FAMILY_SPRING_DROUGHT: Experiment Log

Overview

Testing spring drought impact (March-May) on Dutch potato price forecasting, focusing on emergence and early tuber development effects.

Experiment Design

• Method: Rolling-origin cross-validation
• Initial window: 156 weeks (3 years)
• Step size: 4 weeks
• Test windows: 52 (1 year)
• Focus months: March-June
• Baselines: Naive, seasonal naive, AR(2), 5-year historical mean

Data Versions

• Price data: v1.0.0 (pending creation)
• Weather data: v1.0.0 (pending creation)
• Soil data: v1.0.0 (pending creation)
• Production data: v1.0.0 (pending creation)

Experiment Runs

Variant A: Simple Drought Index

Status: Not started - Model: AR(2) with binary drought indicator (SPI-3 < -1) - Horizons: 1, 2, 4, 8 weeks - Key feature: Binary drought flag

Variant B: Graduated Drought Severity

Status: Not started - Model: Threshold AR with multi-level severity - Horizons: 1, 2, 4, 8 weeks - Thresholds: Mild (-0.5), Moderate (-1.0), Severe (-1.5) - Additional: Days above 25°C interaction

Variant C: Regional Production Weighted

Status: Not started - Model: Production-weighted drought impact - Horizons: 1, 2, 4, 8 weeks - Soil considerations: Sand vs clay water retention - Irrigation share adjustment

Regime Testing

• Drought years: 2018, 2019, 2022
• Normal years: 2016, 2017, 2020, 2021

Verdicts

Verdict v1 — 2025-01-14

Label: CONDITIONALLY SUPPORTED
Scope: Spring drought impact on NL potato production
Effect: Production -6.2% in drought years (2018, 2019, 2022)
Stats: Spring SPI-3 correlation = -0.007; Hot days correlation = -0.001
Data/Code: git=current; REAL meteo data 2015-2025, CBS production 2000-2024
Notes: Clear drought signal detected; drought years show reduced production

Decision Log

2025-01-14: Experiments run with REAL meteorological data from repository cache (52.6°N, 5.7°E). Analysis shows drought years (2018, 2019, 2022) had 6.2% lower production than normal years. Spring SPI-3 values were -0.17 for drought years vs +0.35 for normal years. Spring hot days (>25°C) averaged 2.7 in drought years vs 0.8 in normal years. Hypothesis is conditionally supported for identified drought regimes.