Runoff plots are expensive and usually ineffective, and worldwide the vast majority of plots have produced little or no usable or worthwhile information. That may sound a harsh judgement but is the considered opinion of those who have spent a working lifetime either operating plots or studying other peoples plots, both making mistakes and observing those of others. To quote again from Sir Charles Pereira, "The most absurd pretensions, that half a kilo of soil from two or three square metres can give estimates of loss rates from hill slopes, have been boosted by 'experts'. A vast amount of totally meaningless data has been acquired on the specious excuse that it is 'better than nothing'".
The remarkable thing is that the weaknesses of plot research have been known for more than 50 years, but the lessons have not been learned, and the same mistakes are still being made. When plot work started in the USA in the 1920s, statistical design and analysis was just beginning to be applied to agricultural research. Since then it has been developed and refined for all other aspects of agricultural field research, but somehow just has not caught on in the use of runoff plots. This may be due to the lack of advice on the subject in the literature, which is one of the reasons for this Soils Bulletin. Reports of runoff plots are constantly published in the technical literature, especially recently as they became fashionable, and there are several manuals on how to build and operate plots, but few offer even the most basic advice on how the experiment should be designed, and the results analysed.
But advice is there in the literature. Brandt explained the basic principles of statistical analysis in 1941, and this was reinforced by Hayward who, in 1968, showed that most plot work was seriously flawed, but the standard is not improving as one might expect. A study of the proceedings of the seven meetings of the International Soil Conservation Organization between 1978 and 1992 shows that reports which show poor design and poor analysis greatly exceed studies with good design and analysis.
Can the reason for this be found? Perhaps the copy-cat effect is partly to blame. Academics and research workers see plot work as a high profile activity, and the scramble for publications or meeting a thesis deadline can lead to premature reporting. Some experiments arise from the over-enthusiasm for trying to get local data to rejig the Universal Soil Loss Equation, even in conditions where the USLE is not suitable (as discussed in Chapter 7). Perhaps another factor is that many people trying to operate runoff plots do not have experience of applying statistical design and analysis to land use problems. Administrators may be looking for numbers to justify conservation programmes, or to persuade farmers to change their farming systems.
The general advice to anyone considering setting up runoff plots has to be "don't, unless:
� there is a specific question which can be answered reliably by plot work, and
� there are sufficient resources to design, build and operate the plots efficiently and effectively and to analyse the results".
Where runoff plots are being used, it is important to remember the basic feature of experimentation that a small amount of reliable information is always better than any amount of unreliable and therefore unusable information.
An account of the pitfalls which can beset the operation of plots would fill a book. One of the most common is missing important events. "The data published in tables with an added footnote 'The storage tank overflowed for x percent of rains' are useless because it is the heavy rains that cause the most damage. Runoff and soil-loss equipment must, therefore, be properly designed, installed, and operated." (Lal 1988a).
Some other problems, the sources of which are not quoted so as not to embarrass the authors, are:
� "During 1989, by mistake the plots were treated differently from the year before, and several mistakes in measurements and calculations plus loss of data occurred. However, it has to some extent been possible to collect reliable data."
� "An unforeseen problem was encountered... while the magnitude of the over-estimation is not known, it is believed... to be 100% at higher discharge rates."
� "Due to attacks of crickets and monkeys the assessment of maize yield was limited to measuring the height of growth."
� "Unfortunately the primary data from 1988 have been lost."
� "The automatic logging equipment broke down for 3 days in July 1989, and for one day in August 1990. Rainfall and associated runoff data were lost."
� "The suspended load data set for some events was incomplete because samples were lost due to insects, spiders, or debris blocking the flow-splitting device, or due to leakage or accident."
Other reported problems include:
� metal collecting tanks floating out of saturated ground,
� crops destroyed by baboons,
� runoff entering the top of plots,
� taps in collecting tanks left open,
� measuring flumes or divisors blocked by floating debris,
� termite nests which produce large holes into which the runoff disappears without trace.
One of the best uses for runoff plots is demonstration, where the purpose is to demonstrate known facts. Examples are to demonstrate to farmers that serious erosion is taking place, or to show that erosion is much less from a plot which has a good vegetative cover than from a bare plot. In this case the actual amounts of erosion are not important, so there is no need for replications, nor for complicated collection systems which attempt to catch all of the soil loss. The plots shown in Plate 4 had only simple brick tanks into which the runoff and soil loss flowed, and they overflowed in heavy storms so that only a proportion of the soil lost was caught in the tanks, but they were very effective for demonstrating to large numbers of farmers the essential principle of reducing erosion by better cover.
Another valid use is in comparative studies, for example to test, or demonstrate, or get an approximate indication of the effect on runoff or erosion of a simple comparison such as with or without a surface mulch, or the amount of runoff at the top and at the bottom of a slope.
A third possible use is to obtain data which are to be used to construct or to validate a model or equation to predict runoff or soil loss. But the difficulties in collecting data of sufficient accuracy and reliability are so great and so numerous that only large experimental programmes conducted at great expense over a long period of time can really meet this objective. The classic example is the Universal Soil Loss Equation, which in fact is not at all universal, being only applicable to the eastern half of the USA. It has to be remembered that the USLE was based on a data base of approximately 10 000 plot years, and it is unrealistic to imagine that local variations for different regimes of soil or climate can be constructed from the results of a handful of plots for a year or two.
The problems associated with runoff plots are many and varied.
� Runoff plots are expensive, both in the initial construction and in their maintenance and operating costs.
� They use up a great deal of staff time at several different levels. There is much unskilled manual work in farming the plots and applying the treatments and emptying the tanks, but all of this has to be carefully supervised or things will go wrong. Attempts to train junior staff or local farmers to take measurements or carry out precise operations are usually unsuccessful even after what appears to be an adequate training programme.
� Easy access to the site is important, but very often the conditions to be investigated can only be found in remote areas. Complete reliability is only achieved when professional staff can get to the site quickly at all times of day and night and in all weathers. The most interesting storms are likely to occur in the middle of the night at a holiday weekend and it requires a high level of dedication to get up in the middle of the night and go floundering about on a hillside in torrential rain in the dark. Also it is in the infrequent storm that something will go wrong with the equipment and the results will be lost if there is not someone on site to correct the problem.
� Back-up facilities are needed as well as the immediately operational staff. Laboratory facilities will usually be required for handling the samples, and repairs to electrical or mechanical equipment.
� Runoff plots have all the problems and difficulties of agronomic trials but in addition the much more difficult problems of collecting, catching and recording the soil and water. There is a huge scope for faults and errors.
There are also constraints on what can be investigated on small runoff plots. Cultivation and other farm operations using tractors or oxen are difficult, and so are treatments which involve livestock, but increasing the size of the plot to allow realistic farm operations means dealing with large volumes of runoff and soil loss. Treatments which cannot be investigated on small plots are conservation measures which involve substantial earthmoving such as bench terraces or channel terraces. The hydrologic effects of channel terraces cannot be reproduced on small plots because the mechanics of runoff will be completely different. Similarly the effect of bench terracing is totally artificial on a small plot because lateral movement of surface water is inhibited by the plot boundaries. A partial solution to this problem has sometimes been attempted using a side collector channel as shown in Figure 13 but this is still an artificial situation.
The field-scale loss of soil from land with bench terraces depends on the probability of the structure failing, and that is not something which can be studied on small plots. Measuring the effect of barriers set out on the contour also faces the problem of the unknown probability of failure. Examples are agroforestry practices such as hedgerows and alley cropping, trash lines and contour furrows. An assessment of the effect of these on runoff and erosion can only be made on field-scale plots.
Some of the better technical journals are attempting to raise the standard of field experimentation by rejecting articles for publication which do not have a suitable statistical analysis and a reasonable timespan. Only the main points are briefly dealt with in this Bulletin; for more information see the section Further reading.
There are two basic requirements, replication and randomization.
There must be sufficient replications, that is exactly similar repetitions, to allow a measure of the variance within treatments. This is the experimental error caused by unknown and uncontrollable variations in the soil, or the crop, or the treatment, or the equipment, which cause differences in what theoretically should be identical measurements. For agronomy trials of crop varieties, or fertilizer trials, it may be adequate to have only two replications because one expects that the variation will not be large, but with runoff plots there are the same agronomic possible sources of error and in addition many more which can arise from the installation or operation of the plot equipment. As a result three replications should be considered as an absolute minimum, with more if possible. Theoretically, if time allows, it is possible to carry out exploratory trials which will suggest how many replications are likely to be required. For example, one might start with a set of twelve plots and put them all under the same treatment and measure the variance between the twelve replications. This will allow an estimate of the number of replications required to give an acceptable level of precision. There is no absolute mathematical answer to the number of replications because it is a question of what level of accuracy is expected from the experiment.
Crop rotations are an added complication and best avoided if possible. The effect of a course in the rotation could be different in wet or dry years, so each course should strictly be considered a separate treatment and applied, with replications, in each year of the rotation. This leads to a complicated design with many plots.
Without replication there is no way of comparing the variance between treatments with the variance within treatments - that is there is no way of telling whether different treatments are really causing a measurably different effect. Where replications are used, the usual practice is simply to use the arithmetic mean of the measurements, which is not by itself satisfactory. It is very rare for reports to include any information on the variance of the measurements. Table 1 shows data from one of the few published reports which give the separate results from replications (Hatch 1981), and illustrates the wide variation which can occur.
Randomization is necessary to eliminate bias, that is the differences which can arise from variations in, for example, soil characteristics. A set of runoff plots is usually set out along a contour because this avoids variations up and down the slope, but there are other possible sources of bias. If the sequence of six plots along a contour is three replications of treatment A and then three replications of treatment B, as in the upper line of Figure 14, then a fertility gradient along a line of plots would result in bias since it would affect all three replications of one treatment differently from the three replications of the other treatment.
To avoid this source of bias, in a simple field experiment, say three replications of two treatments, i.e. six plots in total, the best method is to select three blocks of two plots each and then to allocate the two treatments randomly to the two plots of each block. This can be done by using any truly random mechanism such as tossing a dice, dealing cards, or using tables of random numbers, as in the lower line of Figure 14.
Theoretically, large plots will reduce the effect of soil variations because the sample is larger, but collection and measurement of runoff and soil loss from large plots are more complex as discussed in the next section. Another thought is that long thin plots side by side have a longer common boundary than square plots side by side, and so might limit the risk of soil variation, but this introduces other factors such as the minimum desirable width for cropping purposes, and the effect of length of slope.
In addition to lack of homogeneity of the soil, another possible source of bias is effects introduced by the equipment. Some of these effects will apply equally to all plots and so will not be removed by randomization, and examples are the interference with overland flow by the plot boundaries, or the capture of rain into uncovered collecting troughs or tanks. But other effects will vary from plot to plot, such as leakages in or out across plot boundaries, or leakage under the collecting trough. Some of these sources of error can be designed out, such as fitting covers to the collecting system, but chance errors, particularly leakage, can and will occur, so that is another reason for random allocation for treatment within blocks.
Several networks of experimental studies set up to study soil loss have recommended a standard design to ease the comparison and compilation of results. For many years field experiments in the USA have used certain common features of size, shape and treatment, such as the bare fallow plot, but this was far from being a national standard and it was no easy task to combine the huge database for the construction of the USLE.
� The French Overseas Agricultural Research Programme (ORSTOM) encourages the use of a common design in the French-speaking countries of north and west Africa using plots similar to those of the USDA (Roose 1988).
� For a network of runoff plots in Asia, a common style is encouraged by the International Board of Soil Research and Management (IBSRAM 1992).
� A network study of the effect of erosion on productivity is presently sponsored by FAO and since this includes a treatment of artificial desurfacing of the soil, a design for the operation of the treatments is recommended in addition to recommendations for the construction of the plots (FAO 1991).
These moves to standardization are constructive, but they all experience some problems as a result of variations introduced by the local operators of the plots. As an example, one network recommends nine treatments randomly distributed on three blocks which should be as similar as possible to serve as replicates, but one operator chose to put the blocks on three different slopes thinking that this would provide additional information. To some extent it could do this with some complicated analysis, but it undermines the ability to estimate the variance within each treatment.
The cheapest and simplest method is to install the plot and then wait for rain, but the unpredictability of rain can make this frustrating. The alternative is to use artificially manufactured rain through the use of rainfall simulators as discussed in Chapter 6. The main advantages of using a simulator are that it can speed up getting results, and the amount and type of rainfall can be controlled, but repetition of simulator runs on the same plot, while improving matters, is not a substitute for replication because it does not eliminate bias from soil variation. If the plot is non-typical, the bias is fed into all the simulator runs and this can only be avoided by setting up the plot separately for each test. Some methods have been developed to avoid the tedium of moving the simulator. On a small scale, Plate 5 shows a simulator which can conduct six replicated tests on the small octagonal plots. A larger scale solution in Australia was to build a large simulator of lightweight materials so that the whole structure could be lifted by crane and moved to a new position (Plate 6). On small plots the rotating simulator shown in Plate 7 is set up between two plots which can be rained on at the same time.
The disadvantage of simulators is that simulators for large plots are expensive to build and have a high labour requirement to operate. Simpler and cheaper simulators are usually restricted to small plots of a few square metres and these do not reproduce real conditions of surface flow.
The measurement of runoff is much easier than the measurement of soil loss, and easier to scale-up to larger quantities. It is therefore relatively easy to measure the runoff from large plots which are necessary in order to reproduce full-scale farming conditions, but collecting, storing and sampling the large quantities of soil involved is a difficult operation. Devices have been designed which are able to give a continuous monitoring of the rate of soil movement in a flume, but they involve sophisticated instrumentation and are beyond the scope of this Bulletin.
Most plots have boundaries which define the area from which the runoff and soil are being collected, but there are some cases where it is appropriate to use unbounded plots using what are usually referred to as Gerlach Troughs after their inventor. They consist of a small ollecting gutter which is let into the soil surface and connected to a small collecting container on the downstream side. There are various degrees of sophistication in the construction of the gutters and containers but expensive or complicated construction is not justified because what is required is a large number of replications to overcome the variation which arises from the fact that, without any boundaries to direct or limit runoff into the collecting gutter, the amount collected depends on the chance occurrence of minor depressions or rills. Figure 15 illustrates an ingenious application which requires only a plastic dustpan and a bottle.
The size of plots must be related to the purpose of the trial.
� Microplots of one or two square metres may be appropriate if the objective is a simple comparison of two treatments where the effect of the treatments is unlikely to be influenced by scale. An example is illustrated in Plate 8 where the objective was to demonstrate and obtain an approximate figure for the difference in surface runoff when a grass mulch was applied to newly planted tea bushes. The runoff was led into old oil drums which were emptied after each storm. The accuracy was not high but the method was so cheap and simple that ten replications could be used, and gave a useful starting point for investigations on a larger scale. The experiment was continued for ten years to assess how the difference changed with the progressive growth of the tea bushes. Other examples of the use of microplots for simple tests are illustrated in Plate 9 looking at the effect of alternative maize populations at IITA in Nigeria, and Plate 10 showing small container plots to measure the effect of slope on splash. The plots were lifted above ground level so that only splash out of the plots was measured, and while this took out the effect of splash inwards onto the plots, it still allowed a valid comparison of the effect of slope alone.
� Small-scale plots usually of about 100 m� are most commonly used for trials of cropping practices, cover effects, rotations, and any other practice which can be applied to small plots in the same way as it would be on a field scale, and where the effect can be expected to be unaffected by plot size. The original size and shape for this type of plots adopted in the USA was extremely arbitrary - six feet wide seemed to be a suitable width, and an area of one hundredth of an acre sounded like a convenient size for calculations, and so gave a length of 72.6 feet. There is some justification for following a well-established practice so that direct comparisons may be made, but there is no justification for suggesting that these precise measurements should be followed in metric units. In fact plots only six feet wide are liable to have a significant border effect and a more sensible size in metric units would be five metres wide and twenty metres long.
It is important to appreciate that this type of plot is not appropriate for any assessment of the effect of earth-moving practices such as channel terraces, bench terraces, or any structures which depend for their effect on interruption to the surface flow. In such cases the amounts of runoff and soil loss are dominated by the occurrence of weaknesses or failures in the system, i.e., breaks or overtopping in channels or the collapse of banks or walls, and these are not properly reproduced on small plots. The partial solution sometimes attempted by building a collector drain down the side of the plot as shown in Figure 13 is seldom satisfactory, and all practices which involve barriers or structures on the contour should only be investigated using larger plots.
� Field plots of about 1 ha are also appropriate for assessing any treatment which cannot be applied realistically to small plots. It is possible to apply cultivation and similar farm operations through the use of removable plot boundaries, as shown in Plate 11 but this requires large areas between the plots for turning oxen or tractors, and also extreme care in the replacement of the plot boundaries after each operation. A major source of error in runoff plots is leakages across plot boundaries so it is unwise to increase this risk by frequent alterations to the boundaries. Plots of the order of 1 ha are necessary to assess any form of terracing, and also to assess the effect of grazing or livestock management. Plate 12 illustrates an attempt to measure the effect of intensity of grazing on rangeland using small plots with a low concrete boundary. Even such a minor obstruction was sufficient to influence the grazing pattern because the cattle did not like walking over it and the photograph shows that the plot retains more vegetation that the surrounding area.
There is a variation in terminology. British usage is 'catchment' for the area which catches runoff, and 'watershed' for the boundary of the catchment, i.e. the divide which sheds water on either side. American usage is 'watershed' instead of 'catchment'. In this Bulletin the two terms are used interchangeably.
All small-plot studies have two in-built deficiencies. First, at the point of collection there is free-fall discharge over the lip of the collecting trough, a condition which does not occur naturally in field runoff. The second is that small plots measure only movement from the plot, and few studies take account of the real-life situation where deposition takes place farther down the slope. Both these factors lead to an over-estimation of soil loss from plots. A valid estimate of runoff and soil loss from a watershed can only be obtained from measurements at the outlet from the watershed. The ratio of soil lost from the whole watershed to the higher rates of soil movement within the watershed is called the delivery ratio. For agronomists concerned about loss of productivity, the gross movement downslope as measured by field plots may be the most appropriate measure, because sediment trapped lower down the watershed has been permanently lost from the cropped area. But it is misleading to suggest that soil loss measured on small plots can be multiplied up and expressed as tonnes/km� as if this were uniform loss from the whole watershed.
The other point about plot size is that the effect of some major changes in land use, such as those resulting from human settlement, or a programme of terracing or treeplanting, can only be evaluated by measuring what happens to the whole watershed. The instrumentation needed to handle large flows and large soil loss is more complex than needed for field plots and is discussed briefly in Chapter 4. Chapter 1 discussed why watershed studies must be carried out on calibrated paired plots and not by comparing before and after treatment on a single plot.
A technique sometimes used to check on the difference between results from plots of different size is called 'nested plots', where small plots are sited within larger plots. An example is a hydrological study carried out in Sri Lanka, where the purpose was to collect reliable and accurate data on the rainfall and runoff regime under existing farming practices, which included irrigated rice, rain-fed tea, and multiple-crop spice gardens. The design was a single watershed of 5000 ha with measured outflow, comprising 15 separately monitored sub-watersheds from 7 ha to 284 ha. Within the sub-watersheds were replicated plots of 1 ha on each of the main farming systems, and replicated plots of one fiftieth ha to study variations such as different densities of tea planting. The hydrographs of runoff from different sub-watersheds could be compared as in Figure 16 (Hudson 1981a).
Another example comes from China, where plots of various sizes from 60 m� to 1000 m� were installed within monitored sub-watersheds within a 2 ha watershed (Mou Jinze 1981). The conclusion was that plots which extended over the entire slope length from the top of the ridge to the bottom of gully slopes were necessary to measure erosion and runoff on a watershed scale.
Size is dictated both by the treatment, as discussed in the section Size of plots, and the system for collecting soil loss and/or runoff. Microplots may have collecting tanks which store the whole of the runoff and soil, but if it is important to measure extreme events, the size of the tank required becomes excessive for plots of more than a few square metres. As an illustration, a 2 m x 2 m plot, subject to a storm rainfall of 100 mm, with a runoff of 80%, needs a tank storing 320 litres to be practical. But single tanks for plots of 5 m x 20 m would be unmanagably large at 8000 litres, so it is usual to employ some sampling device as discussed in the next section. For plots of 1 ha or more, to sample continuously and store sediment-laden runoff requires complicated instrumentation, and a common approach is to measure continuously only the rate of runoff through a flume, perhaps with intermittent sampling of the sediment, as discussed in Chapter 4.
Two factors must be considered when designing the size and capacity of the collector system. It must be able to handle the maximum probable rate of flow and also to store the maximum probable quantity of runoff. It is well established that infrequent extreme events can contribute a large proportion of the annual runoff and soil loss. In tropical conditions this can amount to three quarters of the annual soil loss occurring during a single storm, so the collecting sampling system must be designed to handle extreme events.
The maximum probable rate of runoff can be computed from the maximum probable runoff ratio from an already saturated soil and the probable maximum intensity over a short period of perhaps five minutes. As an example, a plot of 100 m� receiving a burst of rain at an intensity of 250 mm/h and 100% runoff would generate a rate of flow of about 7 litres per second. To allow for the partial blocking by accumulated sediment or floating debris, a pipe of 100 mm diameter would be appropriate.
The total storage capacity required for the plot of 100 m�, allowing for 80% runoff of 200 mm of rain, would require total storage of 16 m3 and some type of divisor system would probably be appropriate, as discussed in the section Tanks and divisors.
There will be only a few storms where the maximum storage is needed, and many small storms which produce small amounts of runoff. If small containers are placed within the larger tanks, they catch small runoffs and can be more easily emptied and recorded. During big runoff events the small cans overflow but the main tank catches the runoff. Plate 13 shows an example from Kenya (Liniger 1990).
Free discharge from the collector into the storage tanks is required. One possible solution is shown in Figure 17, where 150 mm diameter pipes were used to take the water downhill to a point where the tanks could be close together and above ground for ease of sampling and emptying. The more usual system is to have the collecting tanks below the ground surface, and if they are to be emptied by gravity, a drainage system must be provided. Even if the tanks are to be emptied by pumping, a drainage system may be desirable because monolithic metal tanks have been known to float up when the ground became saturated and the water table rose, and brick built or concrete tanks could suffer from inward leakage in similar conditions.
There is no standard for the ratio of length to width. Short plots are generally considered undesirable because they may interfere with rill development, but the relationship between erosion and length of slope is debatable, in spite of the form assumed in the USLE of E�L0.6. The width should be suitable for the farming systems to be applied, as discussed earlier. For cropped plots, a buffer strip at the sides and top may be used to reduce border effects, and also to allow access without walking on the plots.
Many materials have been used as plot boundaries, including earth banks, brick or concrete walls, timber planks, and strips of metal, asbestos-cement, or plastic. They are usually permanently installed for the life of the plots, but sometimes removable boundaries have been used to permit tractor or ox cultivation across the plot, as illustrated in Plate 11.
Other points about boundaries are:
� Leakage in or out of the plot across the boundaries is a common source of error. There should be a drain above the plot to divert surface water flowing down from higher ground.
� Plots should not have common boundaries otherwise one leakage affects the result of two plots. There should be a buffer strip between plots.
� Boundaries must be carefully installed, inserted deep enough to prevent leakage underneath, tall enough to prevent overtopping (particularly at the top boundary), and lengths of timber or metal must either overlap or be butted tightly together to prevent leaks. The boundaries shown in Plate 11 were strips of asbestos-cement 2 m x 16 cm x 1 cm, inserted 8 cm into the soil, held upright by steel pegs driven vertically on either side, and the joints were butted together and covered with a sheet metal spring clip;
� A small filet of earth heaped against the outside of boundaries stops surface water ponding or flowing against the boundary, but the soil must be drawn against the boundary wall without leaving a channel which could start a rill.
� Similarly, when building earth banks, a channel should not be left where the soil has been taken from. The soil should be taken only from outside the plot.
For small plots, the runoff may spill directly into a tank running the width of the plot, but for larger plots some form of collector trough or channel is required leading to the tanks. Several possible sources of error can occur at this point.
If the edge of the collecting trough is higher than the soil surface, the sediment collected will be reduced as soil is trapped by the sill. If the edge of the collector is lower than the soil surface, there will be excessive erosion at this point and the possibility of initiating small rills. Care is therefore required to ensure that the edge of the collector is at original ground level, and many different methods and devices have been used to ensure this. Sometimes the soil is excavated just uphill from the collector and then replaced and compacted to the right level after the collector has been installed. Another method is to construct a smooth flat approach to the collector trough by installing a strip of sheet metal or plaster or concrete.
If large amounts of soil are eroded, the collecting trough may prevent the natural development of a new profile down the slope - that is the sill of the trough will be higher than the level which the soil would be if the trough were not there. To avoid this possible error, some plots have been constructed with a sill which can be progressively lowered as erosion takes place, for example by making the sill of a weak lime plaster which can be scraped down as required (Hudson 1957).
Another possible error is leakage underneath the collecting trough and if this cannot be controlled by compaction of the soil below the collector, some form of cutoff wall should be installed, perhaps by inserting an impermeable membrane or, if the collecting trough is of brick or concrete, by taking the foundations down deep enough to prevent seepage underneath.
In the case of small plots, all the runoff is led into a single collecting tank where it is stored until it can be measured, sampled and recorded. For larger plots, or when large amounts of runoff are expected, it is impractical to store the whole of the runoff, and some device is used to divide it accurately so that a known fraction can be separated off and stored.
There will always be considerable amounts of floating organic material in the runoff, and this must be caught on screens if any type of divisor or sampler is used. Sometimes a wire mesh screen is placed over the collecting trough as shown in Figure 18, or alternatively one or more screens may be placed in the collecting tanks as in Plate 16. A device widely used in the USA for many years is the GEIB Divisor, which consists of a number of equal rectangular slots. The water passing through the central slot is collected and stored, while that through the other slots runs to waste (Plate 14). This requires a high degree of accuracy in its manufacture and so a number of simpler alternatives have been developed. These include a series of V-notches (Plate 15), or vertical rows of holes drilled in a steel plate (Plate 16), or a series of pipes built into the wall of the tank as in Plates 17 and 18. Any of these can be constructed to take a sample of between one fifth and one twentieth of the total flow.
Another approach is to split the flow successively by dividing the flow in half and then dividing again by half as many times as required. Two splitters which divide in half give a sample of a quarter, three give a one eighth sample and four give a one sixteenth. This means more pipework and channels but the divisors themselves can be very simple if they only have to split the flow in two, and an example is shown in Plates 19 and 20.
It is always desirable that any divisor should be checked to see that the sample is exactly the proportion that it is supposed to be. There are several possible sources of error, such as the different sampling points not being at exactly the same level, blockages or partial blockages of some of the outlets, or that the sampling mechanism interferes with the flow through the divisor. As an example, in the slot divisor shown in Plate 14, at high rates of flow the metal chute taking the sample from the third slot inhibits free flow through that slot, so it is taking a smaller sample than intended. Also the velocity of approach to each slot, or notch, or pipe, must be the same. If the divisor is built into a narrow channel the flow through the end slots can be reduced by friction against the channel walls.
All divisors or splitters are likely to interfere to some extent with the flow conditions so there is a possibility of sediment being deposited within the system. This is not necessarily a problem if the calculation of solids is based on the sum of the sediment retained plus an accurate fraction of what is discharged to waste. Thus in a three tank system shown in Plate 21 with two successive seven-slot divisors, the total runoff, or the total soil loss, is that which is retained in the collecting trough and first tank, plus seven times the sediment in the second tank, plus forty-nine times the sediment in the third tank.
There are several types of divisor which involve moving parts. These are only suitable where there will be constant supervision and maintenance, because there is a high risk of such divisors becoming jammed or choked by debris or subject to mechanical failure. One of the best known is the Coschocton Wheel sampler shown in Plate 22. This is installed under the discharge from a flume, such as an H flume, and the force of the water turns a rotating slot sampler mounted on a vertical axis. A possible disadvantage is that the size of the sample is not constant at all rates of flow, and a possible further refinement is to add a motor to give a constant speed of rotation. Another mechanism is the tilting bucket as used in some automatic rain-gauges. This allows the taking of a sample each time the buckets tilt, and also it is fairly simple to add a counter to record the frequency of the tiltings.
Any recording device which depends upon moving parts or electrical supply is suitable only for use on experimental stations where there will be constant supervision, particularly during the extreme event, which is when mechanical and electrical devices go wrong, and usually occurs in the middle of the night.
First the objective of the experiment must be defined. Will annual records be sufficient for the purpose or do they need to be based on a shorter time interval to monitor changes, or are separate data required for individual storms? There is no point in accumulating daily records if they are going to be lumped together into weekly or monthly totals. When labour is short or expensive, it may be cheaper and more accurate to use larger storage tanks or more divisors to reduce the number of readings.
After defining the kind of data required, a balanced choice is required between how much to use suitably trained staff to take measurements and keep records, and when to use automatic instrumentation.
Good access to the experimental sites is extremely important. If detailed accurate results are required, somebody has to be on site during rainstorms to make sure everything is going to plan. It is unreasonably optimistic to expect good data when these depend on somebody getting out of bed in the middle of the night to go clambering up a steep slippery path through the jungle in torrential rain - but at least one large research programme could be quoted where the operation of runoff plots requires this. Not surprisingly, it has not produced any meaningful results.
There are alternative ways of handling the soil and water in the tanks, where the problem is how to take a representative sample of a mixture of water and soil particles of different sizes. The simplest method is to stir the mixture vigorously and scoop out a sample which is filtered, dried and weighed. This method is almost certain to underestimate the soil loss because large particles of soil settle quickly and are hard to keep in suspension while the sample is being taken. An improvement is to add a flocculant, which causes the suspended matter to settle, so the clear supernatant liquid can be drawn off. This leaves a thick sludge which, when stored, allows a more representative sample to be taken (Jackson 1964). Very effective flocculating chemicals are now commercially available.
Another method avoids the time-consuming drying and weighing in the laboratory by weighing a fixed volume of the sludge and comparing this with the weight of an equal volume of water. After making an allowance for the water displaced by the soil, the weight of dry soil in the sludge can be calculated (Barnett and Holladay 1965). This method avoids the need for laboratory work, and also means that samples of various sizes can be weighed. For example in Zimbabwe containers of 1, 5 and 50 litres were used (Elwell 1976). If divisors are used, it is necessary to calculate the weight of soil in each of the tanks separately because the concentration of sediment will be much greater in the first tank where the heavy particles settle out, and least in the final tank.
If the purpose of the experiments can be served by comparing approximate values of soil loss from different treatments, the runoff from the plot can be passed through a filtering fabric which traps some of the soil particles. Pilot trials are needed to establish the best combination of the required size of the filter and its mesh size.
Even simpler, if it is likely that the relationship between runoff and soil loss will be similar for the treatments being compared, it may suffice to measure runoff only, which is much easier than measuring soil loss.
It was suggested earlier that instruments with moving parts are best avoided if possible, but they do have the advantage that when there is a malfunction one can see exactly what is happening, and perhaps put it right or make a correction. This may not be the case with automatic recorders. Two examples illustrate extreme cases of these two approaches.
For a major hydrological study in the Philippines, the river gauging stations were remote and difficult to get to in the rains, so they were equipped with automatic recorders which stored a month's record of rainfall and river level in a potted plug-in solid-state memory which was to be changed each month. It worked satisfactorily during tests, but when, after the first month of the rainy season, the cassettes were brought back and plugged into the computers, most of them were a complete blank. Apparently the 'tropicalization protection' was not able to cope with the changes of temperature, pressure and humidity in the recorder housing, so damp got in and wrecked the storage. The problem was expensively rectified, but the main loss was having to wait another year to get data of how river flow built up at the beginning of the rainy season.
The other example, also a hydrological network, was in Sri Lanka, with very different conditions. The watershed had a good road network, and all the fifteen recorders were visited daily to check the operation of clockwork pen-and-chart, water-level recorders. The labour market at that time was saturated with skilled clerical workers, so analysing and interpreting the huge number of paper chart hydrographs was done quickly and cheaply with nothing more sophisticated than hand-held calculators.
It should be pointed out that there have been great improvements in instrumentation in recent years. In most circumstances nowadays, electrical or mechanical recorders for water level would be replaced by pressure transducers feeding solid-stated memory packs. So the choice of recording equipment should depend on the special circumstances of each project or experiment.
Measuring rate of runoff is fairly straightforward. There is a wide range of standard flumes for all rates of flow - standard flumes meaning those which, when built and installed according to specified conditions, do not need to be individually calibrated as the flow rate can be read directly from tables or charts if the depth of flow is known. Standard flumes and weirs are discussed in Chapter 4, in the section on Velocity/area method. The most commonly used flume for small plots is the H flume, which was designed for this purpose by the USDA.
Quantity of runoff may be calculated from hydrographs. When the rate of flow is plotted against time, the area under the curve is the quantity of flow.
Measuring rate of soil movement in runoff is much more difficult, and is discussed in Chapter 5. Sophisticated devices have been developed but are not suitable for field experiments. They usually depend on the attenuation of a beam of light or gamma radiation by the solids passing between a source and a sensor on either side of a channel.
The principle which should be applied to all records of experiments is that which is the standard practice of professional surveyors. A surveyor goes into the field with a theodolite or level and records all observations in a notebook. The requirement is that the records in the notebook have to be such that they can be completely understood by other surveyors or draughtsmen at any time. They must not depend upon the person who made the record remembering additional facts, they must not be affected by the passage of time, and they must be neat, clean and readable.
The main source of error in records of field experiment is when rather scrappy records are taken in the field and are later copied out tidily in the office. The trouble is that because the field record is untidy and perhaps wet or muddy it is easy for it to be misread or misinterpreted. The answer is to use specially designed record forms or notebooks in a format which is suitable for producing neat and tidy records in the field. Where possible, subsequent calculations should be done on the same form. When it is necessary to transfer data to summary sheets or into another form, the work should be checked and double checked. There should be frequent inspection of the field notebooks and all the subsequent calculations.
