Would Trigger 2 activate without the greening trend?

Trigger 2 of the Burkina Faso drought framework activates on a JRC ASAP level-3 alert in at least two of Loroum, Oudalan, Séno and Yagha. The biomass side of that alert is ASAP's cumulative-FPAR anomaly, a z-score against the full 2001-to-present record. Biomass in these provinces has risen strongly over that record, so a fixed z-score threshold gets harder to reach every year. This page removes the trend and re-runs the trigger rule for 2001 to 2026 to see what changes.

Bottom line. The trend is real and it is concentrated in the framework provinces: Séno and Loroum have the two steepest biomass trends of Burkina Faso's 45 provinces, and all four are in the top third. But removing it does not unlock any recent activation. In detrended space the years that meet the level-3 rule are 2002, 2009 and 2011, against 2002, 2004, 2009 and 2011 as published. No year after 2011 reaches the threshold either way, and 2026 stays well above it. The trend has shifted recent seasons by roughly 0.3 to 0.8 standard deviations, while the recent seasons sit 1 to 2 standard deviations above the level a warning needs. The gap is larger than the trend.

1 · How strong is the trend, and where #

ASAP publishes, for every province, the dekadal mean of each indicator it thresholds. For the cumulative-FPAR anomaly (zFPARc) on cropland and on rangeland, the seasonal mean (dekads 19 to 30, July to October) was computed for each province and year, and a Theil-Sen slope with a Mann-Kendall test fitted over 2001 to 2025. Figure 1 shows every province.

Figure 1. Trend in seasonal-mean zFPARc per province, 2001 to 2025, in z-score units per decade, sorted by the rangeland trend. Framework provinces in colour. A line joins each province's cropland and rangeland values. Hover for the p-value.

Province	Cropland, z per decade	p	Rangeland, z per decade	p	Rangeland, 2001–05 mean	Rangeland, 2021–25 mean
Séno	+0.56	0.007	+0.68	0.001	−0.24	+1.04
Loroum	+0.41	0.11	+0.58	0.022	−0.25	+1.44
Yagha	+0.30	0.14	+0.41	0.015	+0.35	+0.71
Oudalan	+0.20	0.24	+0.28	0.39	−0.20	+0.94
All 45 provinces, median	−0.12		+0.12

Nationally the picture is mixed: the south and west are browning on cropland (Sissili −0.88 z per decade, Tapoa, Ziro, Balé all below −0.6), while the Sahelian north is greening. Séno ranks first in the country on both land covers, Loroum third and fourth, Yagha eighth and tenth, Oudalan thirteenth. Only five provinces have a significant positive rangeland trend and three of them are framework provinces. Whatever the cause, and ASAP's own documentation warns that FPAR trends mix climate, land use and the MODIS-to-VIIRS sensor splice, the operational effect is the same: a level-3 warning is set against a climatology that these provinces have grown away from.

2 · Detrending and the threshold proxy #

Detrending

For each framework province, land cover and dekad of the year, a Theil-Sen slope was fitted to zFPARc against year over 2001 to 2025 and subtracted, keeping the 2001 to 2025 mean of that dekad unchanged. The same slope is extrapolated to 2026. This is the standard "if there had never been a trend" construction: early years become greener and recent years browner relative to the published values. Figure 2 shows the effect on the Trigger 2 window.

Figure 2. Mean zFPARc over the Trigger 2 window (dekads 21 to 26) for each framework province and land cover, 2001 to 2026, as published by ASAP (grey) and detrended (green). Dashed lines mark the two thresholds used below. 2026 covers dekads 21 to 25 only.

From province means to "would the threshold have been met"

ASAP issues a biomass warning when at least 25 percent of a province's active area has zFPARc below −1. A province mean cannot reproduce that spatial rule exactly, so two mappings are used, as agreed:

Calibrated cut-off. Over every in-season province-dekad from 2001 to 2025 for which both the province mean and the issued warning are known (515 warned dekads in 1,525), the cut-off on the province mean that best reproduces ASAP's biomass warnings is −0.6. It recovers 464 of the 515 warned dekads with 31 false positives (F1 0.92).
ASAP's own −1 applied to the province mean, as a sensitivity. It recovers only 188 of 515 warned dekads (F1 0.54), because a province mean of −1 implies a much larger affected area than 25 percent.

A level-3 alert also needs a meteorological deficit (water balance or SPI-3) in the same dekad. That side is not detrended: the observed meteorological component of ASAP's issued warning is used as-is, so the simulation isolates the biomass question. "Level 3" below therefore means: province-mean zFPARc below the cut-off, cropland or rangeland, in a dekad where ASAP also flagged a meteorological deficit for that land cover. The framework rule is then applied: at least two provinces in the same dekad, dekads 21 to 26.

3 · Results #

Figure 3. For each year, the maximum number of framework provinces meeting the criterion in any one dekad of the Trigger 2 window. First column: ASAP's issued warnings. Next four: the proxy on the published and detrended series, with the calibrated cut-off and with −1. Last two: the biomass condition alone, without the meteorological co-occurrence. An outlined cell (2 or more) would activate Trigger 2. Hover for details.

Method	Years meeting the level-3 rule	Return period	Years meeting the biomass condition alone	Return period
ASAP issued warnings	2002, 2004, 2006, 2009, 2011	5.2 yr	—	—
Proxy, as published, cut-off −0.6	2002, 2004, 2009, 2011	6.5 yr	2001, 2002, 2004, 2006, 2009, 2010, 2011, 2014, 2019	2.9 yr
Proxy, detrended, cut-off −0.6	2002, 2009, 2011	8.7 yr	2002, 2006, 2009, 2010, 2011, 2014, 2016, 2019	3.2 yr
Proxy, as published, −1	2002	26 yr	2002, 2009	13 yr
Proxy, detrended, −1	2002	26 yr	2002	26 yr

Return periods are 26 years of record divided by the number of activating years, 2001 to 2025.

Reading the result

The proxy is faithful enough to trust the comparison. On the published series it reproduces four of ASAP's five activation years. It misses 2006, when Séno and Loroum each crossed the 25-percent-of-area rule with province means that stayed above −0.6. That is the known limitation of using a mean for a spatial rule.
Detrending removes an early activation and adds no late one. With the record-mean baseline, 2004 drops out: Loroum's early-record values are lifted above the cut-off. No year after 2011 comes in. The closest candidates are 2016 (Oudalan), 2019 (Loroum) and 2020 (Yagha), each with a single province meeting the rule. The detrended return period is longer, not shorter.
The biomass condition alone tells the same story. Without requiring a meteorological deficit, detrending adds 2016 and lifts 2019 to three provinces, but 2001 and 2004 fall away, and nothing after 2019 appears. Even biomass-only, no year since 2019 has two provinces below the cut-off in the window.
Why the trend does not flip the answer. Over the framework window the detrending lowers a 2024 or 2025 value by about 0.3 z in Oudalan and 0.7 to 0.8 z in Séno and Loroum rangeland. The published window means for 2021 to 2025 in those provinces run from about +0.5 to +2. Subtracting the trend leaves them near or above zero, still 0.6 to 1.5 z above the calibrated cut-off. The recent seasons have not had biomass deficits of the 2002, 2009 or 2011 kind, with or without the trend.

2026 in detrended space

Figure 4. The 2026 season dekad by dekad for each framework province, as published and detrended, cropland solid and rangeland dashed. The shaded band is the Trigger 2 window; dekad 26, the 2nd dekad of September, remains to be published.

Detrending moves 2026 down by 0.2 to 0.8 z depending on province and land cover. The lowest detrended value in the window so far is Séno cropland at −0.17 in the 1st dekad of September, still 0.4 z above the calibrated cut-off and 0.8 above ASAP's −1. Loroum and Yagha remain above +0.4 detrended. Had the biomass record had no trend, 2026 would still not have met the biomass condition in any province, let alone two, which matches the September comparison with the national bulletins: a rainfall-deficit season with vegetation at or above normal.

4 · What this means for the framework #

The stationarity concern is well founded but, on the biomass side, is not what is holding Trigger 2 back. A fixed z-score threshold over a greening record does get harder to reach, and by 2026 the drift is close to one standard deviation in Séno and Loroum. But the absence of activations since 2011 reflects seasons that were genuinely not biomass droughts on ASAP's terms, not a threshold that has quietly moved out of reach.
A detrended or rolling-baseline biomass indicator would have a longer return period, not a shorter one, under the record-mean baseline agreed here. A present-day baseline (re-centring on the fitted 2025 level) would instead keep recent years where they are and lift early years, which lengthens the return period further. Neither variant produces a recent activation.
If the working group wants Trigger 2 to be reachable in seasons like 2026, the lever is the level, not the trend: the biomass condition alone (ASAP level 2 or above) in two provinces has a return period near 3 years, which is more frequent than the framework's target. A middle option, biomass level 2 in two provinces plus a rainfall deficit somewhere in the AOI, was not tested here and would need its own backtest against the target years.
Worth a conversation with JRC. ASAP's classification document acknowledges the sensor splice and land-use component of FPAR trends. Whether JRC intends to move to a rolling or detrended climatology is a question that affects every framework using ASAP, not only this one.

Caveats. The proxy uses province means for a rule ASAP applies to the fraction of area, so absolute counts are approximate; the comparison between published and detrended is like-for-like and is the point. The slope is fitted per dekad on 25 annual points, so individual dekad slopes are noisy; the window-mean series in Figure 2 average across six of them. The 2026 trend adjustment is an extrapolation. The meteorological side of the level-3 rule is taken from ASAP's issued warnings and is not detrended. That matters: a quick check of the same export shows the rainfall indicators in these provinces also trend wetter, SPI-3 by +0.35 to +0.46 z per decade (p 0.06 to 0.17) and the water-balance index by +0.1 to +0.4 (significant only for Loroum cropland). So the co-occurrence condition has also drifted, and a simulation that detrended both sides could recover more recent level-3 dekads than this one. That is the natural next step; it needs its own threshold calibration for the meteorological proxy. Off-season dekads were excluded using ASAP's own "off season" flag for the framework provinces; the national trend figure uses a fixed July-to-October window for all provinces.

Sources and method. JRC ASAP indicator statistics export, Burkina Faso (country_id 219), GAUL level 2, variables zFPARc cropland and rangeland (variable_id 240, classes 1 and 2, growing-cycle set), dekads 2001-05-21 to 2026-09-01, downloaded 14 September 2026; endpoint as documented in the team's ASAP indicator trends repo (South Sudan analysis, same question). Issued warnings from the ASAP warnings_l2_ts archive dated 3 September 2026, units 3820, 3824, 3825, 3827. Trend: Theil-Sen slope and Mann-Kendall test on seasonal means, 2001 to 2025. Detrending: per province, land cover and dekad of year, Theil-Sen slope removed with the 2001 to 2025 mean preserved, extrapolated to 2026. Threshold calibration: cut-off on the province mean maximising F1 against ASAP's issued biomass warnings (codes 5 to 9), 2001 to 2025. Framework: Cadre de l'action anticipatoire sécheresse au Burkina Faso, 17 avril 2026. Scripts: scripts/detrend_biomass_analysis.py and scripts/detrend_biomass_figs.py in OCHA-DAP/ds-aa-bfa-drought. Page written 14 September 2026 by the OCHA Centre for Humanitarian Data.