The current from Lake Erie to Lake Ontario, driven by the river gauges. Tap the water for its speed.
Nobody publishes a current model of the Niagara, so this one is computed. The riverbed comes from 3,289 depth soundings on NOAA's nautical charts, and the water is routed through it the way friction would spread it: deep water carries more and runs faster.
Two checks came out without tuning. The model sends 63% of the river down the Canadian side of Grand Island (measured: about 58%). And with an ordinary roughness for a big river, it loses the same 2.9 m of height between the Buffalo and Niagara Intake gauges that the river really does. The split between the two falls is set, not found: 10% goes to the American side because that is what is measured.
The flow is live. Environment Canada's gauge at Fort Erie gives the whole river, and NOAA's gauge below the falls gives what actually goes over them. The difference is taken out at the power intakes above the falls and put back at Lewiston and Queenston. The lake gauges at Buffalo and Olcott raise and lower the charted depths. More on that at Around the Falls.
The rapids have no soundings, so their depth is worked out from lidar. The water surface drops 17 m through the Whirlpool Rapids, and only about 15 feet of water running about 13 mph loses height that fast. That rests on one assumed roughness number, so treat rapids speeds as good to about a quarter. The deep pools between the rapids use published depths. The control dam above the Horseshoe is an estimate. Its real gates move all day to hold the pool level and their positions are not published, so the map guesses: about a fifth of the gates shut by day and half at night, which pushes more water around the dam's open end. The model cannot make eddies, so the Whirlpool's spin is added by hand: shaped to the basin, counterclockwise at full flow, and reversing when the gorge drops below about 62,000 cfs, as the real one does. Line speed on screen is exaggerated in slow water so you can see it move.