Use of Machine Learning Methods to Reduce Predictive Error of Groundwater Models

Quantitative analyses of groundwater flow and transport typically rely on a physically‐based model, which is inherently subject to error. Errors in model structure, parameter and data lead to both random and systematic error even in the output of a calibrated model. We develop complementary data‐driven models (DDMs) to reduce the predictive error of physically‐based groundwater models. Two machine learning techniques, the instance‐based weighting and support vector regression, are used to build the DDMs. This approach is illustrated using two real‐world case studies of the Republican River Compact Administration model and the Spokane Valley‐Rathdrum Prairie model. The two groundwater models have different hydrogeologic settings, parameterization, and calibration methods. In the first case study, cluster analysis is introduced for data preprocessing to make the DDMs more robust and computationally efficient. The DDMs reduce the root‐mean‐square error (RMSE) of the temporal, spatial, and spatiotemporal prediction of piezometric head of the groundwater model by 82%, 60%, and 48%, respectively. In the second case study, the DDMs reduce the RMSE of the temporal prediction of piezometric head of the groundwater model by 77%. It is further demonstrated that the effectiveness of the DDMs depends on the existence and extent of the structure in the error of the physically‐based model.     

Distribution and bioaccumulation of polychlorinated biphenyls and organochlorine pesticides residues in sediments and Manila clams (Ruditapes philippinarum) from along the Mid-Western coast of Korea

The distribution and bioaccumulation features of PCBs, DDTs, and HCHs were investigated in the sediments and Manila clams collected from along the Mid-Western coast of Korea. The measured concentrations of ΣPCBs, ΣDDTs and ΣHCHs were 1.08–3.5, 0.12–0.35 and 0.090–0.30 ng g⁻¹ dw in sediment, and 33–390, 7.4–46 and 6.3–27 ng g⁻¹ lipid in Manila clam, respectively. Their levels were found to be relatively lower than those of other contaminated areas and the consumption of Manila clam from these areas seems to be safe for human health according to calculated lifetime cancer risk and hazard indices. The ΣPCBs and ΣDDTs concentrations in sediments showed a significant relationship to those in clams. The significant correlation was observed between BSAF in clams and K_ow for each PCB congeners and DDT metabolites. These findings support that the PCBs and DDTs levels, which are highly hydrophobic chemicals, in clam reflect the sediment pollution through bioaccumulation.     

Analysis of recent climate change over the Arctic using ERA-Interim reanalysis data

This study investigates recent climate change over the Arctic and its link to the mid-latitudes using the ERA-Interim global atmospheric reanalysis data from the European Center for Medium-Range Weather Forecast （ECMWF）. Since 1979, sub- stantial surface warming, associated with the increase in anthropogenic greenhouse gases, has occurred over the Arctic. The great- est warming in winter has taken place offshore in the Kara-Barents Sea, and is associated with the increase in turbulent heat fluxes from the marginal ice zone. In contrast to the marked warming over the Arctic Ocean in winter, substantial cooling appears over Siberia and eastern Asia, linked to the reduction of Arctic sea ice during the freezing season （September-March）. However, in summer, very little change is observed in surface air temperature over the Arctic because increased radiative heat melts the sea ice and the amount of turbulent heat gain from the ocean is relatively small. The heat stored in the upper ocean mixed layer in summer with the opening of the Arctic Ocean is released back to the atmosphere as turbulent heat fluxes during the autumn and through to the following spring. This warming of the Arctic and the reduced sea ice amplifies surface cooling over Siberia and eastern Asia in winter.     

Massive sedimentation of fine sediment with organic matter and enhanced benthic-pelagic coupling by an artificial dyke in semi-enclosed Chonsu Bay, Korea

To assess the impact of an artificial dyke in Chonsu Bay (CBD) on the organic carbon (C(org)) cycle, we measured excess (210)Pb activities, C(org) and nitrogen content in sediment cores. The C(org) oxidation rates (C(ox)) on the surface sediment and benthic nutrient fluxes were also quantified with an in situ benthic chamber. The higher excess (210)Pb inventory, C(org) and nitrogen in cores near the CBD indicated lateral transport and local, massive deposition of particulate matter due to tidal circulation altered by artificial dyke construction. The C(ox) in sediment near the CBD was about twice as high as that out of the bay, suggesting the importance of benthic remineralization of organic matter. The benthic fluxes of dissolved inorganic nitrogen and phosphate were four to six times higher than those outside the bay, corresponding to 141% and 131% respectively, of the requirements for primary production.     

Constructing rainfall depth-frequency curves considering a linear trend in rainfall observations

Comprehensive flood prevention plans are established in large basins to cope with recent abnormal floods in South Korea. In order to make economically effective plans, appropriate design rainfalls are critically determined from the rainfall depth-frequency curves which take the occurrence of abnormal floods into consideration. Conventional approaches to construct the rainfall depth-frequency curves are based on the stationarity assumption. However, this assumption has a critical weak aspect in that it cannot reflect non-stationarities in rainfall observations. As an alternative, this study suggests the non-stationary Gumbel model (NSGM) which incorporates a linear trend of rainfall observations into rainfall frequency analysis to construct the rainfall depth-frequency curves. A comparison of various schemes employed in the model found that the proposed NSGM permits the estimation of the distribution parameters even when shifted in the future by using linear relationships between rainfall statistics and distribution parameters, and produces more acceptable estimates of design rainfalls in the future than the conventional model. The NSGM was applied at several stations in South Korea and then expected the design rainfalls to increase by up to 15–30% in 2050.     

Understanding of the Common Land Model performance for water and energy fluxes in a farmland during the growing season in Korea

The Common Land Model (CLM) is one of the most widely used land surface models (LSMs) due to the practicality of its simple parameterization scheme and its versatility in embracing a variety of field datasets. The improved assessment of land surface water and energy fluxes using CLM can be an alternative approach for understanding the complex land–atmosphere interactions in data‐limited regions. The understanding of water and energy cycles in a farmland is crucial because it is a dominant land feature in Korea and Asia. However, the applications of CLM to farmland in Korea are in paucity. The simulations of water and energy fluxes by CLM were conducted against those from the tower‐based measurements during the growing season of 2006 at the Haenam site (a farmland site) in Korea without optimization. According to the International Geosphere–Biosphere Programme (IGBP) land cover classification, a homogeneous cropland was selected initially for this study. Although the simulated soil moisture had a similar pattern to that of the observed, the former was relatively drier (at 0·1 m³ m^?3) than the latter. The simulated net radiation showed good agreement with the observed, with a root mean squared error (RMSE) of 41 W m^?2, whereas relatively large discrepancies between the simulation and observation were found in sensible (RMSE of 66 W m^?2) and latent (RMSE of 60 W m^?2) heat fluxes. On the basis of the sensitivity analysis, soil moisture was more receptive to land cover and soil texture parameterizations when compared to soil temperature and turbulent fluxes. Despite the uncertainty in the predictive capability of CLM employed without optimization, the initial performance of CLM suggests usefulness in a data‐limited heterogeneous farmland in Korea. Further studies are required to identify the controls on water and energy fluxes with an improved parameterization. Copyright © 2010 John Wiley & Sons, Ltd.     

Understanding the statistical properties of the percent model affinity index can improve biomonitoring related decision making

The percent model affinity (PMA) index is used to measure the similarity of two probability profiles representing, for example, an ideal profile (i.e. reference condition) and a monitored profile (i.e. possibly impacted condition). The goal of this work is to study the effects of sample size, evenness, true value of the index and number of classes on the statistical properties of the estimator of the PMA index. We derive and extend previous formulas of the expectation and variance of the estimator for estimated monitored profile and fixed reference profile. Using the obtained extension, we find that the estimator is asymptotically unbiased, converging faster when the profiles differ. When both profiles are estimated, we calculate the expectation using transformation rules for expectation and in addition derive the formula for the estimator’s variance. Since the computation of the probabilities in the variance formula is slow, we study the behavior of the variance with simulation experiments and assess whether it could be approximated with the variance for the fixed reference profile. Finally, we provide a set of recommendations for the users of the PMA index to avoid the most common caveats of the index.     

Conservation of Korean Rural Heritage through the Use of Ecomuseums

Rural heritage that involves agricultural heritage could be critical to enhancing community identity through a conservational approach. The purpose of this research is to develop conservational methods for rural heritage that include agricultural heritage areas through the use of ecomuseums. To conduct the research, we first reviewed the concepts of ecomuseum and rural heritage. Secondly, we defined the concept of the Korean rural ecomuseum by taking into consideration the concept defined by Riviere, Globally Important Agricultural Heritage Systems (GIAHS), the rural heritage policy of Europe, and the agricultural heritage systems of Korea. Thirdly, we also describe methods that are applicable to Korean rural ecomuseums, which include type classification and components. Finally, we applied the ecomuseum concept to two pilot models in the Cheongsando and Gurae areas, both Korea Important Agricultural Heritage System (KIAHS) sites. We conclude that it is desirable to expand the range of targets for conservation from agricultural heritage to rural heritage areas. From this point of view, the rural ecomuseum can be a highly valuable tool that can help preserve agricultural heritage sites and neighboring areas. Furthermore, the ecomuseum can enhance community identities, and strengthen local economies.     

Increased As load to the Uchen stream due to mine drainage and soils in the abandoned Kangwon mining district of Korea

To determine the appropriate allocation of resources for the future restoration of the abandoned mining district of Kangwon in Korea, identification of the main pollutants and the main sources discharging these pollutants is crucial. Therefore, a 2-year study was undertaken to quantify the amount of pollutants in the Uchen stream (a potential sink for contamination), which runs through the district, and to determine the potential sources of these pollutants, including mine drainage and soil. Arsenic (As) was the main pollutant in mine drainage and soils showing concentrations above the Korean regulated standard levels of 50 μg L⁻¹ and 50 mg kg⁻¹ for water and soil, respectively. In addition, the pollution index (PI) showed that mine drainages were polluted by As to a moderate (2 ≤ PI < 3) or strong (4 ≤ PI < 5) degree. Consequently, As load in mine drainage and soil contributed to increased amounts of As in the stream. The As loads in mine drainages (11 and 587 g month⁻¹ for mine adit 1 and 2, respectively) accounted for only 9% of the total As load to the stream (6,378 g month⁻¹); and the influence of mine drainages on As contents in the stream was more reliant on the total volume of mine drainage generated rather than the As concentration in the mine drainage. Approximately 91% of the As in the stream was derived from the soils within the study area.     

Ensemble-based landslide susceptibility maps in Jinbu area, Korea

Ensemble techniques were developed, applied and validated for the analysis of landslide susceptibility in Jinbu area, Korea using the geographic information system (GIS). Landslide-occurrence areas were detected in the study by interpreting aerial photographs and field survey data. Landslide locations were randomly selected in a 70/30 ratio for training and validation of the models, respectively. Topography, geology, soil and forest databases were also constructed. Maps relevant to landslide occurrence were assembled in a spatial database. Using the constructed spatial database, 17 landslide-related factors were extracted. The relationships between the detected landslide locations and the factors were identified and quantified by frequency ratio, weight of evidence, logistic regression and artificial neural network models and their ensemble models. The relationships were used as factor ratings in the overlay analysis to create landslide susceptibility indexes and maps. Then, the four landslide susceptibility maps were used as new input factors and integrated using the frequency ratio, weight of evidence, logistic regression and artificial neural network models as ensemble methods to make better susceptibility maps. All of the susceptibility maps were validated by comparison with known landslide locations that were not used directly in the analysis. As the result, the ensemble-based landslide susceptibility map that used the new landslide-related input factor maps showed better accuracy (87.11% in frequency ratio, 83.14% in weight of evidence, 87.79% in logistic regression and 84.54% in artificial neural network) than the individual landslide susceptibility maps (84.94% in frequency ratio, 82.82% in weight of evidence, 87.72% in logistic regression and 81.44% in artificial neural network). All accuracy assessments showed overall satisfactory agreement of more than 80%. The ensemble model was found to be more effective in terms of prediction accuracy than the individual model.