Excellent response above.  I have several items on my researchgate that may give you some ideas. Collecting sediment is less of a problem usually than collecting water, due to volumes.  For collecting sediment, see studies on sediment from a small ephemeral gully using filter fences and there is a brief report on Long Nose Wildfire use of filter fences to help estimate if sediment was being delivered to small channels.  For information on small catchments, my Missouri thesis 1975 may be help on methods used, but obviously collecting water is not practical at this scale unless you want to develop a storage pond.  My study for the Hazel Pistol erosion plots in SW Oregon in researchgate sounds much like what Dr. Macedo suggested in his plot study, which I used after clearcutting and high intensity site prep burning decades back.  

If you want to work with catchment size unit of several hectares, installing a flume and using Coshocton sediment and water sampler rotates with flow velocity for sampling and storage into adjacent area or container.  I found a document with a picture of one http://ecoursesonline.iasri.res.in/mod/page/view.php?id=1914

There are many studies on the effects of forest removal on water yield, and somewhat less on sediment but still available.  Alot depends on the situation, details associated with felling and logging practices, BMPs used, soils, slopes, etc.  Be sure you go into adequate detail on these elements.  If you are comparing areas or using controls, make sure the comparibility or differences are brought out in sufficient detail to show consistencies and differences.  

The best size and type of study will depend on your available time, funding, duration of study and conditions.  I have to say though, that surface erosion from forest harvests can be low and difficult to measure if BMPs are employed.  I went to installing the filter dams on catchment scales, because when installed on slopes, typically the amount of sediment was small, difficult to measure.  The effect of the hot burning of unyarded fuels in Oregon essentially fried the soils, leaving them with signs of non-wettability, soil pedastaling, exposure of fine surface roots, slight rilling or flow passage in surface locally from high rainfall of Oregon coast range and erosive soils.  If you go to clearcuts and logging operations, and the sites retain ample organics on soil surface (duff, humus) and stream buffers, and do not see in close inspection any signs of erosion or concentrated flow, transport of sediment may be small and difficult to measure, so probably best to avoid trying to measure on site erosion and trend toward measuring small catchments where the opportunity of concentrated flow and sediment delivery would be easier to measure if it exists.  If trying to use low or unbiased sampling, you may spend much time and effort to not get much from randomizing, even though better statistically and certainly acceptable to report these findings.  Stratifying conditions by soils, slopes, etc. to  specific conditions more likely to have problems or where you see indicators or do not know if there are problems, these are more likely focus areas to invest more time and effort, but then qualify exactly what was done to select or stratify sampling to help identify more of a worse case condition than a wide variety of sites to help determine the average, range, variability, etc.

Thanks to all for answers

But a part of the work is on the road embankments, and in many areas road embankments have a little height.  So I have to choose the small size of the plots.

I'd scale the plot size to the type of process you are measuring.  Road embankments (we call them fillslopes here) can be subject to a variety of erosion processes ranging from sheetwash to debris sliding.  If you have concentrated flow from the road surface to the embankment, you'll have different erosion processes operating than if the embankment is isolated from road runoff.  Walter Megahan did a fair amount of work measuring erosion on granitic fillslopes in the Idaho Batholith. 

I have recently become a fan of using sediment fences with v-notch weirs and pressure tapes (see attached picture).   This method can be applied at multiple scales. 

The initial question posted by Pejman Dalir was about the size of plot to use for collecting runoff and sediment. Some of the responses submitted previously have included additional info on methods/equipment that can be used for assessing sediment transport and discharge. Contributing to this effort, I submit for consideration, an alternative to using a Coshocton wheel, as suggested by William Hansen. The device I am referring to is reported in:

Bathke, G.R.  1987.  A semiportable multislot divisor for erosion and runoff measurements. Journal of the Soil Science Society of America, Volume 51: 485-487.

The key features of this device are:

In theory, bedload sediment is collected in the sediment collection box. Total discharged water volume can be determined by measuring the volume of water collected in the aliquot tank (multiply by factor of 21). Suspended load can be sampled by mixing and sub-sampling the water collected in the aliquot tank. Refer to Bathke's article for more info.

This paper will be useful for u 

al the best 

https://pubs.usgs.gov/sir/2009/5049/pdf/Elliot.pdf

The sizing is apt to depend on your conditions, expectation and length of study.  In circumstance of natural or infrequently managed forest with well developed soils, the size of the plots would need to be greater.  Small catchment size may be needed to capture ephemeral flow response, but in using a catchment, if there is erosion and sediment captured, it may be from surface erosion or from channel erosion.  Does it make a difference?  But if looking at heavily disturbed areas, the plot size can be less.  If your rainfall runoff (R factor of RUSLE) and soil K erodibility factor are high.  If these factors are low, plot size must be larger to have any results.  You want some results, because if you get no runoff or sediment, then you need to increase plot size.  Dr Coe's approach looks similar to the fabric dams that I have used, but lined in plastic to not only capture sediment but capture or gauge flow (with transducer).  If you are capturing flow in small dam, the structure must be stout to avoid failure.  There is probably no perfect size, but I would suggest that if you take a close site look and if you can see some indications of erosion, sediment and/or runoff (ie, pedistaling of soils, exposure of roots, realignment of organics, movement of organic material or scouring), then you are in the approximate size.  If you do not want to capture the effects of channel erosion and ephemeral flow, then you need to avoid catchments and measure effects within slopes. The difficulty in measuring erosion of slopes is the plot must be well confined on top and sides, while in catchment, the catchment surface boundaries can be surveyed and used as the boundaries of the plot.  If you do not need to capture flow, the filter fence approach (~researchgate 2007 paper) can be easy to install to confine plot on all sides as needed, or could be applied to small catchments.  If you are unsure of what size of catchment, the filter fence approach can be installed at several locations down a channel system if necessary.  You may not even need to measure the catchment size until you get some results, and your period of measurement and time of study for a thesis suggests you may catch storms for a year or two, so the likelihood of capturing effects of extreme event is low.  The difficulty we found in the small ephemeral study becomes how to measure the 5 tons of sediment from one storm in 1994 or the 50 tons of sediment captured over a decade (1994-2004).  How sediment is measured may depend on how much is captured, but if you try the non-catchment approach of measuring slope erosion for your thesis and get nothing, if you had installed properly a few of the catchment filter fabric fences, your chances of capturing something to report is increased.  Just make sure you also get some rainfall data also, and a tipping bucket raingauge properly installed in vicinity is a big help to a study as this, where you are unlikely to get extreme event and want to quantify the conditions measured.  

If you go to forest and can absolutely see no signs of recent or past erosion, then go downslope or down topographic depression (pre-channel landform to channel) and seek indicators.   

i agree with Hansen