Skip to content

Latest commit

 

History

History
330 lines (229 loc) · 11.1 KB

File metadata and controls

330 lines (229 loc) · 11.1 KB

Methodology: AWD Policy Transportability Analysis

Overview

This document describes the scientific methodology for assessing the spatial transportability of Vietnam's Alternative Wetting and Drying (AWD) agricultural water management policy to Japan.

Research Question: To what extent is AWD technically feasible in Japanese rice production systems?


1. Water Balance Framework

1.1 MapAWD Algorithm

The core analysis implements the MapAWD (Map-based Assessment of AWD) framework, which evaluates suitability based on dekad-level (10-day period) water availability.

Water Balance Equation (dekad-level):

$$WB_t = P_t - (ET_t + K_t)$$

Where:

  • $WB_t$ = Water balance in dekad $t$ (mm)
  • $P_t$ = Rainfall in dekad $t$ (mm) [CHIRPS]
  • $ET_t$ = Reference evapotranspiration in dekad $t$ (mm) [MODIS]
  • $K_t$ = Percolation loss in dekad $t$ (mm) [SoilGrids-derived]

1.2 Temporal Structure

Seasonal Timing:

  • Season spans dekads 10-28 (May 15 - September 25)
  • Corresponds to main rice growing season in temperate Asia
  • 19 dekads total per season

Exclusion Windows:

  • Exclude first 2 dekads (10-11): Establishment phase, fields need water
  • Exclude last 1 dekad (28): Harvest phase, drying not beneficial
  • Analysis dekads: 12-27 (16 dekads for suitability assessment)

1.3 Suitability Criteria

AWD feasibility requires that fields periodically dry without crop stress. A dekad is considered suitable if it meets both conditions:

  1. Deficit Condition: $WB_t < 0$ (field experiences water deficit)
  2. Threshold Condition: $WB_t \geq T$ (deficit within safe range)

Where $T$ is the deficit threshold, tested across 7 levels: $$T \in {-25, -50, -70, -90, -110, -130, -150} \text{ mm}$$

Interpretation:

  • More negative $T$ = stricter criterion = fewer suitable dekads
  • $T = -150$ mm: Only extreme deficits qualify (most restrictive)
  • $T = -25$ mm: Mild deficits qualify (most lenient)

1.4 Suitability Index Computation

For each pixel and threshold:

$$f_{suit} = \frac{n_{suitable}}{n_{analysis}}$$

Where:

  • $n_{suitable}$ = Count of dekads where $WB_t < 0$ AND $WB_t \geq T$
  • $n_{analysis}$ = Count of analysis dekads (typically 16)
  • $f_{suit}$ = Fraction of suitable dekads (0-1)

1.5 Suitability Classification

Fraction converted to 3-class scale:

Class Name Condition
3 High $f_{suit} \geq 0.66$ (≥66% dekads suitable)
2 Moderate $0.33 \leq f_{suit} < 0.66$ (33-66% suitable)
1 Low $f_{suit} &lt; 0.33$ (<33% suitable)

Rationale: High and moderate classes indicate sufficient dry periods for beneficial AWD. Low class indicates climate insufficiently supports AWD cycles.


2. Input Data

2.1 Rainfall: CHIRPS

Collection: UCSB-CHG/CHIRPS/DAILY
Resolution: 5 km
Temporal: Daily, 1981-present
Processing:

  • Extract 10-day windows for each dekad
  • Sum daily precipitation to dekad totals
  • Units: mm/dekad

Quality Notes:

  • Gold-standard rainfall dataset for agriculture (widely cited)
  • Based on satellite + rain gauge fusion
  • Suitable for regional hydrological studies

2.2 Evapotranspiration: MODIS MOD16A2

Collection: MODIS/061/MOD16A2
Resolution: 500 m
Temporal: 8-day composites, 2000-present
Processing:

  • Extract 8-day ET0 values
  • Reproject to study area grid
  • Convert scale factor 0.1 to mm
  • Dekad interpolation: When dekad boundary crosses 8-day tile boundaries, use weighted average: $$ET_{dekad} = \frac{n_1 \cdot ET_1 + n_2 \cdot ET_2}{10}$$ where $n_1, n_2$ are days in each tile
  • Units: mm/dekad

Quality Notes:

  • MODIS ET widely validated against flux tower networks
  • Accounts for seasonal vegetation dynamics
  • Higher resolution (500m) captures local heterogeneity

2.3 Soil Properties: SoilGrids

Collections:

  • projects/soilgrids-isric/clay_mean (clay %, 0-5cm)
  • projects/soilgrids-isric/sand_mean (sand %, 0-5cm)

Resolution: 250 m
Processing:

  • Extract clay and sand percentages
  • Calculate silt as: silt = 100 - clay - sand
  • Classify into 4 drainage classes based on texture

Drainage Classification:

Class Clay (%) Soil Type Percolation (mm/day)
1 (Well) <20 Sandy 12.0
2 (Moderate) 20-35 Loam 8.0
3 (Imperfect) 35-50 Clay loam 4.0
4 (Poor) ≥50 Clay 3.0

Percolation Rates: Based on USDA Soil Survey Manual estimates for rice-growing regions.

2.4 Elevation: SRTM

Collection: USGS/SRTMGL1_Ellip
Resolution: 30 m
Processing:

  • Compute slope using Sobel filter
  • Classify slopes: feasible if $&lt; 10°$
  • Rationale: Steep slopes complicate water management

2.5 Rice Extent Map

User-provided asset (custom classification)
Content: Binary raster where 1 = rice paddy, 0 = other
Sources (alternatives):

  • User's own field surveys / high-resolution imagery
  • MODIS Land Use Classification (annual)
  • ESA WorldCover (global, 10m resolution)

3. Biophysical Constraints Integration

Beyond water balance, real-world AWD implementation requires favorable terrain and soil conditions.

3.1 Composite Suitability

A pixel is deemed biophysically feasible only if:

$$\text{Feasible} = \text{(Slope OK)} \land \text{(Drainage OK)} \land \text{(WB Moderate or High)}$$

Where:

  • Slope OK: Slope < 10° (flat terrain for uniform water management)
  • Drainage OK: Drainage class ≤ 3 (excludes poorly-drained clay soils)
  • WB Moderate or High: Suitability class ≥ 2 (minimum water balance support)

3.2 Constraint Importance Analysis

Quantifies which constraints most limit feasibility:

  • What % of area fails slope criterion?
  • What % fails drainage criterion?
  • What % fails water balance criterion?
  • How many fail all three?

Identifies whether extension programs should focus on:

  • Training (water management knowledge)
  • Structural change (terracing for steep areas)
  • Subsidy reform (encourage in favorable zones)

4. Spatial Analysis

4.1 Fragmentation Index

Spatial clustering of suitable areas affects extension feasibility. Highly fragmented suitable zones require higher per-unit-area extension costs.

Fragmentation Metric:

$$F = \frac{\bar{A}}{A_{max}}$$

Where:

  • $\bar{A}$ = Mean patch size (hectares)
  • $A_{max}$ = Maximum patch size (hectares)
  • $F \in [0, 1]$, higher = more fragmented

Interpretation:

  • $F = 0.01$: One large contiguous zone (no fragmentation)
  • $F = 0.5$: Patches at half max size (moderate fragmentation)
  • $F = 1.0$: All patches equal size (maximal fragmentation)

4.2 Regional Breakdown

Aggregates suitability statistics by administrative regions:

For each region $r$: $$S_r = \frac{\text{# pixels: class 3 or 2 in region } r}{\text{# rice pixels in region } r}$$

Identifies priority regions for Phase 1 implementation (highest suitability + lowest fragmentation).

4.3 Vietnam vs Japan Comparison

Compares the 3.8× fragmentation ratio to quantify transportability costs:

$$\text{Cost Multiplier} = 1 + 3.8 \times F$$

  • Vietnam (F ≈ 0.1): Multiplier ≈ 1.4×
  • Japan (F ≈ 0.25): Multiplier ≈ 2.0×

Higher multiplier in Japan indicates extension requires more resources per square kilometer.


5. Threshold Sensitivity Analysis

Since deficit threshold ($T$) represents a management choice, we test robustness across 7 thresholds.

Output: DataFrame with columns:

threshold_mm fraction_suitable suitability_class percentage_suitable
-25 0.75 3 (High) 75%
-50 0.60 2 (Moderate) 60%
-70 0.40 2 (Moderate) 40%
... ... ... ...
-150 0.05 1 (Low) 5%

Interpretation:

  • If suitability "robust" across thresholds → confident recommendation
  • If suitability "cliff" (drops sharply) → sensitive to threshold choice

6. Uncertainty & Limitations

6.1 Data Uncertainties

Source Uncertainty Impact
CHIRPS Rainfall ±10-15% regionally Threshold sweep captures this
MODIS ET ±10% over vegetables Acceptable for relative comparison
SoilGrids ±5-10% clay/sand Drainage classes robust to error
SRTM DEM ±10m elevation Slope classification robust
Rice Map ±5-10% area Boundary fuzziness minor

6.2 Methodological Assumptions

  1. Reference ET approximates paddy ET: MODIS ET0 developed for general vegetation; rice-specific adjustments not applied
  2. Dekad aggregation sufficient: Sub-dekad water stress dynamics not captured
  3. No irrigation simulation: Current modeling assumes rain-fed only
  4. Soil texture static: Assumes 2020 soil map applies to analysis year
  5. Climate stationarity: Historical climate patterns repeat in analysis year

6.3 Not Included

  • Groundwater contribution to water balance
  • Farmer risk tolerance and labor availability
  • Market prices and profitability analysis
  • Pest/disease dynamics with increased water stress
  • Climate change impacts on future suitability

7. Quality Assurance

7.1 Validation Approach

  1. Qualitative: Compare results to agronomy literature on AWD feasibility
  2. Quantitative: Sensitivity analysis (vary thresholds, exclusion windows)
  3. Spatial: Inspect maps for unreasonable artifacts or clustered errors
  4. Temporal: Compare multiple years to assess consistency

7.2 Reproducibility

  • All parameters in config/config.yaml
  • All functions vectorized with input validation
  • Logging at each pipeline stage
  • Type hints for interface clarity
  • Unit tests in /tests/ (future implementation)

8. References

AWD Literature:

  • Lampayan, R.M., et al. (2015). "Adoption and economics of alternate wetting and drying in irrigated rice in Southeast Asia." Field Crops Research, 170, 95-108.
  • Bouman, B.A.M., et al. (2007). "Rice and water." Advances in Agronomy, 92, 187-237.

Remote Sensing Data:

  • Funk, C., et al. (2015). "The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes." Scientific Data, 2, 150066.
  • Running, S.W., et al. (2017). "A continuous satellite-derived measure of global terrestrial primary production." BioScience, 54(7), 547-560.

Geospatial Methods:

  • Hijmans, R.J. (2012). "Cross-validation of species distribution models: removing spatial sorting bias." Ecology, 93(3), 679-688.

9. Glossary

Term Definition
AWD Alternate Wetting and Drying: water management practice alternating flooding and drying
Dekad 10-day period; 36 dekads = 360 days per year
ET0 Reference evapotranspiration (standardized to grass/alfalfa)
Percolation Water loss through soil downward movement (mm/day)
Fragmentation Spatial dispersion of suitable patches; higher = more scattered
Suitability Binary classification: suitable or unsuitable for AWD
Threshold Critical water deficit level (negative mm) defining suitability boundary

Document Version: 1.0
Last Updated: January 2026
Corresponding Code: src/water_balance/__init__.py, src/spatial_analysis/__init__.py