What are the best practices for variography in resource estimation?

When it comes to sampling and large data sets it is easy to get swept up in the pure mathematics. You must not forget the geology! It is important for a resource geologist to get some face time with the rocks. Does what you observe in the mining face reflect acurately what the data base is saying? Are there sampling biases, improper methodologies, outliers, bad samples etcetera. Resource estimation is as much a game of building confidence as it is ensuring the overall integrity of the reserve.

Determining the optimal number of domains is of paramount importance. This determination can be guided by geological knowledge and internal clustering validation metrics. Ensuring that established domains are stationary is imperative. The decision of stationarity stands as a foundational prerequisite for geostatistical estimation. Therefore, statistics and variography must substantiate the definition of these domains. Once the domains are defined, understanding the individual contributions of each variable to these domains becomes essential. This knowledge can significantly enhance the interpretation of the domains. This can be achieved through machine learning techniques.

It is required to transform the data before processing to avoid getting outliers in the scenario where it is sparsely distributed like gold grade

while creating experimental variogram, it is always a good practice to check the nugget effect in all three orthogonal directions

When fitting the experimental variogram, it is crucial to remember that the experimental variogram is merely an estimator and not the ground truth. Attempting to achieve a perfect fit between the experimental variogram and a variogram model does not automatically ensure improved predictive performance. This is precisely why conducting sensitivity analysis and cross-validation is paramount when determining the best-fitting variogram model.

The outcome of an experimental variogram, or an attempt to attain a good fit with a corresponding model variogram, shouldn't be treated as the exact representation of spatial parameter variation in a geological domain (as such predictions are susceptible many forms of biases). Through multiple data sensitivity testing and cross validation with ground truth (e.g. Historical/ core data) there is a reduction in uncertainty of outcomes for geologically realistic predictions.

Indeed, variography is always recommended to observe what does the data tell us and how to bridge the gap with concepts and regional information. Disregarding the field case, depositional environment, well sampling or facies modeling methods.

You are missing Data Declustering Without declusturing, variogram models may be biased! Clustered data gives biased variogram results A three dimensional cell declustering gives the best results. Steps: 1: Cell declustering 2:Use weighted data for the optimized cell size for variogram map for anisotropy analysis 3:Use weighted data for calculating the anisotropic variograms 4: Model these variograms (Please don't use these weighted data on kriging.)