Information on the harvest date of crops can help with logistics management in the agricultural industry, planning machinery operations and also with yield prediction modelling. In this study, the determination and prediction of harvest dates for different crops were performed by applying machine learning techniques on C-band synthetic aperture radar (SAR) data. Ground truth data were provided for the Vojvodina region (Serbia), an area with intensive agricultural production, considering winter wheat, maize and soybean fields with exact harvest dates, for the period 2017–2020, including 592 samples in total. Data from the Sentinel-1 satellite were used in the study. Time series of backscattering coefficients for vertical–horizontal (VH) and vertical–vertical (VV) polarisations, both from ascending and descending orbits, were collected from Google Earth Engine. Clustering of harvested and unharvested fields was performed with Principal Component Analysis, multidimensional scaling and t-distributed Stochastic Neighbour Embedding, for initial cluster visualization. It is shown that the separability of unharvested and harvested data in two-dimensional space does not depend on the selected method but more on the crop itself. Support Vector Machine and Multi-layer Perceptron were used as classification algorithms for harvest detection, with the former achieving higher accuracies of 79.65% for wheat, 83.41% for maize and 95.97% for soybean. Finally, regression models were developed for the prediction of the harvest date using Random Forest and the long short-term memory network, with the latter achieving better results: an R2 score of 0.72, mean absolute error of 6.80 days and root mean squared error of 9.25 days, for all crops considered together.

Soil Organic Carbon (SOC) stocks in natural and semi-natural ecosystems remain poorly quantified in intensively cultivated lowland regions, such as Vojvodina, Serbia, which is part of the Pannonian Basin. To address this gap, we developed a general framework for large-scale, multipurpose soil sampling. The study further investigates the potential of machine learning-based regression and classification approaches to predict SOC stocks in forest and grassland ecosystems using diverse land cover and soil-related indicators as key predictors. Spatial clustering for the planning of soil sampling was conducted by combining information from different sources. To implement a systematic and stratified sampling scheme, we followed the LUCAS methodology. Natural and semi-natural forest and grassland areas were delineated using Copernicus LULC data. An exploratory analysis was conducted using climate variables from C3S Copernicus (2015–2024) and soil properties (soil order and type, and silt, sand, and clay proportions) to identify spatial clusters suitable for field sampling. Further, forests and grasslands were clustered separately using an unsupervised K-prototypes approach. Based on this approach, 62 representative locations were identified across forests and grasslands, from which a total of 186 soil samples were collected using composite sampling at three sites per location.Land cover features were collected along 250 m transects at each location, and landscape heterogeneity was quantified using LUCAS-based diversity indicators derived from the same transects. For machine learning–based SOC stock prediction, these indicators were combined with soil descriptors, including soil texture, soil type, and geomorphology-based soil groups, as well as spatial cluster information for forest and grassland areas. SOC stock values were averaged per location, and forest and grassland samples were jointly used in the modeling to capture landscape heterogeneity.Regression modeling aimed to predict continuous SOC stock values, while classification categorized SOC stock into low, medium, and high levels based on thresholds derived from K-means clustering applied to the observed SOC distribution. Among the regression models, Elastic Net achieved the highest performance, with an R² of 0.49 and an RMSE of 13.74 t ha⁻¹, indicating moderate predictive capability given the complexity of SOC stock dynamics and the limited sample size. In contrast, classification models demonstrated higher predictive reliability. Logistic Regression achieved the best performance, with an overall accuracy of 76.9% and a macro F1-score of 77.1%, suggesting that SOC stock can be more robustly distinguished across discrete classes than predicted as a continuous variable. Permutation importance analysis revealed that soil texture was the dominant predictor in both regression and classification models. Overall, the findings highlight the combined importance of soil properties and landscape diversity indicators for SOC stock prediction in natural and semi-natural ecosystems. While continuous SOC stock prediction remains challenging, classification into discrete SOC stock classes provides higher accuracy and more stable performance, suggesting a more reliable framework for SOC stock assessment in heterogeneous landscapes. Independently, this study establishes the first SOC reference framework for natural and semi-natural ecosystems in Vojvodina, providing a conceptual basis for spatial analysis and mapping.

Increasing agricultural production is inevitable in the future since population growth and climate change have led to significant pressure on global food security. One of the ways is to intensify the existing cropland by multi-cropping practice, allowing multiple uses of a single field during one year.  This research aims to identify and map double-cropping land using multi-temporal Sentinel 2 imagery from 2021 and advanced machine learning models. The case study focus is on Bačka and Srem, regions located in the Autonomous Province of Vojvodina, Republic of Serbia. These regions are characterized by fertile land and widespread agriculture production. However, there is a low presence of double-cropping practice due to usually dry summers, but with a tendency to change as the number of irrigation systems increase.Considering the small amount of double-cropping fields, there is a need for direct ground truth data collection. For that reason, the first step was to reduce the area of interest to get insight into the locations of potential double-cropping land. This result was obtained by using the threshold method based on the phenology of crops during the year. The NDVI (Normalized Difference Vegetation Index) time series was utilized to define appropriate thresholds for feature two peak values to discriminate double-cropping within each pixel. The identification of the results was used on-site for collecting ground truth data. Based on the collected data and the analyzed NDVI time series, besides double-crop, three more classes of arable land were distincted and included in the classification: single winter crops, single summer crops and clover. The collected data contained 46 parcels of double crops, 43 single winter crops, 55 single summer crops and 27 parcels of clover. We used time-series images to create a dataset for training the pixel-based Random Forest classification. The results showed a very high overall accuracy of 99% and an  F-score higher than 0.9 for each of the classes.This methodology is a suitable approach for detecting double-cropping systems, with further potential to identify exact crop types and the main practice of combining crops. The findings of this study showed that only about 2% of the study area was under this production. Except for positive economic outcomes, utilizing these systems brings significant environmental benefits and rational use of the soil without expanding physical cropland but with the same advantages. Therefore, the resulting geospatial datasets of double cropping croplands could help solve important questions relevant to food security, irrigation and climate change.