Changes in coral reef communities across a natural gradient in seawater pH

Ocean acidification threatens the survival of coral reef ecosystems worldwide. The negative effects of ocean acidification observed in many laboratory experiments have been seen in studies of naturally low-pH reefs, with little evidence to date for adaptation. Recently, we reported initial data suggesting that low-pH coral communities of the Palau Rock Islands appear healthy despite the extreme conditions in which they live. Here, we build on that observation with a comprehensive statistical analysis of benthic communities across Palau's natural acidification gradient. Our analysis revealed a shift in coral community composition but no impact of acidification on coral richness, coralline algae abundance, macroalgae cover, coral calcification, or skeletal density. However, coral bioerosion increased 11-fold as pH decreased from the barrier reefs to the Rock Island bays. Indeed, a comparison of the naturally low-pH coral reef systems studied so far revealed increased bioerosion to be the only consistent feature among them, as responses varied across other indices of ecosystem health. Our results imply that whereas community responses may vary, escalation of coral reef bioerosion and acceleration of a shift from net accreting to net eroding reef structures will likely be a global signature of ocean acidification.

Transients and asymptotics of natural gradient learning

We analyse natural gradient learning in a two-layer feed-forward neural network using a statistical mechanics framework which is appropriate for large input dimension. We find significant improvement over standard gradient descent in both the transient and asymptotic phases of learning.

Natural gradient matrix momentum

Natural gradient learning is an efficient and principled method for improving on-line learning. In practical applications there will be an increased cost required in estimating and inverting the Fisher information matrix. We propose to use the matrix momentum algorithm in order to carry out efficient inversion and study the efficacy of a single step estimation of the Fisher information matrix. We analyse the proposed algorithm in a two-layer network, using a statistical mechanics framework which allows us to describe analytically the learning dynamics, and compare performance with true natural gradient learning and standard gradient descent.

Computationally efficient natural gradient descent

DUE TO COPYRIGHT RESTRICTIONS ONLY AVAILABLE FOR CONSULTATION AT ASTON UNIVERSITY LIBRARY AND INFORMATION SERVICES WITH PRIOR ARRANGEMENT

Seaweed fails to prevent ocean acidification impact on foraminifera along a shallow-water CO2 gradient

Ocean acidification causes biodiversity loss, alters ecosystems, and may impact food security, as shells of small organisms dissolve easily in corrosive waters. There is a suggestion that photosynthetic organisms could mitigate ocean acidification on a local scale, through seagrass protection or seaweed cultivation, as net ecosystem organic production raises the saturation state of calcium carbonate making seawater less corrosive. Here, we used a natural gradient in calcium carbonate saturation, caused by shallow-water CO2 seeps in the Mediterranean Sea, to assess whether seaweed that is resistant to acidification (Padina pavonica) could prevent adverse effects of acidification on epiphytic foraminifera. We found a reduction in the number of species of foraminifera as calcium carbonate saturation state fell and that the assemblage shifted from one dominated by calcareous species at reference sites (pH 8.19) to one dominated by agglutinated foraminifera at elevated levels of CO2 (pH 7.71). It is expected that ocean acidification will result in changes in foraminiferal assemblage composition and agglutinated forms may become more prevalent. Although Padina did not prevent adverse effects of ocean acidification, high biomass stands of seagrass or seaweed farms might be more successful in protecting epiphytic foraminifera.

Reduced calcification and lack of acclimatization by coral colonies growing in areas of persistent natural acidification

As the surface ocean equilibrates with rising atmospheric CO2, the pH of surface seawater is decreasing with potentially negative impacts on coral calcification. A critical question is whether corals will be able to adapt or acclimate to these changes in seawater chemistry. We use high precision CT scanning of skeletal cores of Porites astreoides, an important Caribbean reef-building coral, to show that calcification rates decrease significantly along a natural gradient in pH and aragonite saturation (Omega arag). This decrease is accompanied by an increase in skeletal erosion and predation by boring organisms. The degree of sensitivity to reduced ?arag measured on our field corals is consistent with that exhibited by the same species in laboratory CO2 manipulation experiments. We conclude that the Porites corals at our field site were not able to acclimatize enough to prevent the impacts of local ocean acidification on their skeletal growth and development, despite spending their entire lifespan in low pH, low Omega arag seawater.

Imaging and quantifying salt-tracer transport in a riparian groundwater system by means of 3D ERT monitoring

Determining groundwater flow paths of infiltrated river water is necessary for studying biochemical processes in the riparian zone, but their characterization is complicated by strong temporal and spatial heterogeneity. We investigated to what extent repeat 3D surface electrical resistance tomography (ERT) can be used to monitor transport of a salt-tracer plume under close to natural gradient conditions. The aim is to estimate groundwater flow velocities and pathways at a site located within a riparian groundwater system adjacent to the perialpine Thur River in northeastern Switzerland. Our ERT time-lapse images provide constraints on the plume's shape, flow direction, and velocity. These images allow the movement of the plume to be followed for 35 m. Although the hydraulic gradient is only 1.43 parts per thousand, the ERT time-lapse images demonstrate that the plume's center of mass and its front propagate with velocities of 2x10(-4) m/s and 5x10(-4) m/s, respectively. These velocities are compatible with groundwater resistivity monitoring data in two observation wells 5 m from the injection well. Five additional sensors in the 5-30 m distance range did not detect the plume. Comparison of the ERT time-lapse images with a groundwater transport model and time-lapse inversions of synthetic ERT data indicate that the movement of the plume can be described for the first 6 h after injection by a uniform transport model. Subsurface heterogeneity causes a change of the plume's direction and velocity at later times. Our results demonstrate the effectiveness of using time-lapse 3D surface ERT to monitor flow pathways in a challenging perialpine environment over larger scales than is practically possible with crosshole 3D ERT.

Gas exchanges in peach palms as a function of the spad chlorophyll meter readings

The close relationship between the chlorophyll-meters readings and the total chlorophyll and nitrogen contents in leaves, has allowed their evaluation both in annual and perennial species. Besides, some physiological events such as the CO2 assimilation have also been estimated by chlorophyll meters. This work was carried out aiming to evaluate the gas exchanges of peach palms as a function of the chlorophyll SPAD-Meter readings. Three year-old peach palms from Yurimaguas, Peru were studied in Ubatuba, SP, Brazil, spaced 2 x 1 m in area under a natural gradient of organic matter which allowed four plots to be considered, according to the peach palms leaves colors, from light yellow to dark green. The SPAD readings and the stomatal frequency of leaflets were evaluated. The photosynthetic photon flux density (PPFD, μmol m-2 s-1), the leaf temperature (Tleaf, ºC), the CO2 assimilation (A, μmol m-2 s-1), the stomatal conductance (g s, mol m-2 s-1), the transpiration (E, mmol m-2 s-1) and the intercellular CO2 concentration (Ci, μmol mol-1) were evaluated with a portable infrared gas analyzer (LCA-4, ADC BioScientific Ltd., Great Amwell, U.K.). A linear increase in the CO2 assimilation as a function of the SPAD readings (y = -0.34 + 0.19x, R² = 0.99), indicates that they can be a rapid and cheap complementary method to evaluate in peach palms some important physiological events, such as CO2 assimilation.

Improving sampling, optimization and feature extraction in Boltzmann machines

L’apprentissage supervisé de réseaux hiérarchiques à grande échelle connaît présentement un succès fulgurant. Malgré cette effervescence, l’apprentissage non-supervisé représente toujours, selon plusieurs chercheurs, un élément clé de l’Intelligence Artificielle, où les agents doivent apprendre à partir d’un nombre potentiellement limité de données. Cette thèse s’inscrit dans cette pensée et aborde divers sujets de recherche liés au problème d’estimation de densité par l’entremise des machines de Boltzmann (BM), modèles graphiques probabilistes au coeur de l’apprentissage profond. Nos contributions touchent les domaines de l’échantillonnage, l’estimation de fonctions de partition, l’optimisation ainsi que l’apprentissage de représentations invariantes. Cette thèse débute par l’exposition d’un nouvel algorithme d'échantillonnage adaptatif, qui ajuste (de fa ̧con automatique) la température des chaînes de Markov sous simulation, afin de maintenir une vitesse de convergence élevée tout au long de l’apprentissage. Lorsqu’utilisé dans le contexte de l’apprentissage par maximum de vraisemblance stochastique (SML), notre algorithme engendre une robustesse accrue face à la sélection du taux d’apprentissage, ainsi qu’une meilleure vitesse de convergence. Nos résultats sont présent ́es dans le domaine des BMs, mais la méthode est générale et applicable à l’apprentissage de tout modèle probabiliste exploitant l’échantillonnage par chaînes de Markov. Tandis que le gradient du maximum de vraisemblance peut-être approximé par échantillonnage, l’évaluation de la log-vraisemblance nécessite un estimé de la fonction de partition. Contrairement aux approches traditionnelles qui considèrent un modèle donné comme une boîte noire, nous proposons plutôt d’exploiter la dynamique de l’apprentissage en estimant les changements successifs de log-partition encourus à chaque mise à jour des paramètres. Le problème d’estimation est reformulé comme un problème d’inférence similaire au filtre de Kalman, mais sur un graphe bi-dimensionnel, où les dimensions correspondent aux axes du temps et au paramètre de température. Sur le thème de l’optimisation, nous présentons également un algorithme permettant d’appliquer, de manière efficace, le gradient naturel à des machines de Boltzmann comportant des milliers d’unités. Jusqu’à présent, son adoption était limitée par son haut coût computationel ainsi que sa demande en mémoire. Notre algorithme, Metric-Free Natural Gradient (MFNG), permet d’éviter le calcul explicite de la matrice d’information de Fisher (et son inverse) en exploitant un solveur linéaire combiné à un produit matrice-vecteur efficace. L’algorithme est prometteur: en terme du nombre d’évaluations de fonctions, MFNG converge plus rapidement que SML. Son implémentation demeure malheureusement inefficace en temps de calcul. Ces travaux explorent également les mécanismes sous-jacents à l’apprentissage de représentations invariantes. À cette fin, nous utilisons la famille de machines de Boltzmann restreintes “spike & slab” (ssRBM), que nous modifions afin de pouvoir modéliser des distributions binaires et parcimonieuses. Les variables latentes binaires de la ssRBM peuvent être rendues invariantes à un sous-espace vectoriel, en associant à chacune d’elles, un vecteur de variables latentes continues (dénommées “slabs”). Ceci se traduit par une invariance accrue au niveau de la représentation et un meilleur taux de classification lorsque peu de données étiquetées sont disponibles. Nous terminons cette thèse sur un sujet ambitieux: l’apprentissage de représentations pouvant séparer les facteurs de variations présents dans le signal d’entrée. Nous proposons une solution à base de ssRBM bilinéaire (avec deux groupes de facteurs latents) et formulons le problème comme l’un de “pooling” dans des sous-espaces vectoriels complémentaires.

On Recurrent and Deep Neural Networks

L'apprentissage profond est un domaine de recherche en forte croissance en apprentissage automatique qui est parvenu à des résultats impressionnants dans différentes tâches allant de la classification d'images à la parole, en passant par la modélisation du langage. Les réseaux de neurones récurrents, une sous-classe d'architecture profonde, s'avèrent particulièrement prometteurs. Les réseaux récurrents peuvent capter la structure temporelle dans les données. Ils ont potentiellement la capacité d'apprendre des corrélations entre des événements éloignés dans le temps et d'emmagasiner indéfiniment des informations dans leur mémoire interne. Dans ce travail, nous tentons d'abord de comprendre pourquoi la profondeur est utile. Similairement à d'autres travaux de la littérature, nos résultats démontrent que les modèles profonds peuvent être plus efficaces pour représenter certaines familles de fonctions comparativement aux modèles peu profonds. Contrairement à ces travaux, nous effectuons notre analyse théorique sur des réseaux profonds acycliques munis de fonctions d'activation linéaires par parties, puisque ce type de modèle est actuellement l'état de l'art dans différentes tâches de classification. La deuxième partie de cette thèse porte sur le processus d'apprentissage. Nous analysons quelques techniques d'optimisation proposées récemment, telles l'optimisation Hessian free, la descente de gradient naturel et la descente des sous-espaces de Krylov. Nous proposons le cadre théorique des méthodes à région de confiance généralisées et nous montrons que plusieurs de ces algorithmes développés récemment peuvent être vus dans cette perspective. Nous argumentons que certains membres de cette famille d'approches peuvent être mieux adaptés que d'autres à l'optimisation non convexe. La dernière partie de ce document se concentre sur les réseaux de neurones récurrents. Nous étudions d'abord le concept de mémoire et tentons de répondre aux questions suivantes: Les réseaux récurrents peuvent-ils démontrer une mémoire sans limite? Ce comportement peut-il être appris? Nous montrons que cela est possible si des indices sont fournis durant l'apprentissage. Ensuite, nous explorons deux problèmes spécifiques à l'entraînement des réseaux récurrents, à savoir la dissipation et l'explosion du gradient. Notre analyse se termine par une solution au problème d'explosion du gradient qui implique de borner la norme du gradient. Nous proposons également un terme de régularisation conçu spécifiquement pour réduire le problème de dissipation du gradient. Sur un ensemble de données synthétique, nous montrons empiriquement que ces mécanismes peuvent permettre aux réseaux récurrents d'apprendre de façon autonome à mémoriser des informations pour une période de temps indéfinie. Finalement, nous explorons la notion de profondeur dans les réseaux de neurones récurrents. Comparativement aux réseaux acycliques, la définition de profondeur dans les réseaux récurrents est souvent ambiguë. Nous proposons différentes façons d'ajouter de la profondeur dans les réseaux récurrents et nous évaluons empiriquement ces propositions.

Species occurrence and richness of lichen, bryophytes and vascular plants near Abisko, Sweden and Toolik lake, Alaska

Little is known about the impact of changing temperature regimes on composition and diversity of cryptogam communities in the Arctic and Subarctic, despite the well-known importance of lichens and bryophytes to the functioning and climate feedbacks of northern ecosystems. We investigated changes in diversity and abundance of lichens and bryophytes within long-term (9-16 years) warming experiments and along natural climatic gradients, ranging from Swedish subarctic birch forest and subarctic/subalpine tundra to Alaskan arctic tussock tundra. In both Sweden and Alaska, lichen diversity responded negatively to experimental warming (with the exception of a birch forest) and to higher temperatures along climatic gradients. Bryophytes were less sensitive to experimental warming than lichens, but depending on the length of the gradient, bryophyte diversity decreased both with increasing temperatures and at extremely low temperatures. Among bryophytes, Sphagnum mosses were particularly resistant to experimental warming in terms of both abundance and diversity. Temperature, on both continents, was the main driver of species composition within experiments and along gradients, with the exception of the Swedish subarctic birch forest where amount of litter constituted the best explanatory variable. In a warming experiment in moist acidic tussock tundra in Alaska, temperature together with soil ammonium availability were the most important factors influencing species composition. Overall, dwarf shrub abundance (deciduous and evergreen) was positively related to warming but so were the bryophytes Sphagnum girgensohnii, Hylocomium splendens and Pleurozium schreberi; the majority of other cryptogams showed a negative relationship to warming. This unique combination of intercontinental comparison, natural gradient studies and experimental studies shows that cryptogam diversity and abundance, especially within lichens, is likely to decrease under arctic climate warming. Given the many ecosystem processes affected by cryptogams in high latitudes (e.g. carbon sequestration, N2-fixation, trophic interactions), these changes will have important feedback consequences for ecosystem functions and climate.

Incorporating curvature information into on-line learning

We analyse the dynamics of a number of second order on-line learning algorithms training multi-layer neural networks, using the methods of statistical mechanics. We first consider on-line Newton's method, which is known to provide optimal asymptotic performance. We determine the asymptotic generalization error decay for a soft committee machine, which is shown to compare favourably with the result for standard gradient descent. Matrix momentum provides a practical approximation to this method by allowing an efficient inversion of the Hessian. We consider an idealized matrix momentum algorithm which requires access to the Hessian and find close correspondence with the dynamics of on-line Newton's method. In practice, the Hessian will not be known on-line and we therefore consider matrix momentum using a single example approximation to the Hessian. In this case good asymptotic performance may still be achieved, but the algorithm is now sensitive to parameter choice because of noise in the Hessian estimate. On-line Newton's method is not appropriate during the transient learning phase, since a suboptimal unstable fixed point of the gradient descent dynamics becomes stable for this algorithm. A principled alternative is to use Amari's natural gradient learning algorithm and we show how this method provides a significant reduction in learning time when compared to gradient descent, while retaining the asymptotic performance of on-line Newton's method.

The theory of on-line learning: a statistical physics approach

In this paper we review recent theoretical approaches for analysing the dynamics of on-line learning in multilayer neural networks using methods adopted from statistical physics. The analysis is based on monitoring a set of macroscopic variables from which the generalisation error can be calculated. A closed set of dynamical equations for the macroscopic variables is derived analytically and solved numerically. The theoretical framework is then employed for defining optimal learning parameters and for analysing the incorporation of second order information into the learning process using natural gradient descent and matrix-momentum based methods. We will also briefly explain an extension of the original framework for analysing the case where training examples are sampled with repetition.

Vegetation:environment relationships and water management in Shark Slough, Everglades National Park

The hydrologic regime of Shark Slough, the most extensive long hydroperiod marsh in Everglades National Park, is largely controlled by the location, volume, and timing of water delivered to it through several control structures from Water Conservation Areas north of the Park. Where natural or anthropogenic barriers to water flow are present, water management practices in this highly regulated system may result in an uneven distribution of water in the marsh, which may impact regional vegetation patterns. In this paper, we use data from 569 sampling locations along five cross-Slough transects to examine regional vegetation distribution, and to test and describe the association of marsh vegetation with several hydrologic and edaphic parameters. Analysis of vegetation:environment relationships yielded estimates of both mean and variance in soil depth, as well as annual hydroperiod, mean water depth, and 30-day maximum water depth within each cover type during the 1990’s. We found that rank abundances of the three major marsh cover types (Tall Sawgrass, Sparse Sawgrass, and Spikerush Marsh) were identical in all portions of Shark Slough, but regional trends in the relative abundance of individual communities were present. Analysis also indicated clear and consistent differences in the hydrologic regime of three marsh cover types, with hydroperiod and water depths increasing in the order Tall Sawgrass , Sparse Sawgrass , Spikerush Marsh. In contrast, soil depth decreased in the same order. Locally, these differences were quite subtle; within a management unit of Shark Slough, mean annual values for the two water depth parameters varied less than 15 cm among types, and hydroperiods varied by 65 days or less. More significantly, regional variation in hydrology equaled or exceeded the variation attributable to cover type within a small area. For instance, estimated hydroperiods for Tall Sawgrass in Northern Shark Slough were longer than for Spikerush Marsh in any of the other regions. Although some of this regional variation may reflect a natural gradient within the Slough, a large proportion is the result of compartmentalization due to current water management practices within the marsh.We conclude that hydroperiod or water depth are the most important influences on vegetation within management units, and attribute larger scale differences in vegetation pattern to the interactions among soil development, hydrology and fire regime in this pivotal portion of Everglades.

Climatic drivers of hourly to yearly tree radius variations along a 6°C natural warming gradient

Climate affects the timing, rate and dynamics of tree growth, over time scales ranging from seconds to centuries. Monitoring how a tree's stem radius varies over these time scales can provide insight into intra-annual stem dynamics and improve our understanding of climate impacts on tree physiology and growth processes. Here, we quantify the response of radial conifer stem size to environmental fluctuations via a novel assessment of tree circadian cycles. We analyze four years of sub-hourly data collected from 56 larch and spruce trees growing along a natural temperature gradient of ∼6 °C in the central Swiss Alps. During the growing season, tree stem diameters were greatest at mid-morning and smallest in the late evening, reflecting the daily cycle of water uptake and loss. Along the gradient, amplitudes calculated from the stem radius cycle were ∼50% smaller at the upper site (∼2200 m a.s.l.) relative to the lower site (∼800 m a.s.l.). We show changes in precipitation, temperature and cloud cover have a substantial effect on typical growing season diurnal cycles; amplitudes were nine times smaller on rainy days (>10 mm), and daily amplitudes are approximately 40% larger when the mean daily temperature is 15–20 °C than when it is 5–10 °C. We find that over the growing season in the sub-alpine forests, spruce show greater daily stem water movement than larch. However, under projected future warming, larch could experience up to 50% greater stem water use, which may severely affect future growth on already dry sites. Our data further indicate that because of the confounding influences of radial growth and short-term water dynamics on stem size, conventional methodology probably overstates the effect of water-linked meteorological variables (i.e. precipitation and relative humidity) on intra-annual tree growth. We suggest future studies use intra-seasonal measurements of cell development and consider whether climatic factors produce reversible changes in stem diameter. These study design elements may help researchers more accurately quantify and attribute changes in forest productivity in response to future warming.