A Survey on Pollution Monitoring Using Sensor Networks in Environment Protection

Table 1

Comparison of different pollution source localization methods.

Methods	Advantages	Deficiencies	Feature properties
The coarse localization algorithms	The coarse localization methods are easily to be implemented in engineering.	The localization result is affected easily by a node’s measurement and nodes’ deployment. Localization precision is comparatively low.	Theoretically, this method can be used to locate a pollution source even when there is only one node. This method is rarely used in multiple-pollution-source localization.
The localization based on contours	An accurate result can be obtained easily if concentration contours has obvious characteristics.	The diffusion contours are irregular in many cases, and the localization methods based on contours cannot be always used.	At least 5 nodes are needed to estimate the plane contour’s shape [74]. This method is not used to locate multiple sources.
The localization methods by solving inverse problems	Even if the pollutant dispersion model is not used, an analytic result is obtained.	The accuracy is determined by the sampling methods of sites which are potential pollution source locations in the field.	At least 3 nodes are needed to locate a pollution source. If pollution sources are far away from each other, this method can be used to locate multiple sources.
The maximum likelihood estimation	The estimation accuracy is improved with the increase in the sample size. The method is a statistical inference method with a comparatively smaller computational complexity.	The likelihood function is hard to be obtained if the prior distribution of the concentration is not known. If the diffusion model is of strong nonlinearity, a globally optimal solution is difficult to be obtained.	Node number requirement is related to the number of unknown parameters and the nonlinear complexity of the diffusion equation. This method can be used to locate multiple pollution sources if an explicit propagation expression can be given.
The Bayesian estimation	Accurate localization results can be obtained more possibly.	The computational complexity in this method is high.	Similar feature properties as the maximum likelihood estimation
Localization methods using a sequence filter	Time-varying characteristics of observations are considered. The computational complexity is not high comparatively.	Choosing proper noise parameters is a problem.	Theoretically, one node’s monitoring values in different times are enough to locate a source. This method can be used to locate multiple pollution sources if an explicit propagation expression can be given.
Localization methods based on least squares	In the least square, there are no complicated statistical calculations. This method has a strong universality.	If the diffusion model is of strong nonlinearity, a globally optimal solution is hard to be obtained.	Node number requirement is related to the number of unknown parameters. In a linear least square problem with two unknown parameters, no less than 3 nodes are needed. This method can be used to locate multiple pollution sources if an explicit propagation expression can be given.