Search

Accurate and reliable prediction of groundwater depth is a critical component in water resources management. In this paper, a new method based on coupling wavelet decomposition method (WA), autoregressive moving average (ARMA) model, and BP neural network (BP) model for groundwater depth forecasting applications was proposed. The relative performance of the proposed coupled model (WA-ARMA-BP) was compared to the regular autoregressive integrated moving average (ARIMA) and BP models for annual average groundwater depth forecasting using leave-one-out cross-validation (LOO-CV). The variables used to develop and validate the models were average groundwater depth data recorded from 1981 to 2010 in Jinghui Canal Irrigation District in the northwest of China. It was found that the WA-ARMA-BP model provided more accurate annual average groundwater depth forecasts compared to the ARIMA and BP models. The results of the study indicate the potential of the WA-ARMA-BP model in forecasting nonstationary time series such as groundwater depth. [ABSTRACT FROM AUTHOR]

To make a reliable reconstruction of past climate from soil-surface modern pollen, it is necessary to reduce the sources of error. In this paper, pollen percentages of the sub-continental scale modern pollen-climate dataset from China and Mongolia (with 68% soil-surface samples) are homogenized at various spatial scales. A tailored calibration-set is then applied to lake sediment-surface pollen assemblages from north-central China to evaluate their predictive power. Results indicate that spatial homogenization of modern pollen percentages can increase the proportion of inertia explained by climatic variables in CCA and improve the model performance of leave-one-out cross-validation using WA-PLS. Soil-surface pollen assemblages can thus be employed into a calibration-set for reliable climate estimation and they perform better when the calibration-set has been locally homogenized. Small-scale (e.g., radii 2, 5, or 10 km) homogenization reduces the local noise in soil-surface pollen assemblages and improves the cross-validated performance, while broader scale homogenization (more than 20 km radius) blurs the pollen-climate relationship. Lake sediment-surface pollen assemblages from close to the shore could contain pollen grains transported by rivers or from the shore vegetation and thus fail to represent regional climate well like the assemblages from the central part and deep-water area of lake. [ABSTRACT FROM AUTHOR]