The reason a cell in the curb strip outperforms one behind a house is geometric rather than horticultural. Right-of-way gardens sit on the routes water already takes as it runs downhill [12], so they intercept flow gathered from many parcels instead of the rain that fell on one. The timing makes the point better than the totals do. In the do-nothing run, flooded area more than tripled between 2 p.m. and 2:45 p.m. as the drainage system gave out; in the street-garden run it leveled off around 2 p.m. and receded from there [11].
Set the areas side by side and the gap is wider than the percentages suggest. Private lots took peak ponding from 8.4 hectares to 6.2 [7][8], removing 2.2 hectares. The right-of-way run took it to 2.0 [9], removing 6.4, roughly 2.9 times as much on less land [2]. Adding private lots on top of street gardens bought a further 0.8 hectares, taking the reduction from 76% to 86% and the deepest water from about 1.5 meters to just over 1 [10][3]. The denominator is modest: 8.4 hectares at 12% puts Crystal Beach at about 70 hectares [1].
The thing this does not tell you is what happened in basements. The storm's signature damage in Ottawa was close to 5,800 basement flooding reports, concentrated in the west end [2], and the model's flood metric is surface depth, water reaching 15 centimeters above ground, chosen as roughly doorstep height [6], with velocity above 0.6 meters per second tracked separately as a hazard [13]. The nearest proxy for sewer surcharge is the pipe result: with gardens in both locations, total runoff volume fell 80% and peak flow through the network was nearly halved [14]. That is consistent with fewer flooded basements. It is not a measurement of them.
The mechanism the authors are arguing for is upstream of all this. Pipes move water faster once it has pooled but do not change how much arrives; a garden intercepts rainfall before the sewer sees it [15]. Their own caveat is the one to hold onto: these are modeled results, and models are kinder to stormwater controls than field monitoring is [18]. Treat 76% as the ceiling of a design study.
What travels here is less the number than the setup. The model was three years in the making before July 1 [3], which means the storm was not selected to suit it, and the 167 millimeters that fell in five hours [1] averages about 33 millimeters an hour [4], well past what standard drainage is sized for [20]. Within one neighborhood and one rainfall record, placement beat coverage; whether that ordering survives different topography is a question the work does not claim to settle [17].
