As noted, sampling efficiency was evaluated in terms of the number of baseline and flood samples taken relative to the sampling opportunities available.
Baseline
The instances of all baseline samples taken by each CT were used to determine baseline sampling efficiency. Each CT should have taken two baseline samples per day throughout the study period. Efficiency, i.e. the percentage of available baseline sampling opportunities actually taken, was derived using Equation 3
P ercen tage baseline sam ples taken = nu m ber o f b a se lin e sam ples taken (num ber o f days)x2 X 100
1 Equation 3
Flood
Flood sampling opportunities
In order to determine flood sampling efficiency, it was necessary to identify the flood sampling opportunities at each site, and then determine if any (NB: not how many) flood samples had been taken close (i.e. within 45 minutes) to the rise and/or the peak of the flood event.
In order to discretise the rises and peaks, and link flood sampling to opportunities, graphs were generated in Excel that depicted all sample instances and water levels over time for each of the four selected sites for the study period of mid-December 2015 to June 2016.
Water level
With the exception of the Tsitsa at Mbelembushe where data from the DWS gauging weir at Xonkonxa was used, water levels were derived from the pressure transducers installed at each of the selected monitoring sites. Water level data were required to indicate the rises in water levels which were the “trigger” for flood sampling by the CTs, and to indicate the peaks in water levels which were the “target” for the flood focused sampling approach. Discharge will be required in future for the determination of SS loads and yields, but was not required for the evaluation of the CT-based SS sampling approach.
Time
Sample times were derived from the ODK forms downloaded from each CT’s smartphone.
The starting time for each ODK form was assumed to be the time at which each sample was taken. Strictly speaking, sampling time should have been derived from the photograph of each sample, which ought to have been captured as soon as each sample was taken.
However, it was both quicker and simpler (and not much less accurate) to use the ODK form start times, as:
• Under normal conditions and according to the sampling protocol, the sample should be taken within ~10 minutes of opening an ODK form;
• All the numeric and text information (including start time) from each ODK form can be rapidly extracted to Excel spreadsheets in batches using the programme ODK Briefcase;
• ODK Briefcase can unfortunately not retain photographs or their time and geo
attributes within the data extracted and batched from each form, storing the photographs separately in an associated folder. Thus, deriving sample times from the photographs would have entailed laboriously accessing each photograph using Windows Explorer in order to transcribe the time.
The water level pressure transducers and the barometric pressure transducers used to compensate them logged total and air pressure respectively at twenty-minute intervals. Time as measured by the pressure transducers was used as the x-axis for all the analytical graphs produced, except at Mbelembushe where time on the x-axis was derived from the six-minute recording intervals from the DWS gauging weir at Xonkonxa weir.
Night periods
Floods that rose or peaked at night were not available as sampling opportunities to the CTs, and needed to be recognised as such in the efficiency and effectiveness analysis. Night periods were defined by ascertaining sunset and sunrise times (Citipedia.info 2017) throughout the year.
Flood event selection
Two levels of high-flow event selection were undertaken at each site using the water level data and graphs. Firstly, a blanket selection of events was made that included ALL water rises and peaks. Rises and peaks occurring at night were distinguished from those occurring in daylight. This left all the potential triggers for flood sampling as perceived by the CTs, who in real-time could not know the eventual duration or amplitude (i.e., the “significance”) of the event they were observing.
Secondly, a subjectively filtered sub-set of “significant” events was selected, comprising those events which from visual analysis of the graphed data were deemed likely to move significant amounts of sediment. Typically, (but not exclusively) these were events separated by a return to near base flow depth, followed by an emphatic rise to a peak of at least twice base flow. Figure 22 illustrates the first and second level of event selection, with eight significant rises/peaks selected from a total of 20 rises/peaks. Again, this subset of
“significant” rises and peaks was divided into “total” and “daylight” rises and peaks.
Figure 22: Graph showing an example of significant rise/peak selection for flood sampling efficiency analysis
The Excel spreadsheets included in Appendix 1 contain the relevant data and graphs used during event selection.
Analysis
The resulting flood event opportunities were analysed as follows:
• Total rises sampled
• Daylight rises sampled
• Total peaks sampled
• Daylight peaks sampled
• Significant rises sampled
• Significant daylight rises sampled
• Significant peaks sampled
• Significant daylight peaks sampled
The efficiency of total and significant flood sampling was thereby determined, not only within the limits imposed on the CTs (daylight sampling) but also against the hypothetical performance of a probe or automated sampler that could monitor or sample continuously.