This study developed a theoretical model for the determination of the Coke Anisotropy Quotient (CAQ) of semi-coke from the properties of its precursor coal. This is an useful parameter to define the resistance and reactivity of semi-coke in the blast furnace. For 36 semi-coke samples, a textural analysis was performed alongside a fluidity test to determine the real CAQ. The main textures observed were: isotropic and circular for high volatile bituminous coals (HVB); lenticular and fine ribbons for the medium volatile bituminous coals (MVB); and medium and thick ribbons for the low volatile bituminous coals (LVB). The CAQ varied in a range from 1 to 11. A principal component analysis (PCA) and multiple regression allowed to discriminated the importance of certain coal properties, in determining the CAQ to be recognized and to estimate parameters of the mathematical model. The statistical analysis suggested that CAQ can be best predicted from the fluidity, volatile matter, and Ro of the parent coals. The veracity of this model result was then tested using a second dataset from Poland. This work optimizes the usefulness of standard datasets in the prediction of CAQ's offering a means of quality control that could be implemented in Colombian coke production.

Context: This paper proposes a method for the prediction of monthly precipitation in the department of Boyacá using models based on deep neural networks (DNNs). These approaches have achieved significant improvements in prediction accuracy when compared to traditional methods. Method: Data with a spatial resolution of 0.5° were extracted from CHIRPS 2.0 and subsequently preprocessed for the implementation of two approaches based on recurrent neural networks (RNNs) with long short-term memory (LSTM) and ConvLSTM architectures, aiming to provide accurate predictions of monthly precipitation in the studied region. Objectives: The goal of this time series analysis is to predict monthly precipitation and develop accurate models that can forecast future rainfall patterns based on historical data. This aids in water resource management and agricultural planning, as well as in mitigating the impacts of droughts or floods. Results: According to the results obtained, the LSTM model stands out for its robustness in terms of performance metrics, such as a lower mean squared error, a lower root mean squared error, and a coefficient of determination closer to 1. This demonstrates its higher accuracy compared to the ConvLSTM model.Conclusions: Deep learning models, especially RNNs with LSTM, are effective tools for predicting crucial climate data.

Monthly precipitation prediction in mountainous terrain is challenging due to orographic forcing and complex atmospheric interactions. We present a systematic framework evaluating hybrid deep learning architectures for the Colombian Andes (Boyacá, 3,965 grid cells, 145-5490 m elevation, forecast horizons from 1 to 12 months). In a companion study, Convolutional Long Short-Term Memory (ConvLSTM) models achieved R²=0.628 with 79K parameters, while our Graph Neural Network with Temporal Attention (GNN-TAT) matched this accuracy with a compact 98K-parameter architecture through explicit topographic encoding. Late fusion via Ridge regression (R²=0.672) achieves a 6.2% improvement over individual models, while early fusion stacking (R²=0.212) and State Space Models (R²=0.200) fail. Elevation-stratified evaluation reveals 16% performance degradation from valleys to high elevations, concentrated in the 1,500-2,800 m orographic transition band, and variogram analysis confirms Late Fusion produces 29% lower spatial error variance. Sub-cell Digital Elevation Model (DEM) feature engineering (10 elevation deciles, 6 Principal Component Analysis components, 5 statistical descriptors) consistently degrades all architectures by 5.6-20.1%, establishing that categorical elevation encoding outperforms continuous terrain statistics under data scarcity. Non-parametric statistical testing (Friedman, Nemenyi, Mann-Whitney) confirms that the top three individual architectures are statistically indistinguishable (p=0.052), supporting the existence of a predictability ceiling for this region. Fusion strategy matters more than architecture choice; documenting negative results advances understanding as effectively as reporting successes.

Forecasting monthly precipitation in mountainous terrain poses challenges that push conventional deep learning approaches to their limits: convective processes operate locally while orographic effects span entire drainage basins. We compare three architecture families on precipitation prediction across the Colombian Andes: ConvLSTM (convolutional recurrent), FNO-ConvLSTM (spectral–temporal), and GNN-TAT (graph attention LSTM). Using CHIRPS v2.0 and SRTM topography for Boyacá department (61 × 65 grid, 3965 nodes), we evaluate 39 configurations across feature bundles (BASIC, KCE elevation clusters, and PAFC autocorrelation lags) and horizons from 1 to 12 months. GNN-TAT matches ConvLSTM accuracy ($R^2$: 0.628 vs. 0.642; RMSE: 82.29 vs. 79.40 mm) with 95% fewer parameters (∼98K vs. 2.1M). Across configurations, GNN-TAT produces a lower mean RMSE (92.12 vs. 112.02 mm; $p=0.015$) and a 74.7% lower variance. The explicit graph structure, with edges weighted by elevation similarity, appears to reduce sensitivity to hyperparameter choices. Pure FNO struggles with precipitation’s spatial discontinuities ($R^2=0.206$), though adding a ConvLSTM decoder recovers much of the lost skill ($R^2=0.582$). Elevation clustering improves GNN-TAT significantly ($p=0.036$) but not ConvLSTM, suggesting that feature design should match the spatial encoding paradigm. ConvLSTM achieves peak accuracy on local patterns; GNN-TAT provides robust predictions with interpretable spatial reasoning. These complementary strengths motivate stacking ensembles that combine grid-based and graph-based representations.

Monthly precipitation prediction in mountainous terrain is challenging due to orographic forcing and complex atmospheric interactions. We present a systematic framework evaluating hybrid deep learning architectures for the Colombian Andes (Boyacá, 3,965 grid cells, 145-5490 m elevation, forecast horizons from 1 to 12 months). In a companion study, Convolutional Long Short-Term Memory (ConvLSTM) models achieved R²=0.628 with 79K parameters, while our Graph Neural Network with Temporal Attention (GNN-TAT) matched this accuracy with a compact 98K-parameter architecture through explicit topographic encoding. Late fusion via Ridge regression (R²=0.672) achieves a 6.2% improvement over individual models, while early fusion stacking (R²=0.212) and State Space Models (R²=0.200) fail. Elevation-stratified evaluation reveals 16% performance degradation from valleys to high elevations, concentrated in the 1,500-2,800 m orographic transition band, and variogram analysis confirms Late Fusion produces 29% lower spatial error variance. Sub-cell Digital Elevation Model (DEM) feature engineering (10 elevation deciles, 6 Principal Component Analysis components, 5 statistical descriptors) consistently degrades all architectures by 5.6-20.1%, establishing that categorical elevation encoding outperforms continuous terrain statistics under data scarcity. Non-parametric statistical testing (Friedman, Nemenyi, Mann-Whitney) confirms that the top three individual architectures are statistically indistinguishable (p=0.052), supporting the existence of a predictability ceiling for this region. Fusion strategy matters more than architecture choice; documenting negative results advances understanding as effectively as reporting successes.