TUNALI, C. Statistická analýza dat z geologického průzkumu [online]. Brno: Vysoké učení technické v Brně. Fakulta strojního inženýrství. 2025.

Posudek vedoucího

Hübnerová, Zuzana

The thesis aims to analyse the results of a geological survey, namely the daily CO measurements taken at a single location in South Moravia, using the Box-Jenkins methodology. Supervision took place through regular meetings focused on methods and R programming. Ms. Tunali worked with challenging real-world data, including missing values and irregular time measurements. The two applied methods required careful coding. Unfortunately, writing progress slowed later on. The scope of the thesis is reasonable, but with more time dedicated to the compilation phase, there could have been space for additional analyses—such as a sensitivity analysis of the cut-off point in Figure 4.42—which might have provided improved the final regression model. The original aim to analyse data from multiple stations was revised in favour of a deeper comparison of two modelling approaches, which provided useful insight into their properties. While both methods are known for single-covariate modelling, the thesis adapted them for multiple covariate series. The main challenge lay in identifying significant lags and parameters of stationary residual procesess in the final regression model. After discussion, Ms. Tunali interpreted the results well and showed a clear effort to understand the details. The proposed algorithms could be useful in other research fields where stochastic models are needed to describe the behaviour of response time series. The text is generally acceptable but contains nonstandard citation formats in some cases. Additionally, certain statements are inaccurate, e.g. page 62, the SARIMA parametrization is incorrect, and equation (4.50) does not correspond to an AR(2) process. Overall, I recommend a final grade of C.

Posudek oponenta

Hrabec, Pavel

This thesis aims to analyze the dependency between CO concentrations and various meteorological variables using advanced time-series modeling techniques. The author applies two frameworks: a transfer function model with prewhitening and a multivariate lagged regression model without prewhitening. While the work demonstrates familiarity with time-series methodology and offers some valid insights into model behavior, several conceptual, formal, and practical issues limit the overall impact and rigor of the thesis. Most noteworthy problems are summarized further. Key modeling decisions—such as choosing AR(*), MA(*) structures or 21 (22) day MA filter for smoothing of cross-correlated time series are presented without statistical testing, or even brief model diagnostics. There is no exploration of alternative configurations or robustness checks. Critical variables such as cloud and gust are used without defined units or physical explanation. The use of citation styles is inconsistent (author–year and numeric references appear mixed), and formatting of equations and figures varies in quality. I thoroughly disagree with a statement in conclusion that introduced models were compared. Only the ability to “explain” (cross-)correlation structure is compared. Thesis gives no estimate for the variability of error term, model fitting quality or at least graphical comparisons of fitted and observed data. While the thesis addresses a relevant and challenging topic with appropriate tools, it falls short of delivering a complete and polished analysis. Several parts suffer from insufficient justification, ambiguous notation, and missing validation steps. Although the work shows mathematical competence, it lacks the depth, interpretability, and practical readiness expected from a top-tier applied mathematics thesis. Judging by a „rushed ending“ i would guess that a few additional weeks could significantly improve presented text, and possibly allowed proper comparison of introduced models and rigorous explanation for some modeling choices. I recommend the thesis for defense with an overall grade D.

eVSKP id 165579