Drought Variability and the Threshold Sensitivity of Cereal Production across Districts in Somalia.

Description

The study draws on two district-level secondary datasets for Somalia, 2001 to 2025: the monthly FAO-SWALIM Combined Drought Index and its precipitation, temperature, and vegetation components for 73 districts (21,827 district-month observations), and FSNAU seasonal maize and sorghum production for 45 crop-growing districts. The monthly index was aggregated to the growing season and joined to production, yielding an analysis panel of 1,777 district-season observations. Data derive from FAO-SWALIM (https://www.faoswalim.org) and FSNAU (https://www.fsnau.org); derived datasets are available from the corresponding author on reasonable request. Two things to confirm before you use these. Check that the district counts and observation totals — 73 districts and 21,827 district-months, 45 crop districts, 1,777 district-seasons, match the final figures in the manuscript exactly, since I am drawing them from our earlier analysis and a late revision could have shifted them. And verify the two source URLs resolve to the actual data pages before submitting.

Steps to reproduce

Here are the reproduction steps with the hyphens and dashes removed. Where a hyphen joined a compound term I have either closed it up, reworded it, or used a space, so the meaning is unchanged. First, assemble the two sources. Pull the monthly FAO SWALIM Combined Drought Index and its precipitation, temperature, and vegetation components for the 73 districts, 2001 to 2025, and remove duplicate records to reach the 21,827 district month panel. Pull FSNAU seasonal maize and sorghum production for the 45 crop growing districts, reconstructing totals for full 2001 to 2025 coverage, and confirm the district names match the drought record exactly. Second, prepare the climate series. Convert the index and each component to deseasonalised anomalies by subtracting, for each district and calendar month, that month's long term average, so the seasonal cycle is removed. Third, run the trend and variability analysis. Apply the Modified Mann Kendall test with the Hamed and Rao autocorrelation correction and the Sen slope per district, with false discovery rate control across districts. Test the change in variability by comparing the early half of the record with the late half using the Levene test, tracking a 24 month rolling standard deviation, and tabulating the frequency of extreme states. Decompose the index onto its components to identify the precipitation origin of the variance change. Fourth, build the impact panel. Aggregate the monthly index to each growing season per district and year, keeping the seasonal mean, minimum, maximum, and within season standard deviation, and join to production on district, year, and season, giving the 1,777 district season panel. Fifth, estimate the dose response. Regress log cereal production on the growing season index and its square, with district, year, and season fixed effects and standard errors clustered by district, run for all cereals and separately by crop and by livelihood system. Sixth, locate the threshold. Run a Hansen style threshold regression, searching over candidate index values for the single split that best divides the response into a steep and a shallow segment, and report the threshold near 0.89 with the frequency of seasons falling below it. Software: R or Python throughout, using pymannkendall or the modifiedmk package for the trend tests, and standard fixed effects panel and threshold routines for the regressions. Each step's output feeds the next, so run them in this order. Same check as before: confirm the figures, 73 districts, 21,827 district months, 45 crop districts, 1,777 district seasons, and the 0.89 threshold, still match the final manuscript before you publish these steps.

Institutions

• Hormuud University

  Banaadir, Mogadishu
• Muzeum a Galerie Orlických Hor

  Královéhradecký kraj, Rychnov nad Kněžnou

Categories

Licence