This paper introduces Sentinel-2 Biophysical Engine implemented on the Google Earth Engine (GEE) platform, addressing the critical need for high-resolution, near-real-time monitoring of Earth's surfaces in response to population growth and climate change. Leveraging Copernicus Ground-Based Observations for Validation (GBOV) data, the tool utilizes neural network models to derive essential biophysical variables such as Leaf Area Index (LAI). Integrating reflectance values, acquisition geometry parameters, and auxiliary data, the GEE implementation demonstrates accuracy through in-situ validation. Results show promising Root Mean Square Error (RMSE), Mean Absolute Error (MAE), correlation, and bias values. With applications in land surface monitoring, agriculture, and hydrological modeling, this tool has strong potential to contribute sustainable resource management and climate change mitigation. Future developments involve expanding the tool to incorporate Landsat-8 data, further enhancing its global applicability.

In this study, a novel approach to map forest attributes has been investigated for boreal forests in Sweden. The methodology relies on machine learning, utilizing a combination of remote sensing data and field data for both training and evaluating the proposed models. To ensure the accuracy in estimating forest attributes at any given time, the approach incorporates a broad range of available remote sensing data including airborne laser scanning (ALS) data, weekly satellite data from Sentinel-1 and Sentinel-2, and global forest map data. However, in this study focus has been on utilizing ALS data. The field data utilized in the study are derived from the Swedish National Forest Inventory and encompass measurements of key forest variables such as above-ground biomass, stem volume, basal area-weighted mean tree height, basal area-weighted mean diameter at breast height, and basal area. The potential of exporting knowledge gained from mapping Sweden to other forested landscapes such as in Latvia, using model updating with limited reference data from the new targeted area will be the next step to investigate. Here, data from Sweden were used to take the first steps towards developing a mapping methodology. The results demonstrate a promising potential of the proposed approach that will showcase new possibilities to share knowledge of updated forest mapping using the increasing flow of high-precision remote sensing data.

This paper outlines the findings of an ongoing research project. We aim to extend the longitudinal analysis of land cover changes (LCC) back to the 1950s using historical aerial photographs, to enhance the quality of land cover (LC) maps for the 1970s and the 1980s by utilizing panchromatic photographic reconnaissance satellite imagery. The project involves the acquisition of previously underutilized historical aerial and satellite reconnaissance imagery, as well as historical demographic data for all the populated places in the selected areas of interest. This study focuses on five diverse regions in Southeast Europe and Anatolia to examine LCC since the 1950s until today.

Building footprint extraction is a crucial task in remote sensing that helps acquire accurate building information for various applications such as city planning, population estimation, and disaster management. In this study, we explored the performance of Unet, Unet++, and DeepLabV3+ segmentation architectures on a very high-resolution Wuhan University Aerial Building dataset. We used InceptionResNetV2 and SE-ResNeXt101 encoders with these segmentation models after conducting pre-experiments with multiple encoder and hyper-parameter combinations. Furthermore, we implemented transfer learning by using the shared weights of a previous building detection study. We converted the raster outputs of deep learning models to vector format to enable a better spatial comparison among different models. All models were trained on the Kaggle platform, utilizing a Tesla P100-PCIe-16GB GPU and the PyTorch library. The F-1 scores for the test dataset range between 0.9867 and 0.9897 for different experiments. As a final assessment, we visually compared our experiment results with the Segment Anything Model.

This paper presents the first results of the ongoing development of new forest mapping methods for the Swedish national forest mapping case using Airborne Laser Scanning (ALS) data, utilizing the recent findings in machine learning (ML) and Artificial Intelligence (AI) techniques. We used Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) as ML models. In addition, Neural networks (NN) based approaches were utilized in this study. ALS derived features were used to estimate the stem volume (V), above-ground biomass (AGB), basal area (B), tree height (H), stem diameter (D), and forest stand age (A). XGBoost ML algorithm outperformed RF 1 % to 3 % in the R² metric. NN model performed similar to ML model, however it is superior in the estimation of V, AGB, and B parameters.

This paper presents the project ForestMap – a project aiming to develop and distribute new methods, which provide the benefits of accurate forest maps to a global audience. Using the recent developments in remote sensing, machine learning, and Artificial Intelligence (AI) the goal is to export the Scandinavian success stories to a wide range of stakeholders in the world.

This study aims to generate a new benchmark dataset from historical panchromatic aerial photographs suitable for deep learning-based Land use/Land cover (LULC) segmentation task. This new benchmark dataset spans a wide geographic area and consists of aerial photographs from various populous areas in Turkey and Bulgaria from the 1950s, 1960s, and 1970s. We implemented U-Net++ and Deeplabv3 segmentation architectures and appropriate hyperparameters and backbone structures to determine the applicability of this dataset, specifically for accurate and fast mapping of past terrain conditions. This unique historical LULC dataset and the different combinations of deep learning experiments proposed can be applied to different geographical regions with similar panchromatic datasets.

https://doi.org/10.1109/igarss46834.2022.9883715

Automatic road extraction from historical maps is an important task to understand past transportation conditions and conduct spatiotemporal analysis revealing information about historical events and human activities over the years. This research aimed to propose the ideal architecture, encoder, and hyperparameter settings for the historical road extraction task. We used a dataset including 7076 patches with the size of$256 \times256$pixels generated from scanned historical Deutsche Heereskarte 1:200 000 Türkei (DHK 200 Turkey) maps and their corresponding digitized ground truth masks for five different roads types. We first tested the widely used Unet++ and Deeplabv3 architectures. We also evaluated the contribution of attention models by implementing Unet++ with the concurrent spatial and channel-squeeze and excitation block and multiscale attention net. We achieved the best results with split-attention network (Timm-resnest200e) encoder and Unet++ architecture, with 98.99% overall accuracy, 41.99% intersection of union, 51.41% precision, 69.7% recall, and 57.72% F1 score values. Our output weights could be directly used for the inference of other DHK maps and transfer learning for similar or different historical maps. The proposed architecture could also be implemented in different road extraction studies.

The task of building footprint segmentation has been well-studied in the context of remote sensing (RS) as it provides valuable information in many aspects, however, difficulties brought by the nature of RS images such as variations in the spatial arrangements and in-consistent constructional patterns require studying further, since it often causes poorly classified segmentation maps. We address this need by designing a joint optimization scheme for the task of building footprint delineation and introducing two auxiliary tasks; image reconstruction and building footprint boundary segmentation with the intent to reveal the common underlying structure to advance the classification accuracy of a single task model under the favor of auxiliary tasks. In particular, we propose a deep multi-task learning (MTL) based unified fully convolutional framework which operates in an end-to-end manner by making use of joint loss function with learnable loss weights considering the homoscedastic uncertainty of each task loss. Experimental results conducted on the SpaceNet6 dataset demonstrate the potential of the proposed MTL framework as it improves the classification accuracy greatly compared to single-task and lesser compounded tasks.

Convolutional neural network (CNN)-based approaches have shown promising results in the pansharpening of the satellite images in recent years. However, they still exhibit limitations in producing high-quality pansharpening outputs. To that end, we propose a new self-supervised learning framework, where we treat pansharpening as a colorization problem, which brings an entirely novel perspective and solution to the problem compared with the existing methods that base their solution solely on producing a super-resolution version of the multispectral image. Whereas the CNN-based methods provide a reduced-resolution panchromatic image as the input to their model along with the reduced-resolution multispectral images and, hence, learn to increase their resolution together, we instead provide the grayscale transformed multispectral image as the input and train our model to learn the colorization of the grayscale input. We further address the fixed downscale ratio assumption during training, which does not generalize well to the full-resolution scenario. We introduce a noise injection into the training by randomly varying the downsampling ratios. Those two critical changes, along with the addition of adversarial training in the proposed PanColorization generative adversarial network (PanColorGAN) framework, help overcome the spatial-detail loss and blur problems that are observed in CNN-based pansharpening. The proposed approach outperforms the previous CNN-based and traditional methods, as demonstrated in our experiments.

The 1999 Izmit earthquake (Mw 7.4) on the North Anatolian Fault Zone resulted in severe damage to the urban areas of Izmit, Adapazari (Sakarya), Golcuk, and Yalova. A semivariogram approach was used to quantify earthquake-induced spatial variation and thereby the degree of damage in the Adapazari inner city. Semivariograms were calculated for 24 transects on SPOT high resolution visible infrared (HRVIR) panchromatic images obtained before and after the earthquake. The differences between the pre- and postearthquake semivariogram shape and measures of shape, range, nugget, and sill were related to the severity of earthquake damage