| 1. |
- Abid, Nosheen, 1993-, et al.
(författare)
-
UCL: Unsupervised Curriculum Learning for Water Body Classification from Remote Sensing Imagery
- 2021
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier. - 1569-8432 .- 1872-826X. ; 105
-
Tidskriftsartikel (refereegranskat)abstract
- This paper presents a Convolutional Neural Networks (CNN) based Unsupervised Curriculum Learning approach for the recognition of water bodies to overcome the stated challenges for remote sensing based RGB imagery. The unsupervised nature of the presented algorithm eliminates the need for labelled training data. The problem is cast as a two class clustering problem (water and non-water), while clustering is done on deep features obtained by a pre-trained CNN. After initial clusters have been identified, representative samples from each cluster are chosen by the unsupervised curriculum learning algorithm for fine-tuning the feature extractor. The stated process is repeated iteratively until convergence. Three datasets have been used to evaluate the approach and show its effectiveness on varying scales: (i) SAT-6 dataset comprising high resolution aircraft images, (ii) Sentinel-2 of EuroSAT, comprising remote sensing images with low resolution, and (iii) PakSAT, a new dataset we created for this study. PakSAT is the first Pakistani Sentinel-2 dataset designed to classify water bodies of Pakistan. Extensive experiments on these datasets demonstrate the progressive learning behaviour of UCL and reported promising results of water classification on all three datasets. The obtained accuracies outperform the supervised methods in domain adaptation, demonstrating the effectiveness of the proposed algorithm.
|
|
| 2. |
- Araza, Arnan, et al.
(författare)
-
Past decade above-ground biomass change comparisons from four multi-temporal global maps
- 2023
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - 1569-8432. ; 118
-
Tidskriftsartikel (refereegranskat)abstract
- Above-ground biomass (AGB) is considered an essential climate variable that underpins our knowledge and information about the role of forests in mitigating climate change. The availability of satellite-based AGB and AGB change (ΔAGB) products has increased in recent years. Here we assessed the past decade net ΔAGB derived from four recent global multi-date AGB maps: ESA-CCI maps, WRI-Flux model, JPL time series, and SMOS-LVOD time series. Our assessments explore and use different reference data sources with biomass re-measurements within the past decade. The reference data comprise National Forest Inventory (NFI) plot data, local ΔAGB maps from airborne LiDAR, and selected Forest Resource Assessment country data from countries with well-developed monitoring capacities. Map to reference data comparisons were performed at levels ranging from 100 m to 25 km spatial scale. The comparisons revealed that LiDAR data compared most reasonably with the maps, while the comparisons using NFI only showed some agreements at aggregation levels <10 km. Regardless of the aggregation level, AGB losses and gains according to the map comparisons were consistently smaller than the reference data. Map-map comparisons at 25 km highlighted that the maps consistently captured AGB losses in known deforestation hotspots. The comparisons also identified several carbon sink regions consistently detected by all maps. However, disagreement between maps is still large in key forest regions such as the Amazon basin. The overall ΔAGB map cross-correlation between maps varied in the range 0.11–0.29 (r). Reported ΔAGB magnitudes were largest in the high-resolution datasets including the CCI map differencing (stock change) and Flux model (gain-loss) methods, while they were smallest according to the coarser-resolution LVOD and JPL time series products, especially for AGB gains. Our results suggest that ΔAGB assessed from current maps can be biased and any use of the estimates should take that into account. Currently, ΔAGB reference data are sparse especially in the tropics but that deficit can be alleviated by upcoming LiDAR data networks in the context of Supersites and GEO-Trees.
|
|
| 3. |
- Axelsson, Arvid, et al.
(författare)
-
Tree species classification using Sentinel-2 imagery and Bayesian inference
- 2021
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 0303-2434. ; 100
-
Tidskriftsartikel (refereegranskat)abstract
- The increased temporal frequency of optical satellite data acquisitions provides a data stream that has the potential to improve land cover mapping, including mapping of tree species. However, for large area operational mapping, partial cloud cover and different image extents can pose challenges. Therefore, methods are needed to assimilate new images in a straightforward way without requiring a total spatial coverage for each new image. This study shows that Bayesian inference applied sequentially has the potential to solve this problem. To test Bayesian inference for tree species classification in the boreo-nemoral zone of southern Sweden, field data from the study area of Remningstorp (58°27′18.35″ N, 13°39′8.03″ E) were used. By updating class likelihood with an increasing number of combined Sentinel-2 images, a higher and more stable cross-validated overall accuracy was achieved. Based on a Mahalanobis distance, 23 images were automatically chosen from the period of 2016 to 2018 (from 142 images total). An overall accuracy of 87% (a Cohen’s kappa of 78.5%) was obtained for four tree species classes: Betula spp., Picea abies, Pinus sylvestris, and Quercus robur. This application of Bayesian inference in a boreo-nemoral forest suggests that it is a practical way to provide a high and stable classification accuracy. The method could be applied where data are not always complete for all areas. Furthermore, the method requires less reference data than if all images were used for classification simultaneously.
|
|
| 4. |
- Axelsson, Christoffer, et al.
(författare)
-
The use of dual-wavelength airborne laser scanning for estimating tree species composition and species-specific stem volumes in a boreal forest
- 2023
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 1872-826X. ; 118
-
Tidskriftsartikel (refereegranskat)abstract
- The estimation of species composition and species-specific stem volumes are critical components of many forest inventories. The use of airborne laser scanning with multiple spectral channels may prove instrumental for the cost-efficient retrieval of these forest variables. In this study, we scanned a boreal forest using two channels: 532 nm (green) and 1064 nm (near infrared). The data was used in a two-step methodology to (1) classify species, and (2) estimate species-specific stem volume at the level of individual tree crowns. The classification of pines, spruces and broadleaves involved linear discriminant analysis (LDA) and resulted in an overall accuracy of 91.1 % at the level of individual trees. For the estimation of stem volume, we employed species-specific k-nearest neighbors models and evaluated the performance at the plot level for 256 field plots located in central Sweden. This resulted in root-mean-square errors (RMSE) of 36 m3/ha (16 %) for total volume, 40 m3/ha (27 %) for pine volume, 32 m3/ha (48 %) for spruce volume, and 13 m3/ha (87 %) for broadleaf volume. We also simulated the use of a monospectral near infrared (NIR) scanner by excluding features based on the green channel. This resulted in lower overall accuracy for the species classification (86.8 %) and an RMSE of 41 m3/ha (18 %) for the estimation of total stem volume. The largest difference when only the NIR channel was used was the difficulty to accurately identify broadleaves and estimate broadleaf stem volume. When excluding the green channel, RMSE for broadleaved volume increased from 13 to 26 m3/ha. The study thus demonstrates the added benefit of the green channel for the estimation of both species composition and species-specific stem volumes. In addition, we investigated how tree height influences the results where shorter trees were found to be more difficult to classify correctly.
|
|
| 5. |
- Georganos, Stefanos, et al.
(författare)
-
A census from heaven : Unraveling the potential of deep learning and Earth Observation for intra-urban population mapping in data scarce environments
- 2022
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 1872-826X. ; 114
-
Tidskriftsartikel (refereegranskat)abstract
- Urban population distribution maps are vital elements for monitoring the Sustainable Development Goals, appropriately allocating resources such as vaccination campaigns, and facilitating evidence-based decision making. Typically, population distribution maps are derived from census data from the region of interest. Nevertheless, in several low-and middle-income countries, census information may be unreliable, outdated or unsuitable for spatial analysis at the intra-urban level, which poses severe limitations in the development of urban population maps of adequate quality. To address these shortcomings, we deploy a novel framework utilizing multisource Earth Observation (EO) information such as Sentinel-2 and very-high-resolution Pleiades imagery, openly available building footprint datasets, and deep learning (DL) architectures, providing end -to-end solutions to the production of high quality intra-urban population distribution maps in data scarce contexts. Using several case studies in Sub-Saharan Africa, namely Dakar (Senegal), Nairobi (Kenya) and Dar es Salaam (Tanzania), our results emphasize that the combination of DL and EO data is very potent and can successfully capture relationships between the retrieved image features and population counts at fine spatial resolutions (100 meter). Moreover, for the first time, we used state-of-the-art domain adaptation methods to predict population distributions in Dar es Salaam and Nairobi (R2 = 0.39, 0.60) that did not require national census or survey data from Kenya or Tanzania, but only a sample of training locations from Dakar. The DL architecture is based on a modified ResNet-18 model with dual-streams to analyze multi-modal data. Our findings have strong implications for the development of a new generation of urban population products that are an output of end-to-end solutions, can be updated frequently and rely completely on open data.
|
|
| 6. |
- Hinsby, Klaus, et al.
(författare)
-
Mapping and understanding Earth : Open access to digital geoscience data and knowledge supports societal needs and UN sustainable development goals
- 2024
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier. - 1569-8432 .- 1872-826X. ; 130
-
Tidskriftsartikel (refereegranskat)abstract
- Open access to harmonised digital data describing Earth 's surface and subsurface holds immense value for society. This paper highlights the significance of open access to digital geoscience data ranging from the shallow topsoil or seabed to depths of 5 km. Such data play a pivotal role in facilitating endeavours such as renewable geoenergy solutions, resilient urban planning, supply of critical raw materials, assessment and protection of water resources, mitigation of floods and droughts, identification of suitable locations for carbon capture and storage, development of offshore wind farms, disaster risk reduction, and conservation of ecosystems and biodiversity. EuroGeoSurveys, the Geological Surveys of Europe, have worked diligently for over a decade to ensure open access to harmonised digital European geoscience data and knowledge through the European Geological Data Infrastructure (EGDI). EGDI acts as a data and information resource for providing wide-ranging geoscience data and research, as this paper demonstrates through selected research data and information on four vital natural resources: geoenergy, critical raw materials, water, and soils. Importantly, it incorporates near realtime remote and in-situ monitoring data, thus constituting an invaluable up -to -date database that facilitates informed decision-making, policy implementation, sustainable resource management, the green transition, achieving UN Sustainable Development Goals (SDGs), and the envisioned future of digital twins in Earth sciences. EGDI and its thematic map viewer are tailored, continuously enhanced, and developed in collaboration with all relevant researchers and stakeholders. Its primary objective is to address societal needs by providing data for sustainable, secure, and integrated management of surface and subsurface resources, effectively establishing a geological service for Europe. We argue that open access to surface and subsurface geoscience data is crucial for an efficient green transition to a net -zero society, enabling integrated and coherent surface and subsurface spatial planning.
|
|
| 7. |
- Hu, Xikun, 1994-, et al.
(författare)
-
Sentinel-2 MSI data for active fire detection in major fire-prone biomes : A multi-criteria approach
- 2021
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 1872-826X. ; 101
-
Tidskriftsartikel (refereegranskat)abstract
- Sentinel-2 MultiSpectral Instrument (MSI) data exhibits the great potential of enhanced spatial and temporal coverage for monitoring biomass burning which could complement other coarse active fire detection products. This paper aims to investigate the use of reflective wavelength Sentinel-2 data to classify unambiguous active fire areas from inactive areas at 20 m spatial resolution. A multi-criteria approach based on the reflectance of several bands (i.e. B4, B11, and B12) is proposed to demonstrate the boundary constraints in several representative biomes. It is a fully automatic algorithm based on adaptive thresholds that are statistically determined from 11 million Sentinel-2 observations acquired over corresponding summertime (June 2019 to September 2019) across 14 regions or countries. Biome-based parameterizations avoid high omission errors (OE) caused by small and cool fires in different landscapes. It also takes advantage of the multiple criteria whose intersection could reduce the potential commission errors (CE) due to soil dominated pixels or highly reflective building rooftops. Active fire detection performance was mainly evaluated through visual inspection on eight illustrative subsets because of unavailable ground truth. The detection results revealed that CE and OE could be kept at a low level with 0.14 and 0.04 as an acceptable trade-off. The proposed algorithm can be employed for rapid active fire detection as soon as the image is obtained without the requirement of using multi-temporal imagery, and can even be adapted to onboard processing in the future.
|
|
| 8. |
- Kratzer, Susanne, et al.
(författare)
-
Integrating mooring and ship-based data for improved validation of OLCI chlorophyll-a products in the Baltic Sea
- 2021
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 1872-826X. ; 94
-
Tidskriftsartikel (refereegranskat)abstract
- A Water Quality Monitor (WQM) equipped with a range of oceanographic sensors was deployed from April 2017 to October 2017 in the North Western (NW) Baltic Sea. We assessed here if the data from a moored chlorophyll-a fluorometer can be used to improve satellite validation in coastal waters. Calibrated mooring data and ship-based chlorophyll-a concentrations from 2017 and 2018 were matched with OLCI (Ocean and Land Colour Instrument) data to validate the C2RCC (Case-2 Regional Coast Colour) processor, a locally-adapted version of C2RCC (LA-C2RCC), as well as the POLYMER processor. Using additional mooring data resulted in a substantial increase in paired observations compared to using ship-based data alone (C2RCC; N = 41-63, LA-C2RCC; N = 37-59, POLYMER; N = 108-166). However, the addition of mooring data only reduced the error and bias of the LA-C2RCC (MNB: from 24 % to 22 %, RMSE: from 60 % to 57 %, APD: both 47 %). In contrast, the statistical errors increased with the addition of mooring data both for C2RCC (MNB: -26 % to -33 %, RMSE: 50 %-51 %, APD 84 %-96 %) and for POLYMER (MNB: 26 %-36 %, RMSE: 79 % to 79 %, APD 64 %-64 %). The results indicate that the locally-adapted version of the C2RCC should be used for the area of investigation. These results are most likely also related to the effect of the System Vicarious Calibration (SVC). As opposed to C2RCC, the locally-adapted version had not been vicariously calibrated. The results indicate that SVC is not beneficial for Baltic Sea data and that more work needs to be done to improve SVC for Baltic Sea waters or for other waters with high CDOM absorption. In order to improve the validation capabilities of moored fluorometers in general, they should be strategically placed in waters with representative ranges of chl-a concentrations for the area of research in question.
|
|
| 9. |
- Lindberg, Eva, et al.
(författare)
-
Classification of tree species classes in a hemi-boreal forest from multispectral airborne laser scanning data using a mini raster cell method
- 2021
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 0303-2434 .- 1569-8432. ; 100
-
Tidskriftsartikel (refereegranskat)abstract
- Classification of tree species or species classes is still a challenge for remote sensing-based forest inventory. Operational use of Airborne Laser Scanning (ALS) data for prediction of forest variables has this far been dominated by area-based methods where laser scanning data have been used for estimation of forest variables within raster cells. Classification of tree species has however not been achieved with sufficient accuracy with area-based methods using only ALS data. Furthermore, analysis of tree species at the level of raster cells with typical size of 15 m ? 15 m is not ideal in the case of mixed species stands. Most ALS systems for terrestrial mapping use only one wavelength of light. New multispectral ALS systems for terrestrial mapping have recently become operational, such as the Optech Titan system with wavelengths 1550 nm, 1064 nm, and 532 nm. This study presents an alternative type of area-based method for classification of tree species classes where multispectral ALS data are used in combination with small raster cells. In this ?mini raster cell method? features for classification are derived from the intensity of the different wavelengths in small raster cells using a moving window average approach to allow for a heterogeneous tree species composition. The most common tree species in the Nordic countries are Pinus sylvestris and Picea abies, constituting about 80% of the growing stock volume. The remaining 20% consists of several deciduous species, mainly Betula pendula and Betula pubescens, and often grow in mixed forest stands. Classification was done for pine (Pinus sylvestris), spruce (Picea abies), deciduous species and mixed species in middle-aged and mature stands in a study area located in hemi-boreal forest in the southwest of Sweden (N 58?27?, E 13?39?). The results were validated at plot level with the tree species composition defined as proportion of basal area of the tree species classes. The mini raster cell classification method was slightly more accurate (75% overall accuracy) than classification with a plot level area-based method (68% overall accuracy). The explanation is most likely that the mini raster cell method is successful at classifying homogenous patches of tree species classes within a field plot, while classification based on plot level analysis requires one or several heterogeneous classes of mixed species forest. The mini raster cell method also results in a high-resolution tree species map. The small raster cells can be aggregated to estimate tree species composition for arbitrary areas, for example forest stands or area units corresponding to field plots.
|
|
| 10. |
- Martinez Garcia, Eduardo
(författare)
-
Artificial intelligence-based software (AID-FOREST) for tree detection: A new framework for fast and accurate forest inventorying using LiDAR point clouds
- 2022
-
Ingår i: International Journal of Applied Earth Observation and Geoinformation. - : Elsevier BV. - 1569-8432 .- 1872-826X. ; 113
-
Tidskriftsartikel (refereegranskat)abstract
- Forest inventories are essential to accurately estimate different dendrometric and forest stand parameters. However, classical forest inventories are time consuming, slow to conduct, sometimes inaccurate and costly. To address this problem, an efficient alternative approach has been sought and designed that will make this type of field work cheaper, faster, more accurate, and easier to complete. The implementation of this concept has required the development of a specifically designed software called "Artificial Intelligence for Digital Forest (AID-FOREST)", which is able to process point clouds obtained via mobile terrestrial laser scanning (MTLS) and then, to provide an array of multiple useful and accurate dendrometric and forest stand parameters. Singular characteristics of this approach are: No data pre-processing is required either pre-treatment of forest stand; fully automatic process once launched; no limitations by the size of the point cloud file and fast computations.To validate AID-FOREST, results provided by this software were compared against the obtained from in-situ classical forest inventories. To guaranty the soundness and generality of the comparison, different tree spe-cies, plot sizes, and tree densities were measured and analysed. A total of 76 plots (10,887 trees) were selected to conduct both a classic forest inventory reference method and a MTLS (ZEB-HORIZON, Geoslam, ltd.) scanning to obtain point clouds for AID-FOREST processing, known as the MTLS-AIDFOREST method. Thus, we compared the data collected by both methods estimating the average number of trees and diameter at breast height (DBH) for each plot. Moreover, 71 additional individual trees were scanned with MTLS and processed by AID-FOREST and were then felled and divided into logs measuring 1 m in length. This allowed us to accurately measure the DBH, total height, and total volume of the stems.When we compared the results obtained with each methodology, the mean detectability was 97% and ranged from 81.3 to 100%, with a bias (underestimation by MTLS-AIDFOREST method) in the number of trees per plot of 2.8% and a relative root-mean-square error (RMSE) of 9.2%. Species, plot size, and tree density did not significantly affect detectability. However, this parameter was significantly affected by the ecosystem visual complexity index (EVCI). The average DBH per plot was underestimated (but was not significantly different from 0) by the MTLS-AIDFOREST, with the average bias for pooled data being 1.8% with a RMSE of 7.5%. Similarly, there was no statistically significant differences between the two distribution functions of the DBH at the 95.0% confidence level.Regarding the individual tree parameters, MTLS-AIDFOREST underestimated DBH by 0.16 % (RMSE = 5.2 %) and overestimated the stem volume (Vt) by 1.37 % (RMSE = 14.3 %, although the BIAS was not statistically significantly different from 0). However, the MTLS-AIDFOREST method overestimated the total height (Ht) of the trees by a mean 1.33 m (5.1 %; relative RMSE = 11.5 %), because of the different height concepts measured by both methodological approaches. Finally, AID-FOREST required 30 to 66 min per ha-1 to fully automatically process the point cloud data from the *.las file corresponding to a given hectare plot. Thus, applying our MTLS-AIDFOREST methodology to make full forest inventories, required a 57.3 % of the time required to perform classical plot forest inventories (excluding the data postprocessing time in the latter case). A free trial of AID -FOREST can be requested at dielmo@dielmo.com.
|
|
