Food-security impact

The primary question: how much of the 2026 cereal recovery did the floods and fires reverse, and where? Damaged cropland hectares become an estimated production loss, then an indicative food-security pressure, referenced against the 2025 drought baseline.

The answer

~25,925 t of cereal production lost across 12 governorates (3 flood AOIs + the national fire set), a lower bound. Range 3,471–39,946 t from cropland-definition and temporal-aggregation sensitivity.
That is 2.16% of the combined 2025 baseline (1,198,549 t), about 2.16% of the national ~1.2 Mt floor.
Indicative food-security pressure (study total): moderate incremental stress, not an IPC phase (see method note).
The damage is concentrated: the two Euphrates flood AOIs (Raqqa, Deir ez-Zor) each lose ~7–8% of their own baseline; the other nine governorates lose under 0.4% each.

Important

First-half-2026 snapshot — figures are lower bounds. Windows cover floods Mar–Jun and fires May 1 – Jun 12. The harvest and fire season peaks in summer (Jun–Aug), so the heaviest crop-fire months are unobserved. Re-run after the season closes for the concluded result; field and expert verification is preferred over remote-sensing self-consistency (DEC-039).

Where the loss falls

Choropleth map of Syria shaded by 2026 cropland-production loss as a percent of the 2025 baseline. Raqqa and Deir ez-Zor are darkest at about 7.6 to 7.8 percent, Hasakah mid-toned at 2.8 percent, all others pale under 1.5 percent, Latakia and Damascus City grey. — Figure 1: Cereal-production loss as a share of each governorate’s 2025 baseline. The Euphrates corridor carries the damage; Latakia and Damascus City are verified-excluded (grey). First-half-2026 lower bound.

The whiplash

The study’s spine is a three-state sequence, drought → tentative recovery → shock, and the question is how much of the recovery the 2026 shock took back.

State	Cereal production	Note
2025 — record drought (baseline)	~1.20 Mt	~60%+ below average; worst in ~60 years (FAO/GIEWS)
2026 — tentative recovery (projected)	~2.30 Mt	more even rainfall; sowing targets not fully met
2026 — observed first-half shock	see below	floods + fires, validated damage only

The recovery figure is an external projection (context). The shock is measured here from human-validated damage records only. The ~25.9 kt loss is about 2.4% of the ~1.1 Mt recovery increment the 2026 harvest was projected to add back: a real dent in the rebound, not a reversal of it.

Production loss by governorate

Code

import matplotlib.pyplot as plt
import seaborn as sns
from viz import apply_theme, caveat_footer, CAVEATS, PHENOMENON_COLORS, bar_value_labels

pal = {k.value: v for k, v in PHENOMENON_COLORS.items()}
apply_theme()
order = (impact.groupby("aoi_id")["production_loss_t_headline"].sum()
               .sort_values(ascending=False).index.tolist())
fig, ax = plt.subplots(figsize=(9, 5))
sns.barplot(
    impact, x="aoi_id", y="production_loss_t_headline", hue="phenomenon",
    hue_order=["flood", "fire"], order=order, palette=pal, ax=ax,
)
bar_value_labels(ax, fmt="{:,.0f}", threshold=200)
ax.set_xlabel("Governorate")
ax.set_ylabel("Estimated production loss — headline (t)")
ax.set_title("2026 cereal-production loss by governorate and phenomenon")
ax.tick_params(axis="x", rotation=35)
for lbl in ax.get_xticklabels():
    lbl.set_ha("right")
ax.legend(title="", loc="upper right", frameon=True)
caveat_footer(fig, CAVEATS["case_study_2026h1"])
plt.show()

Grouped bar chart of estimated 2026 cereal-production loss in tonnes by governorate, split into flood (blue) and fire (red). Raqqa, Hasakah and Deir ez-Zor have by far the largest flood losses (roughly 10,000, 8,600 and 5,000 tonnes); all fire bars are under about 920 tonnes. — Figure 2: Estimated 2026 cereal-production loss (headline t) by governorate and phenomenon. Floods on the Euphrates dominate; the national fire set adds a broad but individually small contribution. First-half-2026 lower bound.

↓ Data behind this figure (impact_by_aoi.csv)

Per-governorate detail

Loss as a share of each governorate’s own 2025 baseline shows where the shock bites hardest in relative terms. The three Euphrates/Khabur AOIs carry essentially all of it; the nine fire-only governorates are individually marginal, so they are summarised into one row below (the full 12-row table is one click away).

Table 1

governorate	baseline 2025 (t)	loss headline (t)	loss low (t)	loss season-ref (t)	loss % of baseline	indicative pressure
raqqa	131,184	10,252.8	405.5	13,234.2	7.82	significant incremental stress
hasakah	345,841	9,561.8	2,218.0	17,847.0	2.77	moderate incremental stress
deir_ez_zor	65,577	4,996.6	291.4	7,750.3	7.62	significant incremental stress
tartus	3,386	51.4	37.5	51.4	1.52	moderate incremental stress
8 other governorates (each <0.4%)	652,561	1,062.8	518.2	1,062.8	0.16	marginal (combined)

↓ Full 12-governorate table (impact_national.csv)

Fire damage by burn severity

The headline counts burned-cropland hectares; this breaks them out by dNBR burn severity (Key & Benson / USGS bins), the depth-of-burn signal behind the area. Most burned cropland sits in the low and moderate-low classes — consistent with stubble and field-edge fires rather than total crop loss — with high-severity hectares a small tail. Severity is shown for the union-cropland headline; the Methods page explains the dNBR ∩ VIIRS chain.

Stacked horizontal bar chart of 2026 burned cropland in hectares by governorate, each bar split by burn severity from low (pale orange) to high (dark red). Hasakah has by far the longest bar at roughly 4,100 hectares, then Raqqa near 1,200, with most governorates under a few hundred hectares; low and moderate-low severity make up the bulk of every bar. — Figure 3: Burned cropland (union-headline ha) by governorate and dNBR burn severity, 2026 study window (May 1 – Jun 12). Hasakah dominates the national fire signal; most hectares are low to moderate-low severity. First-half-2026 lower bound.

↓ Data behind this figure (fire_damage_sensitivity.csv)

Robustness — the headline without the RQ1-flagged hectares

RQ1 flags Hasakah’s June flood hectares as a likely radar harvest artifact and excludes them from attribution. To keep the food-security number and the RQ1 attribution on the same record set, here is the study total recomputed with those flagged dates removed (Hasakah’s flood headline falls back to its attributed March peak).

	Production loss	% of 2025 floor
Headline (all validated records)	~25,925 t	2.16%
Excluding RQ1-flagged Hasakah-June hectares	~24,534 t	2.05%
Difference	−1,391 t	−5.4%

The headline is robust to the flag: removing the flagged hectares moves it only about 5%. That is because the layer already counts each AOI’s peak single flood date, not the sum of dates (DEC-033), so the flagged June dates never compounded, and the total is dominated by the well-supported Euphrates records. We publish both figures rather than choosing one.

Method, and what the numbers are not

Method & provenance (expand)

Validated-only gate. Only human-validated records are read; a single non-validated record makes the layer refuse. No agent set any validation status: the flood and fire gates were closed by a human (DEC-007).
One cropland convention. Union cropland is the headline, intersection the conservative low bound (DEC-015). Flood records (per event-date) are collapsed to the peak single event-date so persistent pixels are not double-counted across dates; the season-reference column is the upper bracket (DEC-033).
Loss = hectares × the 2025 drought-floor yield (~0.223 t/ha, DEC-019). Applying the drought-year yield to 2026 damaged area makes the loss a conservative lower bound: 2026 was a recovery year, so expected yields on the lost hectares were higher.
Damage hectares inherit the pipeline caveats. Flood extent is open-water riverine inundation (flooded vegetation and pluvial upland under-detected, DEC-023); burned cropland is VIIRS-confirmed and a conservative lower bound (DEC-031).

Warning

Not an IPC classification. The indicative pressure is a qualitative band from loss-as-%-of-baseline, not an IPC phase (which needs the full protocol: consumption, livelihoods, nutrition, mortality). GIEWS, FEWS NET and IPC are the authoritative sources; this layer is a production-shock signal that feeds such an assessment. Syria’s 2025 baseline was already a record-drought, Crisis-level context, so the 2026 shock is incremental pressure on top of that.

Syria 2026 Agricultural Shocks · first-half-2026 snapshot (floods Mar–Jun; fires May 1 – Jun 12 2026) · figures are lower bounds (DEC-039). Author: Muhanad Abulhusn. Rendered from human-validated (Tier-2) records only.

Data sources: FAO/GIEWS (baseline & production), GloFAS (discharge), CHIRPS (rainfall), VIIRS/FIRMS (active fire), Sentinel-1/2 & Copernicus EMS (flood/burn extent), ACLED (conflict, cited not republished). License: CC BY 4.0.

Suggested citation:

Abulhusn, M. (2026). Syria 2026 Agricultural Shocks: floods and crop fires against the 2025 record-drought baseline (first-half-2026 snapshot). https://muhanad-husn.github.io/flood-fire/

Provenance & contracts: PRODUCT.md · STRUCTURE.md · decision log · source repo

--- title: "Food-security impact" --- ```{python} #| label: setup #| echo: false import sys, pathlib import pandas as pd REPO_ROOT = None for _cand in (pathlib.Path.cwd(), *pathlib.Path.cwd().parents): if (_cand / "CLAUDE.md").exists() and (_cand / "viz").is_dir(): REPO_ROOT = _cand sys.path.insert(0, str(_cand)) break assert REPO_ROOT is not None, "could not locate repo root (CLAUDE.md + viz/)" OUT = REPO_ROOT / "outputs" nat = pd.read_csv(OUT / "food_security" / "impact_national.csv") impact = pd.read_csv(OUT / "food_security" / "impact_by_aoi.csv") tot = nat.loc[nat["aoi_id"] == "STUDY_TOTAL"].iloc[0] n_govs = (nat["aoi_id"] != "STUDY_TOTAL").sum() ``` The primary question: **how much of the 2026 cereal recovery did the floods and fires reverse, and where?** Damaged cropland hectares become an estimated production loss, then an indicative food-security pressure, referenced against the 2025 drought baseline. ::: {.key-findings} ### The answer ```{python} #| label: headline #| echo: false from IPython.display import Markdown Markdown(f""" - **~{tot.production_loss_t_headline:,.0f} t** of cereal production lost across **{n_govs} governorates** (3 flood AOIs + the national fire set), a lower bound. Range **{tot.production_loss_t_low:,.0f}–{tot.production_loss_t_season_ref:,.0f} t** from cropland-definition and temporal-aggregation sensitivity. - That is **{tot.loss_pct_of_baseline_headline:.2f}%** of the combined 2025 baseline ({tot.baseline_production_2025_t:,.0f} t), about 2.16% of the national ~1.2 Mt floor. - Indicative food-security pressure (study total): **{tot.food_security_pressure_indicative}**, *not* an IPC phase (see method note). - The damage is **concentrated**: the two Euphrates flood AOIs (Raqqa, Deir ez-Zor) each lose ~7–8% of their own baseline; the other nine governorates lose under 0.4% each. """) ``` ::: ::: {.callout-important} **First-half-2026 snapshot — figures are lower bounds.** Windows cover floods Mar–Jun and fires May 1 – Jun 12. The harvest and fire season peaks in summer (Jun–Aug), so the heaviest crop-fire months are unobserved. Re-run after the season closes for the concluded result; field and expert verification is preferred over remote-sensing self-consistency ([DEC-039](glossary.qmd#dec-039)). ::: ## Where the loss falls ```{python} #| label: fig-loss-map #| echo: false #| fig-format: png #| fig-dpi: 200 #| fig-cap: "Cereal-production loss as a share of each governorate's 2025 baseline. The Euphrates corridor carries the damage; Latakia and Damascus City are verified-excluded (grey). First-half-2026 lower bound." #| fig-alt: "Choropleth map of Syria shaded by 2026 cropland-production loss as a percent of the 2025 baseline. Raqqa and Deir ez-Zor are darkest at about 7.6 to 7.8 percent, Hasakah mid-toned at 2.8 percent, all others pale under 1.5 percent, Latakia and Damascus City grey." from viz import choropleth, caveat_footer, CAVEATS import matplotlib.pyplot as plt govs = nat[nat.aoi_id != "STUDY_TOTAL"] vals = dict(zip(govs.aoi_id, govs.loss_pct_of_baseline_headline)) fig = choropleth(vals, title="2026 cropland-production loss (% of 2025 baseline)", cbar_label="loss, % of 2025 baseline", value_fmt="{:.1f}%") caveat_footer(fig, CAVEATS["case_study_2026h1"]) plt.show() ``` ## The whiplash The study's spine is a three-state sequence, **drought → tentative recovery → shock**, and the question is how much of the recovery the 2026 shock took back. | State | Cereal production | Note | |---|---:|---| | 2025 — record drought (baseline) | ~1.20 Mt | ~60%+ below average; worst in ~60 years (FAO/GIEWS) | | 2026 — tentative recovery (projected) | ~2.30 Mt | more even rainfall; sowing targets not fully met | | 2026 — observed first-half shock | see below | floods + fires, validated damage only | The recovery figure is an external projection (context). The shock is measured here from **human-validated damage records only**. The ~25.9 kt loss is about **2.4% of the ~1.1 Mt recovery increment** the 2026 harvest was projected to add back: a real dent in the rebound, not a reversal of it. ## Production loss by governorate ```{python} #| label: fig-loss-by-aoi #| fig-cap: "Estimated 2026 cereal-production loss (headline t) by governorate and phenomenon. Floods on the Euphrates dominate; the national fire set adds a broad but individually small contribution. First-half-2026 lower bound." #| fig-alt: "Grouped bar chart of estimated 2026 cereal-production loss in tonnes by governorate, split into flood (blue) and fire (red). Raqqa, Hasakah and Deir ez-Zor have by far the largest flood losses (roughly 10,000, 8,600 and 5,000 tonnes); all fire bars are under about 920 tonnes." import matplotlib.pyplot as plt import seaborn as sns from viz import apply_theme, caveat_footer, CAVEATS, PHENOMENON_COLORS, bar_value_labels pal = {k.value: v for k, v in PHENOMENON_COLORS.items()} apply_theme() order = (impact.groupby("aoi_id")["production_loss_t_headline"].sum() .sort_values(ascending=False).index.tolist()) fig, ax = plt.subplots(figsize=(9, 5)) sns.barplot( impact, x="aoi_id", y="production_loss_t_headline", hue="phenomenon", hue_order=["flood", "fire"], order=order, palette=pal, ax=ax, ) bar_value_labels(ax, fmt="{:,.0f}", threshold=200) ax.set_xlabel("Governorate") ax.set_ylabel("Estimated production loss — headline (t)") ax.set_title("2026 cereal-production loss by governorate and phenomenon") ax.tick_params(axis="x", rotation=35) for lbl in ax.get_xticklabels(): lbl.set_ha("right") ax.legend(title="", loc="upper right", frameon=True) caveat_footer(fig, CAVEATS["case_study_2026h1"]) plt.show() ``` [↓ Data behind this figure (impact_by_aoi.csv)](https://github.com/Muhanad-husn/flood-fire/blob/main/outputs/food_security/impact_by_aoi.csv){.data-download} ## Per-governorate detail Loss as a share of each governorate's own 2025 baseline shows where the shock bites hardest in *relative* terms. The three Euphrates/Khabur AOIs carry essentially all of it; the nine fire-only governorates are individually marginal, so they are summarised into one row below (the full 12-row table is one click away). ```{python} #| label: tbl-per-aoi #| echo: false g = nat[nat["aoi_id"] != "STUDY_TOTAL"].copy() prominent = g[g["food_security_pressure_indicative"] != "marginal"].copy() marginal = g[g["food_security_pressure_indicative"] == "marginal"].copy() prominent = prominent.sort_values("production_loss_t_headline", ascending=False) summary = pd.DataFrame([{ "aoi_id": f"{len(marginal)} other governorates (each <0.4%)", "baseline_production_2025_t": marginal["baseline_production_2025_t"].sum(), "production_loss_t_headline": marginal["production_loss_t_headline"].sum(), "production_loss_t_low": marginal["production_loss_t_low"].sum(), "production_loss_t_season_ref": marginal["production_loss_t_season_ref"].sum(), "loss_pct_of_baseline_headline": ( marginal["production_loss_t_headline"].sum() / marginal["baseline_production_2025_t"].sum() * 100), "food_security_pressure_indicative": "marginal (combined)", }]) cols = ["aoi_id", "baseline_production_2025_t", "production_loss_t_headline", "production_loss_t_low", "production_loss_t_season_ref", "loss_pct_of_baseline_headline", "food_security_pressure_indicative"] view = pd.concat([prominent[cols], summary[cols]], ignore_index=True).rename(columns={ "aoi_id": "governorate", "baseline_production_2025_t": "baseline 2025 (t)", "production_loss_t_headline": "loss headline (t)", "production_loss_t_low": "loss low (t)", "production_loss_t_season_ref": "loss season-ref (t)", "loss_pct_of_baseline_headline": "loss % of baseline", "food_security_pressure_indicative": "indicative pressure", }) (view.style.hide(axis="index") .format({"baseline 2025 (t)": "{:,.0f}", "loss headline (t)": "{:,.1f}", "loss low (t)": "{:,.1f}", "loss season-ref (t)": "{:,.1f}", "loss % of baseline": "{:.2f}"}) .bar(subset=["loss % of baseline"], color="#9ecae1", vmin=0, vmax=float(g["loss_pct_of_baseline_headline"].max()))) ``` [↓ Full 12-governorate table (impact_national.csv)](https://github.com/Muhanad-husn/flood-fire/blob/main/outputs/food_security/impact_national.csv){.data-download} ## Fire damage by burn severity The headline counts burned-cropland hectares; this breaks them out by **dNBR burn severity** (Key & Benson / USGS bins), the depth-of-burn signal behind the area. Most burned cropland sits in the *low* and *moderate-low* classes — consistent with stubble and field-edge fires rather than total crop loss — with high-severity hectares a small tail. Severity is shown for the union-cropland headline; the [Methods](technical.qmd) page explains the dNBR ∩ VIIRS chain. ```{python} #| label: fig-fire-severity #| echo: false #| fig-cap: "Burned cropland (union-headline ha) by governorate and dNBR burn severity, 2026 study window (May 1 – Jun 12). Hasakah dominates the national fire signal; most hectares are low to moderate-low severity. First-half-2026 lower bound." #| fig-alt: "Stacked horizontal bar chart of 2026 burned cropland in hectares by governorate, each bar split by burn severity from low (pale orange) to high (dark red). Hasakah has by far the longest bar at roughly 4,100 hectares, then Raqqa near 1,200, with most governorates under a few hundred hectares; low and moderate-low severity make up the bulk of every bar." import matplotlib.pyplot as plt from viz import apply_theme, caveat_footer, CAVEATS sev = pd.read_csv(OUT / "tables" / "fire_damage_sensitivity.csv") sev = sev[sev["is_study"]].copy() SEV_ORDER = ["low", "moderate_low", "moderate_high", "high"] SEV_COLORS = {"low": "#fed976", "moderate_low": "#feb24c", "moderate_high": "#fd8d3c", "high": "#bd0026"} SEV_LABEL = {"low": "low", "moderate_low": "moderate-low", "moderate_high": "moderate-high", "high": "high"} wide = (sev.pivot_table(index="aoi_id", columns="severity_class", values="ha_union", aggfunc="sum", fill_value=0.0) .reindex(columns=SEV_ORDER, fill_value=0.0)) wide = wide.loc[wide.sum(axis=1).sort_values().index] # ascending → biggest on top labels = [a.replace("_", " ").title() for a in wide.index] apply_theme() fig, ax = plt.subplots(figsize=(9, 5)) left = pd.Series(0.0, index=wide.index) for cls in SEV_ORDER: ax.barh(labels, wide[cls], left=left.values, color=SEV_COLORS[cls], edgecolor="white", linewidth=0.5, label=SEV_LABEL[cls]) left += wide[cls] totals = wide.sum(axis=1) for y, total in enumerate(totals.values): ax.text(total, y, f" {total:,.0f}", va="center", ha="left", fontsize=7.5, color="#333333") ax.set_xlabel("Burned cropland — union headline (ha)") ax.set_ylabel("") ax.set_title("2026 burned cropland by governorate and dNBR severity") ax.margins(x=0.10) ax.legend(title="burn severity", loc="lower right", frameon=True, fontsize=8) caveat_footer(fig, CAVEATS["case_study_2026h1"]) plt.show() ``` [↓ Data behind this figure (fire_damage_sensitivity.csv)](https://github.com/Muhanad-husn/flood-fire/blob/main/outputs/tables/fire_damage_sensitivity.csv){.data-download} ## Robustness — the headline without the RQ1-flagged hectares [RQ1](rq1.qmd) flags Hasakah's June flood hectares as a likely radar harvest artifact and excludes them from attribution. To keep the food-security number and the RQ1 attribution on the *same* record set, here is the study total recomputed with those flagged dates removed (Hasakah's flood headline falls back to its attributed March peak). ```{python} #| label: excl-flagged #| echo: false events = pd.read_csv(OUT / "floods" / "rq1_attribution" / "event_mechanism.csv") YIELD = float(impact["baseline_yield_t_ha"].iloc[0]) # DEC-019 drought-floor yield attributed = events[events["mechanism"] != "unexplained"] peak_attr = attributed.groupby("aoi_id")["flooded_cropland_ha_union"].max() flood = impact[impact["phenomenon"] == "flood"].set_index("aoi_id") removed_t = 0.0 for aoi, row in flood.iterrows(): new_ha = peak_attr.get(aoi, row["damaged_ha_headline_union"]) removed_t += row["production_loss_t_headline"] - new_ha * YIELD headline = tot.production_loss_t_headline excl = headline - removed_t base = tot.baseline_production_2025_t Markdown(f""" | | Production loss | % of 2025 floor | |---|---:|---:| | **Headline (all validated records)** | ~{headline:,.0f} t | {headline/base*100:.2f}% | | **Excluding RQ1-flagged Hasakah-June hectares** | ~{excl:,.0f} t | {excl/base*100:.2f}% | | *Difference* | *−{removed_t:,.0f} t* | *−{removed_t/headline*100:.1f}%* | """) ``` The headline is **robust to the flag**: removing the flagged hectares moves it only about **5%**. That is because the layer already counts each AOI's **peak single flood date**, not the sum of dates ([DEC-033](glossary.qmd#dec-033)), so the flagged June dates never compounded, and the total is dominated by the well-supported Euphrates records. We publish both figures rather than choosing one. ## Method, and what the numbers are not ::: {.callout-note collapse="true"} ## Method & provenance (expand) - **Validated-only gate.** Only human-validated records are read; a single non-validated record makes the layer refuse. No agent set any validation status: the flood and fire gates were closed by a human ([DEC-007](glossary.qmd#dec-007)). - **One cropland convention.** Union cropland is the headline, intersection the conservative low bound ([DEC-015](glossary.qmd#dec-015)). Flood records (per event-date) are collapsed to the **peak single event-date** so persistent pixels are not double-counted across dates; the season-reference column is the upper bracket ([DEC-033](glossary.qmd#dec-033)). - **Loss = hectares × the 2025 drought-floor yield** (~0.223 t/ha, [DEC-019](glossary.qmd#dec-019)). Applying the *drought-year* yield to 2026 damaged area makes the loss a **conservative lower bound**: 2026 was a recovery year, so expected yields on the lost hectares were higher. - **Damage hectares inherit the pipeline caveats.** Flood extent is open-water riverine inundation (flooded vegetation and pluvial upland under-detected, [DEC-023](glossary.qmd#dec-023)); burned cropland is VIIRS-confirmed and a conservative lower bound ([DEC-031](glossary.qmd#dec-031)). ::: ::: {.callout-warning} **Not an IPC classification.** The indicative pressure is a qualitative band from loss-as-%-of-baseline, **not** an IPC phase (which needs the full protocol: consumption, livelihoods, nutrition, mortality). GIEWS, FEWS NET and IPC are the authoritative sources; this layer is a production-shock signal that *feeds* such an assessment. Syria's 2025 baseline was already a record-drought, Crisis-level context, so the 2026 shock is **incremental** pressure on top of that. ::: {{< include _footer.qmd >}}