I need to determine which temperature measurements are outliers at their depth. Anselin Local Moran's I accounts for 2D autocorrelation, but how do I account for both variables?
Be sure that your interpretation of a bivatiate LISA is correct. The bivariate LISA is the crossproduct of the standardized values of one variable at location i (temperature) with those of the average neighboring values of the second variable (depth). You are testing whether localized correlations between values at location i and those of its neighbors are significantly different from what you would observe under spatial randomness. This would mean that a lo-hi outlier indicates that standardized temperature are significantly correlated with standardized depth.