Roland Stull

The work in this dissertation enhances precipitation forecast skill with a focus on southwest British Columbia (BC), Canada. Local weather predictions in this region can be subject to large uncertainties because of the complex terrain and the upstream data void. Electricity production in BC relies heavily on hydropower. Thus, accurate precipitation forecasts are crucial to manage water resources and mitigate flood risks.There are two major components to a skillful prediction system: (1) numerical weather prediction (NWP), and (2) post-processing of the NWP output. First, this work investigates sensitivities of precipitation performance to configurations of the Weather Research and Forecasting (WRF) model. Skill varies by model grid spacing, parameterization selections, location, season, precipitation intensity, and accumulation period. An evaluation of over 100 systematically varied WRF configurations provides insight to precipitation forecasting challenges and shows that the optimal model setup depends on the weather situation and the verification metric most important to end users.A few of the best performing model configurations are then post-processed using the analog ensemble (AnEn) method. This statistical method derives a future forecast by searching an archive of past model forecasts for similar (analog) conditions, and then collects the corresponding past observations into an ensemble. This dissertation utilizes existing and new optimization techniques to significantly improve AnEn computational efficiency and forecast skill. The detection of good analogs is improved through predictor weighting methods and consideration of temporal predictor trends and supplemental lead times.Finally, this work proposes new ensemble generation techniques. Applying the past analog error to the target forecast reduces the AnEn dry bias and makes prediction of high-impact (heavy-precipitation) events more reliable. A multi-model AnEn further improves predictive skill, but at higher computational cost. The AnEn performance shows larger sensitivity to the grid spacing of the NWP than to the physics configuration.The final ensemble prediction system provides skillful and reliable high-resolution forecasts across all precipitation intensities.

View record

Medium range precipitation forecasts are a crucial input of hydrology models that provide streamflow information for water resource management and flood risk assessments. Generating accurate and timely precipitation forecasts has been a long-standing challenge in British Columbia (BC), Canada, because of its complex terrain and a paucity-of-data problem.In this dissertation, a novel precipitation forecast post-processing routine for BC is developed to convert raw ensembles into bias-corrected, probabilistically calibrated, and downscaled spatiotemporal sequences out to 7 days.The post-processing routine features a hybrid of conventional statistical methods and state-of-the-art Convolutional Neural Networks (CNNs). In the bias-correction and calibration stage, raw ensembles are converted to an Analog Ensemble (AnEn) first and then reconstructed to physically realistic spatiotemporal sequences using the Minimum Divergence Schaake Shuffle (MDSS). These sequences are further bias-corrected by a CNN that considers climatology and terrain information. In the downscaling stage, a CNN pre-trained with high-quality, high-resolution precipitation analysis in the continental US is applied and transferred to BC without acquiring extra training data. It downscales post-processed precipitation sequences into 4-km grid spacing, which resolves small-scale terrain features. Additionally, for operating the post-processing methods on a near-real-time basis, a CNN-based precipitation observation quality control procedure is developed. It removes suspicious observations and returns clean observations that can be used to measure and verify post-processed precipitation forecasts.This post-processing routine is developed for the Global Ensemble Forecast System (GEFS) 3-hourly precipitation forecasts, and it is tested by the GEFS reforecasts from 2017 to 2019. Station-observation-based verification indicates that the post-processed precipitation ensembles are skillful in the BC South Coast, Southern Interior, and Northeast---watersheds with diverse climatological conditions. Compared to conventional statistical post-processing, the methods in this dissertation achieved roughly a 10% increase of Continuous Ranked Probability Skill Score (CRPSS) in all lead times. The Brier Skill Scores (BSS) of heavy precipitation events are increased up to 60% for both 3-hourly lead times and 7-day accumulated totals. In summary, this dissertation pioneers the combination of conventional statistical post-processing and neural networks, and is one of only a few studies pertaining to precipitation ensemble post-processing in BC.

View record

The buoyant rise of wildfire smoke and the resultant vertical distribution of emission products in the atmosphere have a strong influence on downwind pollutant concentrations at the surface, and provide key input into regional and global chemical transport models. Due to inherent complexity of wildfire plume dynamics, smoke injection height predictions are subject to large uncertainties. One of the obstacles to the development of new plume rise parameterizations has been the scarcity of detailed simultaneous observations of fire-generated turbulence, entrainment, smoke concentrations and fire behavior. This thesis makes contributions on two fronts: (i) it demonstrates the feasibility of using coupled fire-atmosphere large-eddy simulations to model wildfire smoke dynamics to produce "synthetic" plume data, and (ii) develops a new energy balance plume rise parameterization to predict the vertical distribution of smoke in the atmosphere. The first part of the thesis focuses on evaluating the large-eddy simulation model used in this work with a detailed observational dataset from a real prescribed burn. The next portion explores the effect of various fire parameters and ambient atmospheric conditions on smoke plume behavior using a range of sensitivity studies. Analysis of flow dynamics shows that the updraft is shaped by complex interactions of fire-induced winds and vorticity generated in response to a near-surface convergence, and does not conform to commonly used mixing and entrainment assumptions.With the knowledge gained through the above numerical experiments, the second half of the thesis introduces a simple parameterization for predicting the mean centerline height for penetrative plumes from fires of arbitrary shape and intensity. Lastly, the proposed parameterization is extended to capture the full vertical distribution of smoke in the atmosphere. The broad goal of this work is to better our understanding of plume rise dynamics and improve smoke dispersion predictions within air quality applications.

View record

The ability to forecast extreme temperature and precipitation events not only helps satisfy the public’s desire to better prepare for such events, but can also provide valuable information about the future risks of such events to emergency managers, regional planners, and policy-makers at all levels of government. This dissertation advances extreme weather forecasting over the complex topography of British Columbia (BC) while accounting for changes in intensity and frequency of extreme events due to nonstationarity.First the problem of finding a dataset to provide climatological distributions is addressed. Weather station data coverage, quality, and temporal completeness across BC degrade outside of population centres, and as one goes back in time. This data paucity motivates the search for the best reanalysis to serve as a climatological reference dataset. The 2-m temperature and daily accumulated precipitation of the reanalyses are compared with observations from meteorological stations distributed over the complex terrain of British Columbia. Upon thorough evaluation, the Japanese 55-year Reanalysis (JRA-55) was found to be best. The second component of this works combines, downscales and bias corrects the best performing reanalysis using the high-spatial-resolution Parameter-Elevation Regressions on Independent Slopes Model (PRISM) dataset and using surface weather station observations. This results in a high-resolution, long-term gridded dataset that is spatially and temporally complete, yielding a very-high-resolution surface analysis (VHRSA).Next, this dataset is used to create a high-resolution, bias-corrected ensemble forecast using the North American Ensemble Forecast System (NAEFS). The post-processed NAEFS is more skillful than the raw NAEFS forecast out to a forecast lead time of 10 days for both 2-m temperature, and daily accumulated precipitation.Statistical temporal stationarity of extreme values of precipitation and temperature are assessed for the 60-year VHRSA period. It is determined that nonstationary distributions should be used to represent annual minima values of daily minimum 2-m temperature during summer months and late winter.Finally, an extreme, or situational awareness index is presented: the Parametric Extreme Index (PEI). It can be used to alert forecasters and other end users of future extreme temperature and precipitation events.

View record

Wind-speed forecasts from numerical-weather-prediction (NWP) models are important for daily wind-resource generation planning. However, NWP models are imperfect. The ability of energy planners to efficiently manage resources is a function of the accuracy of deterministic wind forecasts and of the associated probability estimates of forecast uncertainty. As the amount of energy generated from wind increases to significant levels, improving forecast accuracy and representation of forecast uncertainty is a key area of active research. This dissertation advances wind forecasting over regions of complex topography using the Weather Research and Forecasting (WRF) model. The optimal WRF-model configuration is a function of planetary-boundary-layer (PBL) physics, grid length, and initial-condition choice. The first component of this work determines which of these three factors most influences forecast accuracy over complex terrain. The two largest factors influencing forecast accuracy are the PBL-physics scheme and the grid length, with the dominant factor being a function of location, season, and time of day.The second component of the research addresses the need for probability-based forecast information, which is only recently being used within the industry. Wind forecasts from an ensemble using eight PBL schemes, three grid lengths, and two initial-conditions sources are converted into probability models that are then evaluated. Using the full, empirical ensemble distribution produces uncalibrated probabilistic forecasts. Prescribing a Gaussian probability distribution based on statistical moments of a past training dataset results in calibrated and sharp probabilistic forecasts. Such a method is also computationally cheap. The final aspect of this study evaluates the role of boundary-layer static stability on forecast performance. Traditionally, empirical surface-layer similarity theory has been used to relate surface fluxes of heat, momentum, and moisture to vertical profiles of temperature and wind. To evaluate and improve surface-layer similarity theories over mountain ridges, a year-long field campaign of temperature and wind measurements was conducted at wind farms in British Columbia. New empirical equations for complex terrain are proposed based on the field data, and found to perform well at an independent test location.

View record

Traditional semi-Lagrangian dynamical solvers are widely used in current global numerical weather prediction (NWP) and climate models, but are known to lack inherent mass-conserving properties. Some newer approaches utilize a cell-integrated (conservative) semi-Lagrangian (CISL) semi-implicit solver, which is inherently mass-conserving. However, existing CISL semi-implicit solvers lack consistent formulation among the discrete continuity equation and other discrete conservation equations for scalar tracers such as water vapour and air pollutants. Such inconsistency can lead to spurious generation or removal of scalar mass. In this dissertation, a new cell-integrated semi-Lagrangian (CISL) semi-implicit nonhydrostatic solver is presented with consistent discrete mass conservation equations for air and all tracers, and which preserves the shape of tracer-mass distribution. The discretization does not depend on a mean reference state, but maintains the same framework as typical semi-implicit CISL solvers, where a linear Helmholtz equation is constructed and a single application of the cell-integrated transport scheme is needed for scalar transport. Tests of this new solver are made for a series of increasingly complex flow scenarios. The initial testbed utilizes the hydrostatic, incompressible, shallow-water equations, for which the new solver is shown to be numerically stable. It maintains accuracy comparable to other existing solvers even for a highly nonlinear unstable jet. The second suite of tests are for nonhydrostatic two-dimensional (x-z) fully compressible flows in the atmosphere as governed by the moist Euler equations, which compare well to several idealized benchmark test cases from the literature. The third flow scenario is complex orography, where the nonhydrostatic equations are transformed to use a terrain-following height coordinate. Results from test cases of dry and moist flows over idealized mountain shapes are presented. In summary, the prototype development work presented in this dissertation shows that the proposed CISL nonhydrostatic solver with conservative and consistent transport may be a desirable candidate for a dynamical core in comprehensive global NWP and climate models.

View record

This dissertation presents a reliable probabilistic forecasting system designed to predict inflows to hydroelectric reservoirs. Forecasts are derived from a Member-to-Member (M2M) ensemble in which an ensemble of distributed hydrologic models is driven by the gridded output of an ensemble of numerical weather prediction (NWP) models. Multiple parameter sets for each hydrologic model are optimized using objective functions that favour different aspects of forecast performance. On each forecast day, initial conditions for each differently-optimized hydrologic model are updated using meteorological observations. Thus, the M2M ensemble explicitly samples inflow forecast uncertainty caused by errors in the hydrologic models, their parameterizations, and in the initial and boundary conditions (i.e., meteorological data) used to drive the model forecasts.Bias is removed from the individual ensemble members using a simple degree-of-mass-balance bias correction scheme. The M2M ensemble is then transformed into a probabilistic inflow forecast by applying appropriate uncertainty models during different seasons of the water year. The uncertainty models apply ensemble model output statistics to correct for deficiencies in M2M spread. Further improvement is found after applying a probability calibration scheme that amounts to a re-labelling of forecast probabilities based on past performance. Each component of the M2M ensemble has an associated cost in terms of time and/or money. The relative value of each ensemble component is assessed by removing it from the ensemble and comparing the economic gains associated with the reduced ensembles to those achieved using the full M2M system. Relative value is computed using a simple (static) cost-loss decision model in which the reservoir operator takes action (lowers the reservoir level) when significant inflows are predicted with probability exceeding some threshold. The probabilistic reservoir inflow forecasting system developed in this dissertation is applied to the Daisy Lake hydroelectric reservoir located in the complex terrain of southwestern British Columbia, Canada. The hydroclimatic regime of the case study watershed is such that flashy fall and winter inflows are driven by Pacific frontal systems, while spring and summer inflows are dominated by snow and glacier melt. Various aspects of ensemble and probabilistic forecast performance are evaluated over a period of three water years.

View record

Accurately calculating snow-surface temperature and liquid-water content for a groomed and compacted ski run, known as a ski piste, is crucial to the preparation of fast skis for alpine racing. This dissertation case study focuses on modelling the above variables for a clear-sky intensive observation period in February 2010. An automated weather station collected relevant meteorological data at a point on a ski piste in Whistler, BC, Canada, known as RC Whistler.The radiation budget is fundamental to this problem, and is affected by tall trees dominating the local horizon. Tree temperature was measured using an infrared camera to estimate thermal emissions. This data, along with calculations of sky downwelling longwave radiation by a radiative transfer model, was input to a new model created for this research. Longwave radiation contributions from trees and sky were weighted by their view factors, which had been calculated from a theodolite survey. Model output is total downwelling longwave radiation at the snow surface for RC Whistler, under clear skies.Downwelling solar radiation penetrates the snowpack, while the surface itself undergoes infrared cooling, resulting in a substantial temperature gradient just beneath the snow surface. A new one-dimensional numerical Lagrangian snowpack model has been written, solving the heat-, liquid-water-, and ice-budget equations to calculate the snow-surface temperature. Meteorological measurements from the clear-sky intensive observation period are prescribed as boundary conditions. Model components and parameters are validated and chosen with idealized model runs. In addition to natural atmospheric processes occurring at and just above the snow surface, human factors were considered. These are frequent skiers compacting the snowpack, and grooming snowcats that churn the top layer of the snowpack and work to increase the snow density and hardness, usually once daily. These effects are simulated in the numerical model.The model successfully simulates snow-surface temperature for the RC Whistler clear-sky intensive observation period. This exploratory investigation indicates that the model shows promise. It is a starting point for a more sophisticated version, incorporating complex boundary conditions such as precipitation and cloudiness, and later being driven by numerical weather prediction output.

View record

This dissertation presents a probabilistic weather prediction system for operational (real-time) usage. The proposed system provides complete probability distributions for both continuous weather variables, such as temperature, and mixed discrete-continuous variables like precipitation accumulations.The proposed system decomposes the process of generating probabilistic forecasts into a series of sequential steps, each of which is important in the overall goal of providing probabilistic forecasts of high quality. Starting with an ensemble of input predictors generated by numerical weather prediction models, the system uses the following four components: 1) correction; 2) uncertainty modeling; 3) calibration; and 4) updating. The correction component bias-corrects the input predictors. The uncertainty model converts these predictors into a suitable probability distribution. The calibration component improves this distribution by removing any distributional bias. The update component further improves the forecast by incorporating recently made observations of the true state.The system is designed to be modular. Namely, different implementations of each component can be used interchangeably with any combination of implementations for the other components. This allows future research into probabilistic forecasting to be focused on any one component and also allows new methods to be easily incorporated into the system.The system uses a number of existing correction and uncertainty models, but the dissertation also presents two new methods: Firstly, a new method for calibrating probabilistic forecasts is created. This method is shown to improve probabilistic forecasts that exhibit distributional bias. Secondly, a new method for incorporating recently made observations to existing probabilistic forecasts is developed.The system and its components are tested using meteorological data from daily operational runs of ensemble numerical weather prediction models and their verifying observations from surface weather stations in North America. Each component's contribution to overall forecast quality is analysed.

View record

This dissertation describes research to enhance hydrometeorological forecasts and their application towards clean energy. The secondary objective of this research is exploration of a new evolutionary algorithm as a possible statistical tool to describe some nonlinear aspects of the atmosphere. The products of this work are summarized in four chapters.Motivated by the difficulty in forecasting montane precipitation for hydroelectricity, a novel model output statistical method is introduced to improve numerical daily precipitation forecasts. The proposed method is gene expression programming (GEP). It is used to create a bias-corrected ensemble, called a deterministic ensemble forecast (DEF), which could serve as an alternative to the traditional ensemble average. Comparing the verification scores of GEP DEF vs. an equally- weighted (traditional) ensemble-average DEF, it is found that GEP DEFs were better for about half of the mountain weather stations tested.The need for an enhanced electric load forecasting model with better connections to weather variables is addressed next. GEP is used to forecast relative load minima during nighttime and mid- day, and relative load maxima in the morning and evening. A different method is introduced to use GEP to forecast electric load for the next hour. These methods are verified against independent data for a year of daily load forecasts, and are compared against the operational load forecasts archived by BC Hydro, British Columbia’s largest electric utility company.Also, GEP is used to parametrize two non-iterative approximations for saturated pseudoadiabats (also known as moist adiabats). One approximation determines which moist adiabat passes through a point of known pressure and temperature, such as through the lifting condensation level on a skew- T or tephigram. The other approximation determines the air temperature at any pressure along a known moist adiabat, such as the final temperature of a rising cloudy air parcel. This work can be used to better predict cloudy convection in the atmosphere, which can cause hazardous wind gusts at wind turbines, and can drop heavy precipitation in hydroelectric watersheds.

View record

This dissertation describes research designed to enhance hydrometeorological forecasts. The objective of the research is to deliver an optimal methodology to produce reliable, skillful and economically valuable probabilistic temperature and precipitation forecasts.Weather plays a dominant role for energy companies relying on forecasts of watershed precipitation and temperature to drive reservoir models, and forecasts of temperatures to meet energy demand requirements. Extraordinary precipitation events and temperature extremes involve consequential water- and power-management decisions.This research compared weighted-average, recursive, and model output statistics bias-correction methods and determined optimal window-length to calibrate temperature and precipitation forecasts. The research evaluated seven different methods for daily maximum and minimum temperature forecasts, and three different methods for daily quantitative precipitation forecasts, within a region of complex terrain in southwestern British Columbia, Canada.This research then examined ensemble prediction system design by assessing a three-model suite of multi-resolution limited area mesoscale models. The research employed two different economic models to investigate the ensemble design that produced the highest-quality, most valuable forecasts. The best post-processing methods for temperature forecasts included moving-weighted average methods and a Kalman filter method. The optimal window-length proved to be 14 days. The best post-processing methods for achieving mass balance in quantitative precipitation forecasts were a moving-average method and the best easy systematic estimator method. The optimal window-length for moving-average quantitative precipitation forecasts was 40 days. The best ensemble configuration incorporated all resolution members from all three models. A cost/loss model adapted specifically for the hydro-electric energy sector indicated that operators managing rainfall-dominated, high-head reservoirs should lower their reservoir with relatively low probabilities of forecast precipitation. A reservoir-operation model based on decision theory and variable energy pricing showed that applying an ensemble-average or full-ensemble precipitation forecast provided a much greater profit than using only a single deterministic high-resolution forecast.Finally, a bias-corrected super-ensemble prediction system was designed to produce probabilistic temperature forecasts for ten cities in western North America. The system exhibited skill and value nine days into the future when using the ensemble average, and 12 days into the future when employing the full ensemble forecast.

View record