(749a) Exploration and Forecasting of Water Usage for Treatment Process Development in Small Disadvantaged Agricultural Communities

Remote and disadvantaged communities in agricultural regions are confronted with impairment of their local potable groundwater supplies due to contamination and rising salinity associated with agricultural activities. Nitrate contamination of community potable groundwater sources has been reported in various areas throughout the United States. In California, for example, more than 205 community well water systems have been found (during 2002- 2010) to exceed the maximum nitrate contaminant level (MCL) of 10 mg/L as N and a significant number of communities are also confronted with high salinity above the recommended level of about 500 mg/L total dissolved solids (TDS). These communities rely on local water supplies which must be treated in order to provide safe drinking water.

In order to develop the required local wellhead water treatment and water distribution infrastructure for small remote communities, forecasting of water use information is needed to quantify the temporal (from hourly to seasonal) variability of was use and thus the treatment and delivery capacity. Moreover, information regarding water storage capacity needs and handling of sanitary water in community septic systems must be assessed. Accordingly, the present work demonstrates a data-driven modeling approach to describe and forecast water use patterns in small communities that rely on local well water for their potable water supply. The study included three small communities having population in the range of 16 – 36 located in the midst of agricultural fields in the Salinas Valley (California). These communities rely on their local wells for domestic potable water supply and manage their sanitary wastewater in their local septic systems as they have no access to a centralized sewer system. Groundwater supplies of these communities have been impaired due to nitrate contamination and wellhead treatment via membrane water treatment technology is a viable option. However, the design and operation (which invariably is expected to be intermittent) of such systems has to consider temporal variability of water use patterns. In particular, daily and peak water use data are critical to assess needed water storage capacity, and the volumetric rate of treatment and/or water delivery, as well as establishment of operational protocols. Accordingly, the present study focused on developing data-driven model of water use patterns that is suitable for small agricultural communities. The model was based on water use data compiled for the three study communities, via smart water meters, over a period of six years.

Exploratory data analysis was first conducted to quantify water use patterns at hourly, daily, weekly, monthly and seasonal levels. Self-organizing maps (SOMs) was initially conducted to assess the overall patterns of average daily water use throughout the different months. Subsequently, autoregressive moving average (ARMA) models were developed for water use patterns at different temporal scales (hourly, daily). Climate metrics (i.e., daily and monthly low/high temperature (oC) and rainfall (inches/day) for the region were included in the analysis to assess their relevance to the correlation of water use patterns. Data outliers were retained in the ARMA models given that these models can handle outliers. Water usage patterns for the study communities revealed temporal irregularity (i.e., the variance was non-stationary for the time-series data). Thus, water use data for the study sites followed non-stationary stochastic patterns with non-uniform variance. Accordingly, for ARMA model development for water consumption, non-stationarity was removed via the 2nd order differencing. The ARMA models were based on the combination of auto-regressive (AR) and moving average (MA) polynomials. Combination of the AR and MA polynomials was achieved considering both the variable linear relationship and linear dependence of the forecast errors. The performance of the ARMA models was quantified by the Root Mean Squared Error (RMSE), R-squared, and Trends.

Analyses via ARMA models with good predictive performance (i.e., R2 > 0.85) demonstrated that the predicted changes of water consumption can provide critical information for water supply scheduling, sizing the required water treatment system capacity, the required level of product (treated) water recovery, operation frequency and required local water storage as a buffering capacity of needed water supply. Self-organizing maps (SOM) clustering and Kullback-Leibler (KL) divergence approaches identified similarities among the remote communities in terms of their water use at seasonal, monthly, weekly, and daily levels. This community similarity assessment facilitated the use of transfer learning (i.e., the use of community data for predicting water use for other communities) and data imputation (I.e., filling data gaps). Additionally, in order to detect and report outliers in real-time data (e.g., unexpected spikes in water use due to community garden irrigation or leaks), a Gaussian Mixture Model (GMM) with 2 Gaussian distributions was trained and tested. The outlier detection approach can be useful for operational decision support (i.e., to distinguish between normal and atypical conditions) as well as the development of robust and adaptive predictive model to automatically predict water use with high predictive accuracy and to detect atypical water use. Finally, it is noted that the inclusion of temperature and rainfall as model input attributes improved the model predictive performance in terms of R2 by as high as 29% (from R2 of 0.61 to 0.9) and 27% (from R2 of 0.54 to 0.81) for average daily and hourly water use, respectively. The association of water consumption with meteorological parameters is consistent with the expectation of higher water usage during hotter summer period and lower usage during the colder winter rainy period.

The present study suggests that the ARMA type models can be a useful tool in support of planning regarding water source management, and the design and deployment of local water systems and needed water treatment and wastewater handling. However, development of water use patterns models for causal analysis would require additional model parameters. For example, such parameters could include, but are not limited to, community descriptors such as occupation, income, residents per household, size and number of residential units.