Alex J. Cannon

Downscaling recent streamflow conditions in British Columbia, Canada using ensemble neural network models

Abstract Variations in climate conditions during recent decades in British Columbia, Canada have occurred coincident to significant changes in streamflow conditions in the province. In the current study, the ability of empirical downscaling models to resolve these changes is investigated using ensemble neural networks forced with synoptic-scale atmospheric conditions. Five-day averages of streamflow data from 21 watersheds in the region are modelled using atmospheric data from the NCEP/NCAR reanalysis project as inputs. Ability of the downscaling models to predict streamflow and changes in streamflow between 1975–1986 and 1987–1998 is evaluated using a combination of model performance statistics, comparisons between long-term averages, and results from non-parametric statistical tests. While performance varied between systems, results suggest that empirical downscaling models for streamflow are capable of predicting changes in streamflow observed during recent decades using only large-scale atmospheric conditions as model inputs. Based on comparisons between stepwise linear regression and neural network models, the latter approach is recommended, particularly when trying to model systems with complex non-linear and interactive relationships between inputs and outputs. The use of ensemble averaging as a part of the modelling process is investigated and a number of recommendations are made with respect to this methodology.

Recent Variations in Climate and Hydrology in Canada

Climatic and hydrologic variations between the decades 1976–1985 and 1986–95 are examined at 210 climate stations for temperature, 271 climate stations for precipitation, and 642 hydrology stations from across Canada. The variations in climate are distributed across a broad spatial area. Temperatures were generally warmer in the more recent decade, with many stations showing significant increases during spring and fall. Significant decreases in temperature were found during winter in eastern Canada. Significant increases in temperature were more frequent in western Canada than in the east. Significant decreases in precipitation were also more prevalent in the north, as were increases in the south, except for Ontario and Quebec where little or no change has taken place. The hydrologic responses to these variations in climate are classified into four hydrograph types and six patterns of shifts in streamfiow between the two decades. The 642 hydrologic stations fall into 16 of the potential 24 groups. These 16 classes demonstrate a strong correspondence to the distribution of ecozones in Canada. In addition, these recent variations illustrate the leverage effect of small variations in climate, particularly temperature, on different hydrologic systems.

Quantile regression neural networks: Implementation in R and application to precipitation downscaling ☆

The qrnn package for R implements the quantile regression neural network, which is an artificial neural network extension of linear quantile regression. The model formulation follows from previous work on the estimation of censored regression quantiles. The result is a nonparametric, nonlinear model suitable for making probabilistic predictions of mixed discrete-continuous variables like precipitation amounts, wind speeds, or pollutant concentrations, as well as continuous variables. A differentiable approximation to the quantile regression error function is adopted so that gradient-based optimization algorithms can be used to estimate model parameters. Weight penalty and bootstrap aggregation methods are used to avoid overfitting. For convenience, functions for quantile-based probability density, cumulative distribution, and inverse cumulative distribution functions are also provided. Package functions are demonstrated on a simple precipitation downscaling task.

Bias Correction of GCM Precipitation by Quantile Mapping: How Well Do Methods Preserve Changes in Quantiles and Extremes?

AbstractQuantile mapping bias correction algorithms are commonly used to correct systematic distributional biases in precipitation outputs from climate models. Although they are effective at removing historical biases relative to observations, it has been found that quantile mapping can artificially corrupt future model-projected trends. Previous studies on the modification of precipitation trends by quantile mapping have focused on mean quantities, with less attention paid to extremes. This article investigates the extent to which quantile mapping algorithms modify global climate model (GCM) trends in mean precipitation and precipitation extremes indices. First, a bias correction algorithm, quantile delta mapping (QDM), that explicitly preserves relative changes in precipitation quantiles is presented. QDM is compared on synthetic data with detrended quantile mapping (DQM), which is designed to preserve trends in the mean, and with standard quantile mapping (QM). Next, methods are applied to phase 5 of t...

Downscaling Extremes—An Intercomparison of Multiple Statistical Methods for Present Climate

AbstractFive statistical downscaling methods [automated regression-based statistical downscaling (ASD), bias correction spatial disaggregation (BCSD), quantile regression neural networks (QRNN), TreeGen (TG), and expanded downscaling (XDS)] are compared with respect to representing climatic extremes. The tests are conducted at six stations from the coastal, mountainous, and taiga region of British Columbia, Canada, whose climatic extremes are measured using the 27 Climate Indices of Extremes (ClimDEX; http://www.climdex.org/climdex/index.action) indices. All methods are calibrated from data prior to 1991, and tested against the two decades from 1991 to 2010. A three-step testing procedure is used to establish a given method as reliable for any given index. The first step analyzes the sensitivity of a method to actual index anomalies by correlating observed and NCEP-downscaled annual index values; then, whether the distribution of an index corresponds to observations is tested. Finally, this latter test is...

Probabilistic Multisite Precipitation Downscaling by an Expanded Bernoulli–Gamma Density Network

Abstract A nonlinear, probabilistic synoptic downscaling algorithm for daily precipitation series at multiple sites is presented. The expanded Bernoulli–gamma density network (EBDN) represents the conditional density of multisite precipitation, conditioned on synoptic-scale climate predictors, using an artificial neural network (ANN) whose outputs are parameters of the Bernoulli–gamma distribution. Following the methodology used in expanded downscaling, predicted covariances between sites are forced to match observed covariances through the addition of a constraint to the ANN cost function. The resulting model can be thought of as a regression-based downscaling model with a stochastic weather generator component. Parameters of the Bernoulli–gamma distribution are downscaled from the synoptic-scale circulation, and unresolved temporal variability is generated via an autoregressive noise model. Demonstrated on a multisite precipitation dataset from coastal British Columbia, Canada, the EBDN is capable of sp...

Synoptic Map-Pattern Classification Using Recursive Partitioning and Principal Component Analysis

Abstract A method for classifying synoptic-scale maps into discrete groups is introduced. Tree-based recursive partitioning models are used to develop mappings between synoptic-scale circulation fields and the leading linear and nonlinear principal components (PCs) of weather elements observed at a surface station. Statistically unique but climatically insignificant patterns are avoided by identifying map patterns based on their association with indices related to local weather conditions. The method requires few user-adjustable parameters and includes an algorithm that provides objective guidance for determining the appropriate number of map patterns to retain. The classification method is demonstrated using daily sea level pressure and 500-hPa geopotential height maps from a domain covering British Columbia and the northeastern Pacific Ocean. The linear and nonlinear weather element PCs are derived from daily measurements of surface temperature, dewpoint temperature, cloud opacity, and u and υ wind comp...

Modelling Streamflow in Present and Future Climates: Examples from the Georgia Basin, British Columbia

The Georgia Basin is one of the most hydrologically complex areas of Canada. The form in which precipitation occurs in winter, either snow or rain, is the primary determinant of the hydrologic patterns of rivers. The region consists of mountains, valleys and plains, and has extensive variations in elevation. Variations in temperature and precipitation exert tremendous influence on the amount and form of water that reaches the land surface of the Georgia Basin. The amalgamation of the climatic and geographical factors creates homogenous zones in which distinctly different hydrological processes occur. Climate change could have major regional effects on temperature, precipitation, evapotranspiration and ultimately runoff. In previous work, zones of homogenous hydrologic processes in the Georgia Basin were delineated. In the present work, we used downscaled data from GCMs for future periods as inputs to a hydrologic model. These results are used to forecast the location of these zones in three periods in the 21st century. The results indicate important changes to the three hydrologic types found in the Georgia Basin: rainfall-driven streams demonstrate increased winter flows, snowmelt-driven streams demonstrate an increasingly early onset to spring snowmelt, and hybrid (mixed rain and snow) streams become increasingly rainfall driven. These large changes illustrate some of the impacts to which the Pacific Northwest may have to adapt in the near future.

Modelling Future Streamflow Extremes — Floods and Low Flows in Georgia Basin, British Columbia

The Georgia Basin is one of the most hydrologically complex areas of Canada. Variations in temperature, precipitation and elevation influence the amount and form of water that drives streamflow in its rivers and streams. Climate change could have major regional effects on air temperature, precipitation, evapotranspiration, and ultimately runoff. In previous work, zones of homogenous hydrologic processes were delineated within the basin. Watersheds were separated into three types: rainfall-driven streams, snowmelt-driven streams, and hybrid (mixed rainfall- and snowmelt-driven) streams. Climate change was shown to have major regional effects on each type of watershed, affecting the amounts and patterns of runoff. In the current study we consider changes in extreme hydrologic events, floods and low flows, in these watersheds. Climate data downscaled from the Canadian Coupled General Circulation Model for future time periods are used as inputs to a hydrologic model optimized for mountain watersheds. The discrepancies between observed and modelled streamflows are examined. While the model reproduces central tendency measures well, there are significant biases in the ability of the models to reproduce extremes. Output from the hydrologic model is used to assess relative changes in the frequency, timing, and magnitude of floods and low flows between present and future (2020, 2050 and 2080) climate scenarios. The models suggest that frequency of floods will increase in all watersheds under the projected climate scenarios. In rainfall-driven streams, flood events increase in number, but not in magnitude. In hybrid streams, winter events occur more often while summer snowmelt flood events occur less often. In snowmelt-driven streams, the magnitude and duration of summer floods increase. Low flows in rainfall-driven streams maintain the same frequency and magnitude but occur over an extended period of time during summer. Hybrid streams show an increase in frequency, a decrease in magnitude, and a shift in time of occurrence of low flows to summer rather than winter. In snowmelt-driven streams, low flow events occur less often largely moderated by increased flow due to an overall increase in winter streamflow in a warmer climate.

Forecasting Summertime Surface-Level Ozone Concentrations in the Lower Fraser Valley of British Columbia: An Ensemble Neural Network Approach

ABSTRACT Empirical models for predicting daily maximum hourly average ozone concentrations were developed for 10 monitoring stations in the Lower Fraser Valley (LFV) of British Columbia. According to data from 1991 to 1996, ensemble neural network models increased explained variance an average of 7% over multiple linear regression models using the same input variables. Without modification, all models performed poorly on days when the observed peak ozone concentration exceeded 82 parts per billion, the National Ambient Air Quality Objective. When numbers of extreme events in training data were increased using a histogram equalization process, models were able to forecast exceedances with improved accuracy. Modified generalized additive model (GAM) plots and associated measures of input variable importance and interaction were generated for a subset of the trained models and used to investigate relationships between input variables and ozone levels. The neural network models displayed a high degree of interaction among inputs, and it is likely the ability of these model types to account for interactions, rather than the nonlinearity of individual input variables, that explains their improved forecast skill. Inspection of GAM-style plots indicated that the relative importance of input variables in the ensemble neural network models varied with geographic location within the LFV. Four distinct groups of stations were identified, and rankings of inputs within the groups were generally consistent with physical intuition and results of prior studies.