and beyond the larger drainage of the Bay reveals that the paleosalinity variations reflected regional precipitation.     

Phytoplankton blooms and nitrogen productivity in San Francisco Bay

San Francisco Bay has been considered an HNLC or HNLG (high nutrient low chlorophyll or low growth) region with nonlimiting concentrations of inorganic nutrients yet low standing stocks of phytoplankton. Most of the studies leading to this conclusion come from the South Bay and little is known about nutrient processes and phytoplankton productivity in the northern and central parts of the estuary. Data collected over 3 yr (1999–2003) in Suisun, San Pablo, and Central Bays describe the availability of dissolved inorganic nitrogen (DIN), silicate, and phosphate and the seasonal variability in phytoplankton abundance. Rate measurements of fractionated nitrogen productivity provide the relative contributions of different forms of DIN (ammonium and nitrate) and different sized phytoplankton to the development of seasonal phytoplankton blooms. Regional differences in bloom dynamics are observed with Suisun Bay, the least saline, highest nutrient, most turbid region having less phytoplankton biomass and productivity than San Pablo and Central Bays, except in the abnormally wet spring of 2000. Spring blooms in San Francisco Bay are driven primarily by high rates of nitrate uptake by larger phytoplankton cells following a period of increased ammonium uptake that depletes the ambient ammonium. The smaller occasional fall blooms are apparently flueled mostly by ammonium uptake by small sized phytoplankton. The data suggest that the HNLC condition in the northern and central parts of San Francisco Bay is due primarily to light availability modulated by the interaction between ammonium and nitrate, and the relative amounts of the two forms of the DIN pool available to the phytoplankton.     

Sources of nitrogen and phosphorus to Northern San Francisco Bay

We studied nutrient sources to the Sacramento River and Suisun Bay (northern San Francisco Bay) and the influence which these sources have on the distributions of dissolved inorganic nitrogen (DIN) and dissolved reactive phosphorus (DRP) in the river and bay. We found that agricultural return flow drains and a municipal wastewater treatment plant were the largest sources of nutrients to the river during low river flow. The Sutter and Colusa agricultural drains contributed about 70% of the transport of DIN and DRP by the river above Sacramento (about 20% of the total transport by the river) between August 8 and September 26, 1985. Further downstream, the Sacramento Regional Wastewater Treatment Plant discharged DIN and DRP at rates that were roughly 70% of total DIN and DRP transport by the river at that time. Concentrations at Rio Vista on the tidal river below the Sacramento plant and at the head of the estuary were related to the reciprocals of the river flows, indicating the importance of dilution of the Sacramento waste by river flows. During very dry years, elevated DIN and DRP concentrations were observed in Suisun Bay. We used a steady-state, one-dimensional, single-compartment box model of the bay, incorporating terms for advection, exchange, and waste input, to calculate a residual rate for all processes not included in the model. We found that the residual for DIN was related to concentrations of chlorophylla (Chla). The residual for DRP was also related to Chla at high concentrations of Chla, but showed significant losses of DRP at low Chla concentrations. These losses were typically equivalent to about 80% of the wastewater input rate.     

Trace metals (Cd,Cu, Ni,and Zn) and nutrients in coastal waters adjacent to San Francisco Bay,California

Samples collected in December 1990 and July 1991 show that dissolved Cd, Cu, Ni, and Zn distributions in the Gulf of the Farallones are dominated by mixing of two end-members: (1) metal-enriched San Francisco Bay water and (2) offshore California Current water. The range of dissolved metal concentrations observed is 0.2–0.9 nmol kg^?1 for Cd, 1–20 nmol kg^?1 for Cu, 4–16 nmol kg^?1 for Ni, and 0.2–20 nmol kg^?1 for Zn. Effective concentrations in fresh water discharged into San Francisco Bay during 1990–1991 (estimated by extrapolation to zero salinity) are 740–860 μmol kg^?1 for silicate, 21–44 μmol kg^?1 for phosphate, 10–15 nmol kg^?1 for Cd, 210–450 nmol kg^?1 for Cu, 210–270 nmol kg^?1 for Ni, and 190–390 nmol kg^?1 for Zn. Comparison with effective trace metal and nutrient concentrations for freshwater discharge reported by Flegal et al. (1991) shows that input of these constituents to the northern reaches of San Francisco Bay accounts for only a fraction of the input to Gulf of the Farallones from the estuary system as a whole. The nutrient and trace metal composition of shelf water outside a 30-km radius from the mouth of the estuary closely resembles that of California Current water further offshore. In contrast to coastal waters elsewhere, there is little evidence of Cd, Cu, Ni, and Zn input by sediment diagenesis in continental shelf waters of California.     

Modeling Fate, Transport, and Biological Uptake of Selenium in North San Francisco Bay

Selenium behavior in North San Francisco Bay, the largest estuary on the US Pacific coast, is simulated using a numerical model. This work builds upon a previously published application for simulating selenium in the bay and considers point and non-point sources, transport and mixing of selenium, transformations between different species of selenium, and biological uptake by phytoplankton, bivalves, and higher organisms. An evaluation of the calibrated model suggests that it is able to represent salinity, suspended material, and chlorophyll a under different flow conditions beyond the calibration period, through comparison against long-term data, and the distribution of different species of dissolved and particulate selenium. Model-calculated selenium concentrations in bivalves compared well to a long-term dataset, capturing the annual and seasonal variations over a 15-year period. In particular, the observed lower bivalve concentrations in the wet flow periods, corresponding to lower average particulate selenium concentrations in the bay, are well represented by the model, demonstrating the role of loading and hydrology in affecting clam concentrations. Simulated selenium concentrations in higher organisms including white sturgeon and greater scaup also compared well to the observed data in the bay. Finally, a simulation of changing riverine inflows into the bay that might occur as a consequence of proposed hydrologic modifications indicated significant increases in dissolved and particulate selenium concentrations in the bay. The modeling framework allows an examination of the relationship between selenium loads, variations in inflow, in-bay concentrations, and biota concentrations to support management for limiting wildlife impacts.     

Carbon Sequestration and Sediment Accretion in San Francisco Bay Tidal Wetlands

Tidal wetlands play an important role with respect to climate change because of both their sensitivity to sea-level rise and their ability to sequester carbon dioxide from the atmosphere. Policy-based interest in carbon sequestration has increased recently, and wetland restoration projects have potential for carbon credits through soil carbon sequestration. We measured sediment accretion, mineral and organic matter accumulation, and carbon sequestration rates using ¹³⁷Cs and ²¹⁰Pb downcore distributions at six natural tidal wetlands in the San Francisco Bay Estuary. The accretion rates were, in general, 0.2?C0.5?cm?year^?1, indicating that local wetlands are keeping pace with recent rates of sea-level rise. Mineral accumulation rates were higher in salt marshes and at low-marsh stations within individual sites. The average carbon sequestration rate based on ²¹⁰Pb dating was 79?g?C?m^?2?year^?1, with slightly higher rates based on ¹³⁷Cs dating. There was little difference in the sequestration rates among sites or across stations within sites, indicating that a single carbon sequestration rate could be used for crediting tidal wetland restoration projects within the Estuary.     

Biomass and productivity of three phytoplankton size classes in San Francisco Bay

Primary productivity of three size classes of phytoplankton (<5 μm, 5–22 μm, >22 μm) was measured monthly at six sites within San Francisco Bay throughout 1980. These sites in the three principal embayments were chosen to represent a range of environments, phytoplankton communities, and seasonal cycles in the estuary. Temporal variations in productivity for each size class generaly followed the seasonality of the corresponding fraction of phytoplankton biomass. The 5–22 μm size class accounted for 40 to 50% of the annual production in each embayment, but production by phytoplankton >22 μm ranged from 26% in the southern reach to 54% of total phytoplankton production in the landward embayment of the northern reach. A productivity index is derived that predicts daily productivity for each size class as a function of ambient irradiance and integrated chlorophylla in the photic zone. For the whole phytoplankton community and for each size class, this index was constant and estimated as ?0.76 g C m^?2 (g chlorophylla Einstein)^?1. The annual means of maximum carbon assimilation numbers were usually similar for the three size classes. Spatial and temporal variations in size-fractionated productivity are shown to be primarily due to differences in biomass rather than size-dependent carbon assimilation rates. *** DIRECT SUPPORT *** A01BY034 00005     

Distribution of colloidal trace metals in the San Francisco Bay estuary

The size distribution of trace metals (Al, Ag, Cd, Cu, Fe, Mn, Ni, Sr, and Zn) was examined in surface waters of the San Francisco Bay estuary. Water samples were collected in January 1994 across the whole salinity gradient and fractionated into total dissolved (<0.2 μm colloidal (10 KDa–0.2 μm) and < 10 kDa molecular weight phases. In the low salinity region of the estuary, concentrations of colloidal A1, Ag, and Fe accounted for ≥84% of the total dissolved fraction, and colloidal Cu and Mn accounted for 16–20% of the total. At high salinities, while colloidal Fe was still relatively high (40% of the dissolved), very little colloidal Al, Mn, and Cu (<10%) and no colloidal Ag was detectable. Colloidal Zn accounted for <3% of the total dissolved along the estuary, and colloidal Ni was only detectable (<2%) at the river endmember. All of the total dissolved Cd and Sr throughout the estuary consisted of relatively low molecular weight (<10 kDa) species. The relative affinity of metals for humic substances and their reactivity with particle surfaces appear to determine the amounts of metal associated with colloids. The mixing behavior of metals along the estuary appears to be determined by the relative contribution of the colloidal phase to the total dissolved pool. Metals with a small or undetectable colloidal fraction showed a nonconservative excess (Cd, Cu, Ni, and Mn) or conservative mixing (Sr) in the total dissolved fraction, relative to ideal dilution of river water and seawater along the estuary.

The salt-induced coagulation of colloidal A1, Fe, and Cu is indicated by their highly nonconservative removal along the salinity gradient. However, colloidal metals with low affinity for humic substances (Mn and Zn) showed conservative mixing behavior, indicating that some riverine colloids are not effectively aggregated during their transport to the sea. While colloidal metal concentrations correlated with dissolved organic carbon, they also covaried with colloidal Al, suggesting that colloids are a mixture of organic and inorganic components. Furthermore, the similarity between the colloidal metal:A1 ratios with the crustal ratios indicated that colloids could be the product of weathering processes or particle resuspension. Distribution coefficients for colloidal particles (K_c) and for large, filter-retained particles (K_d) were of the same magnitude, suggesting similar binding strength for the two types of particles. Also, the dependence of the distribution coefficients on the amount of suspended particulate matter (the so-called particle concentration effect) was still evident for the colloids-corrected distribution coefficient (K_p+c) and for metals (e.g., Ni) without affinity for colloidal particles.     

Chemistry and microbiology of a sewage spill in South San Francisco Bay

During three weeks of September 1979, the breakdown of a waste treatment plant resulted in the discharge of a large volume (1.5×10⁷m³) of primary-treated sewage into a tributary of South San Francisco Bay, California. Chemical and microbial changes occurred within the tributary as decomposition and nitrification depleted dissolved oxygen. Associated with anoxia were relatively high concentrations of particulate organic carbon, dissolved CO₂, CH₄, C₂H₄, NH ₄ ⁺ , and fecal bacteria, and low phytoplankton biomass and photosynthetic oxygen production. South San Francisco Bay experienced only small changes in water quality, presumably because of its large volume and the assimilation of wastes that occurred within the tributary. Water quality improved rapidly in the tributary once normal tertiary treatment resumed.     

Specific-lon electrode determinations of sulfide preconcentrated from San Francisco Bay waters

Measurements of low-level dissolved-sulfide concentrations in estuarine water from San Francisco Bay have been made using the sulfide-specific electrode after preservation, separation, and preconcentration of the sulfide species. The separation and preconcentration were acheived by coprecipitation of ZnS with Zn(OH)₂ followed by collection and dissolution of the precipitate, giving concentration factors up to 160-fold Preconcentration provided sulfide solutions that were adequately measurable within the practical working range of the specific-ion electrode The sulfide detection limit with the preconcentration step is 0 02 μg/l Spike recoveries in the range of 81 to 10 1% have been achieved for laboratory-prepared samples having S²⁻ concentrations as low as 0 6 μg/l and 84 to 100% for an estuarine sample spiked in the field with 2 μg/l (S(−II) Positive correlations have been found between dissolved S(−II) concentrations and concentrations of dissolved Cd, Cu, and Ni, negative correlations have been found between bisulfide (HS⁻) activity and activities of Cd²⁺, Cu²⁺, and Ag⁺ species     

基于深度学习的旧金山湾水质预测

传统的水质预测模型计算复杂，且在大对流情况下会引发误差，对于大数据时代下智能化的水质预测问题并不适用。本文针对旧金山湾地表水质研究区的数据资料，利用数据分析、统计检验、深度学习时序模型等技术方法对该研究区的水质进行研究，根据主成分信息构建了长短时记忆（LSTM）循环神经网络模型，对研究区的5个地表水质采样站点进行了水质预测。结果表明：长短时记忆循环神经网络模型通过门控制循环和记忆单元结构，有效控制传入模型的输入特征，从而降低模型的复杂度；双层长短时记忆循环神经网络模型较单层长短时记忆循环神经网络模型的预测精度平均提高5.3%。利用LSTM模型可以对旧金山湾地表水质进行有效评价。