Wildfire spreading prediction using multimodal data and deep neural network approach

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Jan. 31, 2024

Predicting wildfire spread behavior is an extremely important task for many countries. On a small scale, it possible to ensure constant monitoring of the natural landscape through ground means. However, on scale large countries, this becomes practically impossible due remote and vast forest territories. The most promising source data in case that can provide global sensing data. Currently, main challenge development effective pipeline combines geospatial collection application advanced machine learning algorithms. Most approaches focus short-term fire spreading prediction utilize from unmanned aerial vehicles (UAVs) purpose. In study, we address predicting consider forecasting horizon ranging 1 5 days. We train neural network model based MA-Net architecture predict environmental climate data, taking into account spatial distribution features. Estimating importance features another critical issue prediction, so analyze their contribution model's results. According experimental results, significant are wind direction land cover parameters. F1-score predicted burned area varies 0.64 0.68 depending day (from days). study was conducted northern Russian regions shows promise further transfer adaptation other regions. This data-based artificial intelligence (AI) approach be beneficial supporting emergency systems facilitating rapid decision-making.

Language: Английский

---

Practical AI Cases for Solving ESG Challenges

Sustainability, Journal Year: 2023, Volume and Issue: 15(17), P. 12731 - 12731

Published: Aug. 23, 2023

Artificial intelligence (AI) is a rapidly advancing area of research that encompasses numerical methods to solve various prediction, optimization, and classification/clustering problems. Recently, AI tools were proposed address the environmental, social, governance (ESG) challenges associated with sustainable business development. While many publications discuss potential AI, few focus on practical cases in three ESG domains altogether, even fewer highlight may pose terms ESG. The current paper fills this gap by reviewing applications main IT engineering implementations. considered are based almost one hundred publicly available manuscripts reports obtained via online search engines. This review involves study typical production problems each domain, gives background details several selected (such as carbon neutrality, land management, scoring), lists smart algorithms can fake news generation increased electricity consumption). Overall, it concluded that, while already exist, still very far away from reaching its full potential; however, should always remember itself lead some risks.

Language: Английский

---

Forest carbon stock-based bioeconomy: Mixed models improve accuracy of tree biomass estimates

Biomass and Bioenergy, Journal Year: 2024, Volume and Issue: 183, P. 107142 - 107142

Published: March 18, 2024

Language: Английский

---

Pseudo-Labeling Approach for Land Cover Classification Through Remote Sensing Observations With Noisy Labels

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 82570 - 82583

Published: Jan. 1, 2023

Satellite data allows us to solve a wide range of challenging tasks remotely, including monitoring changing environmental conditions, assessing resources, and evaluating hazards. Computer vision algorithms such as convolutional neural networks have proven be powerful tools for handling huge visual datasets. Although the number satellite imagery is constantly growing artificial intelligence advancing, present sticking point in remote sensing studies quality amount annotated Typically, manual labels particular uncertainties mismatches. Also, lot datasets available low resolution. Available representation observed objects can more detailed than annotation. This causes need markup adjustment, which referred pseudo-labeling task. The main contribution this research that we propose pipeline address problem inaccurate low-resolution improvement solving land-cover land-use segmentation task based on from Sentinel-2 satellite. Our methodology takes advantages both classical machine learning (ML) deep (DL) algorithms. We examined random sampling, uniform K-Means sampling compared it with full dataset usage. U-Net, DeepLab, FPN models were trained adjusted dataset. achieved findings show simple yet effective approach preliminary further refinement leads significantly higher results just using raw network pipeline. Moreover, considered technique use less ML model training. experiments involve adjustment up-scaling 30m 10m. verify proposed precise test area annotation F1-score 0.792 0.816.

Language: Английский

---

Advancing forest carbon stocks’ mapping using a hierarchical approach with machine learning and satellite imagery

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Sept. 9, 2024

Remote sensing of forests is a powerful tool for monitoring the biodiversity ecosystems, maintaining general planning, and accounting resources. Various sensors bring together heterogeneous data, advanced machine learning methods enable their automatic handling in wide territories. Key forest properties usually under consideration environmental studies include dominant species, tree age, height, basal area timber stock. Being proxies stand productivity, they can be utilized carbon stock estimation to analyze forests' status proper climate change mitigation measures on global scale. In this study, we aim develop an effective learning-based pipeline using solely freely available regularly updated satellite observations. We employed multispectral Sentinel-2 remote data predict structure characteristics produce detailed spatial maps. Using Extreme Gradient Boosting (XGBoost) algorithm classification regression settings management-level inventory as reference measurements, achieved quality predictions species equal 0.75 according F1-score, area, accuracy 0.75, 0.58 0.56, respectively, R

Language: Английский

---

Deforestation and Forest Monitoring With CNN and RNN

Advances in computational intelligence and robotics book series, Journal Year: 2025, Volume and Issue: unknown, P. 191 - 208

Published: Jan. 10, 2025

Deforestation poses a significant threat to global biodiversity and climate stability, necessitating effective monitoring management strategies. It is highly necessary for an strategy mitigate deforestation as it possesses potential stability biodiversity. A novel deep learning technique with Convolutional Neural Networks (CNNs) Recurrent (RNNs) proposed identify the forest. CNN deployed deforested areas by extracting spatial features RNN are used capture patterns of forest dynamics processing time series satellite data. This mechanism where temporal analysis done prediction.

Language: Английский

---

Exploration of geo-spatial data and machine learning algorithms for robust wildfire occurrence prediction

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

Published: March 28, 2025

Wildfires play a pivotal role in environmental processes and the sustainable development of ecosystems. Timely responses can significantly reduce damages consequences caused by their spread. Several critical issues wildfire behavior analysis include fire occurrence forecasting, early detection, spread prediction. In this study, we focus on which is valuable tool for facilitating earlier intervention. Conventional approaches primarily rely computation indices based weather conditions. However, solutions that utilize more comprehensive data, remote sensing information, artificial intelligence (AI) algorithms may offer substantial advantages rapid decision-making extensive territory monitoring. The wide variety spatial parameters great diversity geographical regions influence complicate task. Consequently, there no unified approach predicting occurrences using data AI techniques. goal study to explore potential various available - meteorological, geo-spatial, anthropogenic machine learning (ML) algorithms. We developed pipeline acquisition subsequent ML-based algorithm development. includes following algorithms: Random Forest, XGBoost, Autoencoder, ConvLSTM, Attention Multilayer Perceptron, RegNetX. addition, several metrics assess quality models case highly imbalanced spatio-temporal data. To conduct collected unique dataset covering large central Russia, incorporating than 17,000 verified events over period 10 years. findings underscore necessity developing individual ML tailored each region, taking into account specific features correlated with probability occurrence. achieved models, as measured F1-score, varies from 0.7 0.87 depending demonstrating integrating such emergency response systems.

Language: Английский

---

Geospatial Data Analysis for Mapping Carbon Sequestration Hotspots

IGI Global eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 193 - 218

Published: April 11, 2025

Geospatial measurement of carbon is required for hotspot identification and precise quantification sinks across various ecosystems. The evolution GIS, remote sensing, LiDAR, spatial modeling using AI has significantly improved the precision extent monitoring. chapter describes techniques examining forest biomass, soil sequestration, ocean through satellite data, geospatial computation, machine learning models. Integration big data enhances flux estimation land-use impact assessment on sequestration capacity. Significant challenges such as resolution, model uncertainty, computational complexity are addressed, along with new solutions. analysis augmented by at core activities maximization, enabling climate change mitigation, sustainable land management, transparent credit systems.

Language: Английский

---

CISA: Context Substitution for Image Semantics Augmentation

Mathematics, Journal Year: 2023, Volume and Issue: 11(8), P. 1818 - 1818

Published: April 11, 2023

Large datasets catalyze the rapid expansion of deep learning and computer vision. At same time, in many domains, there is a lack training data, which may become an obstacle for practical application vision models. To overcome this problem, it popular to apply image augmentation. When dataset contains instance segmentation masks, possible instance-level It operates by cutting from original pasting new backgrounds. This article challenges with objects present various domains. We introduce Context Substitution Image Semantics Augmentation framework (CISA), focused on choosing good background images. compare several ways find backgrounds that match context test set, including Contrastive Language–Image Pre-Training (CLIP) retrieval diffusion generation. prove our augmentation method effective classification, segmentation, object detection different complexity model types. The average percentage increase accuracy across all tasks fruits vegetables recognition 4.95%. Moreover, we show Fréchet Inception Distance (FID) metrics has strong correlation accuracy, can help choose better without training. negative between FID augmented 0.55 experiments.

Language: Английский

---

MineralImage5k: A benchmark for zero-shot raw mineral visual recognition and description

Computers & Geosciences, Journal Year: 2023, Volume and Issue: 178, P. 105414 - 105414

Published: July 20, 2023

Language: Английский

---