Application of Artificial Neural Network for Forecasting Nitrate Concentration as a Water Quality Parameter: A Case Study of Feitsui Reservoir, Taiwan

Water resources play a vital role in various economies such as agriculture, forestry, cattle farming, hydropower generation, fisheries, industrial activity, and other creative activities, as well as the need for drinking water. Monitoring the water quality parameters in rivers is becoming increasingly relevant as freshwater is increasingly being used. In this study, the artificial neural network (ANN) model was developed and applied to predict nitrate (NO3) as a water quality parameter (WQP) in the Feitsui reservoir, Taiwan. For the input of the model, five water quality parameters were monitored and used namely, ammonium (NH3), nitrogen dioxide (NO2), dissolved oxygen (DO), nitrate (NO3) and phosphate (PO4) as input parameters. As a statistical measurement, the correlation coefficient (R) is used to evaluate the performance of the model. The result shows that ANN is an accurate model for predicting nitrate as a water quality parameter in the Feitsui reservoir. The regression value for the training, testing, validation, and overall are 0.92, 0.93, 0.99, and 0.94, respectively.

Facultative amphidromy and pelagic larval duration plasticity of Rhinogobius formosanus (Teleostei, Gobioidei)

Rhinogobius formosanus Oshima, 1919 has long been considered an amphidromous goby. However, a landlocked population recently found in the Jingualiao Creek upstream of the Feitsui Reservoir in Taipei suggests that R. formosanus may complete its life in the river. This study aims to verify the habitat use of the landlocked population of R. formosanus collected from the Feitsui Reservoir and an amphidromous population collected in Malian Creek using otolith Sr:Ca ratio analysis. The hypothesis that early life history varies between the landlocked and migratory gobies was also tested. Genetic analyses show that the Feitsui Reservoir and Malian Creek populations are not genetically different. Rhinogobius formosanus from Malian Creek showed high-to-low otolith Sr:Ca ratios suggesting that these specimens spent a planktonic larval stage in the sea followed by a freshwater life at later stages. In contrast, R. formosanus from the Feitsui Reservoir showed constant lower otolith Sr:Ca ratios, implying a landlocked life history of fish in the creek upstream of the reservoir. In addition, the analysis of growth increments showed a longer pelagic larval duration for the fish in the Malian Creek (58.8 days) than those in the Feitsui Reservoir (38.8). Variation of pelagic larval duration in two genetically homogenous populations implies acclimatization to the reservoir by the landlocked gobies. This study shows that R. formosanus, like some other congeners, is capable of adapting to a freshwater landlocked environment in its early developmental stage and supports the hypothesis that landlocked populations may have a shorter pelagic larval duration.

Variations in triple isotope composition of dissolved oxygen and primary production in a subtropical reservoir

Lakes and reservoirs play an important role in the carbon cycle, and therefore monitoring their metabolic rates is essential. The triple oxygen-isotope anomaly of dissolved O2 [17Δ = ln(1+δ17O) − 0.518 × ln(1 + δ18O)] offers a new, in situ, perspective on primary production, yet little is known about 17Δ from freshwater systems. We investigated the 17Δ together with the oxygen : argon ratio [Δ(O2 ∕ Ar)] in the subtropical Feitsui Reservoir in Taiwan from June 2014 to July 2015. Here, we present the seasonal variations in 17Δ, GP (gross production), NP (net production) and the NP ∕ GP (net to gross ratio) in association with environmental parameters. The 17Δ varied with depth and season, with values ranging between 26 and 205 per meg. The GP rates were observed to be higher (702 ± 107 mg C m−2 d−1) in winter than those (303 ± 66 mg C m−2 d−1) recorded during the summer. The overall averaged GP was 220 g C m−2 yr−1 and NP was −3 g C m−2 yr−1, implying the reservoir was net heterotrophic on an annual basis. This is due to negative NP rates from October to February (−198 ± 78 mg C m−2 d−1). Comparisons between GP rates obtained from the isotope mass balance approach and 14C bottle incubation method (14C–GP) showed consistent values on the same order of magnitude with a GP ∕ 14C–GP ratio of 1.2 ± 1.1. Finally we noted that, although typhoon occurrences were scarce, higher than average 17Δ values and GP rates were recorded after typhoon events.

Variations in triple isotope composition of dissolved oxygen and primary production in a subtropical reservoir

Lakes and reservoirs play an important role in the carbon cycle, and therefore, monitoring their metabolic rates is essential. The triple oxygen isotope anomaly of dissolved O2 [17Δ = ln(1+δ17O) − 0.518×ln(1+δ18O)] offers a new, in situ, perspective on primary production, but is yet to be evaluated in freshwater systems. We investigated the 17Δ together with oxygen-argon ratio (δO2/Ar) in the subtropical Feitsui Reservoir in Taiwan from June 2014 to July 2015. Here, we present the seasonal variations in 17Δ, GP (gross production), NP (net production) and the NP/GP (net to gross ratio) in association with environmental parameters measured. The 17Δ varied with depth and season, with values ranging between 19 and 186 per meg. The 17Δ GP rates were lower from April to September averaging 215±93 mg C m−2 d−1 and higher from October to January averaging 523±66 mg C m−2 d−1. The estimated average annual 17Δ GP was 104 g C m−2 year−1 and the average annual NP was 22 g C m−2 year−1. Overall, the NP/ GP varied slightly between 0.02 and 0.36 and the reservoir was net autotrophic in the mixed layer. Comparisons between 17Δ GP rates and the production rates estimated by in vitro 14C bottle incubation method (14C GP) were consistent on the same order of magnitude, with the 17Δ GP/14C GP ratio of 1±0.8 throughout the study. Although typhoon occurrences were scarce, higher than average 17Δ values and 17Δ GP rates were recorded after typhoon events.

Time series modeling of biochemical oxygen demand at the upstream catchment of Feitsui Reservoir, Taiwan

The adaptive neuro fuzzy inference system (ANFIS) has been proposed to model the time series of water quality data in this study. The biochemical oxygen demand data collected at the upstream catchment of Feitsui Reservoir in Taiwan for more than 20 years are selected as the target water quality variable. The classical statistical technique of the Box-Jenkins method is applied for the selection of appropriate input variables and data pre-processing of using differencing is implemented during the model development. The time series data obtained by ANFIS models are compared to those obtained by autoregressive integrated moving average (ARIMA) and artificial neural networks (ANNs). The results show that the ANFIS model identified at each sampling station is superior to the respective ARIMA and ANN models. The R values at all sampling stations of the training and testing datasets are 0.83–0.98 and 0.81–0.89, respectively, except at Huang-ju-pi-liao station. ANFIS models can provide accurate predictions for complex hydrological processes, and can be extended to other areas to improve the understanding of river pollution trends. The procedure of input selection and the pre-processing of input data proposed in this study can stimulate the usage of ANFIS in other related studies.

Effects of mountain agriculture on nutrient cycling at upstream watersheds

The expansion of agriculture to rugged mountains can exacerbate negative impacts of agriculture activities on ecosystem function. In this study, we monitored streamwater chemistry of four watersheds with varying proportions of agricultural lands (0.4, 3, 17, 22%) and rainfall chemistry of two of the four watersheds at Feitsui Reservoir Watershed in northern Taiwan to examine the effects of agriculture on watershed nutrient cycling. We found that the greater the proportions of agricultural lands, the higher the ion concentrations, which is evident for fertilizer-associated ions (NO3-, K+) but not for ions that are rich in soils (SO42-, Ca2+, Mg2+), suggesting that agriculture enriched fertilizer-associated nutrients in streamwater. The watershed with the highest proportion of agricultural lands had higher concentrations of ions in rainfall and lower nutrient retention capacity (i.e. higher output–input ratio of ions) compared to the relatively pristine watershed, suggesting that agriculture can influence atmospheric deposition of nutrients and a system's ability to retain nutrients. Furthermore, we found that a forested watershed downstream of agricultural activities can dilute the concentrations of fertilizer-associated ions (NO3−, K+) in streamwater by more than 70%, indicating that specific landscape configurations help mitigate nutrient enrichment to aquatic systems. We estimated that agricultural lands at our study site contributed approximately 400 kg ha−1 yr−1 of NO3-N and 260 kg ha−1 yr−1 of PO4-P output via streamwater, an order of magnitude greater than previously reported around the globe and can only be matched by areas under intense fertilizer use. Furthermore, we re-constructed watershed nutrient fluxes to show that excessive leaching of N and P, and additional loss of N to the atmosphere via volatilization and denitrification, can occur under intense fertilizer use. In summary, this study demonstrated the pervasive impacts of agriculture activities, especially excessive fertilization, on ecosystem nutrient cycling at mountain watersheds.