Quote of the day

What we know is a drop, what we don't know is an ocean.

- Isaac Newton

Contact

Multimodal Spatio-Temporal Early Warning of Desert Locust Breeding Zones

Open Agriculture

Multimodal Spatio-Temporal Early Warning of Desert Locust Breeding Zones

Multimodal Spatio-Temporal Early Warning of Desert Locust Breeding Zones

Description

Develop an explainable and probabilistically calibrated multimodal spatio-temporal model to anticipate desert-locust breeding zones from climate and satellite observations.

Details

Context and Problem Statement

The desert locust is a major transboundary threat to food security across North Africa, the Sahel, and East Africa. Existing monitoring systems rely heavily on expert interpretation, while machine-learning approaches may be biased because absence of reporting does not necessarily imply absence of locusts.

Research Question

How can favorable breeding conditions be predicted several weeks in advance while explicitly correcting survey-effort bias and producing uncertainty estimates useful for treatment planning?

Proposed Approach

Fuse rainfall, temperature, soil moisture, and vegetation indices in a ConvLSTM or spatial graph model. Observation bias is addressed through occupancy modeling with imperfect detection, and explainability is provided through variable attribution and spatial contribution maps.

Expected Contribution

A calibrated early-warning framework with observation-bias correction and strict spatio-temporal validation.

Expected Prototype

A dashboard producing weekly risk maps with confidence intervals and export capabilities for surveillance workflows.

Datasets

FAO Locust Hub / DLIS, TerraClimate, ERA5, MODIS, and Sentinel-2.

Challenges

Sampling bias, false negatives, extreme class imbalance, climate-driven non-stationarity, and cross-border data limitations.

Research Question

Can early warning of desert-locust breeding zones be significantly improved by correcting survey-effort bias during learning rather than merely increasing model complexity?

Innovation

The project addresses biased presence-absence data, a methodological issue that can invalidate classical metrics and reduce operational usefulness of alerts.

Expected Deliverable

A calibrated forecasting model, a rigorous spatio-temporal validation protocol, and an interactive mapping demonstrator.

Technologies

Spatio-Temporal Deep Learning (ConvLSTM, GNN) Multispectral Remote Sensing Occupancy Models and Observation-Bias Correction Explainable AI (SHAP) Probabilistic Calibration

Required Skills

  • Deep Learning for Spatio-Temporal Data
  • Remote Sensing and Geomatics
  • Statistics and Observation-Bias Modeling
  • Geospatial Visualization

Datasets

  • FAO Locust Hub / Desert Locust Information Service
  • TerraClimate
  • MODIS and Sentinel-2 vegetation indices

Morocco & Africa Relevance

Morocco is a frontline country in desert-locust control; a calibrated early-warning tool could strengthen coordination across the Maghreb, Sahel, and Horn of Africa.

References

  1. FAO Desert Locust Information Service (Locust Watch)
  2. LocustLens: leveraging environmental data fusion and machine learning for desert locust swarm prediction
  3. A forecasting model for desert locust presence during recession period, using real-time satellite imagery

Project Information

Duration: 4 to 6 months
Academic Level: Engineering, Master's or PhD
Research Area: Agriculture

Keywords

Desert Locust Early Warning Spatio-Temporal Modeling Food Security Explainable AI

Evaluation

Scientific novelty
Impact
Technical feasibility
Data availability
Prototype potential
Industrial relevance

Apply for this topic

To apply for this topic, please send your CV and portfolio to:

contact@yanecode.com
Back to Research