Global climate warming annually leads to a decrease in the agro-climatic potential of farming, many areas of traditional cultivation of land become unsuitable for agriculture due to desertification, swamping and erosion processes. Purpose - the concept of sustainable land use is considered, taking into account foreign experience. Methods are based on the system analysis of works of domestic and foreign scientists and data from the Bureau of National Statistics of the Agency for Strategic Planning and Reforms of the Republic of Kazakhstan, analytical reports of the Committee on Land Resources Management. Results - different approaches to solving the problems of agroindustrial complex are presented, the main trends in the development of the agrarian sector of the republic, the need to combat land degradation, to preserve their productivity and ecosystem functions are shown. Conclusions - the analysis of world practice shows that the principles of land cultivation based on environmental, social and economic criteria should be implemented based on the effective use of agricultural land, that is, through the optimization of agricultural land fund. The research has an interdisciplinary character. The systematic study of this issue contributes to the application of rational methods of land management in agricultural production. The authors note that the use of digital technologies in the sphere of land relations will help to increase the efficiency of production and economic activity, to unite the main factors and efforts of the participants of this process in a single information space, to increase the reliability and completeness of transmitted data. Digital infrastructure in the farming system is the driver of future development. Provision of all types of services in accordance with land legislation should be digitized.
Reported accuracy of unmanned aerial vehicle (UAV)-based yield models depends on whether validation respects repeated plots and seasonal variation. We tested whether temporal descriptors of vegetation indices explain plot-level grain yield of winter triticale and whether it transfers between years. Here, 7 multispectral UAV flights within 190–245 days after sowing were analysed for 20 fixed plots in 2024 and 2025. For normalised difference vegetation index (NDVI), green normalised difference vegetation index (GNDVI), normalised difference red-edge (NDRE), optimised soil-adjusted vegetation index (OSAVI), and wide dynamic range vegetation index (WDRVI), the temporal mean, median, standard deviation, and normalised area under the curve gave 20 candidate predictors. Single-index and multi-index models, with predictors selected by Elastic Net stability selection or a genetic algorithm, were evaluated by nested plot-grouped cross-validation and leave-one-year-out (LOYO) validation against season-mean references. Pooled R2 reached 0.89 (RMSE = 433 kg ha−1; WDRVI–KNN), but a season-mean reference alone gave 0.83. Within 2024, the best models reduced root mean square error (RMSE) from 571 to about 290–380 kg ha−1; within 2025, none improved on the reference. In leave-one-year-out validation, the lowest RMSE (about 2.0 t ha−1) was only 20% below that of the training-season mean (2.5 t ha−1), and all R2 values were negative. Yield differences between control and manured plots reversed direction between seasons (−328 and +300 kg ha−1, both 95% confidence intervals included zero) and are confounded with plot position. The descriptors can rank plot conditions within a season, but two seasons do not support yield forecasting for unseen years.