Abteilung Systemanalyse, Integrated Assessment und Modellierung

We review the main methods used to study spin glasses. In the first part, we focus on methods for fully connected models and systems defined on a tree. These include the replica method, the Thouless-Anderson-Palmer formalism, the cavity method, and the dynamical mean-field theory. In the second part, we deal with the description of low-dimensional systems, mostly in three spatial dimensions, which are mostly studied through numerical simulations. We conclude by mentioning some of the main open problems in the field.

Weunveil the multifractal behavior of Ising spin glasses in their low-temperature phase. Using the Janus II custom-built supercomputer, the spin-glass correlation function is studied locally. Dramatic fluctuations are found when pairs of sites at the same distance are compared. The scaling of these fluctuations, as the spin-glass coherence length grows with time, is characterized through the computation of the singularity spectrum and its corresponding Legendre transform. A comparatively small number of site pairs controls the average correlation that governs the response to a magnetic field. We explain how this scenario of dramatic fluctuations (at length scales smaller than the coherence length) can be reconciled with the smooth, self-averaging behavior that has long been considered to describe spin-glass dynamics.

Recent successes with machine learning (ML) models in catchment hydrology have highlighted their ability to extract crucial information from catchment properties pertinent to the rainfall–runoff relationship. In this study, we aim to identify a minimal set of catchment signatures in streamflow that, when combined with meteorological drivers, enable an accurate reconstruction of the entire streamflow time series. To achieve this, we utilize an explicit noise-conditional autoencoder (ENCA), which, assuming an optimal architecture, separates the influences of meteorological drivers and catchment properties on streamflow. The ENCA architecture feeds meteorological forcing and climate attributes into the decoder in order to incentivize the encoder to only learn features that are related to landscape properties minimally related to climate. By isolating the effect of meteorology, these hydrological features can thus be interpreted as landscape fingerprints. The optimal number of features is found by means of an intrinsic dimension estimator. We train our model on the hydro-meteorological time series data of 568 catchments of the continental United States from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) dataset. We compare the reconstruction accuracy with models that take as input a subset of static catchment attributes (both climate and landscape attributes) along with meteorological forcing variables. Our results suggest that available landscape attributes can be summarized by only two relevant learnt features (or signatures), while at least a third one is needed for about a dozen difficult-to-predict catchments in the central United States, which is mainly characterized by a high aridity index. The principal components of the learnt features strongly correlate with the baseflow index and aridity indicators, which is consistent with the idea that these indicators capture the variability of catchment hydrological responses. The correlation analysis further indicates that soil-related and vegetation attributes are of importance.

The presence or absence of sex can have a strong influence on the processes whereby species arise. Yet, the mechanistic underpinnings of this influence are poorly understood. To gain insights into the mechanisms whereby the reproductive mode may influence ecological diversification, we investigate how natural selection, genetic mixing, and the reproductive mode interact and how this interaction affects the evolutionary dynamics of diversifying lineages. To do so, we analyze models of ecological diversification for sexual and asexual lineages, in which diversification is driven by intraspecific resource competition. We find that the reproductive mode strongly influences the diversification rate and, thus, the ensuing diversity of a lineage. Our results reveal that ecologically-based selection is stronger in asexual lineages because asexual organisms have a higher reproductive potential than sexual ones. This promotes faster diversification in asexual lineages. However, a small amount of genetic mixing accelerates the trait expansion process in sexual lineages, overturning the effect of ecologically-based selection alone and enabling a faster niche occupancy than asexual lineages. As a consequence, sexual lineages can occupy more ecological niches, eventually resulting in higher diversity. This suggests that sexual reproduction may be widespread among species because it increases the rate of diversification.

Modern plankton high-throughput monitoring relies on deep learning classifiers for species recognition in water ecosystems. Despite satisfactory nominal performances, a significant challenge arises from dataset shift, which causes performances to drop during deployment. In our study, we integrate the ZooLake dataset, which consists of dark-field images of lake plankton (Kyathanahally et al. 2021a), with manually annotated images from 10 independent days of deployment, serving as test cells to benchmark out-of-dataset (OOD) performances. Our analysis reveals instances where classifiers, initially performing well in in-dataset conditions, encounter notable failures in practical scenarios. For example, a MobileNet with a 92% nominal test accuracy shows a 77% OOD accuracy. We systematically investigate conditions leading to OOD performance drops and propose a preemptive assessment method to identify potential pitfalls when classifying new data, and pinpoint features in OOD images that adversely impact classification. We present a three-step pipeline: (i) identifying OOD degradation compared to nominal test performance, (ii) conducting a diagnostic analysis of degradation causes, and (iii) providing solutions. We find that ensembles of BEiT vision transformers, with targeted augmentations addressing OOD robustness, geometric ensembling, and rotation-based test-time augmentation, constitute the most robust model, which we call BEsT. It achieves an 83% OOD accuracy, with errors concentrated on container classes. Moreover, it exhibits lower sensitivity to dataset shift, and reproduces well the plankton abundances. Our proposed pipeline is applicable to generic plankton classifiers, contingent on the availability of suitable test cells. By identifying critical shortcomings and offering practical procedures to fortify models against dataset shift, our study contributes to the development of more reliable plankton classification technologies.