Maybe you can start looking at this work:

Title: Estimation of rainfall interception in grassland using eddy flux measurements

Authors: Maruyama, A .; Miyazawa, Y .; Inoue, A.

Affiliation: AA (NARO National Agriculture and Food Research Organization, Ibaraki, Japan [email protected]), AB (Kyushu University, Fukuoka, Japan [email protected]), AC (Prefectural University of Kumamoto, Japan, Kumamoto, Japan [email protected])

Publication: American Geophysical Union, Fall Meeting 2014, abstract # H23S-05

Publication Date: 12/2014

I have not tried this, but had some past experience in looking at forest throughfall and in runoff plot studies, water balance approach (where rainfall, runoff, ET (calculated) and soil moisture change measured), etc.  Your challenge is more difficult than forests, where throughfall can be estimated by directly measuring with substantial point to point variability. If the grass were tall enough, perhaps small throughfall wedge type raingauges could be placed near the ground, but probably difficult to do this.  Another approach might be installing under the grass canopy is home guttering along the slope and have its flow delivered to a storage drum or measured in a calibrated tipping bucket raingauge that would be able to adjust tipping rate based on higher flow amount.   

Another approach I have not tried is to spray a water repellent on the ground surface of some moderate to high sloping land, well contained ground such as very small catchment or confined area(s) (replicated plots) where all the runoff would flow off and could be collected or measured, and preferably this sealing process would eventually degrade so as not to permanently damage the area, In the interim, you capture or otherwise measure the runoff from the experimental plot and compare it with the rain in a raingauge.  I would collect data by storm or daily comparisons and do it long enough to get its essence.  I would be careful at expecting accuracy as there will be variances and differences in storm size, wind, intensity of rain, point to point rainfall data, season, type or species of grass, etc.  If interested in something like this, I have a report on the Hazel Pistol erosion plot study uploaded to researchgate that used plots on sloping land to collect runoff after clearcutting and burning (some nonwettablility of soils from intense burning, but little or no grass present).  I dont know, it may give you some ideas.  

Thanks for your kind responses. I will read the papers you suggest.  I tried spraying the water repellent on the community surface ground. While the plant height of the grassland community is low, and most species are bottom grass, and it is difficult to complete cover the soil surface without any interstice. Thus, during the manupulated rainfall, it is hard to obtain the correct maximum rainfall interception because soil infiltration.