•We compare new and traditional data-driven long-term drought forecast models.•The Standard Precipitation Index (SPI 12 and SPI 24) is forecasted.•Wavelet-ANN and wavelet-SVR models provided better ...results than the other model types used.
Long-term drought forecasts can provide valuable information to help mitigate some of the consequences of drought. Data driven models are suitable forecast tools due to their minimal information requirements and rapid development times. This study compares the effectiveness of five data driven models for forecasting long-term (6 and 12months lead time) drought conditions in the Awash River Basin of Ethiopia. The Standard Precipitation Index (SPI 12 and SPI 24) was forecasted using a traditional stochastic model (ARIMA) and compared to machine learning techniques such as artificial neural networks (ANNs), and support vector regression (SVR). In addition to these three model types, wavelet transforms were used to pre-process the inputs for ANN and SVR models to form WA-ANN and WA-SVR models; this is the first time that WA-SVR models have been explored and tested for long-term SPI forecasting. The performances of all models were compared using RMSE, MAE, R2 and a measure of persistence. The forecast results indicate that the coupled wavelet neural network (WA-ANN) models were better than all the other models in this study for forecasting SPI 12 and SPI 24 values over lead times of 6 and 12months in the Awash River Basin.
► The paper aims to detect trends in the mean flow and total precipitation data. ► Discrete wavelet transforms and Mann–Kendall tests are used for the trend assessment. ► A new criterion is proposed ...to select the mother wavelet and extension border types. ► Most of the trends are upwards and started during the mid-1960s to early 1970s. ► The intra- and inter-annual events are playing major roles behind the observed trends.
This paper aims to detect trends in mean flow and total precipitation data over southern parts of Quebec and Ontario, Canada. The main purpose of the trend assessment is to find out what time scales are affecting the trends observed in the datasets used. In this study, a new trend detection method for hydrological studies is explored, which involves the use of wavelet transforms (WTs) in order to separate the rapidly and slowly changing events contained in a time series. More specifically, this study co-utilizes the Discrete Wavelet Transform (DWT) technique and the Mann–Kendall (MK) trend tests to analyze and detect trends in monthly, seasonally-based, and annual data from eight flow stations and seven meteorological stations in southern Ontario and Quebec during 1954–2008. The combination of the DWT and MK test in analyzing trends has not been extensively explored to date, especially in detecting trends in Canadian flow and precipitation time series. The mother wavelet type and the extension border used in the wavelet transform, as well as the number of decomposition levels, were determined based on two criteria. The first criterion is the mean relative error of the wavelet approximation series and the original time series. In addition, a new criterion is proposed and explored in this study, which is based on the relative error of the MK Z-values of the approximation component and the original time series. Sequential Mann–Kendall analysis on the different wavelet detail components (with their approximation component added) that result from the time series decomposition was also used and found to be helpful because it depicts how harmonious each of the detail components (plus approximation) is with respect to the original data. This study found that most of the trends are positive and started during the mid-1960s to early 1970s. The results from the wavelet analysis and Mann–Kendall tests on the different data types (using the 5% significance level) reveal that in general, intra- and inter-annual events (up to 4years) are more influential in affecting the observed trends.
This study explored the ability of coupled machine learning models and ensemble techniques to predict drought conditions in the Awash River Basin of Ethiopia. The potential of wavelet transforms ...coupled with the bootstrap and boosting ensemble techniques to develop reliable artificial neural network (ANN) and support vector regression (SVR) models was explored in this study for drought prediction. Wavelet analysis was used as a pre-processing tool and was shown to improve drought predictions. The Standardized Precipitation Index (SPI) (in this case SPI 3, SPI 12 and SPI 24) is a meteorological drought index that was forecasted using the aforementioned models and these SPI values represent short and long-term drought conditions. The performances of all models were compared using RMSE, MAE, and R2. The prediction results indicated that the use of the boosting ensemble technique consistently improved the correlation between observed and predicted SPIs. In addition, the use of wavelet analysis improved the prediction results of all models. Overall, the wavelet boosting ANN (WBS-ANN) and wavelet boosting SVR (WBS-SVR) models provided better prediction results compared to the other model types evaluated.
•Coupled machine learning, wavelet, and ensemble methods•Forecasted drought (SPI) in Ethiopia•Compare boosting and bootstrap methods•Wavelet boosting SVR method provided best results.
Data assimilation allows for updating state variables in a model to represent the initial condition of a catchment more accurately than the initial OpenLoop simulation. In hydrology, data ...assimilation is often a prerequisite for forecasting. According to Hornik (1991, https://doi.org/10.1016/0893-6080(91)90009-T) artificial neural networks can learn any nonlinear relationship between inputs and outputs. Here, we hypothesize that neural networks could learn the relationship between the simulated streamflow (from a hydrological model) and the corresponding state variables. Once learned, this relationship can be used to obtain corrected state variables by applying it to observed rather than simulated streamflow. Based on this, we propose a novel, ensemble‐based, data assimilation approach. As a proof of concept and to verify the abovementioned hypothesis, we used an international testbed comprising four hydrologically dissimilar catchments. We applied the new data assimilation method to the lumped hydrological model GR4J, which has two state variables. Within this framework, we compared two types of neural networks, namely, Extreme Learning Machine and the Multilayer Perceptron. Using well‐known metrics such as the continuous ranked probability score, we compared the assimilated streamflow series with the OpenLoop streamflow series and with the observed streamflow. We show that neural networks can be successfully used for data assimilation, with a noticeable improvement over the OpenLoop simulation for all catchments.
Key Points
Neural network‐based methods are proposed for assimilating state variables in conceptual hydrological models
Both multilayer perceptrons and extreme learning machines are shown to generate accurate starting points for streamflow forecasts
Multilayer perceptron ensembles provided more reliable estimates of state variable uncertainty than extreme learning machine ensembles
•The relationships between climate indices and flow in southern Quebec and Ontario were assessed.•The relationships were analyzed at the intra-annual, inter-annual and inter-decadal scales.•Strong ...seasonality reflected by the 6- and 12-month periodicities were seen in the spectra of monthly data.•Significant periodicities at the inter-annual scales were only seen in the spectra of annual data.•The response time of flow to NAO/PDO/ENSO influence ranged from 6–48 months and 1–4years.
The impacts of large-scale climate oscillations on hydrological systems and their variability have been documented in different parts of the world. Since hydroclimatic data are known to exhibit non-stationary characteristics, spectral analyses such as wavelet transforms are very useful in extracting time–frequency information from such data. As Canadian studies, particularly those of regions east of the Prairies, using wavelet transform-based methods to draw links between relevant climate indices e.g., the El Niño Southern Oscillation (ENSO), the North Atlantic Oscillation (NAO), and the Pacific Decadal Oscillation (PDO) and streamflow variability are not common, this study aims to analyze such relationships for the southern regions of Quebec and Ontario. Monthly and annual streamflow data with a record length of 55years were used to capture streamflow variability at intra-annual, inter-annual and inter-decadal scales. The continuous wavelet transform spectra of monthly streamflow data revealed consistent significant 6- and 12-month periodicities, which are likely associated with strong seasonality factors. Its annual counterparts showed four different significant periodicities: up to 4years, 4–6years, 6–8years, and greater than 8years – all of which occurred after the late 1960s/early 1970s. Wavelet coherence analyses show that the influence of ENSO and NAO at the inter-annual scale occurs at 2–6year periodicities, and the influence of PDO occur at periodicities up to 8years and exceeding 16years. Correlations between these climate indices and streamflow were computed to determine the time delay of streamflow response to the influence of ENSO, NAO, and PDO. The lag times ranged from 6–48months (for monthly data) and 1–4years (for annual data). This research contributes to our understanding of streamflow variability over the southern parts of Quebec and Ontario, and the role of ENSO, NAO, and PDO phenomena on this variability. These relationships can be also used to improve hydrological forecasting and water resources management in Ontario and Quebec.
The main purpose of this study is to detect trends in the mean surface air temperature over the southern parts of Ontario and Quebec, Canada, for the period of 1967–2006. This is accomplished by ...determining the most dominant periodic components that affect trends in different temperature data categories (monthly, seasonally-based, seasonal, and annual), which were obtained from a total of five stations. The discrete wavelet transform (DWT) technique, the Mann–Kendall (MK) trend test, and sequential Mann–Kendall analysis were used in this study — co-utilizing these techniques in temperature trend studies has not been explored extensively. The mother wavelet, number of decomposition levels, and boundary condition were determined using a newly proposed criterion based on the relative error of the MK Z-values between the original data and the approximation component of the last decomposition level. This study found that all stations experienced positive trends: significant trends were observed in all of the monthly, seasonally-based, and annual data. For the different seasons, although the trend values were all positive, not all stations experienced significant trends. It was found that high-frequency components ranging from 2 to 12months were more prominent for trends in the higher resolution data (i.e. monthly and seasonally based). The positive trends observed for the annual data are thought to be mostly attributable to warming during winter and summer seasons, which are manifested in the form of multiyear to decadal events (mostly between 8 and 16years).
•Temperature trends were assessed in southern Quebec and Ontario, Canada.•Monthly, seasonally-based, and annual data were used.•Periodicities in the dataset were identified using the discrete wavelet transform.•The Mann-Kendall (MK) test was used to determine important periodicities for trends.•Different versions of the MK tests were used depending on the data characteristics.
Due to inherent uncertainties in measurement and analysis, groundwater quality assessment is a difficult task. Artificial intelligence techniques, specifically fuzzy inference systems, have proven ...useful in evaluating groundwater quality in uncertain and complex hydrogeological systems. In the present study, a Mamdani fuzzy-logic-based decision-making approach was developed to assess groundwater quality based on relevant indices. In an effort to develop a set of new hybrid fuzzy indices for groundwater quality assessment, a Mamdani fuzzy inference model was developed with widely-accepted groundwater quality indices: the Groundwater Quality Index (GQI), the Water Quality Index (WQI), and the Ground Water Quality Index (GWQI). In an effort to present generalized hybrid fuzzy indices a significant effort was made to employ well-known groundwater quality index acceptability ranges as fuzzy model output ranges rather than employing expert knowledge in the fuzzification of output parameters. The proposed approach was evaluated for its ability to assess the drinking water quality of 49 samples collected seasonally from groundwater resources in Iran's Sarab Plain during 2013–2014. Input membership functions were defined as “desirable”, “acceptable” and “unacceptable” based on expert knowledge and the standard and permissible limits prescribed by the World Health Organization. Output data were categorized into multiple categories based on the GQI (5 categories), WQI (5 categories), and GWQI (3 categories). Given the potential of fuzzy models to minimize uncertainties, hybrid fuzzy-based indices produce significantly more accurate assessments of groundwater quality than traditional indices. The developed models' accuracy was assessed and a comparison of the performance indices demonstrated the Fuzzy Groundwater Quality Index model to be more accurate than both the Fuzzy Water Quality Index and Fuzzy Ground Water Quality Index models. This suggests that the new hybrid fuzzy indices developed in this research are reliable and flexible when used in groundwater quality assessment for drinking purposes.
•A new groundwater quality assessment method was developed.•The new method is based on a fuzzy inference system.•Widely accepted indices were also used and compared in this study.•The hybrid fuzzy based indices minimized uncertainties.•The new approaches were useful for groundwater quality assessment.
Drought forecasts can be an effective tool for mitigating some of the more adverse consequences of drought. Data-driven models are suitable forecasting tools due to their rapid development times, as ...well as minimal information requirements compared to the information required for physically based models. This study compares the effectiveness of three data-driven models for forecasting drought conditions in the Awash River Basin of Ethiopia. The Standard Precipitation Index (SPI) is forecast and compared using artificial neural networks (ANNs), support vector regression (SVR), and wavelet neural networks (WN). SPI 3 and SPI 12 were the SPI values that were forecasted. These SPI values were forecast over lead times of 1 and 6 months. The performance of all the models was compared using RMSE, MAE, and R2. The forecast results indicate that the coupled wavelet neural network (WN) models were the best models for forecasting SPI values over multiple lead times in the Awash River Basin in Ethiopia.