An official website of the United States government
Abstract Daily precipitation extremes are projected to intensify with increasing moisture under global warming following the Clausius-Clapeyron (CC) relationship at about $$ 7\% /^\circ {\text{C}} $$ 7 % / ∘ C . However, this increase is not spatially homogeneous. Projections in individual models exhibit regions with substantially larger increases than expected from the CC scaling. Here, we leverage theory and observations of the form of the precipitation probability distribution to substantially improve intermodel agreement in the medium to high precipitation intensity regime, and to interpret projected changes in frequency in the Coupled Model Intercomparison Project Phase 6. Besides particular regions where models consistently display super-CC behavior, we find substantial occurrence of super-CC behavior within a given latitude band when the multi-model average does not require that the models agree point-wise on location within that band. About 13% of the globe and almost 25% of the tropics (30% for tropical land) display increases exceeding 2CC. Over 40% of tropical land points exceed 1.5CC. Risk-ratio analysis shows that even small increases above CC scaling can have disproportionately large effects in the frequency of the most extreme events. Risk due to regional enhancement of precipitation scale increase by dynamical effects must thus be included in vulnerability assessment even if locations are imprecise.
more »
« less
Reconciling Individual Probability Forecasts✱
Individual probabilities refer to the probabilities of outcomes that are realized only once: the probability that it will rain tomorrow, the probability that Alice will die within the next 12 months, the probability that Bob will be arrested for a violent crime in the next 18 months, etc. Individual probabilities are fundamentally unknowable. Nevertheless, we show that two parties who agree on the data—or on how to sample from a data distribution—cannot agree to disagree on how to model individual probabilities. This is because any two models of individual probabilities that substantially disagree can together be used to empirically falsify and improve at least one of the two models. This can be efficiently iterated in a process of “reconciliation” that results in models that both parties agree are superior to the models they started with, and which themselves (almost) agree on the forecasts of individual probabilities (almost) everywhere. We conclude that although individual probabilities are unknowable, they are contestable via a computationally and data efficient process that must lead to agreement. Thus we cannot find ourselves in a situation in which we have two equally accurate and unimprovable models that disagree substantially in their predictions—providing an answer to what is sometimes called the predictive or model multiplicity problem.
more »
« less
- PAR ID:
- 10426401
- Date Published:
- Journal Name:
- ACM Conference on Fairness Accountability and Transparency
- Page Range / eLocation ID:
- 101 to 110
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Reliable probability estimation is of crucial importance in many real-world applications where there is inherent (aleatoric) uncertainty. Probability-estimation models are trained on observed outcomes (e.g. whether it has rained or not, or whether a patient has died or not), because the ground-truth probabilities of the events of interest are typically unknown. The problem is therefore analogous to binary classification, with the difference that the objective is to estimate probabilities rather than predicting the specific outcome. This work investigates probability estimation from high-dimensional data using deep neural networks. There exist several methods to improve the probabilities generated by these models but they mostly focus on model (epistemic) uncertainty. For problems with inherent uncertainty, it is challenging to evaluate performance without access to ground-truth probabilities. To address this, we build a synthetic dataset to study and compare different computable metrics. We evaluate existing methods on the synthetic data as well as on three real-world probability estimation tasks, all of which involve inherent uncertainty: precipitation forecasting from radar images, predicting cancer patient survival from histopathology images, and predicting car crashes from dashcam videos. We also give a theoretical analysis of a model for high-dimensional probability estimation which reproduces several of the phenomena evinced in our experiments. Finally, we propose a new method for probability estimation using neural networks, which modifies the training process to promote output probabilities that are consistent with empirical probabilities computed from the data. The method outperforms existing approaches on most metrics on the simulated as well as real-world data.more » « less
-
We propose a new approach to uncertainty communication: we keep the uncertainty representation fixed, but adjust the distribution displayed to compensate for biases in people’s subjective probability in decision-making. To do so, we adopt a linear-in-probit model of subjective probability and derive two corrections to a Normal distribution based on the model’s intercept and slope: one correcting all right-tailed probabilities, and the other preserving the mode and one focal probability. We then conduct two experiments on U.S. demographically-representative samples. We show participants hypothetical U.S. Senate election forecasts as text or a histogram and elicit their subjective probabilities using a betting task. The first experiment estimates the linear-in-probit intercepts and slopes, and confirms the biases in participants’ subjective probabilities. The second, preregistered follow-up shows participants the bias-corrected forecast distributions. We find the corrections substantially improve participants’ decision quality by reducing the integrated absolute error of their subjective probabilities compared to the true probabilities. These corrections can be generalized to any univariate probability or confidence distribution, giving them broad applicability. Our preprint, code, data, and preregistration are available at https://doi.org/10.17605/osf.io/kcwxmmore » « less
-
Abstract An occupancy model makes use of data that are structured as sets of repeated visits to each of many sites, in order to estimate the actual probability of occupancy (i.e. proportion of occupied sites) after correcting for imperfect detection using the information contained in the sets of repeated observations. We explore the conditions under which preexisting, volunteer-collected data from the citizen science project eBird can be used for fitting occupancy models. Because the majority of eBird’s data are not collected in the form of repeated observations at individual locations, we explore 2 ways in which the single-visit records could be used in occupancy models. First, we assess the potential for space-for-time substitution: aggregating single-visit records from different locations within a region into pseudo-repeat visits. On average, eBird’s observers did not make their observations at locations that were representative of the habitat in the surrounding area, which would lead to biased estimates of occupancy probabilities when using space-for-time substitution. Thus, the use of space-for-time substitution is not always appropriate. Second, we explored the utility of including data from single-visit records to supplement sets of repeated-visit data. In a simulation study we found that inclusion of single-visit records increased the precision of occupancy estimates, but only when detection probabilities are high. When detection probability was low, the addition of single-visit records exacerbated biases in estimates of occupancy probability. We conclude that subsets of data from eBird, and likely from similar projects, can be used for occupancy modeling either using space-for-time substitution or supplementing repeated-visit data with data from single-visit records. The appropriateness of either alternative will depend on the goals of a study and on the probabilities of detection and occupancy of the species of interest.more » « less
-
Cardiomyocytes (CMs), the contractile heart cells that can be derived from human induced pluripotent stem cells (hiPSCs). These hiPSC derived CMs can be used for cardiovascular disease drug testing and regeneration therapies, and they have therapeutic potential. Currently, hiPSC-CM differentiation cannot yet be controlled to yield specific heart cell subtypes consistently. Designing differentiation processes to consistently direct differentiation to specific heart cells is important to realize the full therapeutic potential of hiPSC-CMs. A model that accurately represents the dynamic changes in cell populations from hiPSCs to CMs over the differentiation timeline is a first step towards designing processes for directing differentiation. This paper introduces a microsimulation model for studying temporal changes in the hiPSC-to-early CM differentiation. The differentiation process for each cell in the microsimulation model is represented by a Markov chain model (MCM). The MCM includes cell subtypes representing key developmental stages in hiPSC differentiation to early CMs. These stages include pluripotent stem cells, early primitive streak, late primitive streak, mesodermal progenitors, early cardiac progenitors, late cardiac progenitors, and early CMs. The time taken by a cell to transit from one state to the next state is assumed to be exponentially distributed. The transition probabilities of the Markov chain process and the mean duration parameter of the exponential distribution were estimated using Bayesian optimization. The results predicted by the MCM agree with the data.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3593013.3593980
