The original USLE assumed little slope curvature and no deposition. To account for flow convergence in complex terrain, modifications were made to the LS factor with an equation that incorporates flow accumulation. Because they are now used interchangeably, when you see reference to slope-length most likely it is in reference to the LS factor, as calculated below.
To calculate the LS factor for the RUSLE equation, first calculate flow accumulation (facc) and slope in degrees (slp). Then a bit of map algebra in the raster calculator yields LS.
Power(facc * cell resolution / 22.1, 0.4) * Power(Sin(slp * 0.01745) / 0.09, 1.4) * 1.4
I think this post in the SAGA GIS forum might prove useful in answering your question about how slope is calculated:
https://sourceforge.net/p/saga-gis/discussion/354013/thread/27ecdc6b/
Also, based on my understanding of TWI (as a PhD hydrology student involved in hydrologic modeling), the D-Inf (Tarboton), MFD-md (Qin), DEMON (Costa-Cabral), and MFD (Quinn) with the exponent p=1.1 (Freemann) are the best options for determining the accumulation area 'a' in the TWI calculation.
I think Sorensen's work, and Qin's work, lends credence to my own semi-professional opinion. However, Qin's improved algorithm (MFD-md) hasn't been as throughly tested and used as much as the others have.
When I have used SAGA to calculate TWI, I first calculate the slope using the slope commend in the Terrain Analysis / Morphometry / Slope, Aspect, Curvature module, using the default algorithm mentioned in the forum post. Then I calculate catchment area using the Terrain Analysis / Hydrology / Catchment Area / Catchment Area (Parallel) choice, using either the MFD algorithm or the D-INF algorithm with 1.1 as the convergence factor (p=1.1, from Freeman).
Then I run the TWI option, in Terrain Analysis / Hydrology / Topographic Indices / Topographic Wetness Index (TWI), with the options to convert area to "1/cell size" and use standard calculation. I convert to specific catchment area because this is what the original formulation by Beven and Kirkby called for. As for the difference between the "Standard" and "TOPMODEL", I am not sure what it is - looking into that myself right now.
Reference page 106 onward of the linked pdf for more specific help: http://sourceforge.net/projects/saga-gis/files/SAGA%20-%20Documentation/SAGA%
20Documents/SagaManual.pdf/download
I forgot to add, this all assumes that the DEM has been preprocessed by filling sinks. That is another complicated topic, with two (main) separate options.
I hope this helps!
Tom
P.S. Edits for @reima:
This is something I have only recently dug into, and I can admit I don't think I've reached the bottom yet! I prefer the method of Lindsay and Creed, the minimum impact approach that chooses breeching or filling based on minimizing total topographical impact (officially named "Impact Reduction Algorithm" - IRA) which I thought was implemented in his Terrain Analysis Software tool (formerly TAS, now WhiteBox GAT - link: http://www.uoguelph.ca/~hydrogeo/Whitebox/).
However, even his tool seems to implement other filling schemes:
The sink/depression filling algorithm (basic, but insanely fast) by Wang and Liu (2006) - which I don't believe operates in the IRA manner, but similar to the way ArcMap fills sinks/depressions, just straight up without any breeching.
And the sink/depression filling of Planchon and Darboux (2001), which floods a DEM and then removes the water bit by bit - it can enforce a slope on the filed area, which I think might improve TI calculations.
ArcMap has a new "de-pitting" add-on (http://blogs.esri.com/esri/arcgis/2013/03/05/optimized-tool-for-dem-pit-removal-now-available/) that seems similar to Lindsay and Creeds IRA, but I haven't read the cited paper yet to determine how similar. This method might be worth a look.
I'm also interested in scrutinizing my assumption that TI calculations need filled DEMs. I have three different sized watershed DEMs (<100 sq km, 100-1000 sq km, >1000 sq km), clipped using a shape file from 10 m NED data. These are not filled, since the shape file already provided the watershed delineation. I am going to run the SAGA GIS TI calculation (MFD, p=1.1) on all three watersheds, on both filled and unfilled DEMs, using ArcMaps filling scheme (old and new), and the Wang and Liu algorithm (in Whitebox, maybe in SAGA), and the Planchon and Darboux algorithm (in Whitebox, maybe in SAGA). I will also be calculating the TI values using the TI calculation embedded in my hydrological model.
If you want, I can share these results with you. I might not have them for a month or so though, as I have other more pertinent research that my focus is currently on, but I need to refine my TI calculation process by mid May at the latest.
Best Answer
In the paper mentioned in your question, the authors suggest the use the unit contributing area for the slope length. This unit contributing area can be obtained in ArcGIS using "flow accumulation" tool in Spatial analyst toolbox. Flow accumulation, by default, gives you a value in pixels, so you need to multiply by the pixel size before you divide by 22.13 (coefficient of the RUSLE equation). The best method would consist in creating a weighting raster with the m value and use it when you compute the accumulation.