Crop diversification and rainfed water-risk management: Evidence from smallholder farms in China’s Loess Plateau

Water-related climate shocks threaten rainfed smallholders through root-zone moisture deficits, runoff, erosion, and income disruption. This study examines whether crop diversification is associated with agronomic recovery, livelihood recovery, and interannual income stability, while separating temporally ordered recovery estimates from a secondary stability analysis with weaker timing. A multistage stratified survey conducted from July to November 2025 produced 1580 valid household questionnaires in eight counties and 56 villages of the hilly-gully Loess Plateau. The recovery sample comprises 894 households for which a station-linked event was meteorologically ranked among respondent-confirmed agriculturally relevant years, crop allocation was measured in the immediately preceding agricultural year, and a complete first post-event year was available. Drought exposure is represented by the minimum March-October SPEI-6, with SPEI-3 and SPEI-9 sensitivity analyses; extreme rainfall is represented by station-specific R95p-style days and Rx1day, with 90th- and 99th-percentile checks. County fixed-effect OLS models with village-clustered standard errors are complemented by fractional logit, beta regression, weighting, recall-window, and land-size specifications. A one-standard-deviation increase in the Simpson diversification index is associated with 5.1 %age points higher yield recovery (95% CI: 4.5–5.6), 4.4 points higher income recovery (3.9–4.9), and 2.8 points lower income volatility (2.4–3.1 points lower). The volatility result is secondary because the 2024 diversification measure overlaps the 2020–2024 income window. Diversification-by-shock interactions are interpreted as effect modification rather than confirmation of a causal mechanism. Subgroup associations are smaller for female-headed households. Overall, the estimates are modest, context-dependent associations that require validation with prospective household panels, plot-level soil-water measurements, and process-based crop models.