Search

Many applications in science and engineering involve data defined at specific geospatial locations, which are often modeled as random fields. The modeling of a proper correlation function is essential for the probabilistic calibration of the random fields, but traditional methods were developed with the assumption to have observations with evenly spaced data. Available methods dealing with irregularly spaced data generally require either interpolation or computationally expensive solutions. Instead, we propose a simple approach based on least square regression to estimate the autocorrelation function. We first tested our methodology on an artificially produced dataset to assess the performance of our method. The accuracy of the method and its robustness to the level of noise in the data indicate that it is suitable for use in realistic problems. In addition, the methodology was used on a major application, the modeling of animal species connected with zoonotic diseases. Understanding the population dynamics of reservoirs of zoonotic diseases, such as bats, is a crucial first step to predict and prevent potential spillover of deadly viruses like Ebola. Due to the limited data on bats across Africa, their density and migrations can only be studied with probabilistic numerical models based on samples of the ecological bare carrying capacity (K0$$ {K}_0 $$). For this purpose, the bare carrying capacity was modeled as a random field and its statistics calibrated with the available data. The bare carrying capacity of bats was found to be denser in central Africa. This is because climatic and environmental conditions are more suitable for the survival of bats. The proposed methodology for random field calibration was shown to be a promising approach, which can cope with large gaps in data and with complex applications involving large geographical areas and high resolution. [ABSTRACT FROM AUTHOR]