Combined Wavelet Transform With Long Short-Term Memory Neural Network for Water Table Depth Prediction in Baoding City, North China Plain

Accurate estimation of water table depth dynamics is essential for water resource management, especially in areas where groundwater is overexploited. In recent years, as a data-driven model, artificial neural networks (NNs) have been widely used in hydrological modeling. However, due to the non-stationarity of water table depth data, the performance of NNs in areas of over-exploitation is challenging. Therefore, reducing data noise is an essential step before simulating the water table depth. This research proposed a novel method to model the non-stationary time series data of water table depth through combing the advantages of wavelet analysis and Long Short-Term Memory (LSTM) neural network (NN). A typical groundwater over-exploitation area, Baoding, North China Plain (NCP), was selected as a study area. To reflect the impact of anthropogenic activities, the variables harnessed to develop the model includes temperature, precipitation, evaporation, and some socio-economic data. The results show that decomposing the time series of the water table depth into three sub-temporal components by Meyer wavelets can significantly improve the simulation effect of LSTM on the water table depth. The average NSE (Nash-Sutcliffe efficiency coefficient) value of all the sites increased from 0.432 to 0.819. Additionally, a feedforward neural network (FNN) is used to compare forecasts over 12-months. As expected, wavelet-LSTM outperforms wavelet-FNN. As the prediction time increases, the advantages of wavelet-LSTM become more evident. The wavelet-LSTM is satisfactory for forecasting the water table depth at most in 6 months. Furthermore, the importance of input variables of wavelet-LSTM is analysed by the weights of the model. The results indicate that anthropogenic activities influence the water table depth significantly, especially in the sites close to the Baiyangdian Lake, the largest lake in the North China Plain. This study demonstrates that the wavelet-LSTM model provides an option for water table depth simulation and predicting areas of over-exploitation of groundwater.

Water table depth prediction based on Deep learning models in Electrical Power Transmission Lines area

Background: Predicting water table depth in Electrical Power Transmission Lines area presents great importance and helps the decision makers do the safety analysis during the project. The present study predicts the water table depth with observed weather data and hydrologic data. Method: The study first compared the results of LSTM, GRU, LSTM-S2S, and FFNN models in daily data simulation. Moreover, two scenarios (S1 and S2) were set to identify the effect of the water component on water table depth simulation. In addition, in order to analyze how data time scale influences the model simulation results, the monthly scale data was simulated by LSTM, GRU, and LSTM-S2S models. Result: The result indicated that LSTM-S2S was the best model for predicting daily water table depth among the four models. By contrast, FFNN performed the worst. LSTM and GRU model performed equally well both in daily data and monthly data simulation. S1 performed better than S2 in the water table depth simulation. The average daily performance of R2 and NSE was both higher than that in the monthly results with LSTM, GRU, and LSTM-S2S models. Conclusion: As a result, the method in the present study can be used to simulate the water table depth in the future in Electrical Power Transmission Lines area.

Global CO2 fertilization of Sphagnum peat mosses via suppression of photorespiration during the twentieth century

Natural peatlands contribute significantly to global carbon sequestration and storage of biomass, most of which derives from Sphagnum peat mosses. Atmospheric CO2 levels have increased dramatically during the twentieth century, from 280 to > 400 ppm, which has affected plant carbon dynamics. Net carbon assimilation is strongly reduced by photorespiration, a process that depends on the CO2 to O2 ratio. Here we investigate the response of the photorespiration to photosynthesis ratio in Sphagnum mosses to recent CO2 increases by comparing deuterium isotopomers of historical and contemporary Sphagnum tissues collected from 36 peat cores from five continents. Rising CO2 levels generally suppressed photorespiration relative to photosynthesis but the magnitude of suppression depended on the current water table depth. By estimating the changes in water table depth, temperature, and precipitation during the twentieth century, we excluded potential effects of these climate parameters on the observed isotopomer responses. Further, we showed that the photorespiration to photosynthesis ratio varied between Sphagnum subgenera, indicating differences in their photosynthetic capacity. The global suppression of photorespiration in Sphagnum suggests an increased net primary production potential in response to the ongoing rise in atmospheric CO2, in particular for mire structures with intermediate water table depths.

Development of a Deep Learning Emulator for a Distributed Groundwater–Surface Water Model: ParFlow-ML

Integrated hydrologic models solve coupled mathematical equations that represent natural processes, including groundwater, unsaturated, and overland flow. However, these models are computationally expensive. It has been recently shown that machine leaning (ML) and deep learning (DL) in particular could be used to emulate complex physical processes in the earth system. In this study, we demonstrate how a DL model can emulate transient, three-dimensional integrated hydrologic model simulations at a fraction of the computational expense. This emulator is based on a DL model previously used for modeling video dynamics, PredRNN. The emulator is trained based on physical parameters used in the original model, inputs such as hydraulic conductivity and topography, and produces spatially distributed outputs (e.g., pressure head) from which quantities such as streamflow and water table depth can be calculated. Simulation results from the emulator and ParFlow agree well with average relative biases of 0.070, 0.092, and 0.032 for streamflow, water table depth, and total water storage, respectively. Moreover, the emulator is up to 42 times faster than ParFlow. Given this promising proof of concept, our results open the door to future applications of full hydrologic model emulation, particularly at larger scales.

Alleviating peatland fire risk using water management trinity and community involvement

Conventionally, agriculture in peatland requires soil drainage to enable the crops to grow. This often results in being over-drained and makes it vulnerable to fires. The risk can be contained by applying water management trinity (WMT), which creates canals for water regulation and reservoirs instead of drainage. This study aimed to examine, elaborate, and validate the WMT effect and community involvement in minimizing fire risk in peatland. We collected water table depth every two weeks from 1 April 2017 to 31 December 2020 in a coconut plantation under WMT and employed Focus Groups Discussions (FGD) in five villages in Pulau Burung District, Indragiri Hilir Regency, Riau. The result showed that the existence of WMT for more than three decades has successfully maintained water table depth between 30 and 70 cm that is influenced by seasons. The fire occurrence based on the FGD interview has been validated with hotspot data from NASA’s FIRMS. This research also employed SWOT analysis to examine the local fire mitigation strategy. The progress in lowering fire incidents and risk should be intervened with finding long-term solutions to increase farmers’ capability on sustainable agriculture. Our finding reveals that the main strength in lowering fire risk is people’s awareness in every village on the negative impact of land burning, along with the existence of WMT.

High Resolution Water Table Modeling of the Shallow Groundwater Using a Knowledge-Guided Gradient Boosting Decision Tree Model

Detailed knowledge of the uppermost water table representing the shallow groundwater system is critical in order to address societal challenges that relate to the mitigation and adaptation to climate change and enhancing climate resilience in general. Machine learning (ML) allows for high resolution modeling of the water table depth beyond the capabilities of conventional numerical physically-based hydrological models with respect to spatial resolution and overall accuracy. For this, in-situ well and proxy observations are used as training data in combination with high resolution covariates. The objective of this study is to model the depth of the uppermost water table for a typical summer and winter condition at 10 m spatial resolution over entire Denmark (43,000 km2). CatBoost, a state of the art implementation of gradient boosting decision trees, is employed in this study to model the water table depth and the associated uncertainties. The groundwater domain has not been the most prominent field of applications of recent hydrological ML advances due to the lack of big data. This study brings forward a novel knowledge-guided ML framework to overcome this limitation by integrating simulation results from a physically-based groundwater flow model. The simulation data are utilized to (1) identify wells that represent the uppermost water table, (2) augment missing training data by accounting for simulated water level seasonality, and (3) expand the list of covariates. The curated training dataset contains around 13,000 wells, 19,000 groundwater proxy observations at lakes, streams and coastline as well as 15 covariates. Cross validation attests that the ML model generalizes well with a mean absolute error of around 115 cm considering solely well observations and a MAE of <50 cm taking also the proxy observations into consideration. Quantile regression is applied to estimate confidence intervals and the estimated uncertainty is largest for moraine clay soils that are characterized with a distinct geological heterogeneity. This study highlights a novel research avenue of knowledge-guided ML for the groundwater domain by efficiently supporting a ML model with a physically-based hydrological model to predict the depth of the water table at unprecedented spatial detail and accuracy.