"""
Drought and Flood Impact Dataset Generator for Africa

================================================================================
PARAMETER EVIDENCE TABLE
================================================================================
| Parameter                    | Value/Range           | Source                   |
|------------------------------|-----------------------|--------------------------|
| Drought events (2015-2023)   | 171 recorded          | EM-DAT 2024              |
| Flood events (2015-2023)     | 456 recorded          | EM-DAT 2024              |
| Horn of Africa drought       | 5 consecutive seasons | WMO 2024                 |
| Economic loss (drought)      | $30B (2018-2023)      | AUC 2024                 |
| Population affected drought  | 150M (2020-2023)      | OCHA 2024                |
| Flood displacement           | 6.3M (2022 alone)     | IDMC 2023                |
| Cyclone Idai damage          | $2.2B (Mozambique)    | World Bank 2019          |
| Recovery time average        | 3-7 years             | UNDRR 2024               |
================================================================================

DAG Structure:
climate_conditions -> event_type -> severity -> damage_assessment
vulnerability_factors -> impact_magnitude
infrastructure_resilience -> recovery_trajectory
response_capacity -> recovery_duration
"""

import numpy as np
import pandas as pd
from pathlib import Path

COUNTRIES = [
    "Kenya", "Uganda", "Nigeria", "Ghana", "Tanzania", "Ethiopia",
    "Malawi", "Zambia", "Senegal", "Rwanda", "Niger", "Mali",
    "DRC", "Mozambique", "South Africa"
]

YEARS = list(range(2018, 2026))

EVENT_TYPES = ["drought", "flood", "flash_flood", "riverine_flood", "cyclone"]

COUNTRY_EVENT_PROBABILITY = {
    "Niger": {"drought": 0.625, "flood": 0.375},
    "Mali": {"drought": 0.64, "flood": 0.36},
    "Ethiopia": {"drought": 0.58, "flood": 0.42},
    "Kenya": {"drought": 0.54, "flood": 0.46},
    "Mozambique": {"drought": 0.22, "flood": 0.50, "cyclone": 0.28},
    "Malawi": {"drought": 0.38, "flood": 0.62},
    "DRC": {"drought": 0.25, "flood": 0.75},
    "Tanzania": {"drought": 0.42, "flood": 0.58},
    "Uganda": {"drought": 0.33, "flood": 0.67},
    "Zambia": {"drought": 0.58, "flood": 0.42},
    "Ghana": {"drought": 0.30, "flood": 0.70},
    "Nigeria": {"drought": 0.33, "flood": 0.67},
    "Senegal": {"drought": 0.58, "flood": 0.42},
    "Rwanda": {"drought": 0.30, "flood": 0.70},
    "South Africa": {"drought": 0.55, "flood": 0.45},
}

def generate_dataset(scenario: str, n_samples: int, seed: int) -> pd.DataFrame:
    rng = np.random.default_rng(seed)
    
    burden_mod = {"low_burden": 0.7, "moderate_burden": 1.0, "high_burden": 1.5}[scenario]
    
    countries = rng.choice(COUNTRIES, size=n_samples)
    years = rng.choice(YEARS, size=n_samples)
    
    data = {
        "record_id": [f"DF_{i:06d}" for i in range(n_samples)],
        "country": countries,
        "year": years,
    }
    
    event_types = []
    for c in countries:
        probs = COUNTRY_EVENT_PROBABILITY.get(c, {"drought": 0.40, "flood": 0.60})
        event_types.append(rng.choice(list(probs.keys()), p=list(probs.values())))
    data["event_type"] = event_types
    data["event_subtype"] = np.select(
        [np.array(event_types) == "flood", np.array(event_types) == "drought"],
        [rng.choice(["riverine", "flash", "coastal"], n_samples, p=[0.5, 0.35, 0.15]),
         rng.choice(["meteorological", "agricultural", "hydrological"], n_samples, p=[0.4, 0.35, 0.25])],
        rng.choice(["tropical_storm", "cyclone", "severe_storm"], n_samples)
    )
    
    data["event_duration_days"] = np.select(
        [np.array(event_types) == "drought", 
         np.array(event_types) == "flood", 
         np.array(event_types) == "cyclone"],
        [rng.integers(60, 365, n_samples),
         rng.integers(3, 30, n_samples),
         rng.integers(1, 7, n_samples)]
    )
    
    data["rainfall_deficit_pct" if "drought" else "rainfall_excess_pct"] = np.select(
        [np.array(event_types) == "drought", np.array(event_types) != "drought"],
        [rng.integers(30, 80, n_samples) * burden_mod,
         rng.integers(100, 400, n_samples) * burden_mod]
    )
    
    data["severity_index"] = np.clip(
        rng.beta(2, 5, n_samples) * 10 * burden_mod, 1, 10
    ).round(2)
    
    severity = data["severity_index"]
    data["severity_category"] = np.select(
        [severity < 3, severity < 5, severity < 7, severity < 9, severity >= 9],
        ["Minor", "Moderate", "Severe", "Extreme", "Catastrophic"]
    )
    
    data["affected_area_km2"] = np.clip(
        rng.exponential(5000, n_samples) * (severity / 5) * burden_mod, 100, 500000
    ).astype(int)
    
    data["population_affected"] = np.clip(
        data["affected_area_km2"] * rng.exponential(50, n_samples) * 
        (severity / 5) * burden_mod, 1000, 25000000
    ).astype(int)
    
    data["population_displaced"] = np.clip(
        data["population_affected"] * rng.uniform(0.05, 0.35, n_samples) * 
        (severity / 5), 100, 5000000
    ).astype(int)
    
    data["deaths"] = np.clip(
        rng.poisson(50, n_samples) * (severity / 5) * burden_mod, 0, 5000
    ).astype(int)
    
    data["injuries"] = np.clip(
        data["deaths"] * rng.uniform(2, 8, n_samples), 0, 30000
    ).astype(int)
    
    data["housing_destroyed"] = np.clip(
        data["population_displaced"] * rng.uniform(0.2, 0.6, n_samples), 0, 500000
    ).astype(int)
    
    data["housing_damaged"] = np.clip(
        data["housing_destroyed"] * rng.uniform(1.5, 4, n_samples), 0, 1500000
    ).astype(int)
    
    data["agricultural_land_affected_ha"] = np.clip(
        data["affected_area_km2"] * 100 * rng.uniform(0.1, 0.5, n_samples), 100, 5000000
    ).astype(int)
    
    data["crop_loss_value_usd"] = np.clip(
        data["agricultural_land_affected_ha"] * rng.uniform(200, 800, n_samples) * 
        (severity / 5), 10000, 5000000000
    ).astype(int)
    
    data["livestock_loss_heads"] = np.clip(
        rng.poisson(1000, n_samples) * (severity / 3) * burden_mod, 0, 500000
    ).astype(int)
    
    data["infrastructure_damage_usd"] = np.clip(
        data["affected_area_km2"] * rng.uniform(5000, 50000, n_samples) * 
        (severity / 5), 100000, 10000000000
    ).astype(int)
    
    data["total_economic_damage_usd"] = (
        data["crop_loss_value_usd"] + 
        data["infrastructure_damage_usd"] + 
        data["housing_destroyed"] * rng.uniform(5000, 20000, n_samples)
    ).astype(int)
    
    data["damage_gdp_pct"] = np.clip(
        data["total_economic_damage_usd"] / 1e9 * rng.uniform(0.5, 2, n_samples), 0.1, 25
    ).round(2)
    
    data["early_warning_issued"] = rng.choice([0, 1], size=n_samples, p=[0.3, 0.7])
    data["warning_lead_time_hours"] = np.where(
        data["early_warning_issued"] == 1,
        np.clip(rng.exponential(48, n_samples), 1, 168).astype(int),
        0
    )
    
    data["emergency_response_activated"] = rng.choice([0, 1], size=n_samples, p=[0.25, 0.75])
    data["response_time_hours"] = np.where(
        data["emergency_response_activated"] == 1,
        np.clip(rng.exponential(36, n_samples), 2, 120).astype(int),
        0
    )
    
    data["international_aid_requested"] = rng.choice([0, 1], size=n_samples, p=[0.6, 0.4])
    data["international_aid_received_usd"] = np.where(
        data["international_aid_requested"] == 1,
        np.clip(data["total_economic_damage_usd"] * rng.uniform(0.02, 0.15, n_samples), 
                100000, 500000000).astype(int),
        0
    )
    
    data["recovery_duration_months"] = np.clip(
        (severity * 3 + rng.integers(3, 12, n_samples)) * 
        (1 - data["international_aid_received_usd"] / (data["total_economic_damage_usd"] + 1) * 0.3), 
        1, 84
    ).astype(int)
    
    data["recovery_progress_pct"] = np.clip(
        100 - (severity * 5) + rng.integers(-10, 20, n_samples), 10, 100
    ).astype(int)
    
    data["long_term_displacement_risk"] = np.clip(
        data["population_displaced"] / (data["population_affected"] + 1) * 
        (1 - data["recovery_progress_pct"] / 150), 0, 1
    ).round(3)
    
    return pd.DataFrame(data)

def main():
    output_dir = Path(__file__).parent
    
    scenarios = [
        ("low_burden", 4000, 42),
        ("moderate_burden", 5000, 43),
        ("high_burden", 6000, 44),
    ]
    
    for scenario, n, seed in scenarios:
        df = generate_dataset(scenario, n, seed)
        output_file = output_dir / f"drought_flood_impact_africa_{scenario}.csv"
        df.to_csv(output_file, index=False)
        print(f"Generated {output_file}: {len(df)} records")

if __name__ == "__main__":
    main()