breeding slower than the climate

Problem statement

CIMMYT's Heat and Drought Wheat Improvement Consortium (HeDWIC) breeds wheat varieties for the warming conditions that South Asian and Sub-Saharan African farmers will face in 2035–2050. The breeding pipeline — from initial crosses through multi-environment testing to national variety release — requires 12–15 years. But climate projections indicate that the temperature regime a variety is bred for will have shifted by the time that variety reaches farmers' fields. CIMMYT is breeding for conditions that will no longer exist when the product arrives. Every breeding cycle is chasing a moving target, and the target is accelerating: global mean temperatures are rising, and heat extremes (which cause the most wheat damage) are increasing in frequency and intensity.

Why this matters

Wheat provides 20% of the calories and protein consumed by humans worldwide and is the most widely grown crop on Earth. South Asia's Indo-Gangetic Plain produces over 100 million tons of wheat annually from 30 million hectares across Bangladesh, India, Nepal and Pakistan; globally, each degree Celsius of mean-temperature increase reduces wheat yields by an average of 6.0% (Zhao et al. 2017). CIMMYT-led multi-model simulations project average wheat yield declines of 16% in South Asian countries (and 15% in African countries) by mid-century (Pequeno et al. 2021). But the breeding pipeline bottleneck means that varieties released in 2035 were crossed in 2020–2023 and selected under current conditions — they will arrive already partially obsolete. The mismatch between biological development timelines and climate trajectories is structural, not solvable by working faster within the existing pipeline.

What’s been tried and why it hasn’t worked

CIMMYT has invested heavily in shuttle breeding (testing in multiple heat environments simultaneously to compress cycle time), genomic selection (using DNA markers to predict performance and skip some field testing), and speed breeding (accelerated generation cycling under controlled conditions). Conventional breeding takes more than 10 years from initial cross to cultivar release; speed breeding can achieve 3 to 9 generations per year versus 1 to 2 under conventional cycling (Wanga et al. 2021), but it compresses only the generation-advance stages of the pipeline — significant but insufficient. Genomic selection's accuracy for heat tolerance is limited because heat tolerance is polygenic and involves genotype-by-environment interactions that markers capture poorly. Speed breeding accelerates generation time but not the multi-environment testing that validates real-world performance. The fundamental constraint is that validating a variety's performance under heat stress requires exposing it to heat stress across multiple seasons and locations — and this cannot be compressed below the time it takes to grow multiple crop cycles in multiple environments.

What would unlock progress

Two complementary approaches could help. First, improved crop simulation models that reliably predict variety performance under future climate scenarios could allow breeding programs to select for conditions that don't yet exist — breeding for 2045 temperatures using 2025 data. CIMMYT's own modelers flag the limits of current models: the Pequeno et al. (2021) heat-damage projections considered only changes in daily absolute temperatures, not changes in the frequency of heat events. Second, pre-breeding with wild wheat relatives (Aegilops, Triticum dicoccoides) that evolved under extreme heat could introduce novel heat tolerance mechanisms — but introgression from wild relatives adds years to the pipeline. The speed-versus-diversity tension is unresolved: faster pipelines favor elite × elite crosses with predictable outcomes; climate adaptation may require wild germplasm with unpredictable but wider adaptation.

Entry points for student teams

A computational team could analyze the gap between CIMMYT's crop models and observed heat impacts in recent extreme seasons (the record March–April 2022 heat wave across India and Pakistan) to identify which physiological processes the models fail to capture. A breeding science team could map the decision points in CIMMYT's pipeline where genomic prediction accuracy is lowest and evaluate whether alternative prediction methods (machine learning on phenomic data, envirotyping) could improve accuracy at those points. A systems team could model the timeline mismatch quantitatively: for a variety released in year N, what proportion of its target environment's heat profile will have shifted beyond the validation range by the time it reaches peak adoption?