Decision system · early warning

SIRAS / EWIS

Shrimp Index Risk Alert System — an early warning and information system that flags trouble in a pond while there is still time to act on it.

Role: In-situ correlation
& alert system design
Inputs: Sentinel-2 spectral trends
+ in-situ ABW samples
Method: Rule-based classification
on rate of change
Output: PCS / PHS scores
& combined alert label
Result: 80% overall accuracy
across multiple regions

The problem

Shrimp farming is vulnerable to disease, poor water quality and environmental change, any of which can cause significant production loss. Most farmers identify these problems only once they are already severe, which sharply limits what corrective action can still do.

An early warning and information system detects risk at an early stage and issues a timely alert — enabling proactive measures, reducing losses, and improving overall farm productivity.

My role

Collecting and correlating in-situ data from shrimp culture farms with the patterns observed in spectral signatures, and developing the risk alert system that turns that correlation into an action a farmer can take.

Objectives

Use alerts observed within ponds together with remotely sensed data to develop a rule-based alert system.
Generate early risk alerts for disease outbreaks, water quality deterioration and abnormal growth conditions — enabling timely intervention and more sustainable shrimp farming.

End-to-end methodology workflow diagram for the early warning information system — **Fig. 01**End-to-end methodology, from spectral time series through state classification to alert labelling.

How it works

The system takes the temporal pattern of each pond's spectral signature and computes its rate of change over time, using that to classify ponds into health states. It then compares average body weight (ABW) sampled on site against the standard ABW for that day of culture, and uses the gap to set the alert label.

Stage 01Generate input, upload to S3

Stage 02SIRAS detects risk from spectral signature patterns

Stage 03Compare sampled ABW against standard ABW

Stage 04Flag alerts, return JSON with metadata

Sample SIRAS and EWIS risk maps across a farm cluster — **Fig. 02**Sample SIRAS / EWIS maps — pond-level risk state across a farm cluster.

What I built

A correlation study linking in-situ farm observations to the spectral behaviour of the same ponds over the culture cycle.
A rule set that classifies pond state from the rate of change of the spectral signature, rather than from any single observation.
A growth check that compares sampled ABW against the standard ABW curve for the pond's day of culture.
Two separate scores — pond culture state (PCS) and pond health state (PHS) — plus a combined alert label, so a farm manager can see which of the two is driving the warning.
A FastAPI service over the alert engine, returning JSON alerts with the metadata needed to act on them.

API contract

Request	Type
pondid	str
stocking_date	str
harvest_date	str, optional

Response	Meaning
time	Timestamp of the assessment.
pondid / pond	Pond identifier.
PCS	Pond culture state score.
PHS	Pond health state score.
Alert_PCS	Alert raised on culture state.
Alert_PHS	Alert raised on health state.
Alert_Combined	Combined signal across both.
Alert_Label	Final label surfaced to the farm.
DOC	Day of culture.
STD_abw	Standard average body weight for that DOC.
Sampling_abw	Average body weight measured on site.

Request sequence

01Client calls FastAPI

02Validate request

03Fetch satellite and in-situ data

04Generate alerts

05Return result

Results

The proposed rule-based method achieved 80% overall accuracy across multiple regions.
Risk is surfaced from the trajectory of the spectral signature, so an alert can arrive before the problem is visible at the pond.

The index this system watches is documented in Shrimp Index (SI), and the pond geometries it runs over come from Aqua-Boundary.

← Shrimp Index (SI) Next: Palm Tree Classification →