Ordinary Rainfall Triggered Taiwan's Record Matai'an Landslide, Not the Later Dam-Breach Flood

The rain that collapsed the slope came to 573 mm over three days, which the Landslide Blog at Eos, summarising the paper, calls not a particularly exceptional level for Taiwan ^[7]. The precursors were in the ground: He et al. date instability on that slope to before 1985 ^[4], name the M=7.4 Hualien earthquake of 3 February 2024 as a possible further destabilisation ^[5], and record renewed movement with a large tension crack during Typhoon Kong-rey in November 2024 ^[6].

The earthquake link is hedged in the source and should stay hedged: the wording is that the slope "may have been further destabilised" ^[5]. About 17 months separate that earthquake from the failure ^[7]. The post's account of the pre-1985 movement comes from the paper's reconstruction and does not say the instability had been logged in a hazard inventory before July 2025 ^[19]. He et al. published in Landslides under a Creative Commons licence ^[18].

Wipha's failure on 19 July put 285.90 million cubic metres into the valley and left a barrier 200 m high ^[8][3]. Summer rain, including Typhoon Podul in August, filled the impoundment to 67 million cubic metres ^[9]. Sixty-six days passed between the slide and the breach ^[1].

Then the dam went in two pulses. The first peak came at 15:38 and the second at 16:24, 46 minutes later ^[11][3], the second about 41 per cent larger than the first ^[2]. Both peak figures in the post are cubic metres and carry no time interval ^[11]; read as cubic metres per second, the larger one would empty a 67 million cubic metre lake in roughly two hours ^[5]. The flood damaged or destroyed 1,606 buildings ^[12].

Records cited in the post put the relocation at 8,524 people ^[13]. Add the 24 dead and missing and the exposed population is 8,548, so the fatalities are 0.28 per cent of it, about one in 356 ^[4]. Those who died probably ignored the evacuation order, according to the post, which concludes that "the loss of life does not indicate a failure by the authorities" ^[15][14].

So the detection and the warning worked here, and what failed was compliance. The post's argument is that surveillance over a much larger area, for an event with no clearly defined trigger, with flooding reaching much further and in a country with a less developed emergency management system, is "infinitely harder" ^[16]. Its closing line is that "the focus should be on vulnerability and resilience rather than on magic bullet warning systems" ^[17].