F1 - Smart Irrigation
Algorithm / Method
RandomForest & HistGradientBoosting
Dataset / Scope
31 Years of Udawalawe Hydrology Data
Performance Metric
0.71 MCM Inflow RMSE
Field Impact
20-35% water savings vs static scheduler
4th Year Software Engineering Research Project
Adaptive Smart Irrigation &
Crop Optimization Platform
An integrated decision-support system for Sri Lankan irrigation schemes, combining IoT telemetry, remote-sensing crop health, rainfall forecasting, and resource-constrained agricultural optimization.
11,687
Udawalawe daily records
10m
Sentinel-2 analysis scale
71,737
Retail price observations
1-14
Day forecast horizon
Challenges Resolved
Overcoming fragmentation and water waste in irrigation schemes.
ASICOP replaces manual, static schedules and isolated agricultural tools with an end-to-end integrated solution. Here is how we resolve key domain inefficiencies:
Inefficient Scheduling & High Water Waste
The Problem
Traditional flood irrigation is applied on fixed 7-10 day schedules or on manual observation, ignoring weather forecasts and leading to estimated water losses of 35-50%.
The Resolution
ASICOP uses live soil moisture telemetry and a RandomForestClassifier combined with HistGradientBoosting reservoir release predictions, reducing field-level water waste by 20-35%.
Rigid Seasonal Water Quota Allocation
The Problem
Sri Lankan command-area schemes allocate top-down water quotas before the season. These quotas are highly rigid and cannot adapt to crop changes or weather fluctuations mid-season.
The Resolution
The F4 ACA-O service formulates a mathematical PuLP Mixed-Integer programming model that optimizes crop and area allocation under water limits, and offers Plan B re-planning if quotas change mid-season.
Delayed Field Stress & Disease Alerts
The Problem
Agronomists and officers manually survey fields to detect stress and disease. By the time visual symptoms are identified, the damage to crop health is often irreversible.
The Resolution
F2 integrates 10m Sentinel-2 satellite indices (NDVI/NDWI) for early zone-level stress detection, and a fine-tuned MobileNetV2 leaf image classifier for instant 38-class edge disease diagnosis.
Market Price Shocks & Price Volatility
The Problem
Farmers decide what crops to cultivate based on current prices, leading to crop gluts and severe market price crashes at harvest time.
The Resolution
F4 runs a LightGBM price predictor trained on 71,737 retail price observations, feeding anticipated farmgate prices into a Fuzzy-TOPSIS scorer to ensure crops are selected for economic viability.
Project Architecture
Four research streams feeding one water-aware decision loop.
The platform is more than a dashboard. Each service provides a decision signal that improves another service: stress influences irrigation, forecasts influence optimization, and field telemetry keeps the whole plan grounded.
F1 Irrigation • Hesara
IoT Smart Water Management
Field telemetry, crop thresholds, reservoir safety gates, and ML predictions are fused into valve actions for quota-based irrigation fields.
11,687 Udawalawe daily hydrological records0.71 MCM RMSE for next-day release prediction
- ESP32 telemetry for soil moisture, temperature, humidity, and field water level
- Random Forest field valve classifier plus HistGradientBoosting reservoir release predictor
- Manual officer queue when reservoir limits or quota rules block automated opening
F2 Crop Health • Abishek
Hybrid Satellite Crop Health Monitoring
Remote-sensing zone health and plant image diagnosis identify stress early enough to prioritize irrigation and adjust crop planning.
10 m Sentinel-2 Level-2A pixel scale95.43% MobileNetV2 validation accuracy
- NDVI and NDWI stress labels for Good, Moderate Stress, and High Stress zones
- Five-stage vegetation validation rejects cloud, water, urban, and non-crop requests
- Field stress index feeds F1 irrigation priority and F4 suitability penalties
F3 Forecasting • Trishni
ML Time-Series Forecasting and Alerting
Reservoir and rainfall forecasts expose drought, flood, and uncertainty signals before they affect field schedules or seasonal plans.
12 engineered time-series featuresP10/P50/P90 water-risk bands
- Gradient Boosting, Random Forest, LSTM, ARIMA, ensemble, and anomaly layers
- Open-Meteo weather integration and water-level risk endpoints
- Forecast scenarios suppress irrigation after rain and constrain F4 water budgets
F4 ACA-O • Dilruksha
Adaptive Crop and Area Optimization
Crop suitability, price signals, water quotas, field stress, and policy rules are combined into practical crop-area recommendations.
71,737 Hector retail price observations5-criterion Fuzzy-TOPSIS crop ranking
- Fuzzy-TOPSIS balances soil, water, yield history, sensitivity, and crop duration
- LightGBM and neural price models provide market-risk signals for crop ranking
- Greedy/PuLP allocation logic handles water quota, area, profit, and paddy policy
Closed-Loop Architecture
The technical idea: dynamic service co-operation.
Traditional research trains models in isolation. ASICOP connects models at runtime: forecasts alter valve schedules, crop stress penalizes area suitability, and remaining water quotas drive optimization.
Closed-Loop Telemetry & Control Signal Flow
Generating diagram...
Signal Integrations
Inter-service Data Exchanges
F3→F1
Rainfall suppression
Expected rain of ≥5mm within 24h triggers valve closure advice, saving reservoir storage.
F2→F1 & F4
Stress priority & suitability penalty
Severe field stress escalates valve priority, while repeated stress acts as a penalty weight in crop suitability planning.
F1→F4
Water budget feedback
Live remaining water quotas act as hard mathematical constraints in the mixed-integer optimization solver.
Model Specifications
Algorithm & Dataset Details
F2 - Crop Health
Algorithm / Method
MobileNetV2 Transfer Learning & Sentinel-2
Dataset / Scope
54,306 images (38 plant-disease classes)
Performance Metric
95.43% disease classification accuracy
Field Impact
Automated vegetation validation & zone stress map
F3 - Water Forecasting
Algorithm / Method
ARIMA/SARIMA & LSTM Ensembles
Dataset / Scope
1994-2025 Udawalawe level time series
Performance Metric
P10/P50/P90 risk scenarios & alerts
Field Impact
1-14 day lookahead for reservoir/rainfall
F4 - Crop Optimization
Algorithm / Method
Fuzzy-TOPSIS & PuLP MIP Solver
Dataset / Scope
71,737 Hector farmgate price records
Performance Metric
Expected profit vs water budget optimization
Field Impact
Top-3 recommendations & dynamic Plan B updates
Evidence First
The visuals come from the project research assets.
The project evidence is grounded in generated research outputs from the repo: hydrology records, NDVI health maps, ensemble forecasts, price modelling, and crop allocation architecture. Click on any diagram below to inspect it in full scale.
F1 - Release prediction
HistGradientBoosting on 2023-2025 test data from the 32-sheet Udawalawe workbook.
F2 - Disease classification
MobileNetV2 transfer learning across 38 PlantVillage crop health classes.
F3 - Best current forecast
Gradient Boosting on the limited 1994 notebook subset; full multi-year training is documented as the fix.
F4 - Price prediction
PricePredictorNN result on Hector-derived crop price data, used as a relative market signal.
View Larger
F1 Irrigation
View Larger
F2 Crop Health
View Larger
F3 Forecasting
View Larger
F4 ACA-O
Submission Structure
Research material organized for evaluation.
Our complete research logs, timeline milestones, slide decks, and final report drafts have been archived inside structured pages for departmental evaluation.
Abstract
Integrated project overview and research highlights
Domain
Literature, gap, problem, objectives, methodology, and technologies
Milestones
Assessment timeline with dates and details
Documents
Project-wide and individual final submission files
Slides
Proposal, progress, final, and stream deep-dive decks
Team
Member profiles, stream ownership, and contact details