In this study, we examine the potential of continuous ground moisture monitoring over a mining site using a combination of in-situ soil moisture sensors and multi-sensor SAR images. We focus on examining and improving methodologies for surface soil moisture (or ground moisture) retrieval from SAR measurements focusing on detailed in situ reference observations for several key sediments types in the study area. The mining site represents a limestone quarry located in southeastern Finland. We hypothesize that sediment-specific well-calibrated models can be instrumental in improving soil moisture retrieval under different weather conditions to produce spatially explicit ground moisture estimates at high resolution compared to baseline approaches. Studied SAR data are represented by Copernicus Sentinel-1 C-band images, and methodologies will be expanded later to L-band images from ALOS-2 PALSAR-2 and the upcoming NISAR mission.

Cloud detection is a necessary step before the application of remote sensing images. However, the radiation intensity similarity between artificial lights and clouds is higher in nighttime remote sensing images than in daytime remote sensing images, making it difficult to distinguish artificial lights from clouds. This article proposes a deep learning method called multifeature fusion for cloud detection network (MFFCD-Net) to detect clouds in daytime and nighttime remote sensing images. A dilated residual upsampling module was designed for upsampling feature maps while enlarging the receptive field. A multiscale feature-extraction fusion module (MFEF) was designed to enhance the ability to distinguish regular textures of artificial lights from random textures of clouds. Moreover, an adaptive feature-fusion module (AFF) was designed to select and fuse the feature in the encoding stage and decoding stage, thus improving the cloud detection accuracy. To the best of our knowledge, this is the first time that a method is designed for cloud detection in both daytime and nighttime remote sensing images. The experimental results on Suomi-NPP Visible Infrared Imaging Radiometer Suite (VIIRS) of the panchromatic (PAN) day/night band (DNB) images show that MFFCD-Net could obtain a better balance in commission and omission rates than baseline methods (92.3% versus 90.5% on the F1-score) in daytime remote sensing images. Although artificial lights introduce strong interference in nighttime remote sensing images, MFFCD-Net can better distinguish artificial lights from clouds than baseline methods (90.8% versus 88.4% on the F1-score). The results indicate that MFFCD-Net is promising for cloud detection both in daytime and nighttime remote sensing images. The source code and dataset are available athttps://github.com/Neooolee/MFFCD-Net.

In this study, we examine the potential of several self-supervised deep learning models in predicting forest attributes and detecting forest changes using ESA Sentinel-1 and Sentinel-2 images. The performance of the proposed deep learning models is compared to established conventional machine learning approaches. Studied use-cases include mapping of forest disturbance (windthrown forests, snowload damages) using deep change vector analysis, forest height mapping using UNet+ based models, Momentum contrast and regression modeling. Study areas were represented by several boreal forest sites in Finland. Our results indicate that developed methods allow to achieve superior classification and prediction accuracies compared to traditional methodologies and mimimize the amount of necessary in-situ forestry data.

Clouds are a very important factor in the availability of optical remote sensing images. Recently, deep learning (DL)-based cloud detection methods have surpassed classical methods based on rules and physical models of clouds. However, most of these deep models are very large, which limits their applicability and explainability, while other models do not make use of the full spectral information in multispectral images, such as Sentinel-2. In this article, we propose a lightweight network for cloud detection, fusing multiscale spectral and spatial features (CD-FM3SFs) and tailored for processing all spectral bands in Sentinel-2A images. The proposed method consists of an encoder and a decoder. In the encoder, three input branches are designed to handle spectral bands at their native resolution and extract multiscale spectral features. Three novel components are designed: a mixed depthwise separable convolution (MDSC) and a shared and dilated residual block (SDRB) to extract multiscale spatial features, and a concatenation and sum (CS) operation to fuse multiscale spectral and spatial features with little calculation and no additional parameters. The decoder of CD-FM3SF outputs three cloud masks at the same resolution as input bands to enhance the supervision information of small, middle, and large clouds. To validate the performance of the proposed method, we manually labeled 36 Sentinel-2A scenes evenly distributed over mainland China. The experiment results demonstrate that CD-FM3SF outperforms traditional cloud detection methods and state-of-the-art DL-based methods in both accuracy and speed.

The spectral and spatial resolutions of modern optical Earth observation data are continuously increasing. To fully utilize the data, integrate them with other information sources, and create applications relevant to real-world problems, extensive training data are required. We present TAIGA, an open dataset including continuous and categorical forestry data, accompanied by airborne hyperspectral imagery with a pixel size of 0.7 m. The dataset contains over 70 million labeled pixels belonging to more than 600 forest stands. To establish a baseline on TAIGA dataset for multitask learning, we trained and validated a convolutional neural network to simultaneously retrieve 13 forest variables. Due to the size of the imagery, the training and testing sets were independent, with strictly no overlap for patches up to$45\times 45$pixels. Our retrieval results show that including both spectral and textural information improves the accuracy of mapping key boreal forest structural characteristics, compared with an earlier study including only spectral information from the same image. TAIGA responds to the increased availability of hyperspectral and very high resolution imagery, and includes the forestry variables relevant for forestry and environmental applications. We propose the dataset as a new benchmark for spatial–spectral methods that overcomes the limitations of widely used small-scale hyperspectral datasets.

Very high resolution remote sensing data of forests, where individual tree crowns are separable, contains structural information on tree size and density. Such information is complementary to the spectral signatures currently used in forestry applications. Advanced machine learning methods, e.g. convolutional neural networks (CNNs), offer an automated and standardized way of retrieving both spectral and structural information from imagery. A key characteristic in CNNs is patch size, which should be large enough to include dominant structural scale, yet as small as possible to avoid unnecessary averaging. Our results show that the patch should be larger than one tree, but increasing it excessively reduces retrieval accuracy. Furthermore, large patch sizes can cause loss of independence between training and validation data, leading to overestimating model performance.

Cloud detection is an important step in the processing of remote sensing images. Most methods based on convolutional neural networks (CNNs) for cloud detection require pixel-level labels, which are time-consuming and expensive to annotate. To overcome this challenge, this letter proposes a novel semisupervised algorithm for cloud detection by training a self-attentive generative adversarial network (SAGAN) to extract the feature difference between cloud images and cloud-free images. Our main idea is to introduce visual attention into the process of generating “real” cloud-free images. The training of SAGAN is based on three guiding principles: expansion of attention maps of cloud regions which will be replaced with translated cloud-free images, reduction of attention maps to coincide with cloud boundaries, and optimization of a self-attentive network to handle the extreme cases. The inputs for SAGAN training are the images and image-level labels, which are easier, cheaper, and more time-saving than the existing methods based on CNN. To test the performance of SAGAN, experiments are conducted on the Sentinel-2A Level 1C image data. The results show that the proposed method achieves very promising results with only the image-level labels of training samples.

Spatial-spectral approaches applied on hyperspectral images (HSI) with limited labels suffer from overfitting when the size of input filters and the percentage of training data increases. In those cases, pixel values corresponding to testing sets are partly or completely seen during training phase, reducing the number independent testing pixels and leading to overoptimistic accuracy assessment. These effects have been demonstrated in several previous works but still require attention. In this work we propose additional visulizations and measures of the overlapping and overfitting effects, demonstrated on common HSI datasets, to increase awareness on these issues.

Citizen Science or participatory sensing can help filling in the gaps of in-situ reference forestry data needed to analyze satellite images. In this study, we show that the Relasphone, a biomass measuring application previously developed and tested in boreal forests of Finland and temperate-cold pine forests in Durango, Mexico, can be adapted to pine forests in Asturias, Spain. Relasphone measurements were in good agreement with reference data over 20 plots (R2= 0.89 for stem volume) and performed well against 3 traditional relas-cope measuring devices. The results suggest the Relasphone can be easily deployed in other temperate biomes.

In this study, we utilized all available Landsat images over two adjacent orbits between 1997 and 2015 for the quasi automatic detection of clearcuts and storm damages in boreal forest of Finland. Landsat time series modelling and analysis was done utilizing the Continuous Change Detection and Classification (CCDC) algorithm with a slight modification for rapid operative conditions. The change maps derived from dense time series analysis showed a good agreement compared to reference maps of clearcuts and storm damages obtained from visual interpretation of Very High Resolution image pairs, by lack of reliable reference in temporal and or spatial domain.

Many cloud and shadow detection methods have been proposed already, but improvements can be made on accuracy or automation. In this study, we propose a Fully Convolutional Network model for the detection of clouds and shadows in optical satellite images. The proposed model was trained on 165 Landsat images in Finland, and tested on an independent set of images. The cloud and shadow detection accuracy reached 95%, outperforming both quantitatively and qualitatively a selection of other deep learning architectures.

This is a selection of results of the North State project, that demonstrate how innovative methods applied to the new Sentinel data streams can be combined with models to monitor carbon and water fluxes for pan-boreal Europe.

Citizen Science, propelled by the growing popularity of smartphones, can provide valuable reference information for remote sensing image analysis. We demonstrate that the Relasphone, a biomass measuring application previously developed and tested in boreal forests of Finland, can be adapted to temperate-cold pine forests in Durango State, Mexico. Relasphone measurements were in good agreement with reference data over 55 plots (R2= 0.94), and have been used to produce a biomass map over Durango from a Landsat 8 image. The results suggest the Relasphone can be easily deployed in other biomes.

Citizen observations, environmental data gathered by volunteers without professional observation capabilities, have been extensively used for Finnish water quality monitoring tasks. Recently, mobile smartphones and their digital cameras have enabled more direct measurements of transparency related water quality variables with inexpensive technology suitable for volunteers. These “Secchi3000” ideas of measurement technology by viewing known targets through multiple viewing path lengths within measured water were used to develop an iQwtr measurement device for water transparency related citizen observations. Past experiences with crowdsourcing and use of in situ water transparency data with satellite observations are reviewed and future challenges outlined.

Storms and heavy winds can induce degradations on wide areas of forest, which have a severe impact on the ecological habitat and on economical value. During ten days in July and August 2010, the thunderstorms named Asta, Veera, Lahja and Sylvi raged in southern and middle Finland. About 8.1 million m3of forest was damaged, and the storms caused severe damage to the country infrastructure (e.g. buildings, power lines and railroads). The estimated financial losses of Asta storm alone were over 20 million euros.

The availability of ground reference forest data can be a bottleneck in remote sensing studies. Data may be available in only limited areas because of the cost and lengthy process of traditional forest inventory data collection by professionals. In certain cases, forest inventory data may not be easy to obtain, if not impossible.

This paper compares ALOS PALSAR fully polarimetric and dual-polarized data in the application area of land cover mapping. To assure versatile comparison of the data, different classification methods and different features of data are used. Two of the classification methods used are based on supervised classification and two on unsupervised classification. Polarimetric data are used in three ways: (1) as fully polarimetric data; (2) features calculated from fully polarimetric data; and (3) intensity data of selected channels. Combinations of six (water, field, sparse forest, dense forest, peat land, and urban areas), five, four, and three classes were used for classification. Fully polarimetric data gave better results (87.5%-84.7% with three classes; open land areas, forest, and water) than intensity data only (83.6%-78.6%), but the differences in the overall accuracies between the methods were not more than 7.6%. Kappa coefficients of agreement are moderate for all the classifications. Supervised classification can be expected to perform better than unsupervised classification, given that the training areas can be selected accurately. Dual polarization data were found to be an attractive alternative in cases where fully polarimetric data are not available or it is of low resolution. With intensities of selected polarimetric features, it was possible to obtain a high classification accuracy as with fully polarimetric data. This also opens possibilities for nonspecialist users to benefit from polarimetric information in classification.

The objective of ForSe - Season Monitoring study was to develop an automatic method to analyze web-camera images of nature. As the outcome the image analysis produces indices that indicate the seasonal development stage of the forest (e.g. degree of autumn colour of deciduous trees). IP web-cameras of a pilot camera network were programmed to take one image in 15 minute interval on daylight hours during autumn period. One camera was used as a source of the training data (Enontekio), and one for testing data (Oulanka). The image data was preprocessed to reduce noise and to and spectral angle feature was calculated to compensate the illumination variations between consequential images and within a single image. Selected areas of the training site camera images of autumn season were classified into six classes describing the seasonal status of the leaves (green, light green, yellow, red, brown, fallen). The spectral angle features were calculated for these areas and clustered by K-means into 30 clusters. Class labels were assigned to the cluster centres using k-NN method (k= 5). To see the progress of a certain colour class in the time series of images of a test site camera, the classified pixels within selected regions of interest (ROI) were used to produce a continuous season colour index (SCI). The behaviour of the index was compared with a reference classification supplied by phenology experts from Finnish Forest Research Institute (Metla).

Six polarimetric change detection indices were tested on L band ALOS PALSAR data over Kuortane, Finland. Tests included quantitative evaluation of change indices compared to reference change maps, and qualitative evaluation by visual inspection. Results suggest an additional accuracy using fully polarimetric data in change detection. Contrast Ratio and the Wishart test gave the best results among the tested indices.

The potential of TerraSAR-X imagery for forest cutting monitoring was experimentally investigated in images over Kuortane, Finland. In this study, 10 dual-polarised change detection indices were tested versus a ground truth of clearcuts done between the acquisition dates. Preliminary results suggest the decrease of entropy and the increase of coherence at X-band could be good change detection indices. Results were probably affected by adverse weather conditions during the acquisition of the second scene.

In this study, we extended the potential of a Content- Based Image Retrieval (CBIR) system based on Self-Organizing Maps (SOMs), for the analysis of remote sensing data. A database was artificially created by splitting each image to be analyzed into small images (orimagelets). Content-based image retrieval was applied to fully polarimetric airborne SAR data, using a selection of polarimetric features. After training the system on this imagelet database, automatic queries could detect changes. Results were encouraging on airborne SAR data and may be more useful for spaceborne polarimetric data.

This article aims at providing a comparison of polarimetric change detection indices from a practical point of view. Six polarimetric change detection indices were tested on L band EMISAR data over Norway. Tests included quantitative evaluation of change maps compared to a ground truth of changes, and qualitative evaluation by visual inspection. Contrast ratio and the Wishart test gave the best results among the tested indices.

Methods for terrain correction of polarimetric SAR data were studied and developed. Ortho-rectification resampling and amplitude correction utilized Stokes matrix data. The Stokes matrix of thermal noise was subtracted before amplitude normalization. Application of an azimuth-slope correction algorithm resulted in slightly narrower distribution of orientation angles compared to input data.

The increasing amount and resolution of satellite sensors demand new techniques for browsing remote sensing image archives. Content-based querying allows an efficient retrieval of images based on the information they contain, rather than their acquisition date or geographical extent. Self-organizing maps (SOMs) have been successfully applied in the PicSOM system to content-based image retrieval in databases of conventional images. In this paper, we investigate and extend the potential of PicSOM for the analysis of remote sensing data. We propose methods for detecting man-made structures, as well as supervised and unsupervised change detection, based on the same framework. In this paper, a database was artificially created by splitting each satellite image to be analyzed into small images. After training the PicSOM on this imagelet database, both interactive and off-line queries were made to detect man-made structures, as well as changes between two very high resolution images from different years. Experimental results were both evaluated quantitatively and discussed qualitatively, and suggest that this new approach is suitable for analyzing very high resolution optical satellite imagery. Possible applications of this work include interactive detection of man-made structures or supervised monitoring of sensitive sites

Trees were located and classified to pine and spruce classes using features computed from Ikonos multispectral and panchromatic channels at study site in Eastern Finland. Spectral signatures were sampled from the extracted tree locations, and a set of contextual features was computed in the neighborhood around each located tree from the Ikonos panchromatic channel. Circular masks of five different sizes were used. The contextual features included higher order statistical features (skewness, kurtosis), and additional features obtained by fitting either two Gaussian distributions or a Weibull distribution to the intensity histogram. The contextual features aimed at capturing differences in the distribution of intensities of pine and spruce crowns. Pure (100%) pine and spruce plots with medium stem volume (100 - 200 mVha), and with a minimum distance of 15 m from stand borders, were used in the study. The training data contained 5 plots of pine and 5 plots of spruce, from which 196 trees were located (96 pine, 100 spruce). The separate validation data set consisted of 6 plots (3 plots both pine and spruce) containing 116 trees (54 pine, 62 spruce). Stepwise linear discriminant analysis was used to select the best separating features and for classification. From the multispectral channels, the best separating feature was the blue channel. From the contextual features the best separating features were the Weibull shape parameter, the ratio of sample mean and median, kurtosis, and skewness, the set of best features being slightly different for different sampling radii. For the validation data set, the percentage of correctly classified trees was 87.9% when using spectral channels only, and increased from 81.9% to 98.3% along with increasing sampling radius using only the contextual features. The classification accuracy reached 99.1% when both spectral and contextual features were used.

Content-based querying allows efficient retrieval of images based on the information they contain, rather than acquisition date or geographical extent. We extend the potential of a content-based image retrieval (CBIR) system based on Self- Organizing Maps (SOMs), to the analysis of remote sensing data. A database was artificially created by splitting each satellite image to be analyzed into small images. After training the CBIR system on this imagelet database, both interactive and off-line queries were made to detect man-made structures, as well as changes. Experimental results suggest that this new approach is suitable for analyzing very high-resolution optical satellite imagery. Possible applications include interactive detection of man-made structures and supervised monitoring of sensitive sites.