Improving the forecasts of extreme streamflow by support vector regression with the data extracted by self‐organizing map

During typhoons or storms, accurate forecasts of hourly streamflow are necessary for flood warning and mitigation. However, hourly streamflow is difficult to forecast because of the complex physical process and the high variability in time. Furthermore, under the global warming scenario, events with extreme streamflow may occur that leads to more difficulties in forecasting streamflows. Hence, to obtain more accurate hourly streamflow forecasts, an improved streamflow forecasting model is proposed in this paper. The computational kernel of the proposed model is developed on the basis of support vector machine (SVM). Additionally, self‐organizing map (SOM) is used to analyse observed data to extract data with specific properties, which are capable of providing valuable information for streamflow forecasting. After reprocessing, these extracted data and the observed data are used to construct the SVM‐based model. An application is conducted to clearly demonstrate the advantage of the proposed model. The comparison between the proposed model and the conventional SVM model, which is constructed without SOM, is performed. The results indicate that the proposed model is better performed than the conventional SVM model. Moreover, as regards the extreme events, the result shows that the proposed model reduces the forecasting error, especially the error of peak streamflow. It is confirmed that because of the use of data extracted by SOM, the improved forecasting performance is obtained. The proposed model, which can produce accurate forecasts, is expected to be useful to support flood warning systems. Copyright © 2012 John Wiley & Sons, Ltd.     

A multivariate non‐parametric model for synthetic generation of daily streamflow

A p‐order multivariate kernel density model based on kernel density theory has been developed for synthetic generation of multivariate variables. It belongs to a kind of data‐driven approach and is able to avoid prior assumptions as to the form of probability distribution (normal or Pearson III) and the form of dependence (linear or non‐linear). The p‐order multivariate kernel density model is a non‐parametric method for synthesis of streamflow. The model is more flexible than conventional parametric models used in stochastic hydrology. The effectiveness and satisfactoriness of this model are illustrated through its application to the simultaneous synthetic generation of daily streamflow from Pingshan station and Yibin‐Pingshan region (Yi‐Ping region) of the Jinsha River in China. Copyright © 2007 John Wiley & Sons, Ltd.     

Comparison of local and global approximators in multivariate chaotic forecasting of daily streamflow

ABSTRACT

Although it is conceptually assumed that global models are relatively ineffective in modelling the highly unstable structure of chaotic hydrologic dynamics, there is not a detailed study of comparing the performances of local and global models in a hydrological context, especially with new emerging machine learning models. In this study, the performance of a local model (k-nearest neighbour, k-nn) and, as global models, several recent machine learning models – artificial neural network (ANN), least square-support vector regression (LS-SVR), random forest (RF), M5 model tree (M5), multivariate adaptive regression splines (MARS) – was analysed in multivariate chaotic forecasting of streamflow. The models were developed for Australia’s largest river, the River Murray. The results indicate that the k-nn model was more successful than the global models in capturing the streamflow dynamics. Furthermore, coupled with the multivariate phase-space, it was shown that the global models can be successfully used for obtaining reliable uncertainty estimates for streamflow.     

Incorporating synoptic-scale climate signals for streamflow modelling over the Mediterranean region using machine learning models

ABSTRACT

Understanding streamflow patterns by incorporating climate signal information can contribute remarkably to the knowledge of future local environmental flows. Three machine learning models, the multivariate adaptive regression splines (MARS), the M5 Model Tree and the least squares support vector machine (LSSVM) are established to predict the streamflow pattern over the Mediterranean region of Turkey (Besiri and Baykan stations). The structure of the predictive models is built using synoptic-scale climate signal information and river flow data from antecedent records. The predictive models are evaluated and assessed using quantitative and graphical statistics. The correlation analysis demonstrates that the North Pacific (NP) and the East Central Tropical Pacific Sea Surface Temperature (Niño3.4) indices have a substantial influence on the streamflow patterns, in addition to the historical information obtained from the river flow data. The model results reveal the utility of the LSSVM model over the other models through incorporating climate signal information for modelling streamflow.     

基于随机森林模型的GRACE数据3种空间降尺度对比

陆地水储量是赋存在陆地上各种形式水的综合体现,研究其时空变化对认识区域水循环过程和水资源调控等具有重要意义。然而现有陆地水储量变化数据实际分辨率较低,限制了其在中小流域或地区中的应用。针对这一问题,本文基于GRACE重力卫星和其后续卫星GRACE-FO反演的陆地水储量变化数据,首先采用随机森林模型,分别基于格点、区域(流域)和区域(全国)3种空间降尺度思路将GRACE数据降尺度至0.25°×0.25°,后结合GLDAS模型数据,基于水量平衡原理计算得到地下水储量变化数据,最后基于降尺度模型模拟效果和实测地下水位数据评估3种降尺度思路在全国的适用性。结果表明:随机森林模型能够较好地模拟驱动数据(降水、气温、植被条件指数和土壤水储量)与GRACE数据的统计关系,验证期格点降尺度思路的平均相关系数总体在0.6左右,区域降尺度思路的平均纳什效率系数、相关系数和均方根误差分别>0.5、>0.75和<6.6 cm,3种空间降尺度思路的模拟精度均满足基本要求;2003—2021年间,GRACE数据、格点降尺度、区域降尺度(流域)和区域降尺度(全国)得到的我国陆地水储量亏缺量分别约为...     

Suitability of satellite-based hydro-climate variables and machine learning for streamflow modeling at various scale watersheds

ABSTRACT

Streamflow modeling is essential to investigate processes in the hydrologic cycle and important for water resource management application. However, in-situ hydrologic data paucity, because of various factors such as economic, political, instrument malfunctioning, and poor spatial distribution, makes the modeling process challenging. To overcome this limitation, we introduced a satellite remote sensing-based machine learning approach – boosted regression tree (BRT) – that integrates spatial land surface and climate variables that describe the sub-units, and applied it in three variable size watersheds in the Upper Mississippi River Basin (UMRB), USA. The model simulation results were tested using an independent dataset and showed Nash–Sutcliffe efficiency values of 0.80, 0.76, and 0.69 for the UMRB, Illinois River Watershed, and Raccoon River Watershed, respectively. In addition, we compared the performance of the machine learning models with existing process-based modeling results. Overall performance is comparable with the process-based approaches, but with significantly less modeling effort and resources.     

Water quality monitoring at a virtual watershed monitoring station using a modified deep extreme learning machine

ABSTRACT

A new deep extreme learning machine (ELM) model is developed to predict water temperature and conductivity at a virtual monitoring station. Based on previous research, a modified ELM auto-encoder is developed to extract more robust invariance among the water quality data. A weighted ELM that takes seasonal variation as the basis of weighting is used to predict the actual value of water quality parameters at sites which only have historical data and no longer generate new data. The performance of the proposed model is validated against the monthly data from eight monitoring stations on the Zengwen River, Taiwan (2002–2017). Based on root mean square error, mean absolute error, mean absolute percentage error and correlation coefficient, the experimental results show that the new model is better than the other classical spatial interpolation methods.     

Monthly suspended sediment load prediction using artificial intelligence: testing of a new random subspace method

ABSTRACT

The predictive capability of a new artificial intelligence method, random subspace (RS), for the prediction of suspended sediment load in rivers was compared with commonly used methods: random forest (RF) and two support vector machine (SVM) models using a radial basis function kernel (SVM-RBF) and a normalized polynomial kernel (SVM-NPK). Using river discharge, rainfall and river stage data from the Haraz River, Iran, the results revealed: (a) the RS model provided a superior predictive accuracy (NSE = 0.83) to SVM-RBF (NSE = 0.80), SVM-NPK (NSE = 0.78) and RF (NSE = 0.68), corresponding to very good, good, satisfactory and unsatisfactory accuracies in load prediction; (b) the RBF kernel outperformed the NPK kernel; (c) the predictive capability was most sensitive to gamma and epsilon in SVM models, maximum depth of a tree and the number of features in RF models, classifier type, number of trees and subspace size in RS models; and (d) suspended sediment loads were most closely correlated with river discharge (PCC = 0.76). Overall, the results show that RS models have great potential in data poor watersheds, such as that studied here, to produce strong predictions of suspended load based on monthly records of river discharge, rainfall depth and river stage alone.     

Assessing the effects of forest biomass reductions on forest health and streamflow

Forest biomass reductions in overgrown forests have the potential to provide hydrologic benefits in the form of improved forest health and increased streamflow production in water-limited systems. Biomass reductions may also alter evaporation. These changes are generated when water that previously would have been transpired or evaporated from the canopy of the removed vegetation is transferred to transpiration of the remaining vegetation, streamflow, and/or non-canopy evaporation. In this study, we combined a new vegetation-change water-balance approach with lumped hydrologic modelling outputs to examine the effects of forest biomass reductions on transpiration of the remaining vegetation and streamflow in California's Sierra Nevada. We found that on average, 102 mm and 263 mm (8.0% and 20.6% of mean annual precipitation [MAP]) of water were made available following 20% and 50% forest biomass-reduction scenarios, respectively. This water was then partitioned to both streamflow and transpiration of the remaining forest, but to varying degrees depending on post-biomass-reduction precipitation levels and forest biomass-reduction intensity. During dry periods, most of the water (approximately 200 mm [15.7% on MAP] for the 50% biomass-reduction scenario) was partitioned to transpiration of the remaining trees, while less than 50 mm (3.9% on MAP) was partitioned to streamflow. This increase in transpiration during dry periods would likely help trees maintain forest productivity and resistance to drought. During wet periods, the hydrologic benefits of forest biomass reductions shifted to streamflow (200 mm [15.7% on MAP]) and away from transpiration (less than 150 mm [11.8% on MAP]) as the remaining trees became less water stressed. We also found that streamflow benefits per unit of forest biomass reduction increased with biomass-reduction intensity, whereas transpiration benefits decreased. By accounting for changes in vegetation, the vegetation-change water balance developed in this study provided an improved assessment of watershed-scale forest health benefits associated with forest biomass reductions.     

Modelling long-term groundwater fluctuations by extreme learning machine using hydro-climatic data

The ability of the extreme learning machine (ELM) is investigated in modelling groundwater level (GWL) fluctuations using hydro-climatic data obtained for Hormozgan Province, southern Iran. Monthly precipitation, evaporation and previous GWL data were used as model inputs. Developed ELM models were compared with the artificial neural networks (ANN) and radial basis function (RBF) models. The models were also compared with the autoregressive moving average (ARMA), and evaluated using mean square errors, mean absolute error, Nash-Sutcliffe efficiency and determination coefficient statistics. All the data-driven models had better accuracy than the ARMA, and the ELM model’s performance was superior to that of the ANN and RBF models in modelling 1-, 2- and 3-month-ahead GWL. The RMSE accuracy of the ANN model was increased by 37, 34 and 52% using ELM for the 1-, 2- and 3-month-ahead forecasts, respectively. The accuracy of the ELM models was found to be less sensitive to increasing lead time.     

Estimating daily streamflow in the Congo Basin using satellite-derived data and a semi-distributed hydrological model

ABSTRACT

The application of remotely-sensed data for hydrological modeling of the Congo Basin is presented. Satellite-derived data, including TRMM precipitation, are used as inputs to drive the USGS Geospatial Streamflow Model (GeoSFM) to estimate daily river discharge over the basin from 1998 to 2012. Physically-based parameterization was augmented with a spatially-distributed calibration that enables GeoSFM to simulate hydrological processes such as the slowing effect of the Cuvette Centrale. The resulting simulated long-term mean of daily flows and the observed flow at the Kinshasa gauge were comparable (40 631 and 40 638 m³/s respectively), in the 7-year validation period (2004–2010), with no significant bias and a Nash-Sutcliffe model efficiency coefficient of 0.70. Modeled daily flows and aggregated monthly river outflows (compared to historical averages) for additional sites confirm the model reliability in capturing flow timing and seasonality across the basin, but sometimes fails to accurately predict flow magnitude. The results of this model can be useful in research and decision-making contexts and validate the application of satellite-based hydrological models driven for large, data-scarce river systems such as the Congo.     

Roles of forest disturbance and climate variability on streamflow components in snow-dominated paired watersheds at multiple temporal scales

The paired watershed experimental (PWE) approach has long been used as an effective means to assess the impacts of forest change on hydrology in small watersheds (<100 km²). Yet, the effects of climate variability on streamflow are not often assessed in PWE design. In this study, two sets of paired watersheds, (1) Camp and Greata Creeks and (2) 240 and 241 Creeks located in the Southern Interior of British Columbia, Canada, were selected to explore relative roles of forest disturbance and climate variability on streamflow components (i.e., baseflow and surface runoff) at different time scales. Our analyses showed that forest disturbance is positively related to annual streamflow components. However, this relationship is statistically insignificant since forest disturbance can either increase or decrease seasonal streamflow components, which eventually limited the positive effect on streamflow at the annual scale. Interestingly, we found that forest disturbance consistently decreased summer streamflow components in the two PWEs as forest disturbance can augment earlier and quicker snow-melt processes and hence reduce soil moisture to maintain summer streamflow components. More importantly, this study revealed that climate variability played a more significant role than forest disturbance in both annual and seasonal streamflow components, for instance, climate variability can account for as much as 90% of summer streamflow components variation in Camp, suggesting the role of climate variability on streamflow should be highlighted in the traditional PWE approach to truly advance our understanding of the interactions of forest change, climate variability and water for sustainable water resource management.     

Simulation of streamflow with statistically downscaled daily rainfall using a hybrid of wavelet and GAMLSS models

ABSTRACT

A rainfall–streamflow model is proposed, in which a downscaled rainfall series and its wavelet-based decomposed sub-series at optimum lags were used as covariates in GAMLSS (Generalized Additive Model in Location, Scale and Shape). GAMLSS is applied in climate change impact assessment using CMIP5 general climate model to simulate daily streamflow in three sub-catchments of the Onkaparinga catchment, South Australia. The Spearman correlation and Nash-Sutcliffe efficiency between the observed and median simulated streamflow values were high and comparable for both the calibration and validation periods for each sub-catchment. We show that the GAMLSS has the capability to capture non-stationarity in the rainfall–streamflow process. It was also observed that the use of wavelet-based decomposed rainfall sub-series with optimum lags as covariates in the GAMLSS model captures the underlying physics of the rainfall–streamflow process. The development and application of an empirical rainfall–streamflow model that can be used to assess the impact of catchment-scale climate change on streamflow is demonstrated.     

Potential of support vector regression for prediction of monthly streamflow using endogenous property

In the recent past, a variety of statistical and other modelling approaches have been developed to capture the properties of hydrological time series for their reliable prediction. However, the extent of complexity hinders the applicability of such traditional models in many cases. Kernel‐based machine learning approaches have been found to be more popular due to their inherent advantages over traditional modelling techniques including artificial neural networks(ANNs ). In this paper, a kernel‐based learning approach is investigated for its suitability to capture the monthly variation of streamflow time series. Its performance is compared with that of the traditional approaches. Support vector machines (SVMs) are one such kernel‐based algorithm that has given promising results in hydrology and associated areas. In this paper, the application of SVMs to regression problems, known as support vector regression (SVR), is presented to predict the monthly streamflow of the Mahanadi River in the state of Orissa, India. The results obtained are compared against the results derived from the traditional Box–Jenkins approach. While the correlation coefficient between the observed and predicted streamflows was found to be 0·77 in case of SVR, the same for different auto‐regressive integrated moving average (ARIMA) models ranges between 0·67 and 0·69. The superiority of SVR as compared to traditional Box‐Jenkins approach is also explained through the feature space representation. Copyright © 2009 John Wiley & Sons, Ltd.     

Changes in streamflow components following logging and regeneration in the southern forest of Western Australia

Two small experimental catchments were established in the south-west of Western Australia to study the effects of logging and subsequent regeneration on the mechanism of streamflow generation. Following a six year pre-treatment calibration period (1976–1981), one catchment (March Road) was logged and reforested in 1982 and the other (April Road South) remained as a control. Logging resulted in an increase in groundwater levels and subsequently groundwater discharge area. The deep, permanent groundwater levels in the valley and upslope areas rose until 1986 and then began to decline. The maximum rise was 5 m in the upslope areas. The duration of shallow, intermittent groundwater system, perched on underlying clay, was extended from 2–3 months in winter before logging to 5–6 months after logging. The shallow groundwater level rose in the valley and began to discharge at the ground surface in 1986. Logging resulted in an increase in streamflow. The maximum increase (≈18% of annual rainfall) was in 1983, one year after logging. The increase in streamflow was due to a substantial decrease in interception and evapotranspiration, increased recharge to the shallow groundwater system, decreased soil moisture deficit and consequently an increase in throughflow. The increase in base flow was about twice that of quick flow. The changes in streamflow and its components in the subsequent years were closely related to the groundwater discharge area. Most of the quick flow was generated as saturation excess overland flow from the groundwater discharge area in the valley. The expansion of the groundwater discharge area, increased soil moisture content, higher groundwater level and the presence of the shallow groundwater system for the extended periods were responsible for the process of streamflow generation.     

Analysis of the interannual variability of annual daily extreme water levels in the St Lawrence River and Lake Ontario from 1918 to 2010

We compared the interannual variability of annual daily maximum and minimum extreme water levels in Lake Ontario and the St Lawrence River (Sorel station) from 1918 to 2010, using several statistical tests. The interannual variability of annual daily maximum extreme water levels in Lake Ontario is characterized by a positive long‐term trend showing two shifts in mean (1929–1930 and 1942–1943) and a single shift in variance (in 1958–1959). In contrast, for the St Lawrence River, this interannual variability is characterized by a negative long‐term trend with a single shift in mean, which occurred in 1955–1956. As for annual daily minimum extreme water levels, their interannual variability shows no significant long‐term change in trend. However, for Lake Ontario, the interannual variability of these water levels shows two shifts in mean, which are synchronous with those for maximum water levels, and a single shift in variance, which occurred in 1965–1966. These changes in trend and stationarity (mean and variance) are thought to be due to factors both climatic (the Great Drought of the 1930s) and human (digging of the Seaway and construction of several dams and locks during the 1950s). Despite this change in means and variance, the four series are clearly described by the generalized extreme value distribution. Finally, annual daily maximum and minimum extreme water levels in the St Lawrence and Lake Ontario are negatively correlated with Atlantic multidecadal oscillation over the period from 1918 to 2010. Copyright © 2013 John Wiley & Sons, Ltd.     

Assessment and evaluation of potential climate change impact on monsoon flows using machine learning technique over Wainganga River basin,India

In this study, classification- and regression-based statistical downscaling is used to project the monthly monsoon streamflow over the Wainganga basin, India, using 40 global climate model (GCM) outputs and four representative concentration pathways (RCP) scenarios. Support vector machine (SVM) and relevance vector machine (RVM) are considered to perform downscaling. The RVM outperforms SVM and is used to simulate future projections of monsoon flows for different periods. In addition, variability in water availability with uncertainty and change point (CP) detection are accomplished by flow–duration curve and Bayesian analysis, respectively. It is observed from the results that the upper extremes of monsoon flows are highly sensitive to increases in temperature and show a continuous decreasing trend. Medium and low flows are increasing in future projections for all the scenarios, and high uncertainty is noticed in the case of low flows. An early CP is detected in the case of high emissions scenarios.     

多波地震深度学习的油气储层分布预测案例

有机并有效利用纵波与转换横波在油气储层敏感度上存在的差异,有助于突出地震油气储层特征,有助于提高地震油气储层分布边界刻画的精度.基于此,本文设计了一种卷积神经网络与支持向量机方法相结合的多波地震油气储层分布预测的深度学习法（Deep Learning Method）.首先,利用莱特准则剔除所生成的多波地震属性中可能存在的异常值降低网络变体数量.然后,通过能突出多波地震油气储层特征的聚类算法和无监督学习算法构建隐藏层,用于增加网络共享,提取油气特征.最后,将增加网络罚值后的井点样本作为支持向量机预测的输入样本,以降采样后的C3卷积层属性作为学习集,进行从已知到未知的地震油气储层的预测.本方案应用于HG地区晚三叠统HGR组的碳酸盐岩油气储层预测,所预测的地震油气储层边界更加清晰,预测结果与实际情况基本吻合.应用结果表明：本论文方案不仅具有可行性,且具有有效性.

     

基于随机森林的内陆湖泊水体有色可溶性有机物(CDOM)浓度遥感估算

水体中的有色可溶性有机物(CDOM)是湖泊生态系统中氮、磷等有机营养物质的重要来源,利用卫星遥感数据反演内陆水体中CDOM浓度一直是个挑战.因此本文基于滇池2009年9月、2017年4月以及太湖2016年7月的现场原位观测和室内实验,在分析水体固有光学特性的基础上,引入机器学习算法,建立了基于哨兵-3A OLCI传感器的我国内陆湖泊水体CDOM浓度随机森林反演模型.利用独立的验证数据集对所构建的随机森林模型及常用的波段比值模型、一阶微分模型、半分析模型、BP神经网络模型等的反演精度进行评价.结果表明:随机森林模型的均方根误差为0.14 m-1,平均相对误差为21%,与反演效果相对较好的BP神经网络模型相比,均方根误差降低了50%,平均相对误差降低了38%,反演精度得到了显著的提高.根据随机森林算法的特征重要性参数提供的各自变量影响力结果,发现B11(709 nm)和B6(560 nm)波段贡献率最大,是反演CDOM的敏感波段.最后将随机森林模型应用到滇池2017年4月12日、太湖2017年5月18日的哨兵-3A OLCI影像上,得到滇池、太湖水体CDOM浓度分布图.滇池CDOM浓度的分布特征大致符合东北、西南高,中西部低的趋势,且河口处的CDOM浓度高于湖泊水体,表明径流的输入给滇池水体带来了大量的CDOM.太湖CDOM浓度的分布特征大致符合西部高,湖心区和东部低的趋势.太湖西部以及北部梅梁湾受入湖河流影响较大,CDOM浓度较高,太湖开敞区远离河口处,受外源河流的影响逐渐减小,且由于湖水的不断稀释,CDOM浓度不断降低.太湖东部水生植物很多,湖水较为清澈,CDOM浓度较低.