After Cyclone Pam, drought finished what the storm had started in Anauda Johnson’s household garden.
“There’s nothing left in the garden.”
Vanuatu opens the story; the stress test that follows uses six Pacific countries with longer crop records. See Methods.
These dry years are real history, not projections — but they now land in a warmer climate. Hotter air raises evaporative demand, so the same dry spell can stress crops more. Resilience matters even where rainfall trends are uncertain.
Surface-temperature anomalies for the six countries in this stress test. Pacific Data Hub, SPC:DF_CLIMATE_CHANGE(1.0).
Split one hectare between cassava, taro and sweet potato. Your mix carries through every test below.
This mix is now your garden for every test below.
Shares normalize automatically. Hypothetical hectare × real national yield histories — not an observed household garden.
Tonga’s driest year in this rainfall record. See each crop vs trend — and what your mix would have harvested.
Ten driest years in each of six countries — sixty events, none hand-picked. The big number is your worst harvest across them.
One of those sixty tests was much worse than the rest.
Fiji · 1992
In the worst historical stress test for the equal-third hectare, expected harvest was about 9.16 tonnes.
The historical crop outcomes imply about 5.79 tonnes.
3.37 tonnes short
≈ 3,365 kg per hectare · hypothetical equal-third hectare
Implied harvest versus a fitted benchmark — not an observed household garden, and not a claim that drought caused this exact tonne loss.
That is the physical shock.
But what was it worth?
Not honestly.
Complete producer prices for all three crops are not available for that stress year.
So we keep the worst physical year in tonnes — and move to the years we can price consistently.
We price only the years where all three producer prices are observed.
Fiji · 2004
Each strategy faces the same historical year. Only the crop allocation changes.
If you knew the winner beforehand, one crop would be enough.
But you don't.
Among the 10 dry stress events for which all three producer prices were available, this was the largest percentage downside for the equal-third diversified hectare.
One year tells us what the loss can look like.
What happens across every stress event we can price?
Ten dry years. Same question. Did diversification make losing less costly?
It reduced the average downside.
Across the 10 price-covered dry stress events, average farm-gate-value downside was 6.6% under an equal-probability single-crop benchmark and 3.1% for the equal-third diversified hectare.
Price coverage is limited to 10 events in Fiji and New Caledonia. Treat this economic result as exploratory.
Lower average downside is a risk principle — not a universal planting recipe.
The historical data support diversification as a risk-management principle. They do not prescribe one Pacific crop mix.
Different crops move together differently across countries, and national harvested-area patterns also differ. Principle ≠ prescription.
Different crops do not always struggle together.
Hover or tap the image to unveil the number
Yield is only one dimension of resilience. In Pacific communities, crop choices can also carry cultural, ceremonial, exchange, labour, taste and place-based significance that tonnes alone cannot capture. In places including Samoa and Fiji, taro also carries important ceremonial and exchange value.
Crop as more than yield — dignity in the harvest.
“There’s nothing left in the garden.”
At the beginning of this story, those words described the stakes.
Our analysis cannot tell us what a different planting strategy would have done for Anauda Johnson’s household.
But his story shows why the difference between a total loss and a partial one can matter.
It is to give a garden more ways to have something left.
Diversification cannot make every crop survive.
It reduces how much the whole hectare depends on one crop being the winner.
56 / 60
In 56 of the 60 selected dry country-years, at least one of the three national crop-yield series remained at or above its fitted trend.
The crops usually did not all move below trend together.
6.6% → 3.1%
53.3% lower average downside
Across the 10 price-covered stress events, average farm-gate-value downside was 6.6% under the equal-probability single-crop benchmark and 3.1% for the equal-third diversified hectare.
10 events · Fiji + New Caledonia · exploratory
Some harvest can remain.
When crop outcomes differ, a poor result in one crop need not translate into the same proportional loss across the entire diversified hectare.
Some harvest value can remain.
The price-covered stress test shows how retaining crop output can also retain part of the hectare’s gross farm-gate value.
The whole bet does not rest on one crop.
Spreading land across crops can reduce exposure to discovering, after the shock, that the crop you relied on was the one that struggled.
For a household that depends on a food garden, retaining part of a harvest can matter for food availability, exchange, sale and livelihood stability.
Those household outcomes are not measured by this stress-test model; they are the wider pathways that make agricultural resilience matter.
ABC News documents what total garden loss meant in Anauda Johnson’s 2015 story on Tanna Island. That account establishes the human stakes of crop loss. Vanuatu is not one of the six countries in the 60-event stress test, and this model does not estimate his household counterfactual.
Pacific food-system and adaptation sources already listed in Methods — including Nunn et al. (2024), Kaoh, Lebot, McGregor, Bourke et al. (2016), and the PLOS Pacific smallholder food-security scoping review (2022) — provide contextual evidence that crop diversity and agricultural resilience can matter for food availability and livelihoods. They do not prove this model’s quantitative results.
He began this story with nothing left.
We cannot know whether diversification would have changed his year.
The lesson is not that diversity prevents loss.
It is that resilience can mean having something left to fall back on.
Planting options forward.
There may not even be a climate-proof garden.
Diversity cannot guarantee the harvest. It gives a garden more than one way through a bad year.
Fund the test before funding the prescription.
Examples of relevant climate-resilience infrastructure and investment — not validation of the exact cassava+taro+sweet-potato portfolio analysed here:
Optional technical depth. These tools sit outside the main story — they do not prescribe a Pacific planting mix.
Correlation describes whether the detrended national yield series tend to move together. It is not a causal drought-response coefficient.
Capping anomalies at ±50% checks whether a few extreme crop-year observations drive the lower-tail conclusion. The broad pattern does not disappear under the cap — that is a robustness check, not proof.
Worst-year harvest % — shorter bars (less severe) are better. Click a row to apply that mix.
In-sample historical optimisation. Not a planting recommendation. Never “the optimal Pacific garden.”
Gold = the mix with the least-severe worst year in this historical sample (10% cassava / 40% taro / 50% sweet potato). Teal = near-equal thirds. Circle size grows with how many of the 60 dry years were severe losses. This is in-sample history, not a recommended planting mix.
Across 60 unusually dry country-years, no root stays on top. Which crop you would have wanted depends on the place and the year — information you only learn after the shock.
Only 4 / 60 had all three below fitted trend together. This is a national crop-series result — not a claim that every household garden followed the same pattern.
The same 60 dry stress events applied to each strategy. Look at how far the worst harvest losses stretch. Diversification isn’t trying to pick the winner — it’s trying to make the loser hurt less.
Event-level bootstrap support for a small price-covered sample — not causal or Pacific-wide inference.