Thoughts on the two disasters, sixteen days apart, and perspective on monitoring.

On 10 August, a Mw 7.4 earthquake struck beneath Chocó, Colombia, at around 110 km depth. On 26 August, part of a glacier on Langtang Lirung failed and sent roughly 100 million cubic metres of ice, rock and water down the Lende Khola and Trishuli valleys in Nepal.Both are already heavily mapped. The more interesting question sits outside the frame of those images. 

1. The headline number said little about where people died

The Nepal collapse registered as a magnitude 5.2 signal, catalogued as a landslide rather than an earthquake. Colombia's event was M7.4, but deep enough that the energy spread wide instead of concentrating near the source. Chocó, at the epicentre, accounted for a small share of the fatalities; Pereira and Cali, hundreds of kilometres away, for most of them. Damage is a function of exposure and building stock, and neither shows up in a seismogram.

CNN published before-and-after satellite images showing the scale of devastation in the city of Pereira after the 7.4-magnitude earthquake struck western Colombia on August 10, published August 13, 2026.

2. The warning system was not missing. It was calibrated for a different flood

Flash flood warning has run on this river system for years. It did not catch this event because the water level gauges were set for monsoon flooding, where the hydrograph rises over hours. What arrived was a wave moving at tens of metres per second. Detection thresholds encode an assumption about which hazard is coming. Trained on the historical record, they become the weak point rather than the safeguard.

3. The hazard was not on any list

Glacial risk in High Mountain Asia is largely inventoried as lakes: moraine-dammed water bodies, ranked by outburst potential, watched for growth. This was not that. A slope holding ice failed and dropped into a valley, and the lake only appeared afterwards, behind the debris. Risk registers are catalogues of known objects. A hazard with no object to catalogue stays invisible to them, however well the monitoring is run.

4. Both events landed where the standard products degrade

Chocó is among the wettest, cloudiest places on Earth, under dense forest: optical is limited, and coherence is low before anything happens. Damage proxy maps read coherence loss, and forest decorrelates on its own. The upper Trishuli is steep enough for radar shadow and layover. The places that most need rapid mapping are often the places where the default methods perform worst.

5. The archive from before decides what can be measured

Open imagery releases after a disaster get attention, and they earn it. But damage mapping is different, and a post-event scene is only useful against a comparable scene from before. That archive is the result of tasking decisions made years earlier, and commercial tasking follows economic value. Remote, forested, mountainous districts are rarely priority targets until the week they become one. The real gap in Earth observation is not who can image a place today. It is how deep the record goes for places nobody had a reason to watch.

6. The flood is over. The sediment is not

Flood extent maps expire within days. Early open-data reconstructions suggest metres of fresh sediment on the Trishuli valley floor, a widened channel, and a lake held back by debris. In Chocó, slopes are loaded with landslide material waiting for the next heavy rain. The most valuable observation window is probably six to twenty-four months out, once the activations close.That last point is where I think a useful transfer sits. Space4Good works at the other end of this timescale, on forest and carbon monitoring, where the product is a multi-year series verified against conditions on the ground. MRV is built on what emergency mapping lacks by design: a stable baseline, a fixed revisit rhythm, and a definition of change agreed before the change happens.

Before-After animation using ESA Sentinel-2 imagery from Copernicus Hub.

A valley buried under fresh sediment is that kind of problem. The same multi-year satellite time series used for forest, biodiversity and carbon monitoring can also support recovery: tracking how landscapes change after a disaster, where sediment moves, how vegetation returns, whether slopes continue to destabilise, and how the landscape responds long after the initial event has disappeared from the news. The technology is already there. What changes is what we choose to keep watching.

The conclusion is simple. Satellites showed what had happened in Nepal and Colombia within days, and they did it well. What neither valley has is anything that keeps watching for the next ten years, while the sediment moves, the slopes fail again and the rebuilt roads are tested. That is a question of attention, planning and funding, not of technology. The instruments are already in orbit.