The Lag

从被记录,到被行动

Flood hazard in the Msimbazi basin has been on the official record for decades. Prevention eventually arrived, at very large scale, and very late. This paper measures how much exposure accumulated in between — and asks whether that interval is something an instrument can see.

Authors: Guqing Yu (presenting) · Zhijun He
Full title: The Lag: Documented Flood Hazard and Continued Exposure Growth in Dar es Salaam, 1995–2025
Venue: AAG Annual Meeting 2027, virtual session · 8–12 February
Status: On the 2027 programme — abstract 53417
Companion: Attributive Ambiguity — the theoretical argument this paper is the empirical floor beneath
Funding: None. No grant or institutional support. The presenting author's conference registration was paid privately by the co-author.

01The question underneath

Why a paper about a time interval

Prevention capacity succeeds by producing nothing. A flood that does not arrive leaves no trace, no ribbon-cutting, no disbursement to credit. So the indices that decide where adaptation money goes measure prevention by whether programmes exist rather than by whether they work, and in aggregation read response capacity as adaptive capacity.

Correcting that requires answering a prior question, and it is an empirical one: does prevention, or its absence, leave any observable trace at all? If it does not, no diagnostic built on this distinction can exist, however well argued.

This paper proposes one candidate trace — the interval between the year a hazard entered the official record and the year anything was built against it — and tests whether it can be measured.

02Why Dar es Salaam, and not a cleaner case

The case was chosen for its documents, not for its outcome

The Msimbazi basin is home to roughly 1.6 million people and the road links between the central business district and the rest of the city. Its lower reaches flood severely at least twice in most rainy seasons.

What makes it usable is that its hazard has a dated paper trail. A catchment-level flood risk assessment in 2016. City-wide community hazard mapping from 2017, produced openly and locally. A participatory basin plan in 2018, assembled through a nine-month process across dozens of institutions. And from 2022, a basin development project carrying quantified baselines and relocation targets against fixed dates.

It would be more convenient for our argument to pick a place where prevention never came. Here it did come, at a scale of hundreds of millions of dollars, after decades of documentation.

So this is not a case of prevention withheld. It is a case of prevention arriving very late, in a city where nobody was ever able to call the flooding an act of nature. That makes it the harder test, and the more honest one: the question stops being whether prevention happened and becomes how long the interval was, and what filled it.

03The abstract, as submitted

Verbatim

Prevention capacity succeeds by producing nothing, so composite indices of disaster capacity measure it by whether programmes exist rather than by whether they work. Correcting that requires a prior answer: does prevention, or its absence, leave an observable trace?

This paper proposes the lag between documentation and action as that trace, and tests it in the Msimbazi basin of Dar es Salaam. Flood hazard here has been formally recorded for decades, with successive World Bank and municipal instruments dating the moment each area entered the official record: catchment risk assessment in 2016, the participatory Msimbazi Opportunity Plan in 2018, city-wide community hazard mapping from 2017, and a basin development project from 2022 carrying quantified baselines and relocation targets.

Against this dated documentary series we set the exposure trajectory, reconstructing built-up growth inside the mapped hazard corridor from annual Landsat and Sentinel composites. The question is not whether prevention happened; it did, at scale, and late. The question is how much exposure accumulated between the year a corridor was documented and the year anything was built to protect it.

We state the falsification condition in advance. If exposure growth inside documented corridors is indistinguishable from growth outside them, documentation leaves no measurable shadow and the proposed trace fails. Existing land-cover work for this city establishes that the measurement is tractable; what is new here is the pairing with a dated record of official knowledge.

04What is not ours

Stated before someone points it out

Measuring built-up expansion into the Dar es Salaam floodplain is not new. Land-cover change in this city has been mapped repeatedly with Landsat, channel change and settlement dynamics in the Msimbazi have been assessed in the flood risk literature, and the hazard mapping we rely on was produced by a community mapping project, not by us.

We say this plainly because the contribution sits elsewhere, and would be easy to misrepresent. The imagery half is deliberately conventional; we reuse established methods because they work.

What has not been done is the other half: establishing, document by document, the year each corridor entered the official record, at what spatial resolution and with what authority behind it — and then aligning that series against the exposure curve.

The documentary timeline is the new object. The satellite record is the instrument that reads it.

05How this fails

Falsification conditions, published before results

No shadow. If exposure growth inside documented hazard corridors is indistinguishable from growth outside them, documentation leaves no measurable trace and the proposed proxy fails.
Confounding we cannot strip. If growth inside corridors is fully explained by land price, tenure informality and distance to work, then what we are measuring is the housing market, not the shadow of a document.
Dating that will not hold. If the year a corridor entered the official record cannot be fixed with confidence — because documents are undated, unpublished, or contradict each other — the independent variable does not exist.
Cloud. Coastal tropical optical imagery is heavily obscured. If annual composites cannot yield a stable classification, the exposure series is too noisy to carry the test.

Any one of these would end the paper. The first would also constrain the companion argument, which is the point of running the test at all.

06Where it sits in the programme

Specialty groups, and the dates that matter

Hazards, Risks and Disasters
Africa
Remote Sensing
Urban Geography
Development Geographies

15 October 2026 — abstract submission closes.
5 November 2026 — session organising closes; 6–12 November, open add period.
After 12 November — paper abstracts not assigned to a session convert to posters.
17 December 2026 — last date abstracts can be edited.

This paper is in the virtual programme and will be placed in a virtual session, which is a separate pool from the in-person one. It is not in a session yet.

If you are organising a virtual session on disaster risk, adaptation finance, or urban exposure, we would rather join yours than run our own. Abstract code 53417. Write to the presenting author.

Think the lag is unmeasurable?

Say so now rather than in February. Zhijun He — zhijun.he@yale.edu

The companion paper: Attributive Ambiguity · How we see the problem: Premises · Who is working on this: People

播种光明，收获永恒