Modeling land use/land cover changes using quad hybrid machine learning model in Bangweulu wetland and surrounding areas, Zambia

Environmental Challenges, Journal Year: 2024, Volume and Issue: 14, P. 100866 - 100866

Published: Jan. 1, 2024

Wetlands are among the most productive natural ecosystems globally, providing crucial ecosystem services to people. Regrettably, a substantial 64% –71% of wetlands have been lost worldwide since 1900, mainly due changes in land use and cover (LULC). This issue is not unique Zambia's Bangweulu Wetland System (BWS), which faces similar challenges. However, there limited information about LULC BWS. Furthermore, finding accurate cost-effective methods understand dynamics complicated by multitude available techniques for classification. Non-parametric like Machine Learning (ML) offer greater accuracy, but different ML models come with distinct strengths weaknesses. Combining multiple has potential create more precise classification model. Open-source software QGIS spatial data Landsat also play significant role this endeavour. The primary objective study was enhance accuracy modeling wetland areas. Six models: Support Vector (SVM), Naive Bayes (NB), Decision Tree (DT), Artificial Neural Network (ANN), Random Forest (RF), K-Nearest Neighbour (KNN) were used image 8 (2020 image) 5 (1990, 2000, 2010 images) QGIS. Four SVM, NB, DT, KNN, performed better than other models. Consequently, Quad (4) hybrid model created fusing maps from these four highest performance. Results revealed that fusion classified KNN (Quad model) showcased superior performance compared individual Kappa Index scores 0.87, 0.72, 0.84 0.87 years 1990, 2020, respectively. analysis 1990 2020 showed yearly decline -1.17%, -1.01%, -0.12% forest, grassland, water body coverage, In contrast, built-up areas cropland increased at rates 1.70% 2.70%, underscores consistent growth alongside reduction forest grassland. Although experienced gradual decrease over period, minimal. Long-term monitoring will be essential evaluating success interventions, guiding conservation efforts, mitigating negative impacts on ecosystem, determining whether bodies sustained trend or short-term phenomenon.

Language: Английский

---

A performance evaluation of random forest, artificial neural network, and support vector machine learning algorithms to predict spatio-temporal land use-land cover dynamics: a case from lusaka and colombo

Frontiers in Environmental Science, Journal Year: 2024, Volume and Issue: 12

Published: Sept. 12, 2024

Reliable information plays a pivotal role in sustainable urban planning. With advancements computer technology, geoinformatics tools enable accurate identification of land use and cover (LULC) both spatial temporal dimensions. Given the need for precise to enhance decision-making, it is imperative assess performance reliability classification algorithms detecting LULC changes. While research on application machine learning evaluation widespread many countries, remains limited Zambia Sri Lanka. Hence, we aimed support vector (SVM), random forest (RF), artificial neural network (ANN) changes taking Lusaka Colombo City as study area from 1995 2023 using Landsat Thematic Mapper (TM), Operational Land Imager (OLI). The results reveal that RF ANN models exhibited superior performance, achieving Mean Overall Accuracy (MOA) 96% 94% Lusaka, respectively. Meanwhile, SVM model yielded (OA) ranging between 77% years 2023. Further, algorithm notably produced slightly higher OA kappa coefficients, 0.92 0.97, when compared models, across areas. A predominant change was observed expansion vegetation by 11,990 ha (60.4%), primarily through conversion 1,926 bare lands into during 1995–2005. However, noteworthy shift built-up areas experienced significant growth 2005 2023, with total increase 25,110 (71%). despite entire period there still net gain over 11,000 (53.4%) cover. In case Colombo, expanded 1,779 (81.5%), while decreased 1,519 (62.3%) concerned period. simulation also indicated 160-ha 2023–2035 Lusaka. Likewise, saw rise 337 within same Overall, outperformed algorithms. Additionally, prediction indicate an upward trend scenarios. resultant maps provide crucial baseline will be invaluable planning policy development agencies countries.

Language: Английский

---

ENVINet5 deep learning change detection framework for the estimation of agriculture variations during 2012–2023 with Landsat series data

Environmental Monitoring and Assessment, Journal Year: 2024, Volume and Issue: 196(3)

Published: Feb. 5, 2024

Language: Английский

---

Interchangeability of Cross-Platform Orthophotographic and LiDAR Data in DeepLabV3+-Based Land Cover Classification Method

Land, Journal Year: 2025, Volume and Issue: 14(2), P. 217 - 217

Published: Jan. 21, 2025

Riverine environmental information includes important data to collect, and the collection still requires personnel’s field surveys. These on-site tasks face significant limitations (i.e., hard or danger entry). In recent years, as one of efficient approaches for collection, air-vehicle-based Light Detection Ranging technologies have already been applied in global research, i.e., land cover classification (LCC) monitoring. For this study, authors specifically focused on seven types LCC bamboo, tree, grass, bare ground, water, road, clutter) that can be parameterized flood simulation. A validated airborne LiDAR bathymetry system (ALB) a UAV-borne green System (GLS) were study cross-platform analysis LCC. Furthermore, visualized using high-contrast color scales improve accuracy methods through image fusion techniques. If high-resolution aerial imagery is available, then it must downscaled match resolution low-resolution point clouds. Cross-platform interchangeability was assessed by comparing interchangeability, which measures absolute difference overall (OA) macro-F1 interchangeability. It noteworthy relying solely photographs inadequate achieving precise labeling, particularly under limited sunlight conditions lead misclassification. such cases, plays crucial role facilitating target recognition. All digital imagery, LiDAR-derived fusion) present results over 0.65 OA around 0.6 macro-F1. The found vegetation (bamboo, grass) road species comparatively better performance compared with clutter ground species. Given stated conditions, differences derived from different years (ALB year 2017 GLS 2020) are main reason. Because identification all items except relative RGB-based features cannot substituted easily because 3-year gap other Derived reconstruction, also has further change between ALB leads decreased case individual species, without considering seasons platforms, classify bamboo trees higher F1 scores especially proved high types. photography (UAV), high-precision measurement (ALB, GLS), satellite used. equipment expensive, opportunities limited. Based this, would desirable if could continuously classified Artificial Intelligence, investigated unique aspect exploring models across platforms.

Language: Английский

---

Evaluation of future land use change impacts on soil erosion for holota watershed, Ethiopia

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 25, 2025

Soil erosion is a critical global challenge that degrades land and water resources, leading to reduced soil fertility, pollution of bodies, sedimentation in hydraulic structures reservoirs. In Ethiopia, where agriculture forms the backbone economy, unplanned LULC changes have intensified erosion, posing significant threat food security sustainable development. Holota watershed rapid population growth urbanization accelerated use cover (LULC) changes, significantly affecting patterns. This study aims assess spatiotemporal their impact on from 2000 2050. Using Landsat imagery 2000, 2010, 2020, supervised classification with maximum likelihood algorithm was applied Google Earth Engine (GEE) map five classes: forest, cropland, built-up areas, shrubland, grassland. The future for 2050 predicted using CA–Markov chain model. 2020 maps estimated Revised Universal Loss Equation (RUSLE). Results indicate annual loss 13.3 t ha − 1 yr increasing 15.9 by Cropland, grassland are expected be major contributors while forest shrubland likely play mitigating role. novelty this research lies its integration cutting-edge remote sensing technologies, such as GEE CA-Markov model, predict combined data-scarce region, providing actionable insights conservation planning Ethiopian highlands. These findings offer essential guidance planners implement management practices aimed at reducing including promoting restoration, adopting contour farming, enforcing regulations limit expansion cropland areas erosion-prone zones.

Language: Английский

---

Prediction of Urban Surface Water Quality Scenarios Using Water Quality Index (WQI), Multivariate Techniques, and Machine Learning (ML) Models in Water Resources, in Baitarani River Basin, Odisha: Potential Benefits and Associated Challenges

Earth Systems and Environment, Journal Year: 2025, Volume and Issue: unknown

Published: April 9, 2025

Language: Английский

---

Multi-Temporal Passive and Active Remote Sensing for Agricultural Mapping and Acreage Estimation in Context of Small Farm Holds in Ethiopia

Land, Journal Year: 2024, Volume and Issue: 13(3), P. 335 - 335

Published: March 6, 2024

In most developing countries, smallholder farms are the ultimate source of income and produce a significant portion overall crop production for major crops. Accurate distribution mapping acreage estimation play role in optimizing resource allocation. this study, we aim to develop spatio–temporal, multi-spectral, multi-polarimetric LULC approach assess Oromia Region Ethiopia. The study was conducted by integrating data from optical radar sensors sentinel products. Supervised machine learning algorithms such as Support Vector Machine, Random Forest, Classification Regression Trees, Gradient Boost were used classify area into five first-class common land use types (built-up, agriculture, vegetation, bare land, water). Training validation collected ground high-resolution images split 70:30 ratio. accuracy classification evaluated using different metrics accuracy, kappa coefficient, figure metric, F-score. results indicate that SVM classifier demonstrates higher compared other algorithms, with an Sentinel-2-only integration microwave 90% 94% value 0.85 0.91, respectively. Accordingly, Sentinel-1 Sentinel-2 resulted alone. findings demonstrate remarkable potential multi-source remotely sensed agricultural small farm holdings. These preliminary highlight active passive remote sensing estimation.

Language: Английский

---

Machine-Learning Algorithms for Mapping LULC of the uMngeni Catchment Area, KwaZulu-Natal

Published: April 10, 2024

Analysis of land use/land cover (LULC) in the catchment areas is first action toward safeguarding freshwater resources. The LULC information watershed has gained popularity natural science field as it helps water resource managers and environmental health specialists develop conservation strategies based on available quantitative information. Thus, remote sensing cornerstone addressing environmental-related issues at level. In this study, performance four machine learning algorithms (MLAs), such Random Forests (RF), Support Vector Machine (SVM), Artificial Neural Networks (ANN), Naïve Bayes (NB) was investigated to classify into nine relevant classes undulating landscape using Landsat 8 Operational Land Imager (L8-OLI) imagery. assessment MLAs were visual inspection analyst commonly used metrics, user’s accuracy (UA), producers’ (PA), overall (OA), kappa coefficient. produced good results, where RF (OA= 97.02%, Kappa= 0.96), SVM 89.74 %, 0.88), ANN 87%, 0.86), NB 68.64 0.58). results show outstanding model over with a small margin. While yielded satisfactory which could be primarily influenced by its sensitivity limited training samples. contrast, robust due an ability high-dimensional data data.

Language: Английский

---

Evaluating Machine-Learning Algorithms for Mapping LULC of the uMngeni Catchment Area, KwaZulu-Natal

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(12), P. 2219 - 2219

Published: June 19, 2024

Analysis of land use/land cover (LULC) in catchment areas is the first action toward safeguarding freshwater resources. LULC information watershed has gained popularity natural science field as it helps water resource managers and environmental health specialists develop conservation strategies based on available quantitative information. Thus, remote sensing cornerstone addressing environmental-related issues at level. In this study, performance four machine learning algorithms (MLAs), namely Random Forests (RFs), Support Vector Machines (SVMs), Artificial Neural Networks (ANNs), Naïve Bayes (NB), were investigated to classify into nine relevant classes undulating landscape using Landsat 8 Operational Land Imager (L8-OLI) imagery. The assessment MLAs was a visual inspection analyst commonly used metrics, such user’s accuracy (UA), producers’ (PA), overall (OA), kappa coefficient. produced good results, where RF (OA = 97.02%, Kappa 0.96), SVM 89.74%, 0.88), ANN 87%, 0.86), NB 68.64%, 0.58). results show outstanding model over with significant margin. While yielded satisfactory its sensitivity limited training samples could primarily influence these results. contrast, robust be due an ability high-dimensional data data.

Language: Английский

---

Provision of land use and forest density maps in semi-arid areas of Iran using Sentinel-2 satellite images and vegetation indices

Advances in Space Research, Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 1, 2024

Language: Английский

---