evidoria

← Back to browse

Good practice Imported

ShoShin AI Early Warning — Kenya's Sensor-Based Alert System for Lake Victoria Fish Kills

Kenya · Kisumu · See the Kenya profile · See the Kisumu profile

Evidence: Observational / pre–post Top 66% 53/100 · Ask Evidence Copilot about this practice

Underwater sensors and AI models built by Kenya's KMFRI and Nairobi's ShoShin Innovation Hub flagged a dissolved-oxygen crash at Dunga Beach, Kisumu in February 2026; over 300 farmers moved 450+ tilapia cages by SMS alert, averting a repeat of $1m in 2024-25 losses.

300+
Farmers alerted and relocated cages (February 2026)
450+
Tilapia cages relocated (February 2026)
~1 USD million
Prior losses averted (2024-25 baseline, Dunga Beach) (2024-2025)
75%+
Kenyan fish farmers reporting mortalities (2022 survey)
15
Additional high-risk hotspots identified for expansion
ShoShin AI Early Warning — Kenya's Sensor-Based Alert System for Lake Victoria Fish Kills

Details

Maturity
Pilot
Promoter
Kenya Marine and Fisheries Research Institute (KMFRI), with ShoShin Innovation Hub
Period
February 2026–present (pilot)
Keywords
aquaculture, environmental monitoring, disaster early warning, fisheries management, IoT sensors

Context

Cage-based tilapia farming on Lake Victoria at Dunga Beach, Kisumu, has suffered episodic mass fish kills from dissolved-oxygen crashes linked to algal blooms fed by untreated urban waste and agricultural runoff. A 2022 KMFRI survey found more than three-quarters of Kenyan fish farmers had experienced mortalities, and 2024-25 oxygen crashes destroyed roughly US$1 million in fish stock at Dunga Beach alone.

Objectives

Detect dissolved-oxygen crashes early enough to give farmers time to move cages and avert mass fish kills.

Activities

Kenya's Marine and Fisheries Research Institute (KMFRI) and the Nairobi-based ShoShin Innovation Hub deployed underwater sensors and cloud-hosted AI models, trained on KMFRI's historical fish-kill research combined with live sensor readings; alerts trigger SMS messages with simple, actionable guidance ('slow feeding', 'move cages') to the feature phones of more than 300 tilapia farmers.

Results

In February 2026 the system flagged a dissolved-oxygen crash below the 2.0 mg/L threshold associated with fish kills; more than 300 farmers relocated more than 450 cages within hours. No further mass fish deaths have been recorded at Dunga Beach since the alert, against a 2024-25 baseline of roughly $1 million in losses at the same site.

Conclusions

Researchers and reporters describe the system as buying farmers the hours between an oxygen crash and a total loss, not as addressing the underlying pollution driving the crashes. Open concerns include rural connectivity gaps, the sustainability of funding once government support tapers off, and unresolved questions about who controls farmers' sensor data over time. KMFRI has identified 15 further high-risk hotspots for possible expansion, with two more sites expected to go live in the second half of 2026, and the NGO WorldFish is adapting a similar approach (E-Samaki) for coastal mariculture sites in Kilifi and Kwale counties.

Implementation

Indicative cost
Low (< €50k) — The Kenyan government currently covers the platform's data and dissemination costs; a low-cost farmer subscription model is under consideration as the network scales beyond the single pilot site.
Time to results
Short (< 1 year) — System went live and produced its first documented alert in February 2026; two further sites are expected to go live in the second half of 2026, with 15 additional hotspots identified for possible future expansion.
Staffing & skills
Kenya Marine and Fisheries Research Institute (KMFRI) researchers, ShoShin Innovation Hub technical/AI team, SMS-based farmer alert network coordinators

Conditions for success

  • Underwater dissolved-oxygen sensors with reliable data transmission
  • AI models trained on historical fish-kill research combined with live sensor readings
  • Simple, actionable SMS alerts reaching farmers' feature phones
  • Farmers able to act quickly (move cages) once alerted

Common failure modes

  • Cannot address the underlying pollution (untreated urban waste and agricultural runoff feeding algal blooms) driving oxygen depletion
  • Rural connectivity gaps could prevent alerts reaching all farmers
  • Long-term funding depends on a subscription model still under consideration once government support tapers
  • Unresolved questions over who controls farmers' sensor data over time

Where it fits

Governance type
government research institute with innovation-hub and NGO partners
Scale
single-site pilot (Dunga Beach), expanding to 2 more sites in H2 2026 and 15 identified hotspots
Income level
low-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful