[Paper Review] Predicting Regional Locust Swarm Distribution with Recurrent Neural Networks

Hadia Mohmmed Osman Ahmed Samil, Annabelle Martin+3 more|arXiv (Cornell University)|Nov 29, 2020

Insect and Arachnid Ecology and Behavior4 citations

TL;DR

This study proposes a Long Short-Term Memory (LSTM) recurrent neural network to predict regional locust swarm distributions one month in advance using historical locust attack data and environmental variables like soil moisture and vegetation density. The model achieves 81% macro-averaged recall and 60% macro-averaged precision, demonstrating strong performance in forecasting swarm locations and infestation intensity with reasonable accuracy.

ABSTRACT

Locust infestation of some regions in the world, including Africa, Asia and Middle East has become a concerning issue that can affect the health and the lives of millions of people. In this respect, there have been attempts to resolve or reduce the severity of this problem via detection and monitoring of locust breeding areas using satellites and sensors, or the use of chemicals to prevent the formation of swarms. However, such methods have not been able to suppress the emergence and the collective behaviour of locusts. The ability to predict the location of the locust swarms prior to their formation, on the other hand, can help people get prepared and tackle the infestation issue more effectively. Here, we use machine learning to predict the location of locust swarms using the available data published by the Food and Agriculture Organization of the United Nations. The data includes the location of the observed swarms as well as environmental information, including soil moisture and the density of vegetation. The obtained results show that our proposed model can successfully, and with reasonable precision, predict the location of locust swarms, as well as their likely level of damage using a notion of density.

Motivation & Objective

• To develop a machine learning model that forecasts locust swarm locations one month in advance to enable early intervention.
• To address the challenge of limited predictive capability for locust swarms, which can destroy food supplies equivalent to 35,000 people's daily consumption.
• To leverage environmental data—soil moisture and vegetation density—alongside historical swarm observations for improved prediction.
• To evaluate model performance using precision, recall, and qualitative heatmap analysis of predicted vs. actual swarm density.
• To explore the potential of integrating satellite data and field observations to enhance long-term forecasting and disaster preparedness.

Proposed method

• The model uses a univariate LSTM network to process sequential time-series data of locust swarm occurrences and environmental conditions.
• Input features include monthly locust attack counts, soil moisture levels, and vegetation density across regional grids.
• The network is trained to predict the next month’s swarm distribution by learning temporal dependencies in the data.
• A binary classification head is applied to convert continuous predictions into presence/absence labels for precision and recall evaluation.
• Density levels (low, medium, high) are defined based on the number of attacks per region to assess prediction accuracy across infestation intensities.
• Model training is performed on a single NVIDIA V100 GPU for 50 epochs, with a sequence length of 12 months.

Experimental results

Research questions

• RQ1 Can an LSTM-based model effectively predict the geographic distribution of locust swarms one month in advance using environmental and historical swarm data?
• RQ2 How well does the model generalize across regions with varying swarm intensity, particularly in high-density infestation zones?
• RQ3 To what extent can the model capture the temporal and spatial movement patterns of locust swarms, such as migration from Oman to western India?
• RQ4 How do precision and recall metrics perform on a skewed dataset with few swarm events per region?
• RQ5 What is the impact of incorporating additional data sources like satellite imagery on future forecasting accuracy?

Key findings

• The model achieves a macro-averaged recall of 81%, indicating strong ability to detect actual locust attacks in advance.
• The macro-averaged precision is 60%, showing a reasonable balance between true positives and false positives.
• Recall for high-density infestation regions (more than 4 attacks) reaches 86%, demonstrating superior performance in predicting severe outbreaks.
• The model successfully captures locust migration patterns, such as the movement from Oman in June 2020 to western India in July 2021, as shown in heatmap comparisons.
• The model predicts both the location and intensity of swarm attacks with good qualitative alignment to ground truth, particularly in regions with high swarm activity.
• The use of LSTM enables effective learning of long-term temporal dependencies, improving prediction accuracy over static models.