Model Weighting for One-Dependence Estimators by Measuring the Independence Assumptions

AbstractUncertainty in long-term projections of future climate can be substantial and presents a major challenge to climate change adaptation planning. This is especially so for projections of future precipitation in most tropical regions, at the spatial scale of many adaptation decisions in water-related sectors. Attempts have been made to constrain the uncertainty in climate projections, based on the recognised premise that not all of the climate models openly available perform equally well. However, there is no agreed ‘good practice’ on how to weight climate models. Nor is it clear to what extent model weighting can constrain uncertainty in decision-relevant climate quantities. We address this challenge, for climate projection information relevant to ‘high stakes’ investment decisions across the ‘water-energy-food’ sectors, using two case-study river basins in Tanzania and Malawi. We compare future climate risk profiles of simple decision-relevant indicators for water-related sectors, derived using hydrological and water resources models, which are driven by an ensemble of future climate model projections. In generating these ensembles, we implement a range of climate model weighting approaches, based on context-relevant climate model performance metrics and assessment. Our case-specific results show the various model weighting approaches have limited systematic effect on the spread of risk profiles. Sensitivity to climate model weighting is lower than overall uncertainty and is considerably less than the uncertainty resulting from bias correction methodologies. However, some of the more subtle effects on sectoral risk profiles from the more ‘aggressive’ model weighting approaches could be important to investment decisions depending on the decision context. For application, model weighting is justified in principle, but a credible approach should be very carefully designed and rooted in robust understanding of relevant physical processes to formulate appropriate metrics.

Download Full-text

Mediterranean climate change projections: an update from CMIP5 and CMIP6.

10.5194/egusphere-egu21-12803 ◽

2021 ◽

Author(s):

Josep Cos ◽

Francisco J Doblas-Reyes ◽

Martin Jury

Keyword(s):

Climate Change ◽

21St Century ◽

Hot Spot ◽

Weighting Scheme ◽

Temperature And Precipitation ◽

High Emission ◽

The Mediterranean ◽

Model Ensembles ◽

Projected Changes ◽

Model Weighting

The Mediterranean has been identified as a climate change hot-spot due to increased warming trends and precipitation decline. Recently, CMIP6 was found to show a higher climate sensitivity than its predecessor CMIP5, potentially further exacerbating related impacts on the Mediterranean region.To estimate the impacts of the ongoing climate change on the region, we compare projections of various CMIP5 and CMIP6 experiments and scenarios. In particular, we focus on summer and winter changes in temperature and precipitation for the 21st century under RCP2.6/SSP1-2.6, RCP4.5/SSP2-4.5 and RCP8.5/SSP5-8.5 as well as the high resolution HighResMIP experiments. Additionally, to give robust estimates of projected changes we apply a novel model weighting scheme, accounting for historical performance and inter-independence of the multi-member multi-model ensembles, using ERA5, JRA55 and WFDE5 as observational reference.&#160;Our results indicate a significant and robust warming over the Mediterranean during the 21st century irrespective of the used ensemble and experiments. Nevertheless, the often attested amplified Mediterranean warming is only found for summer. The projected changes vary between the CMIP5 and CMIP6, with the latter projecting a stronger warming. For the high emission scenarios and without weighting, CMIP5 indicates a warming between 4 and 7.7&#186;C in summer and 2.7 and 5&#186;C in winter, while CMIP6 projects temperature increases between 5.6 and 9.2&#186;C in summer and 3.2 to 6.8&#186;C in winter until 2081-2100 in respect to 1985-2005. In contrast to temperature, precipitation changes show a higher level of uncertainty and spatial heterogeneity. However, for the high emission scenario, a robust decline in precipitation is projected for large parts of the Mediterranean during summer. First results applying the model weighting scheme indicate reductions in CMIP6 and increases in CMIP5 warming trends, thereby reducing differences between the two ensembles.

Download Full-text

Effect of diversified performance metrics and climate model weighting on global and regional trend patterns of precipitation and temperature

Erdkunde ◽

10.3112/erdkunde.2019.04.04 ◽

2019 ◽

Vol 73 (4) ◽

pp. 303-322

Author(s):

Christoph Ring ◽

Felix Pollinger ◽

Luzia Keupp ◽

Irena Kaspar-Ott ◽

Elke Hertig ◽

...

Keyword(s):

Performance Metrics ◽

Climate Model ◽

Regional Trend ◽

Model Weighting ◽

Precipitation And Temperature

Download Full-text

Taiwan-Specific Model for VS30 Prediction Considering Between-Proxy Correlations

Earthquake Spectra ◽

10.1193/061217eqs113m ◽

2018 ◽

Vol 34 (4) ◽

pp. 1973-1993 ◽

Cited By ~ 12

Author(s):

On Lei Annie Kwok ◽

Jonathan P. Stewart ◽

Dong Youp Kwak ◽

Pang-Li Sun

Keyword(s):

Ground Motion ◽

Prediction Models ◽

Site Response ◽

Specific Model ◽

Weighted Mean ◽

Response Models ◽

Database File ◽

A Site ◽

Gradient Effects ◽

Model Weighting

Ergodic site response models are generally conditional on the time-averaged shear wave velocity in the upper 30 m ( V S30). Ground motion databases contain many recordings from Taiwan, and because of site characterization efforts, 56% of recording sites have V S30 derived from measurements. We develop proxy-based V S30 prediction models, one application of which is for the remaining 44% of Taiwan sites. Our approach, which can be suitable for other regions, differs from previous studies in which proxies are based on detailed geologic categories and possible within-category topographic gradient effects. Instead, we use three broad, age-based geologic categories, and for the youngest category of Holocene and Quaternary undivided sediments, we propose models conditioned on gradient and elevation. We also adapt a geomorphic terrain-based method, thus providing two V S30-prediction models. We describe a model weighting scheme that combines the models in consideration of their relative dispersions and correlation, producing a weighted mean and standard deviation natural-log V S30. Included as an electronic supplement is a profile database file and a site database with site parameters for Taiwan ground motion stations.

Download Full-text

Note on the output of a queuing system

Journal of Applied Probability ◽

10.1017/s0021900200032976 ◽

1969 ◽

Vol 6 (02) ◽

pp. 459-461 ◽

Cited By ~ 4

Author(s):

Duane C. Boes

Keyword(s):

Exponential Distribution ◽

Queuing System ◽

Service Completion ◽

Service Times ◽

Independence Assumptions

Consider a queuing system that has c servers and d waiting positions. Assume that the input is Poisson with rate α and the service times are exponential with mean β –1. Further assume the following: (i) a customer arriving when all servers are busy and all waiting positions are occupied is “cleared” from the system; (ii) a customer arriving when all servers are busy and not all waiting positions are occupied waits with probability 1 – ζ and “balks” or “clears” with probability ζ; (iii) a customer arriving when not all servers are busy commences service immediately (never balks); and, (iv) a customer who is waiting for service may “defect”, the distribution of time until a waiting customer defects being given by an exponential distribution with mean γ–1. Also, the usual independence assumptions, which make the process that is described by the number in the system at time t Markov, are assumed. An “output” of this queuing system is defined to occur whenever a service completion occurs, or whenever an arrival “clears” or “balks”, or whenever a waiting customer “defects”. Thus the output is a pooling of service completion epochs, the epochs when arrivals are cleared, the epochs when arrivals balk, and the defection epochs.

Download Full-text

On search for a Brownian target

Journal of Applied Probability ◽

10.1017/s0021900200046970 ◽

1980 ◽

Vol 17 (01) ◽

pp. 243-247

Author(s):

Thomas L. Corwin

Keyword(s):

Wiener Process ◽

Failure Probability ◽

Search Effort ◽

Exponential Functional ◽

Independence Assumptions

A target is assumed to move according to a Wiener process in ℝ1. The probability of detecting the target is computed in terms of the search effort which accumulates along the target's path. Under certain independence assumptions this probability is given by the expectation of an exponential functional of the process. It is shown in this note that the failure probability in a search for a Wiener target is asymptotically proportional to , where T is the accumulated time spent searching. The asymptotic failure probability is also shown to be independent of the position of the search in ℝ1. In a similar fashion, it is shown that the failure probability in a search for a Wiener target in ℝ2 is independent of the position of the search and asymptotically proportional to (c log T + l)–1, c >0

Download Full-text