Water Treatment Technologies and Their Costs - Global Drinking Water Management and Conservation - page 38

flow rates were obtained from the actual
manufacturers (see footnotes for details). For all estimated models, we

The raw data for costs for different

nd average
costs per volume of treated water, where the costs are (1) capital costs, amortized
(by straight-line depreciation) over a 20 year period, and (2) O&M costs, that
include labor, materials, and energy costs for given

ow rates.
In the case of surface water, UV-based technologies would most likely require
that source water be pretreated using a filtration or sediment removal process before

being disinfected by UV. For communities that are concerned about pesticides and
other micro-pollutants, advanced oxidation processes (AOPs) may be worth con-
sidering. AOPs may not be practical for small systems, but with the implementation
of new regulations on drinking water quality in the future, it may be worthwhile for
small systems to include UV-oxidation-based treatment technologies in their menu
of possible technology options. We note that there are some small communities that
are already using AOPs for surface water treatment and also for groundwater
remediation, even at a small scale. 2
We brie

y describe each technology in Table 3.3 3 and illustrate the statistically
modeled costs associated with each of them thereafter. All the technologies con-
sidered here produce municipal standard drinking water, and most assume that the
raw source water is surface water, which is easily contaminated by animals and/or
human activity.
We use the nonlinear least squares (NLLS) estimation process since it can
capture a wider range of functional forms than the ordinary least squares (OLS)
method. Simple linear models may not describe certain data generating processes
very well especially if the functional form changes over its domain. For instance,
our cost data for UV (see Fig. 3.2 ) shows that a much better description of the data
can be had if a nonlinear approach (solid line) is used instead of a strictly linear one
(dashed line). In fact, since most of our data followed the same format as in Fig. 3.2 ,
we used the NLLS method to estimate cost functions for the different classes of
technology. The NLLS technique has the added advantage of yielding better esti-
mates when the amount of data is limited. 4

2 Stockton California remediates its groundwater using Trojan UV Environmental Contamination
Treatment, an AOP; their flow rate is 1,100 cubic meters per day.
3 This information was collected from a number of companies that produce each technology.
4 Of course if we make the error term multiplicative, then we could estimate the model by simply
taking logarithms. But then we would have to assume that the logarithm of the error term is
normally distributed. There is no justi cation for such an assumption. Here we follow the practice
of standard statistical models in which the error term is always additive, representing all omitted
variables. The objective is to estimate economies of scale given by the estimated exponent in the
nonlinear least squares model. This estimated exponent is the constant elasticity, as is well known.