Heat and drought in the framework provinces

A colleague asked whether high temperatures have led to or worsened drought in Burkina Faso, and what the current temperature outlook implies. This page looks at the four framework provinces (Loroum, Oudalan, Séno, Yagha) using the same per-province indicator series behind the ASAP trigger: temperature, rainfall, water balance and cumulative-FPAR biomass, 1989 to the 1st dekad of September 2026, plus the September 2026 temperature forecasts of three North American Multi-Model Ensemble (NMME) models.

1 · Is temperature "always going up" here? #

It depends on the season. Figure 1 shows the framework-province mean temperature for the hot pre-season months and for the growing season, from ASAP's ECMWF-based series.

The pre-season has warmed by about a degree since the early 1990s, three of the four provinces significantly. The growing season has warmed by a quarter of a degree, and the year-to-year swings (a full degree between a wet and a dry season) dwarf it. For the question that matters to the trigger, growing-season heat, the trend is small enough that anomalies can be read against the long-term mean almost directly. The analysis below removes it anyway, so that the "heat" being tested is heat relative to what the year would have been expected to bring.

2 · How heat and drought relate, 2001 to 2025 #

Hot seasons are drought seasons

• Correlations. Detrended July to September temperature correlates −0.48 with seasonal rainfall, −0.61 with SPI-3, −0.44 with the water satisfaction index and −0.42 with detrended biomass, all pooled over the four provinces and highly significant. Seasonal rainfall total, by contrast, correlates only +0.07 with biomass in these provinces: the total hides the timing and dry spells that matter, whereas temperature integrates them (a dry spell is also a cloudless, hot spell).
• Activation years were hot. The five years Trigger 2 would have activated averaged +0.43 °C above trend in July to September, against −0.05 °C for the other years (p < 0.001). 2002, 2011 and 2009 are the first, third and fourth hottest growing seasons of 2001 to 2025 once the trend is removed. The framework's target bad years (2011, 2014, 2017, 2019, 2022) averaged +0.23 °C.
• But hot is not sufficient. 2023 and 2021 were the second and fifth hottest growing seasons and ended with detrended biomass 0.3 to 0.5 z above normal. Rain was near normal and the vegetation did not suffer. Heat without a rainfall deficit did not produce a drought.

How much does heat add, beyond rainfall?

A regression of detrended window biomass on seasonal rainfall and detrended temperature, both standardised, separates the two:

Read the middle columns: a growing season one degree hotter than expected goes with biomass about one standard deviation lower, once seasonal rainfall is held fixed. That is roughly the distance from a normal season to the biomass-warning cut-off. Temperature explains about a fifth of the year-to-year biomass variance in these provinces; seasonal rainfall total, on its own, explains almost none.

Does the water balance already capture this?

ASAP's water satisfaction index is a crop water-balance model driven by CHIRPS rainfall and ECMWF reference evapotranspiration, so temperature enters it through crop water demand. In these provinces, though, its year-to-year variation is almost entirely rainfall: rainfall explains 44 percent of the window water balance and adding temperature raises that to 46 (p 0.06). That is a statement about the water balance, not about biomass. The question that matters is whether temperature still explains biomass once the water balance, rather than raw rainfall, is the other predictor:

It does. The water balance on its own is a weak predictor of biomass here (r = 0.27), and once temperature is added the water-balance coefficient loses significance while temperature keeps its size. Temperature alone predicts biomass better than the water balance alone. By province the effect is strongest in Loroum, present but weaker in Oudalan and Séno, and absent in Yagha. So although temperature is an input to the water balance, it carries biomass information the water balance does not, most likely because a seasonal-total bucket model does not see the dry-spell timing that heat marks.

Two readings are consistent with this. Heat may worsen drought directly, by raising evaporative demand and crop water stress during the grain-filling weeks. Or heat may be the better measurement of a bad season, since the same clear skies that dry the soil also heat the air, so that temperature carries the information about dry-spell timing that the seasonal total loses. The data here cannot separate the two, and for a trigger it does not need to: either way, a hot growing season is a strong sign of a poor biomass season.

Does heat come first?

• Within the season, no. At the ten-day scale (dekads 16 to 30, 1991 to 2025), the temperature anomaly correlates −0.36 with the rainfall anomaly of the same dekad, −0.19 with the previous dekad's rain, and +0.04 with the next dekad's. A hot dekad is a dry dekad, not a warning of one. When a dekad is more than one standard deviation hot, the chance that it is also dry (rain below −0.5 sd) is 56 percent against a base rate of 35 percent.
• Before the season, weakly. April to June temperature, detrended, has no relationship with the coming season's rainfall (r = +0.04) and only a modest one with biomass (r = −0.32, p = 0.001; −0.22 z per sd in the regression). Pre-season and growing-season heat are themselves uncorrelated (r = 0.05). A hot April to June is not a useful early signal of drought in these provinces, though it may leave soils and pastures with less to start from.

3 · 2026: the hottest growing season on record, so far #

The pre-season was unremarkable: April to June 2026 ranked 14th of 38 years, +0.3 °C above the 1991 to 2020 normal, with Séno and Yagha half a degree warm and Oudalan slightly cool. The growing season is another matter. From the 1st dekad of July to the 1st dekad of September, the framework-province mean is 1.4 °C above normal, 3.6 standard deviations, and the hottest in the record by a wide margin: the previous highest, 2002, was +0.7 °C. Every one of the seven dekads is warm, from +0.6 to +2.2 °C, and the anomaly is the same whether computed over the growing-cycle cropland mask or the full cropland mask. Loroum leads at +1.6 °C, Yagha is lowest at +1.2 °C.

Put 2026 on Figure 2 and it sits off the historical relationship. The four provinces are +0.8 to +1.4 °C above their trend lines, where the 2001 to 2025 fit predicts biomass between −0.6 and −1.1 z, at or below the warning cut-off. Instead biomass stands at +0.0 to +1.5 z. Two things may be going on. The late onset concentrated the rain into a shorter, later window, and where the crop and pasture did establish they grew well; the national bulletins describe delayed but not failed vegetation. And the biomass series carries the strong upward trend documented on the companion page, so that a season which would once have registered as poor now registers as normal. Either way, the heat this season has not yet translated into the biomass deficit that Trigger 2 needs, and the ASAP water-balance index shows it: the framework-province mean over dekads 21 to 25 is 87 percent, the second lowest in the 36-year record behind only 1996 and below 2002 and 2009, so the water stress is real even while the vegetation index holds up.

4 · The temperature forecast for the rest of the season #

The intended source was ECMWF's SEAS5, but the Copernicus data store request sat in the queue for more than a day. Three NMME models with a September 2026 issuance are available openly through the IRI Data Library instead, and they agree closely enough that the message does not depend on the choice.

Reading the forecast

• Direction and confidence are not in doubt. Every model puts every month in the warm tercile with far more than a third of its members, and in every case the 2026 ensemble mean ranks first or third among the 21 to 31 hindcast years. For September, which is already one third observed, the forecast is consistent with the +1.3 °C the framework provinces have recorded in the first dekad.
• Magnitude is uncertain and the largest numbers should be discounted. CFSv2's +3.3 °C for September is 5.8 standard deviations of its hindcast; the model has a documented warm drift in real-time forecasts relative to its 1991 to 2010 hindcast, and that older baseline sits below the others by about the warming since. CanSIPS, the largest ensemble with the full 1991 to 2020 baseline, gives the most conservative picture: +1.2 °C in September easing to +0.6 to +0.8 °C in October and November, with 57 percent of members warm in those months. Observed October and November anomalies in the framework provinces have reached +1.3 and +2.4 °C (both in 2023), so a +1 °C October is well within recent experience, not a record.
• Part of the signal is the trend itself. Against a 1991 to 2020 baseline, any recent year starts roughly half a degree warm in these months. The forecasts are warmer than that, so they carry a genuine seasonal signal on top, but the excess over recent years is smaller than the raw anomaly suggests.

What it implies for the season

• For the last two Trigger 2 dekads. On the 2001 to 2025 relationship, a growing season 1 °C hotter than expected goes with biomass about 0.9 z lower. The season has already run 1.4 °C hot without the biomass deficit appearing, so the historical rate cannot be applied mechanically. Still, a hot first half of September with the rains withdrawing early is the combination most likely to pull the cumulative-FPAR anomaly down late; the 2nd dekad of September ASAP update is the one to watch.
• For the end of season. The forecast heat falls mostly after the Trigger 2 window closes. It coincides with ANAM's outlook of long end-of-season dry spells and a late end of season over Liptako and Yaadga, and with crops sown a month late that will be filling grain in late September and October. Higher temperatures raise crop water demand and shorten grain filling; for pasture they hasten drying of the standing biomass that herds depend on through the dry season. That is an end-of-season risk the framework, as designed, will not signal.
• For monitoring. Temperature is cheap to add to the verification checklist: the ASAP province series updates every dekad with the warnings, and the CanSIPS and GEOS forecasts are refreshed monthly on the IRI Data Library without an account.

5 · What this means for the framework #

• Temperature is a good drought indicator here, and not a leading one. It tracks the biomass outcome better than seasonal rainfall does, but it moves with the dry spells rather than ahead of them. As a trigger it would sit in the same August to September window as ASAP, not earlier.
• It could strengthen Trigger 2 rather than replace it. A quick test: ASAP biomass warnings (level 2 or above) in two provinces plus a growing season at least 0.2 °C above trend by the 2nd dekad of August selects 2002, 2006, 2009 and 2011, four of the five level-3 activation years, and excludes the cool level-2 years 2001, 2010 and 2016. It misses 2004, which was only 0.13 °C warm by then. The heat condition does the work the meteorological co-occurrence does in the level-3 rule, without depending on ASAP's water-balance flag. That is a design idea for the 2027 revision, to be backtested properly against impact data, not a change for this season.
• The trend question is manageable. Because growing-season warming is slow, a temperature threshold expressed as an anomaly from a rolling or trend-adjusted baseline stays meaningful for years; the pre-season is where the baseline shifts fast, and the pre-season is not where the drought signal is.
• 2026 is worth watching precisely because it breaks the pattern. Record growing-season heat with normal-or-better vegetation has no precedent in this record. If the late-sown crop meets the long end-of-season dry spells ANAM forecasts, the vegetation index may fall late, after the Trigger 2 window closes.