Multi-source environmental raster dataset of FAO desert locust survey records in Ethiopia (2015–2023) for machine-learning prediction of locust presence
- 1Department of Computer Science, Addis Ababa University, Addis Ababa, Ethiopia
Abstract
This article describes a multi-source environmental raster dataset built around 18,868 georeferenced desert locust (Schistocerca gregaria) survey records reported for Ethiopia to the FAO Desert Locust Information Service (DLIS) between 2015 and 2023: 12,056 records in which locusts were present (label 1) and 6812 survey stops at which no locusts were recorded (label 0). For each record, a patch of about 10 × 10 km centered on the survey location was extracted in Google Earth Engine at ≈250 m resolution and stored as a 60-band GeoTIFF with 59 predictor bands and one constant label band. The predictor bands combine MODIS NDVI, EVI, land surface temperature, a near-infrared–shortwave-infrared water index and evapotranspiration, CHIRPS precipitation, ERA5 10 m wind, SMAP surface soil moisture and TerraClimate actual and reference evapotranspiration, each aggregated over the 30, 60 and 90 days before the observation, together with static terrain, soil and land-cover layers and variables derived in the processing pipeline. The 18,871 files (including three duplicate files) are distributed by year and class in 32 ZIP archives on Harvard Dataverse. The article provides a band-level data dictionary, the label definition, a quantitative quality-control summary, including bands filled with zeros where a source collection had no data and groups of identical rasters, and guidance for leakage-aware evaluation. The data support spatial prediction, temporal forecasting, multimodal and transfer learning, and analysis of the environmental conditions associated with desert locust reports in the Horn of Africa.
Related
- A Curated Phenotype-Linked HIV-1 Sequence Dataset for Machine Learning Approaches to Tropism Prediction, Disease Progression and Proviral Reservoir Studies.
- StableProt: Structure-Aware Deep Learning for Protein Thermostability (Tm) and Environmental Adaptation (OGT) Prediction
- 2,234 attempts to harmonize a multi-platform B-cell lymphoma bulk transcriptomic dataset: ComBat is superior by cross-batch prediction quality
- Sentiment analysis in Bangla: A three-class dataset and comparative evaluation of machine learning, deep learning, and transformer approaches
- Cross-dataset few-shot deep learning for landmark detection in biological morphometrics
BibTeX
@article{hierpa2026multisource,
title = {{Multi-source environmental raster dataset of FAO desert locust survey records in Ethiopia (2015–2023) for machine-learning prediction of locust presence}},
author = {Rabra Hierpa and Solomon Gizaw},
journal = {Data in Brief},
year = {2026},
month = oct,
doi = {10.1016/j.dib.2026.113376},
url = {https://doi.org/10.1016/j.dib.2026.113376},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - Multi-source environmental raster dataset of FAO desert locust survey records in Ethiopia (2015–2023) for machine-learning prediction of locust presence
AU - Rabra Hierpa
AU - Solomon Gizaw
PY - 2026
DA - 2026/10/23
JO - Data in Brief
DO - 10.1016/j.dib.2026.113376
UR - https://doi.org/10.1016/j.dib.2026.113376
AB - This article describes a multi-source environmental raster dataset built around 18,868 georeferenced desert locust (Schistocerca gregaria) survey records reported for Ethiopia to the FAO Desert Locust Information Service (DLIS) between 2015 and 2023: 12,056 records in which locusts were present (label 1) and 6812 survey stops at which no locusts were recorded (label 0). For each record, a patch of about 10 × 10 km centered on the survey location was extracted in Google Earth Engine at ≈250 m resolution and stored as a 60-band GeoTIFF with 59 predictor bands and one constant label band. The predictor bands combine MODIS NDVI, EVI, land surface temperature, a near-infrared–shortwave-infrared water index and evapotranspiration, CHIRPS precipitation, ERA5 10 m wind, SMAP surface soil moisture and TerraClimate actual and reference evapotranspiration, each aggregated over the 30, 60 and 90 days before the observation, together with static terrain, soil and land-cover layers and variables derived in the processing pipeline. The 18,871 files (including three duplicate files) are distributed by year and class in 32 ZIP archives on Harvard Dataverse. The article provides a band-level data dictionary, the label definition, a quantitative quality-control summary, including bands filled with zeros where a source collection had no data and groups of identical rasters, and guidance for leakage-aware evaluation. The data support spatial prediction, temporal forecasting, multimodal and transfer learning, and analysis of the environmental conditions associated with desert locust reports in the Horn of Africa.
SN - 2352-3409
ER -