Introduction: The nomadic livestock sector is a pillar of food security, employment, and economic stability in the rural and nomadic regions of Iran, particularly in Kerman Province. In this arid and semi-arid region, the livelihoods of nomadic communities are intrinsically linked to natural rangelands. However, this dependence renders them highly vulnerable to climate fluctuations and recurring droughts. In recent years, Kerman has experienced consecutive and severe drought periods, significantly reducing the quality and quantity of rangeland vegetation, which serves as the primary fodder source for traditional livestock farming. To mitigate these risks, agricultural insurance has long been employed as a safety net. Nevertheless, traditional indemnity-based insurance systems in Iran face substantial challenges, including high administrative costs for field assessments, moral hazard, adverse selection, and significant delays in claims settlement.
To overcome these inefficiencies, Index-Based Livestock Insurance (IBLI) has emerged as a promising innovation. Unlike traditional insurance, IBLI bases payouts on an objective external indicator—such as rainfall or vegetation indices derived from satellite data—rather than actual loss verification. While IBLI offers transparency and speed, its primary weakness is "basis risk", which occurs when the index payout does not perfectly correlate with the actual losses experienced by farmers. To minimize basis risk, it is critical to identify the most accurate proxy for livestock mortality and to determine the precise "trigger point" or threshold at which payments should be activated. While indices like the Normalized Difference Vegetation Index (NDVI) are widely used, few studies have systematically compared multiple vegetation indices to find the optimal non-linear threshold for arid environments. This study aims to fill this gap by employing a Threshold Regression Model to identify the most effective vegetation index—among NDVI, the Vegetation Condition Index (VCI), and the NDVI Anomaly Index (NDVIA)—for predicting livestock mortality and designing efficient insurance contracts in Kerman Province.
Materials and Methods: The study area is Kerman Province, a region characterized by extensive nomadic livestock farming and high climatic variability. The research utilizes a dataset covering ten years from 2013 to 2023 (1392 to 1402). The dependent variable is the seasonal mortality rate of small livestock (sheep and goats), with data obtained from the Veterinary Organization, Agricultural Insurance Fund, and General Administration of Nomadic Affairs of Kerman.
The core independent variables are three remotely sensed vegetation indices derived from MODIS and Landsat 8 satellite imagery via the Google Earth Engine platform:
1. NDVI (Normalized Difference Vegetation Index): A standard measure of vegetation health based on near-infrared and red band reflectance.
2. VCI (Vegetation Condition Index): Compares the current NDVI to its historical minimum and maximum, effectively highlighting drought stress relative to local potential.
3. NDVIA (NDVI Anomaly Index): Measures the deviation of the current NDVI from the long-term mean, normalized by the historical standard deviation, to identify statistical anomalies.
To isolate the effect of vegetation on mortality, the study also incorporated control variables, including the Standardized Precipitation Index (SPI) for drought intensity, temperature changes (T) representing heat stress, and livestock vaccination rates (V) to account for veterinary interventions.
The analytical approach employs a Threshold Regression (TR) model. Unlike linear models that assume a constant relationship between variables, the TR model allows for regime switching. It tests for the existence of non-linear relationships and identifies specific threshold values (γ) that split the data into "normal" and "crisis" regimes. This is crucial for insurance design, as livestock mortality is hypothesized to remain stable until vegetation drops below a critical survival level, after which mortality spikes exponentially. The study applied Phillips-Perron unit root tests for stationarity, linearity tests to justify the use of TR, and Akaike (AIC) and Schwarz (BIC) information criteria to select the best-fitting model.
Results and Discussion
Descriptive statistics revealed that the average livestock mortality rate during the study period was 2.31%, with significant seasonal fluctuations ranging from 1% to 5%, reflecting the impact of drought cycles. The linearity tests strongly rejected the null hypothesis of a linear relationship for all three vegetation indices, confirming that the relationship between rangeland health and livestock survival is inherently non-linear and regime-dependent.
Among the three models tested, the model based on the NDVIA (NDVI Anomaly Index) demonstrated superior performance. It achieved the highest coefficient of determination (R²=0.95), compared to 0.92 for NDVI and 0.89 for VCI. Furthermore, the NDVIA model yielded the lowest AIC and BIC values, indicating it provides the best balance between model complexity and explanatory power.
The TR analysis for the NDVIA model identified a critical threshold (γ) of -0.03. This value represents the "tipping point" for the ecosystem. The results distinguished two distinct regimes:
1. Regime 1 (Above Threshold/Normal Conditions): When NDVIA is above -0.03, the relationship between vegetation and mortality is negative but moderate (slope coefficient of -2.40). In this state, livestock are generally resilient, and minor fluctuations in forage do not lead to mass deaths.
2. Regime 2 (Below Threshold/Crisis Conditions): When NDVIA drops below -0.03, the slope coefficient dramatically increases to 63.32. This indicates a structural break which mortality rates skyrocket in response to even small further declines in vegetation. This steep slope quantitatively captures the collapse of herd resilience due to starvation.
The control variables offered further insights. The Standardized Precipitation Index (SPI) and temperature (T) showed significant correlations with mortality in the crisis regime, confirming that heat stress and lack of water exacerbate the effects of forage scarcity. Interestingly, the vaccination rate variable showed low statistical significance in the threshold models (P>0.1). This suggests that during severe environmental crises (drought and famine), veterinary interventions are insufficient to prevent mortality caused primarily by starvation and dehydration. The findings imply that "climate factors" dominate "health management factors" during extreme events.
The superior performance of NDVIA over NDVI and VCI is attributed to its ability to capture relative anomalies. In an arid region like Kerman, where absolute vegetation cover is naturally low, raw NDVI values might not clearly distinguish between a normal dry season and a catastrophic drought. NDVIA, by accounting for historical deviations and volatility, provides a more sensitive signal of deviations from the "norm", making it the most reliable proxy for the actual risk faced by nomads.
Conclusions: This study concludes that index-based insurance offers a viable solution to the challenges of traditional livestock insurance in Kerman Province, provided the correct index and thresholds are utilized. The research demonstrates that the NDVIA is the most accurate predictor of livestock mortality among the tested indices. The identified threshold of -0.03 serves as a scientifically backed "trigger point" for insurance payouts.
The implications for policy and insurance design are significant. Relying on linear models or simple NDVI averages may lead to basis risk, where payouts do not align with actual catastrophic losses. By adopting the non-linear NDVIA model, the Agricultural Insurance Fund can design contracts that trigger payments exactly when the ecosystem crosses the critical threshold from manageable stress to catastrophic mortality. This ensures that funds are released when nomads need them most-before the mortality rate climbs the steep slope identified in the "crisis regime".
The study recommends that future insurance schemes in Kerman should transition to using NDVIA as the primary basis for contracts. Additionally, combining vegetation indices with temperature and rainfall data could further refine the model. While the study highlighted the dominance of climatic factors over veterinary interventions during droughts, it suggests moving towards a hybrid "climate-health" insurance model in the future. Integrating socio-economic factors, such as fodder prices and market access during droughts, into future models could further enhance the precision of payout thresholds and ultimately improve the resilience of the nomadic communities against climate change.