Stochastic Downscaling Method Research Articles

Stochastic downscaling of gridded precipitation to spatially coherent subgrid precipitation fields using a transformed Gaussian model

AbstractClimate impact models often require unbiased point‐scale observations, but climate models typically provide biased simulations at the grid scale. While standard bias adjustment methods have shown to generally perform well at adjusting climate model biases, they cannot overcome the gap between grid‐box and point scale. To overcome this limitation, combined bias adjustment and stochastic downscaling methods have been developed. These methods, however, are single‐site methods and cannot represent spatial dependence. Here we propose a multisite stochastic downscaling method that can be applied to bias‐adjusted climate model output for generating spatially coherent time series of daily precipitation at multiple stations, conditional on the driving climate model. The method is based on a transformed truncated multivariate Gaussian model and can also be used to downscale to a full field at finer‐grid resolution. An evaluation for stations across selected catchments in Austria demonstrates the good performance of the stochastic model at representing marginal, temporal and spatial aspects of daily precipitation, including extreme events.

High-resolution stochastic downscaling method for ocean forecasting models and its application to the Red Sea dynamics

Abstract. High-resolution modelling of a large ocean domain requires significant computational resources. The main purpose of this study is to develop an efficient tool for downscaling the lower-resolution data such as those available from Copernicus Marine Environment Monitoring Service (CMEMS). Common methods of downscaling CMEMS ocean models utilise their lower-resolution output as boundary conditions for local, higher-resolution hydrodynamic ocean models. Such methods reveal greater details of spatial distribution of ocean variables; however, they increase the cost of computations and often reduce the model skill due to the so called “double penalty” effect. This effect is a common problem for many high-resolution models where predicted features are displaced in space or time. This paper presents a stochastic–deterministic downscaling (SDD) method, which is an efficient tool for downscaling of ocean models based on the combination of deterministic and stochastic approaches. The ability of the SDD method is first demonstrated in an idealised case when the true solution is known a priori. Then the method is applied to create an operational Stochastic Model of the Red Sea (SMORS), with the parent model being the Mercator Global Ocean Analysis and Forecast System at 1/12∘ resolution. The stochastic component of the model is data-driven rather than equation-driven, and it is applied to the areas smaller than the Rossby radius, within which distributions of ocean variables are more coherent than over a larger distance. The method, based on objective analysis, is similar to what is used for data assimilation in ocean models and stems from the philosophy of 2-D turbulence. SMORS produces finer-resolution (1/24∘ latitude mesh) oceanographic data using the output from a coarser-resolution (1/12∘ mesh) parent model available from CMEMS. The values on the fine-resolution mesh are computed under conditions of minimisation of the cost function, which represents the error between the model and true solution. SMORS has been validated against sea surface temperature and ARGO float observations. Comparisons show that the model and observations are in good agreement and SMORS is not subject to the “double penalty” effect. SMORS is very fast to run on a typical desktop PC and can be relocated to another area of the ocean.

Deriving high spatiotemporal rainfall information over Singapore through dynamic-stochastic modelling using ‘HiDRUS’

The study of climate change adaptation plans for drainage infrastructure in a small country as that of Singapore, rainfall projections on the time scale of minutes and on the spatial scales of 1 km are deemed appropriate In this paper, we introduce an application of radar-based stochastic downscaling for rainfall projections at high temporal and spatial resolutions. The input for stochastic model is derived from a Regional Climate Model. The sub-hourly extreme rainfall intensity derived from stochastic model outputs was validated against observed rain-gauge data over the historical period. Considering the advantage in computational efficiency of the stochastic downscaling method, thousand scenarios of rainfall projections at very high temporal and spatial resolution were generated. The implication of this approach is that, from these stochastically downscaled time series of rainfall, it is possible to study future sub-hourly extreme rainfall intensities which would be useful to address issue of flash floods/drainage systems.

PreciPatch: A Dictionary-based Precipitation Downscaling Method

Climate and weather data such as precipitation derived from Global Climate Models (GCMs) and satellite observations are essential for the global and local hydrological assessment. However, most climatic popular precipitation products (with spatial resolutions coarser than 10km) are too coarse for local impact studies and require “downscaling” to obtain higher resolutions. Traditional precipitation downscaling methods such as statistical and dynamic downscaling require an input of additional meteorological variables, and very few are applicable for downscaling hourly precipitation for higher spatial resolution. Based on dynamic dictionary learning, we propose a new downscaling method, PreciPatch, to address this challenge by producing spatially distributed higher resolution precipitation fields with only precipitation input from GCMs at hourly temporal resolution and a large geographical extent. Using aggregated Integrated Multi-satellitE Retrievals for GPM (IMERG) data, an experiment was conducted to evaluate the performance of PreciPatch, in comparison with bicubic interpolation using RainFARM—a stochastic downscaling method, and DeepSD—a Super-Resolution Convolutional Neural Network (SRCNN) based downscaling method. PreciPatch demonstrates better performance than other methods for downscaling short-duration precipitation events (used historical data from 2014 to 2017 as the training set to estimate high-resolution hourly events in 2018).

Downscaling climate change of water availability, sediment yield and extreme events: Application to a Mediterranean climate basin

A robust hydrological impact assessment is indispensable for mitigation and adaptation planning. This study presents an integrated modelling methodology for evaluating climate change impacts on water availability, sediment yield and extreme events at the catchment scale. We propose the use of the spatial–temporal Neyman–Scott Rectangular Pulses (STNSRP) model—RainSim V3 and the rainfall conditioned daily weather generator—ICAAM‐WG, as well as the physically based spatially distributed hydrological model—SHETRAN. The change factor approach was applied for obtaining unbiased rainfall and temperature statistics. The ICAAM‐WG was developed based on the modified Climate Research Unit daily Weather Generator (CRU‐WG). The methodology is proposed to generate synthetic series of hourly precipitation, daily temperature and potential evapotranspiration, hourly runoff and hourly sediment discharge. We demonstrated a possible application in a 705‐km2 Mediterranean climate basin in southern Portugal. The case study showed the evaluation of future climate change impacts on annual and monthly water balance components and sediment yield, annual and seasonal flow duration curves, empirical extreme value distributions and the theoretical fits. It did not consider the possible uncertainty due to the limit of computational resources.The methodology can be well justified as follows: (a) the use of synthetic hourly instead of daily precipitation enables SHETRAN to be more capable of reproducing reliable storm runoff processes and the consequent sediment transport processes; (b) the use of SHETRAN makes possible the impact assessment to be accessible for any model grid square within the study basin; (c) the use of a statistical–stochastic downscaling method facilitates the generation of the synthetic series with unlimited length. It makes possible robust hydrological impact assessments if uncertainties related to the global climate model, regional climate model, greenhouse gas emission scenario, downscaling method, hydrological model and observational data are considered.

A combined statistical bias correction and stochastic downscaling method for precipitation

Abstract. Much of our knowledge about future changes in precipitation relies on global (GCMs) and/or regional climate models (RCMs) that have resolutions which are much coarser than typical spatial scales of precipitation, particularly extremes. The major problems with these projections are both climate model biases and the gap between gridbox and point scale. Wong et al. (2014) developed a model to jointly bias correct and downscale precipitation at daily scales. This approach, however, relied on pairwise correspondence between predictor and predictand for calibration, and, thus, on nudged simulations which are rarely available. Here we present an extension of this approach that separates the downscaling from the bias correction and in principle is applicable to free-running GCMs/RCMs. In a first step, we bias correct RCM-simulated precipitation against gridded observations at the same scale using a parametric quantile mapping (QMgrid) approach. In a second step, we bridge the scale gap: we predict local variance employing a regression-based model with coarse-scale precipitation as a predictor. The regression model is calibrated between gridded and point-scale (station) observations. For this concept we present one specific implementation, although the optimal model may differ for each studied location. To correct the whole distribution including extreme tails we apply a mixture distribution of a gamma distribution for the precipitation mass and a generalized Pareto distribution for the extreme tail in the first step. For the second step a vector generalized linear gamma model is employed. For evaluation we adopt the perfect predictor experimental setup of VALUE. We also compare our method to the classical QM as it is usually applied, i.e., between RCM and point scale (QMpoint). Precipitation is in most cases improved by (parts of) our method across different European climates. The method generally performs better in summer than in winter and in winter best in the Mediterranean region, with a mild winter climate, and worst for continental winter climate in Mid- and eastern Europe or Scandinavia. While QMpoint performs similarly (better for continental winter) to our combined method in reducing the bias and representing heavy precipitation, it is not capable of correctly modeling point-scale spatial dependence of summer precipitation. A strength of this two-step method is that the best combination of bias correction and downscaling methods can be selected. This implies that the concept can be extended to a wide range of method combinations.

Modeling the wind circulation around mills with a Lagrangian stochastic approach

This work aims at introducing model methodology and numerical studies related to a {Lagrangian} stochastic approach applied to the computation of the wind circulation around mills. We adapt the Lagrangian \textit{stochastic downscaling method} that we have introduced in [3] and [4] to the atmospheric boundary layer and we introduce here a Lagrangian version of the \textit{actuator disc} methods to take account of the mills. We present our numerical method and numerical experiments in the case of non rotating and rotating actuator disc models. First, for validation purpose we compare some numerical experiments against wind tunnel measurements. Second we perform some numerical experiments at the atmospheric scale and present some features of our numerical method, in particular the computation of the probability distribution of the wind in the wake zone, as a byproduct of the fluid particle model and the associated PDF method.

Sampling the uncertainty associated with segmented normal fault interpretation using a stochastic downscaling method

A large-scale normal fault may be composed of several overlapping fault segments separated by relay zones at a finer scale. Fault segmentation may be critical to the understanding and the forecast of physical phenomena at the reservoir or basin scale (e.g. fluid flow, seismic rupture propagation). In this paper, we propose an automatic and stochastic method to sub-divide (downscale) below the data resolution a segmented normal fault into en-echelon segments that may be linked by connecting faults, based on variations in the orientation of the fault. The downscaling algorithm is composed of three main steps. (1) The first step involves detecting the segments using geometrical criteria. (2) Then the overlapping segments are modeled using isolated fault descriptions and statistics. (3) Lastly, the maturity of the simulated relay zones is evaluated based on relay geometry. If a given maturity threshold is reached, the relay ramp is breached. These three downscaling steps depend on seven parameters that can be defined as constant or randomly chosen from probability distributions to sample uncertainties. The method was applied to a large normal fault that laterally limits a hydrocarbon reservoir and is poorly imaged from seismic data. This resulted in one hundred possible 3D fault array models. The analysis of the emerging parameters shows a strong variability of the number and length of segments, and of the overlap and spacing between segments.

Stochastic Lagrangian method for downscaling problems in computational fluid dynamics

This work aims at introducing modelization, theoretical and numerical studies related to a new downscaling technique applied to Computational Fluid Dynamics. Our method consists in building a local model, forced by large scale information computed thanks to a classical numerical weather predictor. The local model, compatible with the Navier-Stokes equations, is used for the small scale computation (downscaling) of the considered fluid. It is inspired by S.B. Pope's works on turbulence, and consists in a so-called Langevin system of stochastic differential equations. We introduce this model and exhibit its links with classical RANS models. Well-posedness, as well as mean-field interacting particle approximations and boundary condition issues are addressed. We present the numerical discretization of the stochastic downscaling method and investigate the accuracy of the proposed algorithm on simplified situations.

Stochastic downscaling method: application to wind refinement

In this article, we propose a new stochastic downscaling method: provided a numerical prediction of wind at large scale, we aim to improve the approximation at small scales thanks to a local stochastic model. We first recall the framework of a Lagrangian stochastic model borrowed from Pope. Then, we adapt it to our meteorological framework, both from the theoretical and numerical viewpoints. Finally, we present some promising numerical results corresponding to the simulation of wind over the Mediterranean Sea.

Meteorological uncertainty and rainfall downscaling

Abstract. We explore the sources of forecast uncertainty in a mixed dynamical-stochastic ensemble prediction chain for small-scale precipitation, suitable for hydrological applications. To this end, we apply the stochastic downscaling method RainFARM to each member of ensemble limited-area forecasts provided by the COSMO-LEPS system. Aim of the work is to quantitatively compare the relative weights of the meteorological uncertainty associated with large-scale synoptic conditions (represented by the ensemble of dynamical forecasts) and of the uncertainty due to small-scale processes (represented by the set of fields generated by stochastic downscaling). We show that, in current operational configurations, small- and large-scale uncertainties have roughly the same weight. These results can be used to pinpoint the specific components of the prediction chain where a better estimate of forecast uncertainty is needed.

Current regional climate change studies in Hungary : a review

Abstract. After focusing on the changes in Hungarian temperature and preeipitation during this Century, possible hydrological, agricultural and ecological consequences of a future climate change are described. These results have been obtained using a modified version of empirical downscaling techniques, developed to analyse the local effects of global climate change in a twofold concentration of atmospheric greenhouse gases scenario. In addition, regional changes in temperature and precipitation were examined with the help of the more specific stochastic downscaling method. The climate of Hungary has become warmer and drier over the last Century. It is to be expected that an increasing concentration of atmospheric greenhouse gases will enhance the tendency towards aridification.