Hybrid deep learning with a random forest system for sustainable agricultural land cover classification using DEM in Najran, Saudi Arabia

The resolution of digital elevation model (DEM) used for watershed delineation and data extracted from the delineated watershed relevant to hydrography plays an important role in the calibration of hydrological models. The effect of DEM resolution on the drainage length, watershed area and average slope are studied in many earlier works. DEM resolution can have impact on hydrologic modeling parameters, water quality simulation results. The impacts of DEM resolution on results of the watershed response and water quality modeling are studied using HSPF (Hydrologic Simulation Program Fortran) model applied to a watershed in Kentucky in this study. Three major streams in this watershed are impaired due to pathogen pollution according to the current state of Kentucky water quality standard set for recreation. The effects of DEM resolution on water quality modeling and calibration for fecal coliform bacteria are studied using a HSPF model. Results from this study indicate that the DEM resolution affects hydrography parameters of the delineated watershed, influence the calibration process, and the non-point and point source loadings to the streams in the watershed.

Land‐use maps produced with high accuracy are extremely important as a basis for better use of the land as they are widely utilized in many areas such as agricultural policies, natural resources management, and environmental operations. This study aimed to produce a high accuracy land‐use map using digital surface models (DSMs) produced using different interpolation methods and a random forest (RF) classifier. High spatial resolution Triplesat‐2 images, Worldview‐2 (WV‐2) images, and unmanned aerial vehicle (UAV) images were used in this study. First, DSMs were produced from the point cloud using different interpolation methods. The image fusion process was applied and these fused images were classified using an RF classifier together with the DSMs. Overall results showed that the DSM obtained by the kriging method yielded better results than the other methods by increasing the classification accuracy by 7%–13%. In addition, the McNemar test was applied to the images resulting from the classification of fused images with and without using DSMs to investigate the statistical significance of the differences. McNemar test demonstrated that the values were greater than 3.84, which revealed that the DSM produced via kriging interpolation had significantly increased the classification accuracy at a 95% confidence interval.

Flood inundation maps are dependent on the topographic and geomorphologic features of a wadi (drainage basin) in arid regions, which are most susceptible for potential flash flood occurrences, such as in the southwestern part of the Kingdom of Saudi Arabia. It is not possible to control the potential flood hazards by using only technological instruments that forewarn the occurrences or imminence. Additionally, it would be better to prepare flood risk maps so as to delineate the risky areas to educate the administrators and local settlers. The availability of these maps is the key requirement for any urban development that entails land use allocation, identification of dam, tunnel, highway, bridge sites, and infrastructure locations for sustainable future. This paper suggests the necessary steps in flood inundation map preparation after determining the possible flood discharge. For this purpose, a set of critical cross-sections along the possible flood plain are taken in the field with surveying methods and measurements. The calculation of the average flow velocity in each section is calculated according to the cross-section geometric, hydraulic, and material properties. Synthetic rating curves (SRC) are prepared for each cross section, which are very useful especially in arid and semi-arid regions where there are no perennial surface water flows for natural rating curve measurements. All the SRCs appear in the form of power function which relates the flow depth to discharge in a given cross section. It is then possible to calculate the flood depth in the cross section through its SRC. Depending on the cross-section shape, the flood width can be calculated. The connection of a series of widths on a scaled topographic map delineates the flood inundation area. If digital elevation map (DEM) is available, then the SRCs can be integrated with these maps and the flood inundation delineation can be achieved automatically. Since DEMs are not available, the topographic maps are used for this purpose in order to delineate flood inundation areas within wadis Hali and Yiba from the southwestern Kingdom of Saudi Arabia.

Detection of radioactive waste sites in the Chornobyl exclusion zone using UAV-based lidar data and multispectral imagery

A draft Standard International Classification (SIC) of land use has been proposed and adopted by the Conference of European Statisticians. The classification scheme has been critically assessed for its suitability for application to different forms of land use survey in the United Kingdom. and also for its compatibility with land cover categories derived from data acquired by remote sensing techniques. At the most general level there is a high degree of agreement with the categories proposed in the USGS classification for use with remote sensor data. In the context of British requirements, the most suitable portion of the SIC appears to be in the classification of' forest and other wooded land'. An alternative classification of 'agricultural land' is offered which is better suited to the needs of the UK agricultural census. The SIC treatment of 'natural terrain' is criticised for being out of step with current ecological practice. especially in the treatment of 'wetlands'. It is suggested that the SIC should be open to modification in the light of its application in the next few years, especially with the likelihood of more widespread adoption for land cover mapping of high resolution satellite data. 1. Background Arising originally out of a meeting on land use statistics held in Geneva in March 1980 under the auspices of the Economic Commission for Europe Con­ ference of European Statisticians, an amended version of the draft Standard International classification (SIC) of land use was approved at the 31st plenary session of the Conference of European Statisticians in June, 1983, [2] and Table 1. In the United Kingdom, at about the same time, the Land Applications Working Group (LAWG) of the National Remote Sensing Centre was deliberat­ ing on the matter of land use classification, with particular reference to identify­ ing a classification suitable for use with remotely sensed images, from aircraft or satellite, as data sources. Robert Wright is a graduate of Glasgow University (B.Sc. Honours in Geography), the International Institute for Aerospace Survey and Earth Sciences (ITC) in the Netherlands (B.Sc. in Photogram­ metric Engineering) and the University of Michigan (M.Sc. in Remote Sensing). He is currently a lecturer in the Geography department at Aberdeen University, where he is responsible for teaching and research activity in the field of environmental remote sensing. He is a member of the Land Applications Working Group of the U.K. National Remote Sensing Centre, Royal Aircraft Establishment, Farnborough, England.

Abstract. Classifying land use and land cover (LULC) is essential for monitoring change and protecting the environment, especially in arid regions where ecosystems are fragile and vulnerable to climate change. This research focuses on Errachidia, an arid region in Morocco known for its natural oases, which are vital to the ecological ecosystems and the living conditions of the local population.To study changes in land use and land cover (LULC), 10-metre resolution Sentinel 2 satellite imagery was used from 2017 to 2023. Based on ground truth samples, classification was made by using Google Earth classifiers, namely Random Forest (RF), Support Vector Machine (SVM) and Cartesian Regression Trees (CART). Among these models, the Random Forest (RF) classifier outperformed the others, achieving an impressive accuracy of 94% and a kappa coefficient of 0.89, proving its strength in dealing with the challenges associated with arid environments. Looking to the future, we wanted to predict how land use would change between 2023 and 2030. Using the geemap Python package with GEE, RF and Cellular Automata-Markov (CA-Markov) were compared to predict future trends. While RF excelled in categorizing current land use with an accuracy of 93.17% and a kappa coefficient of 0.71, CA-Markov proved to have an accuracy of 93.82% and a kappa coefficient of 0.75 for long-term predictions. The predictions revealed alarming trends: The risk of desertification in Errachidia is increasing as agricultural areas decrease and desert-like sandy areas increase. The ability of the CA-Markov system to model how land changes over space and time made it particularly effective in capturing these changes. This study highlights the strengths of using GEE-RF classification and CA-Markov prediction for long-term future changes. This study underscores the efficacy of RF classification and CA-Markov in predicting long-term future changes. By identifying these changes, policymakers and land managers can take steps to combat land degradation and ensure the long-term health of these fragile ecosystems. By detecting these changes, policymakers and land managers can take steps to address land changes and ensure the long-term health of these fragile ecosystems.

Comprehensive land consolidation is essential for safeguarding arable land, promoting efficient land use, and enhancing regional ecological environments. Accurate land use classification provides the fundamental basis for reliable potential assessment. In karst regions such as the Lijiang River Basin, complex topography and geomorphological heterogeneity present significant challenges for traditional land cover classification, often resulting in greater uncertainty and reduced reliability in consolidation potential analyses. To address the challenges of heterogeneous karst geomorphology and fragmented land use, we propose an enhanced Random Forest (RF) classification algorithm that integrates recursive feature elimination, sequential backward selection, and a weighted voting mechanism. This approach enables optimal multi-scale feature selection and significantly improves classification performance. Building on this foundation, we developed a comprehensive evaluation model for land consolidation potential that simultaneously considers agricultural, construction, and ecological land types. We systematically introduced complex network analysis to quantify spatial associations among these land types, thereby enabling optimized spatial zoning of consolidation potential. The results indicate that: (1) The enhanced RF algorithm achieved the highest classification accuracy in the Lijiang River Basin, with overall accuracy and Kappa coefficient improved by 3% and 4%, respectively, compared to the conventional RF. This method effectively addresses misclassification issues related to small and fragmented land parcels. (2) The spatial distribution of consolidation potential grades for various land types in the Lijiang River Basin shows clear differentiation, ranking from highest to lowest as ecological land, agricultural land, and construction land. The proposed model increased the coverage of comprehensive consolidation areas for multiple land types to 94.97% of the basin area. The proposed method provides a theoretical foundation and technical pathway for holistic, full-cycle, and spatially optimized land consolidation management in karst and other complex ecological regions, demonstrating strong scalability and practical value.

Semiarid Tunisia is characterized by agricultural production that is delimited by water availability and degraded soil. This situation is exacerbated by human pressure and the negative effects of climate change. To improve the knowledge of long-term (1980 to 2020) drivers for Land Use and Land Cover (LULC) changes, we investigated the semiarid Rihana region in central Tunisia. A new approach involving Google Earth Engine (GEE) was used to map LULC using Landsat imagery and vegetative indices (NDVI, MSAVI, and EVI) by applying a Random Forest (RF) classifier. A Rapid Participatory Systemic Diagnosis (RPSD) was used to consider the relation between LULC changes and their key drivers. The methodology relied on interviews with the local population and experts. Focus groups were conducted with practicians of the Regueb Agricultural Extension Services, followed by semi-structured interviews with 52 households. Results showed the following: (1) the RF classifier in Google Earth Engine had strong performance across diverse Landsat image types resulting in overall classification accuracy of ≥0.96 and a kappa coefficient ≥0.93; (2) rainfed olive land increased four times during the study period while irrigated agriculture increased substantially during the last decade; rangeland and rainfed annual crops decreased by 58 and 88%, respectively, between 1980 and 2021; (3) drivers of LULC changes are predominately local in nature, including topography, local climate, hydrology, strategies of household, effects of the 2010 revolution, associated increasing demand for natural resources, agricultural policy, population growth, high cost of agricultural input, and economic opportunities. To summarize, changes in LULC in Rihana are an adaptive response to these various factors. The findings are important to better understand ways towards sustainable management of natural resources in arid and semiarid regions as well as efficient methods to study these processes.

Timely identification of sugarcane cultivation is essential for precision agriculture, policy formulation, and sustainable land management. Our study presents a multimodal framework that integrates spectral, morphological, and textural features from Sentinel-2 imagery with hybrid deep learning techniques for sugarcane mapping. A comprehensive feature space combining NDVI, geometrical descriptors, and GLCM-based texture metrics was used to train Random Forest (RF) and Support Vector Machine (SVM) classifiers. A hybrid convolutional neural network (CNN+MLP) architecture further merged visual embeddings from image patches with tabular descriptors. The RF model achieved a macro F1-score of 0.91, a Kappa coefficient of 0.86, and a macro-AUC of 0.94 across agricultural regions in Brazil. DBSCAN clustering was applied to filter noisy labels and improve the quality of the training dataset. A domain-specific confusion cost matrix illustrated the economic impact of misclassification, showing that the RF and CNN+MLP models improved accuracy and financial efficiency by reducing crop identification errors. Although the analysis was limited to São Paulo and Mato Grosso do Sul, the framework showed consistent and robust performance within these regions. Overall, it highlights the benefits of combining morphospectral, textural, and deep features for crop classification, providing a scalable foundation for time-series modeling and real-time land-use analytics. • Sugarcane regions were accurately mapped using Sentinel-2 satellite imagery across different agricultural zones in Brazil. • A hybrid CNN+MLP model combined visual embeddings and tabular features to improve classification. • Morphological, spectral (NDVI), and GLCM-based texture features were fused to improve crop discrimination. • Random Forest and Support Vector Machine classifiers were benchmarked and compared with the deep learning approach. • Unsupervised DBSCAN clustering was applied to detect and remove noisy samples, improving training set quality.

<div> <p>The Hydrologic Simulation Program-FORTRAN (HSPF) model is widely used to develop management strategies for water resources and to evaluate the hydrologic effect of various management scenarios. The spatial resolution of the input data used to parameterize HSPF model may induce uncertainty in model outputs. In this study, the impact of spatial resolutions of Digital Elevation Model (DEM) and land use map on the uncertainty of HSPF predicted flow and sediment were evaluated. DEM resolution can affect stream length, watershed area and average slope, whereas land use data resolution can lead to redistribution of land use information. Results showed that finer resolution DEM and land use maps can generate higher flow volumes and sediment loads compared to modelling scenarios using inputs of coarse resolution. The relative change in model performance between the baseline scenario (high-resolution) and scenarios of coarser resolution described uncertainties due to DEM and land use spatial information, and the probability density function of these uncertainties was used to estimate these uncertainties. Modelled flow and sediment uncertainty due to DEM resolution seems to follow a log-normal and a general extreme value distribution respectively, whereas modelled flow and sediment uncertainty due to land use resolution seems to both follow a general extreme value distribution. Overall, results highlight the need for a high-resolution DEM and land use maps in the application of the HSPF model while they provide useful information for reducing the model uncertainties.</p> </div> <p> </p>

This chapter explains how land use in the Loess Plateau area was classified using TM (Thematic Mapper) remote sensing data. Slope fields were extracted from DEMs (Digital Elevation Models) at different scales and the effects of different scales on slope field precision were also analysed. Results show that a DEM with a scale of 1:50,000 (with a resolution of 25 m) was basically sufficient to extract slope fields with the required precision. Although the accuracy of slopes extracted from a scale of 1:10,000 is higher than DEMs of other scales, the detailed land appearance did not show obvious improvement on the slope field precision because of the special land use in the studied area, especially the considerable effect of land use type on plantation layout (plantations focused on Loess Mao of small slopes and large areas). Furthermore, with the constraints of using the resolution of 30 m for the land use, DEMs with the scale of 1:10,000 could not show its advantage of high precision and increased the redundancy and computing load; thus it was not appropriate for obtaining slope fields of large areas. The DEM with a scale of 1:250,000 could not be used to extract slope field graphs effectively either, because of its large generalizations over the study area.

Yamula Dam that has been constructed in Kayseri is one of the most important projects regarding the irrigation of agricultural lands. The total area of the collected water body was reached in 2005 as planned before, and electricity production started. Together with the realization of the project, it is planned to use agricultural lands in basin more efficiently and productively. In this study, it was aimed to determine the hydrological structure of Yamula Basin and potentially irrigable lands by using geographic information systems and remote sensing technologies. The hydrological structure was determined using digital elevation model. The land use map was prepared by using Landsat satellite image for the year 2016. The analysis and queries were carried out by overlapping the land use, land use capacity, topographic maps and sub-basin layers. The potential agricultural lands were determined in accordance with the results obtained from the spatial analysis.

The digital elevation model (DEM) matrix allows to reveal the relationship of the soil cover with morphometric parameters. Therefore, in the absence of the possibility of a large-scale field survey of soils, for territories with a high degree of erosion hazard, the data on the relief make it possible to carry out predictive large-scale soil maps. The aim of the work is to create a cartographic model of the soil cover of agricultural land based on the extrapolation of the results of DEM processing and to compare it with the existing large-scale soil map in similar natural conditions. The object of research is the territory of LLC “Selskoe” located in the Solikamsk urban district of the Perm region. Agricultural land use belongs to the northernmost agricultural lands in the region. The total area of research was 429 hectares of arable land. The plot includes 8 fields. For soil mapping, a digital elevation model ALOS 30 and a large-scale soil map of the key site were used, which characterizes part of the land use of the subsidiary farm “Voskhod”. Using the results of the classification of the relief according to the GIS SAGA TPI based landform classification algorithm as a contour base and the existing soil map of the key site, a soil map of LLC “Selskoe” was made by the extrapolation method. The steepness of the slopes and the topographic moisture index were used as auxiliary data. In conditions of complex relief, a clear dependence of the location of soils on relief elements has been established. By extrapolating data from a large-scale soil survey, 10 soil cartographic units were identified. According to the relief elements, podzolic, sod-podzolic, bog-podzolic and alluvial soils were identified. In terms of granulometric composition, light soils prevail, a small area is occupied by medium loamy soils.

Digital elevation models (DEMs) are the most obvious data sources in landslide susceptibility assessment. Many landslide casual factors are often generated from DEMs. Most studies on landslide susceptibility assessments rely on freely available DEMs. However, very little is known about the performance of different DEMs with varying spatial resolutions on the accurate assessment of landslide susceptibility. This study compared the performance of four different DEMs including 30 m Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM), 30–90 m Shuttle Radar Topographic Mission (SRTM), 12.5 m Advanced Land Observation Satellite (ALOS) Phased Array Type L band Synthetic Aperture Radar (PALSAR), and 25 m Survey of Bangladesh (SOB) DEM in landslide susceptibility assessment in the Rangamati district in Bangladesh. This study used three different landslide susceptibility assessment techniques: modified frequency ratio (bivariate model), logistic regression (multivariate model), and random forest (machine-learning model). This study explored two scenarios of landslide susceptibility assessment: using only DEM-derived causal factors and using both DEM-derived factors as well as other common factors. The success and prediction rate curves indicate that the SRTM DEM provides the highest accuracies for the bivariate model in both scenarios. Results also reveal that the ALOS PALSAR DEM shows the best performance in landslide susceptibility mapping using the logistics regression and the random forest models. A relatively finer resolution DEM, the SOB DEM, shows the lowest accuracies compared to other DEMs for all models and scenarios. It can also be noted that the performance of all DEMs except the SOB DEM is close (72%–84%) considering the success and prediction accuracies. Therefore, anyone of the three global DEMs: ASTER, SRTM, and ALOS PALSAR can be used for landslide susceptibility mapping in the study area.

The purpose of this article is to develop an algorithm for modelling the above-ground biomass density (AGBD) of small urban geosystems using the example of the urban geosystem of the city of Bucha based on lidar data from the Global Ecosystem Dynamics Investigation (GEDI) mission. Methodology. Modelling was performed using machine learning methods, namely Extreme Gradient Boosting and Random Forest. The research algorithm included loading geospatial data, processing it, and performing modelling with accuracy assessment of the results. Results. Various analysis predictors were tested, including a digital elevation model (DEM) and spectral indices based on Sentinel-2 satellite data. Secondly, two AGBD modelling methods were tested, namely Extreme Gradient Boosting and Random Forest, taking into account data on land use and land cover (LULC) systems. Under these conditions, it was determined that the best modelling method is Random Forest using DEM and spectral indices with a geospatial resolution of 10 m. Data on above-ground biomass density for the small urban geosystem of Bucha were obtained. The highest above-ground biomass density was recorded in green areas with tree vegetation. Scientific novelty. Analysis of previous studies indicates the absence of a generally accepted list of predictors for above-ground biomass density modelling. The study uses a unique set of predictors, determined taking into account the characteristics of a small urban geosystem LULC. The use of a layer mask to take into account data on LULC is an innovative approach. It should be noted that no assessment of above-ground biomass density based on GEDI lidar data has been conducted for the territory of Ukraine, which also contributes to the scientific novelty of the study. Practical significance. The use of Earth observation data to estimate above-ground biomass density enable the replacement of direct in situ measurements, which are difficult and, in some cases, practically impossible. Data on above-ground biomass density are important for understanding the carbon cycle, which in turn affects the climate system. Based on the data obtained, ecosystem services to achieve climate neutrality in urban geosystems can be assessed.