Methodology Decision Matrix

Use this guide to select the right AquaScope methodology for your research question.

Quick Decision Tree

What's your goal?
│
├─ Understand my data → EDA (aquascope eda)
│   ├─ Find trends? → Mann-Kendall / Sen's Slope
│   ├─ Find outliers? → IQR / Isolation Forest
│   └─ Assess quality? → aquascope quality
│
├─ Predict future values → Forecasting
│   ├─ Simple/fast? → ARIMA
│   ├─ Seasonal data? → Prophet
│   ├─ Complex patterns? → Random Forest / XGBoost
│   └─ Long sequences? → LSTM
│
├─ Assess flood risk → FloodChallenge
│   ├─ Return periods? → GEV / LP3 (aquascope hydro --analysis flood-freq)
│   ├─ Forecast floods? → aquascope solve "forecast flooding..."
│   └─ Flow statistics? → FDC (aquascope hydro --analysis fdc)
│
├─ Monitor drought → DroughtChallenge
│   ├─ SPI computation → aquascope solve "assess drought..."
│   └─ Water balance → DroughtChallenge.water_balance()
│
├─ Assess water quality → WaterQualityChallenge
│   ├─ WHO compliance? → check_who_guidelines()
│   ├─ Detect anomalies? → detect_anomalies()
│   └─ Trend analysis? → trend_analysis()
│
└─ Classify/compare → Statistical methods
    ├─ Compare stations? → ANOVA / Kruskal-Wallis
    ├─ Cluster stations? → K-means / Hierarchical
    └─ Water Quality Index? → WQI pipeline

Detailed Matrix

By Data Characteristics

Your Data	Records	Temporal	Spatial	Best Methods
Short time-series (<1 yr)	<500	✅	❌	ARIMA, basic stats
Long time-series (>3 yr)	>1000	✅	❌	Prophet, LSTM, Mann-Kendall, GEV
Multi-station snapshot	>100	❌	✅	Kriging, IDW, clustering
Multi-station time-series	>5000	✅	✅	Full suite — EDA → recommend
Single parameter	any	any	any	Direct analysis or forecast
Multi-parameter	any	any	any	Correlation, PCA, WQI

By Research Goal

Goal	Method ID	Pipeline	Min Data
Trend detection	mann_kendall	trend_analysis	30 records
Seasonal decomposition	stl_decomposition	seasonal_decomposition	2 years
Spatial interpolation	idw_kriging	spatial_interpolation	10 stations
Water quality index	wqi	water_quality_index	5 parameters
Clustering	clustering	cluster_analysis	50 records
Time-series forecast	arima / prophet	forecast CLI	365 records
Anomaly detection	isolation_forest	Quality challenge	100 records
Flood frequency	gev / lp3	hydro --analysis flood-freq	5 years
Low-flow analysis	7q10 / 30q5	hydro --analysis low-flow	3 years
Baseflow separation	lyne_hollick / eckhardt	hydro --analysis baseflow	1 year
Flow duration curve	fdc	hydro --analysis fdc	1 year

Model Selection for Forecasting

Scenario	Best Model	Why
Quick forecast, small data	ARIMA	Fast, interpretable, handles trends
Seasonal patterns (rain, temp)	Prophet	Built-in seasonality + holidays
Non-linear relationships	Random Forest	Handles interactions, robust
High-accuracy needed	XGBoost	State-of-art for tabular data
Long-term dependencies	LSTM	Captures temporal patterns
Anomaly detection	Isolation Forest	Unsupervised, handles multivariate
Drought monitoring	SPI Model	Standard WMO approach

CLI Quick Reference

# Let AI choose for you
aquascope recommend --from-file data.json --goal "your research question"

# Or use the NL agent
aquascope solve "your research question in plain English"

# List all methods
aquascope list-methods

# Run a specific pipeline
aquascope run --method trend_analysis --file data.json
aquascope forecast --model prophet --file timeseries.csv --days 30
aquascope hydro --analysis fdc --file discharge.csv