SEARCH WITHIN CONTENT
Citation Information : International Journal on Smart Sensing and Intelligent Systems. Volume 13, Issue 1, Pages 1-7, DOI: https://doi.org/10.21307/ijssis-2020-005
License : (BY-NC-ND-4.0)
Received Date : 20-December-2019 / Published Online: 15-February-2020
Corrosion of external cast-iron pipe surfaces, a major contributor to pipe failure, has been attributed to the free water in the soil surrounding the pipe. Because observation at pipe depth is difficult, a potential proxy is the soil surface moisture. Herein highly accurate elevation data derived from airborne light detection and ranging is used to model the distribution of soil water in urban catchments containing pipe infrastructure. The results are compared with local soil moisture Theta Probe measurements along the pipe. The results show potential to identify wetter spots above underground infrastructure, which may inform its corrosion potential, without digging up the asset.
Many major cities such as Sydney have extensive networks of water supply pipelines, including major trunk lines that are ageing and increasingly prone to failure. Many of these pipes are made of cast iron. This material is traditional and has been shown to have excellent long-term durability, with many pipes surviving well past 100 years. Others, however, fail much earlier, and often this is through corrosion of the external surfaces of pipes. Such corrosion usually leads to weakening of pipe walls and eventually to pipe failure (Melchers, 2017). The mechanisms involved have been the subject of an extensive recent research programme (Melchers et al., 2018, 2019; Petersen and Melchers, 2014, 2019). Previous work identified that, apart from workmanship issues associated with the initial installation of the pipes and their burial, a major factor in the corrosion of the pipes at depth, typically one metre but sometimes much more, is the free water that can make its way from the surface through the backfill soil to the pipe exterior walls. The major contributor to this free water is rainfall, and the associated accumulation of water in the soil surrounding the pipe.
Terrain indices are commonly used to map the spatial variability of soil moisture and identify those areas prone to saturation (Kemppinen et al., 2018; Kim, 2009; Tenenbaum et al., 2006; Western et al., 1999). These indices describe the spatial distribution of soil moisture through a catchment based on elevation change (i.e. local slope) and the contributing area. Originally these techniques are used in natural catchments, although they show potential for testing in urban catchments, similar to that described in this work. For example, a topographic wetness index (TWI) (Beven and Kirkby, 1979) has been used successfully to assess the stability of a rail line corridor in the United Kingdom based on the soil moisture distribution (Hardy, 2010). For any assessment with terrain indices, accurate elevation data is required. Airborne light detection and ranging (LiDAR) technology measures return laser pulses, typically from an aircraft. This technique has the ability to estimate distances, and hence elevations, extremely accurately at a high spatial resolution. LiDAR data are commonly used for terrain indices analysis and have been shown to provide accurate results (Kemppinen et al., 2018; Tenenbaum et al., 2006; Hardy, 2010; Thomas et al., 2017; Lang et al., 2012; Vaze and Teng, 2007; Vaze et al., 2010).
The work described herein shows that terrain indices derived from airborne LiDAR can be used to map the spatial variability of soil moisture in the corridors along pipe infrastructure. This will be of assistance in the assessment of corrosion potential of buried infrastructure and aid in mitigating its impacts, both economically for the water authority and for the community.
The study area for the present project was limited to a single pipeline. It is located within Jesmond Park, a public park located in Newcastle, NSW, Australia (Figure 1). The pipeline is part of the water supply infrastructure managed by Hunter Water Corporation (HWC), the local water authority. The park is managed by the City of Newcastle and subject to typical management practices and maintenance, including watering. The catchment was selected because it has a mixture of pervious and impervious surfaces, each of which contribute in a different way to surface and subsurface water runoff and routing. This was considered desirable to test the effectiveness of the LiDAR terrain index methodology and also to permit comparison with previous work that involved measurements of corrosion pitting depth (Melchers, 2019).
LiDAR data were provided by HWC for their area of operations. This data set was originally collected by NSW Land and Property Information over NSW (now held by NSW Spatial Services). It was provided in a point-cloud format that allowed different point classifications to be included in the creation of the elevation model. The data set used in this study was collected with a Leica ALS50-II device on 14 September 2014 and was provided with points already classified. The accuracy of the data is 0.3 m vertically and 0.8 m horizontally with 95% confidence (Batchelor, 2015).
Terrain indices are used to calculate the accumulation of soil moisture for each cell, into which the catchment of interest has been divided. For a rectilinear regular grid this means that each cell is surrounded by eight neighbouring cells, including those at the corners. The water accumulation is derived from the catchment contributing area and its slope, to obtain the spatial distribution of relative wetness. An important aspect of each terrain index is the flow routing algorithm associated with it since this algorithm calculates the water flow direction based on elevation changes in neighbouring cells.
The terrain index used in the present study is the SAGA wetness index (SWI) (Conrad et al., 2015). According to Lang et al. (2012), this uses the FD8 flow routing algorithm (Freeman, 1991). The FD8 algorithm routes water to all eight neighbouring cells determined by slope, with steeper slopes causing more water to move in that direction. Other algorithms such as D8 or D∞ route to only one or two neighbouring cells (Lang et al., 2012).
The SWI is recommended for use on valley floors as it represents more accurately the lateral redistribution of moisture, at least to compare the TWI (Millard and Richardson, 2014). This might be expected as SWI was designed for use where there are small vertical distances between channels and the base of the valley (Olaya and Conrad, 2009; Boehner et al., 2002). The SWI has been preferred in other studies (Kemppinen et al., 2018). Since the present study is in an urban setting, many of the flow paths are altered from their natural conditions and to better fit those of a valley floor with little difference in vertical elevation between neighbouring cells. The SWI is defined as:Boehner and Selige, 2006). Higher values of the SWI represent wetter conditions.
For the present study, the SWI was calculated using the notion that depressions in the (local) topography are not filled with water. This is considered more representative of local-scale elevation depressions (Lang et al., 2012) and how free moisture may accumulate around pipe infrastructure.
The LiDAR point cloud data set was processed to remove high and medium vegetation from the data set to create an elevation model that shows minor vegetation noise on the ground as well as urban infrastructure (i.e. houses, buildings and roads). This was done to allow the flow routing of the SWI function to portray realistic flow directions. The point cloud data set was processed using ArcMap v10.6.1 to create an elevation data set with a 1 m resolution over the study catchment. The 1 m resolution was adopted for the present study as it has been reported to provide accurate results for larger-scale topographic indices and hydrologic modelling (Vaze and Teng, 2007; Vaze et al., 2010). The resulting elevations were used to delineate the catchment using functions from SAGA v2.3.2 (Conrad et al., 2015) via the QGIS v3.4.9 long-term release interface. This was done to reduce the number of cells used in computations without removing any that affect the contributing area (i.e. SCAm) from the SWI calculations. The catchment delineation was completed using a catchment depression filling routine (Wang and Liu, 2006) which fills localised depressions to ensure hydrologically connected contributing upland areas (i.e. SCA m ) are derived. It is important to note, although the catchment was derived using a filled elevation, the unfilled catchment elevation was used for the SWI calculation for a given point, as discussed in the previous section.
Soil moisture data were collected using a Delta-T Theta Probe (ML3), with a Delta-T HH2 Moisture Meter for reading values (Delta-T Devices Ltd, 2013). The probe can penetrate approximately 6 cm into the soil, and works by measuring the dielectric constant of the soil (Figure 2). More detailed information about the procedure and the measurement technique is available, as reported by Matula et al. (2016). Soil moisture observations were collected on three transects along the specified pipeline. This was to allow various moisture conditions to be observed. At each point, the probe was inserted into the ground at three locations within a 0.5 m × 0.5 m quadrat. To account for the localised spatial variability of soil moisture, three readings were taken and then averaged for the comparison with the SWI. The soil moisture transects were completed on 23 August 2019, 5 September 2019 and 24 September 2019.
The location of each sampling point was recorded with a Garmin eTrex H GPS unit for georeferencing to compare with the SWI. The location of each measurement along transects 2 is plotted in Figures 1 and 3 (lower).
The soil moisture readings taken at transects 1, 2 and 3 had average values of 14.25, 30.59 and 31.68%, respectively, with standard deviations 5.56, 10.69 and 9.64%. For transect 1, the average moisture readings are consistent with the occurrence of some minor recent rainfall events but no significant rainfall for many weeks before the soil moisture readings were taken. However, for transects 2 and 3, large storm events occurred in the week leading up to the sampling date. Although the minimum soil moisture content readings remained relatively constant (6.0, 8.1 and 9.0%) the maximum and range varied quite significantly between the transects. The maximum soil moisture of 29.0% in transect 1 compares with 53.7 and 55.0% in transects 2 and 3. The differences in maximum and minimum soil moisture follow a similar pattern, with transects 1, 2 and 3 returning values of 23.0, 45.6 and 46.0%. It should be noted that the transects were to be performed at various times after rain, in order to allow estimation of the changes in moisture conditions and their spatial distributions between sampling dates. In turn, this was done to allow assessment of the effect differing moisture contents and thus the suitability of SWI for predicting soil moisture distributions in various degrees of wetness.
Figure 3 shows the output of the SWI calculations for the catchment area and the observation points for transect number 2. The locations marked in red show depressions with relatively large contributing area, which would be expected to permit more soil moisture accumulation. The point-wise variation demonstrates the spatial variability of wetness identified across the pipeline. Extremely low and negative values are seen at houses and for steep slopes (i.e. high β, see Equation (1)) with small catchment areas (i.e. low SCA m ), indicating these cells have low soil moisture accumulation.
Figure 4 shows the regression equations of the three transects considered herein. The derived regression coefficients R indicate a reasonable correlation between the SWI and the soil moisture in all conditions. Additionally, each regression equation returned a p-value of less than 0.05, indicating all correlations are significant. Better results are seen for transects 2 and 3, when the average soil moisture increases (to 30.59 and 31.68%) leading to an R value of 0.81 and 0.87 (p-value < 0.05). This trend agrees with similar results in the literature using terrain indices that observed wetter conditions providing better correlations with soil moisture (Tenenbaum et al., 2006; Western et al., 1999; Hardy, 2010). It is considered likely that the lower accuracy during drier conditions is caused by a continuous lack of rain, and hence very low levels of soil moisture, as well as the drying of soil across horizontal planes, without runoff routing occurring for redistribution of stored soil water (i.e. for transect 1, when the average soil moisture was 14.25% the correlations returned an R value of 0.68 (p-value < 0.05). This indicates that the LiDAR-based procedure still provided a reasonable estimation of the soil moisture content and its spatial distribution. Under the dry conditions experienced when observations for transect 1 were performed, the reduced soil moisture range could be responsible for the poorer correlation results (compared to transects 2 and 3), as a smaller range prevents the SWI from establishing the relative moisture as is seen in wetter conditions.
It is likely that in extremely wet conditions the terrain indices will reduce in effectiveness as the spatial distribution of water will show high moisture at all points being assessed. This aspect will need further investigation.
In addition to the work described above using SWI, the standard TWI technique was tested over the three transects using the D8 and D∞ flow routing algorithm. The results from these tests are not presented herein but showed poor relationships with the soil moisture probe measurements. This suggests that for urban environments such as studied herein, SWI may be better suited to soil wetness modelling.
Although the Theta Probe only measures soil moisture 6 cm deep, previous studies indicate that these values are likely to have closely similar mean, variance and frequency distribution to soil moisture values at depths up to 30 cm (Wilson et al., 2003; Bretreger et al., under review). The correlation of the 6 cm soil layer to deeper layers needs further investigation.
As mentioned, the measurement of moisture at pipe depth (around 1 m deep or more) is problematic. However, the 6 cm validation as reported herein may be sufficient to give confidence to inform asset managers of the relative accumulation of soil moisture at practical pipe depths, although further investigation with conditions at greater depth is considered warranted.
Although the SWI is providing a good correlation with soil moisture for the three transects considered herein, it is clearly a relative measure that must be interpreted such as through correlations as shown in Figure 4. Finally, it is important to note that the SWI is a static measure of relative wetness for the specific catchment for which it was determined. As such it cannot be directly used to compare soil moisture in the time domain (i.e. temporally).
The SWI was found to give an indication of the spatial distribution of relative soil moisture around a localised catchment, based on elevation changes, and subsequent runoff and routing flow paths and potential moisture accumulation. The presented results show that SWI is effective in providing trends of soil moisture distribution at up to a 6 cm depths along a pipeline in this catchment. The effectiveness of the SWI based method was reduced when using it in drier conditions. This is likely due to less variability in soil moisture.
The results presented herein provide support for further work to assess the ability of SWI for predicting spatial soil moisture variability and patterns for the localised catchments relevant for underground pipelines and other infrastructure and their condition and corrosion assessment.
The authors would like to thank Ian Hiles from Hunter Water Corporation for providing the LiDAR and pipe spatial data. The authors would also like to thank Chun Yu Wong and Ankur Srivastava for their help when collecting the field data used in this paper. Sydney Water Corporation and Hunter Water Corporation supported this work through the NSW Smart Sensing Network’s Critical Underground Pipe Sensing project.