viscover

viscover implements empirical Bayes small-area estimation for zero-inflated, skewed agricultural variables using a zero-inflated lognormal mixed-effects model to predict quantities such as RUSLE2 sheet and rill erosion.

Key Features:

• Zero-Inflated Lognormal Mixed Effects Model: Employs a two-part unit-level model combining a lognormal model for positive RUSLE2 responses and a logistic mixed-effects model for the binary indicator of nonzero responses.
• Correlated Random Area Effects: Models correlated random effects across geographic areas (e.g., counties) to capture the association between probability of nonzero responses and means among positive values.
• Empirical Bayes Small Area Predictors: Produces empirical Bayes predictors for small-area estimation and uses a bootstrap estimator for mean squared error (MSE) to assess prediction uncertainty.
• Integration with Auxiliary Data: Constructs auxiliary covariates by overlaying satellite-derived land cover maps with geographic soil property databases to support prediction across areas such as South Dakota counties.

Scientific Applications:

• Small-area erosion estimation: Estimates RUSLE2 sheet and rill erosion at county or other small-area levels for agricultural surveys and conservation assessments.
• Conservation Effects Assessment: Applies to datasets such as the Conservation Effects Assessment Project (CEAP) to quantify soil erosion and inform land management and policy decisions.

Methodology:

Combines unit-level lognormal and logistic mixed-effects models for zero-inflation and skewness; incorporates correlated random area effects; applies empirical Bayes prediction with a bootstrap MSE estimator; constructs covariates via overlay of satellite-derived land cover and soil property databases.