Forest Mapping Research Articles

Using quantile mapping and random forest for bias‐correction of high‐resolution reanalysis precipitation data and CMIP6 climate projections over Iran

AbstractClimate change is expected to cause important changes in precipitation patterns in Iran until the end of 21st century. This study aims at evaluating projections of climate change over Iran by using five climate model outputs (including ACCESS‐ESM1‐5, BCC‐CSM2‐MR, CanESM5, CMCC‐ESM2 and MRI‐ESM2‐0) of the Coupled Model Intercomparison Project phase 6 (CMIP6), and performing bias‐correction using a novel combination of quantile mapping (QM) and random forest (RF) between the years 2015 and 2100 under three shared socioeconomics pathways (SSP2‐4.5, SSP3‐7.0 and SSP5‐8.5). First, bias‐correction was performed on ERA5‐Land reanalysis data as reference period (1990–2020) using the QM method, then the corrected ERA5‐Land reanalysis data was considered as measured data. Based on the corrected ERA5‐Land reanalysis data (1990–2020) and historical simulations (1990–2014), the future projections (2015–2100) were also bias‐corrected utilizing the QM method. Next, the accuracy of the QM method was validated by comparing the corrected ERA5‐Land reanalysis data with model outputs for overlapping years between 2015 and 2020. This comparison revealed persistent biases; hence, a combination of QM‐RF method was applied to rectify future climate projections until the end of the 21st century. Based on the QM result, CMCC‐ESM2 revealed the highest RMSE in both SSP2‐4.5 and SSP3‐7.0 amounting to 331.74 and 201.84 mm·year−1, respectively. Particularly, the exclusive use of the QM method displayed substantial errors in projecting annual precipitation based on SSP5‐8.5, notably in the case of ACCESS‐ESM1‐5 (RMSE = 431.39 mm·year−1), while the RMSE reduced after using QM‐RF method (197.75 mm·year−1). Obviously, a significant enhancement in results was observed upon implementing the QM‐RF combination method in CMCC‐ESM2 under both SSP2‐4.5 (RMSE = 139.30 mm·year−1) and SSP3‐7.0 (RMSE = 151.43 mm·year−1) showcasing approximately reduction in RMSE values by 192.43 and 50.41 mm·year−1, respectively. Although each bias‐corrected model output was evaluated individually, multi‐model ensemble (MME) was also created to project the annual future precipitation pattern in Iran. By considering that combination of QM‐RF method revealed the lower errors in correcting model outputs, we used the QM‐RF technique to create the MME. Based on SSP2‐4.5, the MME climate projections highlight imminent precipitation reductions (>10%) across large regions of Iran, conversely projecting increases ranging from 10% to over 20% in southern areas under SSP3‐7.0. Moreover, MME projected dramatic declines under SSP5‐8.5, especially impacting central, eastern, and northwest Iran. Notably, the most pronounced possibly decline patterns are projected for arid regions (central plateau) and eastern areas under SSP2‐4.5, SSP3‐7.0 and SSP5‐8.5.

Enhancing Forest Conservation Efforts: Mapping of High Conservation Value Forests in the Republic of Moldova

Enhancing Forest Conservation Efforts: Mapping of High Conservation Value Forests in the Republic of Moldova

Comparative Analysis of Different Forest Fire Susceptibility Models in Benguet, the Philippines

Fire is a common threat to forests in our country. It is usually uncontrolled and consumes the natural fuels of the forest, affecting both the vegetation and the soil. In the Philippines, forest fires are mostly frequent in regions with naturally occurring vegetation that is highly susceptible to forest fire, and one of them is the Cordillera Administrative Region. This study was conducted to [1] determine forest fire occurrences in Benguet province, [2] generate forest fire hazard maps using three indices [structural fire index (SFI), fire risk index (FRI), and hybrid fire index (HFI)], and [3] recommend potential forest management strategies to reduce fire susceptibility or manage fire occurrence in the province. NDMI, elevation, slope, aspect, distance from roads, and vicinity to settlements were the influencing factors used in the occurrence of forest fires. Moreover, forest fire occurrences in the province were collected from DENR records. Results revealed that from 2019–2021, a total of 90 forest fire incidents were observed in the area. In terms of the indices, results showed that maps using HFI and FRI revealed better results than the SFI model. Furthermore, there were forest fire incidences that coincided with the FRI model, particularly in those areas identified as having high and very high susceptibility ratings.

High-resolution operational soil moisture monitoring for forests in central Germany

Abstract. The forests of central Germany (Saxony, Saxony-Anhalt, and Thuringia) are vital components of the local ecosystems, the economy, and recreation. However, in recent years, these forests have faced significant challenges due to prolonged climate-change-induced droughts, causing water shortages, tree stress, and pest outbreaks. One of the key components of the forests' vitality and productivity is the availability of soil moisture. Given the anticipated increase in the frequency and severity of drought events, there is a growing demand for accurate and real-time soil moisture information. This underscores the need for development of an appropriate monitoring tool to make forest management strategies more effective. The article introduces an operational high-resolution soil moisture monitoring framework for the forests in central Germany. The key components of this system include the advanced LWF-BROOK90 1D water balance model, a large database of the National Federal Forest Inventory, high-resolution forest soil maps, real-time climate data from the German Meteorological Service, and a web information platform for the presentation of daily updated results. This system informs the public and empowers forest managers and other decision-makers to take targeted, local measures for sustainable forest management, aiding in both drought mitigation and long-term forest health in the face of climate change. The validation of the system using soil moisture measurements from 51 stations with various sensor depths (up to 100 cm) showed an overall good agreement (0.76 median Pearson correlation), which was found to be higher for deciduous rather than coniferous forests. Finally, the framework is discussed against the background of the main limitations of existing monitoring systems and how operational soil moisture measurements contribute to better interpretation of simulations.

UAV survey mapping of illegal deforestation in Madagascar

Societal Impact StatementUnmanned aerial vehicle (UAV) imagery highlights the extent of illegal deforestation in protected areas for the biodiverse humid forest of central Madagascar. The ultra‐high‐resolution (<10‐cm pixel) images enable the creation of detailed forest 3D base maps and provide the means to quantify forest stand losses. To help communities safeguard their forests, local non‐governmental organisations can use UAV maps in combination with weekly deforestation alerts to facilitate an immediate on‐ground response that significantly restricts illegal activity. Integrating ultra‐high‐resolution UAV mapping and coarse‐resolution freely available satellite imagery should have much wider applications in Madagascar and the humid tropics for community‐based conservation.Summary This study of the Ambohimahamasina humid forest shows that small UAVs offer a detailed (<10‐cm pixel), rapid and cost‐effective solution to provide maps of detailed deforestation patterns not visible in satellite imagery. Calculating forest extent and volume are valuable ways to rapidly assess forest losses and prioritise areas for ground patrols. The use of 3‐dimensional measurements for above ground carbon estimates indicate how, in the future, these metrics could be used to calculate carbon payments for conservation programs. By combining UAV and free satellite imagery, an effective alert system has been developed that supports community initiatives in the protection of their natural forest resources. The wealth of ultra‐high‐resolution UAV data collected in this study provides insights into forest dynamics, supports local community forest management, and has the potential to measure the value of the forest.

Estradiol and testosterone associated with risk of breast cancer: a meta-analysis

Background: To investigate the correlation between estradiol and testosterone in patients with breast cancer. Methods: The literatures on the correlation between estradiol and testosterone on the risk of breast cancer were searched and collected. The time limit is that each database is established until December 2023. After screening, the modified Jadad scale was used to evaluate the quality of the research literature. NoteExpress 3.2 was used for literature management, and Excel 2003 was used for data collection and extraction. Statistical analysis was performed using RevMan 5.4.1 software to determine whether there was heterogeneity in the study according to the size of Q test (P-value), and then the OR value of combined effects was calculated using fixed or random effects models and forest maps were drawn. At the same time, the literatures with the greatest weight were excluded for sensitivity analysis, and the literature bias was evaluated by drawing funnel plot. Results: A total of 628 literatures were retrieved, and 11 case-control trials met the criteria for inclusion. Meta-analysis results showed that the level of E2 in breast cancer patients was higher than that in non-breast cancer control group, but the difference was not statistically significant (OR=121.56, 95%CI (-3.32-264.44), P=0.06). The level of E2 in premenopausal patients with breast cancer was higher than that in non-breast cancer control group, but the difference was not statistically significant (OR=8.26, 95%CI (-2.83-19.34), P=0.14). The level of E2 in postmenopausal patients with breast cancer was higher than that in non-breast cancer control group, and the difference was statistically significant (OR=20.36, 95%CI (7.04-33.68), P=0.003). Preoperative T level was higher in patients with breast cancer than in non-breast cancer control group, but the difference was not statistically significant (OR=14.77, 95%CI (-14.11-43.65), P=0.32). The T level before and after surgery in breast cancer patients was higher than that in non-breast cancer control group, and the difference was statistically significant (OR=12.91, 95%CI (4.43-21.39), P=0.003). Sensitivity analysis showed that the combined effect size results were stable and reliable OR (95%CI) was 24.41 (10.21~38.61), P=0.0007. Funnel plot results showed publication bias. Conclusion There is a positive correlation between the levels of estradiol and testosterone and the occurrence and development of breast cancer after menopause.

Mapping Forest Growing Stock and Its Current Annual Increment Using Random Forest and Remote Sensing Data in Northeast Italy

The role of forests in providing multiple goods and services has been recognized worldwide. In such a context, reliable spatial predictions of forest attributes such as tree volume and current increment are fundamental for conducting forest monitoring, improving restoration programs, and supporting decision-making processes. This article presents the methodology and the results of the wall-to-wall spatialization of the growing stock volume and the current annual increment measured in 273 plots of data of the Italian National Forest Inventory over an area of more than 3260 km2 in the Friuli Venezia Giulia region (Northeast Italy). To this aim, a random forest model was tested using as predictors 4 spectral indices from Sentinel-2, a high-resolution Canopy Height Model derived from LiDAR, and geo-morphological data. According to the Leave One Out cross-validation procedure, the model for the growing stock shows an R2 and an RMSE% of 0.67 and 41%, respectively. Instead, an R2 of 0.47 and an RMSE% of 57% were obtained for the current annual increment. The validation with an independent dataset further improved the models’ performances, yielding significantly higher R2 values of 0.84 and 0.83 for volume and for increment, respectively. Our results underline a relatively higher importance of LiDAR-derived metrics compared to other covariates in estimating both attributes, as they were even twice as important as vegetation indices for growing stock. Therefore, these metrics are promising for the development of a national LiDAR-based model.

Is Your Training Data Really Ground Truth? A Quality Assessment of Manual Annotation for Individual Tree Crown Delineation

For the accurate and automatic mapping of forest stands based on very-high-resolution satellite imagery and digital orthophotos, precise object detection at the individual tree level is necessary. Currently, supervised deep learning models are primarily applied for this task. To train a reliable model, it is crucial to have an accurate tree crown annotation dataset. The current method of generating these training datasets still relies on manual annotation and labeling. Because of the intricate contours of tree crowns, vegetation density in natural forests and the insufficient ground sampling distance of the imagery, manually generated annotations are error-prone. It is unlikely that the manually delineated tree crowns represent the true conditions on the ground. If these error-prone annotations are used as training data for deep learning models, this may lead to inaccurate mapping results for the models. This study critically validates manual tree crown annotations on two study sites: a forest-like plantation on a cemetery and a natural city forest. The validation is based on tree reference data in the form of an official tree register and tree segments extracted from UAV laser scanning (ULS) data for the quality assessment of a training dataset. The validation results reveal that the manual annotations detect only 37% of the tree crowns in the forest-like plantation area and 10% of the tree crowns in the natural forest correctly. Furthermore, it is frequent for multiple trees to be interpreted in the annotation as a single tree at both study sites.

How important is Forest Age in explaining the species composition of Near-natural Spruce Forests?

Multiple ecological drivers, along with forest age, determine the species composition of boreal forest ecosystems. However, the role of age in successional changes in forests cannot be understood without taking site conditions, the disturbance regime and forest structure into account. In this study, we ask two research questions: 1. What is the relationship between forest age and overall species composition in older near-natural spruce forests, i.e. forests of age beyond harvest maturity? 2. Do species associated with different forest habitats respond similarly to variation in forest age? Data were collected in 257 Norway spruce dominated 0.25 ha plots from three study areas in Southeastern and Central Norway. Species inventories were conducted for lichens and bryophytes on trees and rocks, vascular plants on the forest floor, and for deadwood-associated bryophytes and polypore fungi. Although NMDS ordination analyses of the total species composition identified a main axis related to the age of the oldest trees in two of the study areas, variation partitioning analyses showed that age explained a small fraction of variation of the species composition compared to site conditions, logging history, forest structure, and differences between the sites in all habitats. The unique variation explained by forest age species was, however, significant for all habitats. The fraction of variation in species composition explained by forest age was the largest for lichens and bryophytes on trees, and for deadwood-associated bryophytes and polypore fungi. Our results suggest that practical mapping of near-natural forests for management purposes inventories should include site conditions, forest structure and between site differences in addition to forest age.

Forest Fire Risk Assessment and Mapping Using Remote Sensing and GIS Techniques: A Case Study in Nghe An Province, Vietnam

Forest Fire Risk Assessment and Mapping Using Remote Sensing and GIS Techniques: A Case Study in Nghe An Province, Vietnam

Spatial Pattern of Forest Age in China Estimated by the Fusion of Multiscale Information

Forest age is one of most important biological factors that determines the magnitude of vegetation carbon sequestration. A spatially explicit forest age dataset is crucial for forest carbon dynamics modeling at the regional scale. However, owing to the high spatial heterogeneity in forest age, accurate high-resolution forest age data are still lacking, which causes uncertainty in carbon sink potential prediction. In this study, we obtained a 1 km resolution forest map based on the fusion of multiscale age information, i.e., the ninth (2014–2018) forest inventory statistics of China, with high accuracy at the province scale, and a field-observed dataset covering 6779 sites, with high accuracy at the site scale. Specifically, we first constructed a random forest (RF) model based on field-observed data. Utilizing this model, we then generated a spatially explicit forest age map with a 1 km resolution (random forest age map, RF map) using remotely sensed data such as tree height, elevation, meteorology, and forest distribution. This was then used as the basis for downscaling the provincial-scale forest inventory statistics of the forest ages and retrieving constrained maps of forest age (forest inventory constrained age maps, FIC map), which exhibit high statistical accuracy at both the province scale and site scale. The main results included the following: (1) RF can be used to estimate the site-scale forest age accurately (R2 = 0.89) and has the potential to predict the spatial pattern of forest age. However, (2) owing to the impacts of sampling error (e.g., field-observed sites are usually located in areas exhibiting relatively favorable environmental conditions) and the spatial mismatch among different datasets, the regional-scale forest age predicted by the RF model could be overestimated by 71.6%. (3) The results of the downscaling of the inventory statistics indicate that the average age of forests in China is 35.1 years (standard deviation of 21.9 years), with high spatial heterogeneity. Specifically, forests are older in mountainous and hilly areas, such as northeast, southwest, and northwest China, than in southern China. The spatially explicit dataset of the forest age retrieved in this study encompasses synthesized multiscale forest age information and is valuable for the research community in assessing the carbon sink potential and modeling carbon dynamics.

Forestry Extension Performance in Enhancing Forest Farmer Knowledge and Skills: A Case Study in Tompobulu, Maros Regency

Forestry Extension Workers, as one of the important parties in maintaining and preserving forest resources in Indonesia, are expected to have an understanding of good extension methods and techniques so that farmers around the forest can improve their knowledge and skills in managing and utilising forest products. This study aimed to assess the performance of forestry extension workers in Tompobulu District. And also evaluate their role in enhancing the knowledge and skills of forest farmer groups in Maros Regency. It was located in six villages, namely Bonto Somba Village, Bonto Manurung, Bonto Matinggi, Bonto Manai, Tompubulu, and Pucak Maros Regency with the number of samples in the study was 57. The study employed qualitative descriptive analysis using a Likert scale, and quantitative analysis using the t-test. The performance of forestry extension workers that significantly influences increasing the knowledge and skills of forest farmer groups consists of six indicator components, namely: facilitating forest farmer groups to prepare forest product utilisation maps; facilitating forest farmer groups to prepare annual work programs and plans; facilitating forest farmer group members to improve human resource quality; facilitating forest farmer groups to disseminate or spread information on forest product technology; facilitating forest farmer groups to grow their empowerment and independence; and facilitating forest farmer groups to increase production results of superior forest product commodities. Meanwhile, the performance of forestry extension workers that does not significantly influence increasing the knowledge and skills of forest farmer groups is Facilitating Business Partnerships Between Forest Farmer Groups, Entrepreneurs, and the Government, Facilitating Forest Farmer Groups to Access Business Capital with Financial Institutions, Banks, and the Government; and Facilitating Increasing the Income and Welfare of Forest Farmer Groups.

Data driven forest fire susceptibility mapping in Bangladesh

Forests are essential natural resources that serve to facilitate economic activity while also providing an essential impact on climate regulation and the carbon cycle. In the Chittagong hill tracts, forest fires significantly contribute to the eradication of forests and the depletion of biodiversity. As a result, addressing the forest fire hazard is crucial for mitigating its potential negative effects. By implementing machine learning (ML) and deep learning (DL) algorithms, the primary aim of this research is to determine which areas within the CHT are most prone to forest fires. Artificial Neural Network (ANN), Classification and Regression Trees (CART), Categorical Boosting (Catboost), Convolutional Neural Network (CNN), Deep Neural Network (DNN), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), Support Vector Machine (SVM), eXtreme Gradient Boosting (XGboost) are among these algorithms. This research investigates 267 past forest fire incidents while incorporating fifteen predictive parameters for forest fires. The utmost accuracy is demonstrated by the Catboost model, which has an AUC value of 0.83. Applied models the Bandarbon and Rangamati regions of the Chittagong hill tracts have been thoroughly analyzed and discovered to be extremely and moderately vulnerable to forest fires due to their intricate physical and social characteristics. The Catboost model indicates that 67.38% of the total regions and 0.59% of the regions in Chittagoong hill tracts are classified as moderate and high susceptible zones, respectively. The results of this study present the models utilized in the present analysis exhibit a high level of effectiveness. Furthermore, the findings indicated that the NDVI and annual temperature had the most influence. The application of both ML and DL approaches in the Chittagong hill tracts for the first time resulted in models that exhibited excellent accuracy and precision, combining perfectly with the actual conditions on the ground. This study will undeniably assist the local government in implementing efficient governance and preservation strategies for the sustainable maintenance of forest resources, while enhancing the well-being of individuals living in and around the forest. Forest fire susceptibility assessment can be utilized in various locations worldwide that include significant natural and human-induced characteristics.

MAPPING THE RISK OF FOREST FIRES IN THE SOUK AHRAS MOUNTAINS

Forests fire is the most devastating factor of degradation of the Algerian forest, despite numerous efforts to limit it. Traditional fire prevention and firefighting methods have shown their limitations due to their slowness… , is it therefore useful to adopt new techniques in order to react quickly and effectively. The main objective of this study is to model the «forest fire» risk, in order to produce a risk map, which is indispensable for preventing and informing on fire-vulnerable areas. To achieve this objective, a forest fire risk map was elaborated, on the basis of the scientific methods using geographical information systems and satellite imagery. The adopted approach is an empirical model, involving three parameters controlling the behavior of fire, the topo-morphology of the ground, the combustibility of the vegetation cover and human activity. The study area chosen is one of the most affected areas by this phenomenon (Souk Ahras Mountains). It belongs to the densest forest in Algeria, known for its sensitivity to fire, particularly in relation to its physical conditions (nature of the vegetation cover, climatic conditions (strong Sirocco wind) and socio-economic conditions (high population density).

Assessment of Carbon Sink and Carbon Flux in Forest Ecosystems: Instrumentation and the Influence of Seasonal Changes

Accurate measurement and estimation of forest carbon sinks and fluxes are essential for developing effective national and global climate strategies aimed at reducing atmospheric carbon concentrations and mitigating climate change. Various errors arise during forest monitoring, especially measurement instability due to seasonal variations, which require to be adequately addressed in forest ecosystem research and applications. Seasonal fluctuations in temperature, precipitation, aerosols, and solar radiation can significantly impact the physical observations of mapping equipment or platforms, thereby reducing the data’s accuracy. Here, we review the technologies and equipment used for monitoring forest carbon sinks and carbon fluxes across different remote sensing platforms, including ground-based, airborne, and spaceborne remote sensing. We further investigate the uncertainties introduced by seasonal variations to the observing equipment, compare the strengths and weaknesses of various monitoring technologies, and propose the corresponding solutions and recommendations. We aim to gain a comprehensive understanding of the impact of seasonal variations on the accuracy of forest map data, thereby improving the accuracy of forest carbon sinks and fluxes.

Index-based forest degradation mapping using high and medium resolution multispectral sensors

ABSTRACT Monitoring dense forest ecosystems, such as the laurel forest in Garajonay National Park, is vital for biodiversity conservation, carbon storage, and ecological balance. This study employs satellite remote sensing technologies to introduce a novel methodology, based on vegetation indices, aiming to assess and protect the health of the forest. Utilizing the Jeffries-Matusita distance and a histogram-based method, optimal indices to map forest degradation, like Wide Dynamic Range Vegetation Index (WDRVI) and Modified Simple Ratio (MSR), were identified among 19 generated indices. The study processed imagery from three satellite sensors (WorldView-2, PlanetScope and Sentinel-2), producing maps distinguishing healthy and degraded areas. The study’s practical significance lies in offering a method to assess the suitability of sensors and indices for effectively mapping forest degradation. This approach aids conservation efforts and provides valuable insights for environmental managers and policymakers, facilitating the implementation of targeted strategies to safeguard Garajonay National Park's unique laurel forest ecosystem. Emphasizing the role of remote sensing in practical vegetation protection endeavors, the study contributes to on-the-ground initiatives, ensuring the preservation and sustainability of the park's rich biodiversity.

Coupled zoning and spatial heterogeneity of human activities and natural endowments based on self-organizing map and random forest: A case study of the agro-pastoral ecotone in Gansu, China

Regional zoning is an important way for humans to understand geographical space，but the existing research on regional zoning focuses on natural conditions, ignoring the interdependent characteristics of human and nature, and the driving mechanism of regional differentiation is not sufficiently explained. In this study, using algorithms such as self-organizing feature map, clustering quality index, structural similarity index measure (SSIM) and random forest, the human activity (HA)–natural endowment (NE) coupled zoning in the agro-pastoral ecotone in Gansu, China (AEGC), was conducted to explore the driving mechanism of regional differentiation. Results show the following: (1) The HA-NE coupled zoning has the best clustering effect when implementing the partitioning scheme with a classification number of 4. Accordingly, the AEGC can be divided into four regions with significant differences between HAs and NEs. (2) From the perspective of spatial distribution, HA–NE coupled zoning, climate zoning, geographic zoning, and vegetation zoning are similar, whereas HA zoning is different. The results of SSIM showed that HA–NE coupled zoning takes all types of factors into consideration (SSIM mean 0.708), and the zoning results are better than single type zonings. HAs have strong independence (SSIM mean 0.576), and geographic conditions are closely related to all other factors (SSIM mean 0.671). (3) Elevation is the most important driver of regional differentiation in the AEGC, with a contribution degree of 22.36%; other important drivers include land use intensity, precipitation, and normalized difference vegetation index, with contribution degree distributions of 17.38%, 16.34%, and 15.79%. Furthermore, the dominant factors of regional differentiation in each sub-region are different. This study emphasizes that regional characteristic factors and uneven spatial distribution factors are important drivers of regional differentiation, and land use intensity has become an important force influencing geographic space. Policy recommendations for zonal governance are made based on the inherent conditions of different regions. This study may provide a reference for scientific regional zoning and cognitive regional differentiation.

U-Net Convolutional Neural Network for Mapping Natural Vegetation and Forest Types from Landsat Imagery in Southeastern Australia.

Accurate and comparable annual mapping is critical to understanding changing vegetation distribution and informing land use planning and management. A U-Net convolutional neural network (CNN) model was used to map natural vegetation and forest types based on annual Landsat geomedian reflectance composite images for a 500 km × 500 km study area in southeastern Australia. The CNN was developed using 2018 imagery. Label data were a ten-class natural vegetation and forest classification (i.e., Acacia, Callitris, Casuarina, Eucalyptus, Grassland, Mangrove, Melaleuca, Plantation, Rainforest and Non-Forest) derived by combining current best-available regional-scale maps of Australian forest types, natural vegetation and land use. The best CNN generated using six Landsat geomedian bands as input produced better results than a pixel-based random forest algorithm, with higher overall accuracy (OA) and weighted mean F1 score for all vegetation classes (93 vs. 87% in both cases) and a higher Kappa score (86 vs. 74%). The trained CNN was used to generate annual vegetation maps for 2000-2019 and evaluated for an independent test area of 100 km × 100 km using statistics describing accuracy regarding the label data and temporal stability. Seventy-six percent of pixels did not change over the 20 years (2000-2019), and year-on-year results were highly correlated (94-97% OA). The accuracy of the CNN model was further verified for the study area using 3456 independent vegetation survey plots where the species of interest had ≥ 50% crown cover. The CNN showed an 81% OA compared with the plot data. The model accuracy was also higher than the label data (76%), which suggests that imperfect training data may not be a major obstacle to CNN-based mapping. Applying the CNN to other regions would help to test the spatial transferability of these techniques and whether they can support the automated production of accurate and comparable annual maps of natural vegetation and forest types required for national reporting.

High-dimensional detection of Landscape Dynamics: a Landsat time series-based algorithm for forest disturbance mapping and beyond

ABSTRACT Time series analysis of medium-resolution multispectral satellite imagery is critical to investigate forest disturbance dynamics at the landscape scale. In particular, the spatial, temporal, and radiometric consistency of Landsat time series data provides unprecedented insight into past disturbances that occurred over the last four decades. Several Landsat time series-based algorithms have been developed to automate the detection of forest disturbances. However, automated detection of non-stand-replacing disturbances based on Landsat time series remains a challenging task due to the difficulty of effectively separating them from spectral noise. Here, we present the High-dimensional detection of Landscape Dynamics (HILANDYN) algorithm, which exploits spatial and spectral information provided by Landsat time series to detect forest disturbance dynamics retrospectively. A novel and unsupervised procedure for changepoint detection in high-dimensional time series allows HILANDYN to perform the temporal segmentation of inter-annual time series into linear trends. The algorithm embeds a noise filter to remove spurious changepoints caused by residual spectral noise in the time series. We tested HILANDYN to detect disturbances that occurred in the forests of the European Alps over a period of 39 years, i.e. between 1984 and 2022, and evaluated its accuracy using a validation dataset of 3000 plots randomly located inside and outside the disturbed patches. We compared HILANDYN with the Bayesian Estimator of Abrupt change, Seasonality, and Trend (BEAST), which is a well-established and high-performing time series-based algorithm for changepoint detection. The quantitative results highlighted that the number of bands, i.e. original Landsat bands and spectral indices, included in the high-dimensional time series and the threshold controlling the significance of changepoints strongly influenced the user’s accuracy (UA). Conversely, changes in the combinations of bands primarily affected the producer’s accuracy (PA). HILANDYN achieved an F1 score of 0.801, which increased to 0.833 when we activated the noise filter, allowing the algorithm to balance UA (83.1%) and PA (83.5%). The qualitative results showed that disturbed forest patches detected by HILANDYN were characterized by a high spatio-temporal consistency, regardless of the disturbance severity. Furthermore, our algorithm was able to detect forest patches associated with secondary disturbances, such as salvage logging, that occur in close succession with respect to the primary event. The comparison with BEAST evidenced a similar sensitivity of the algorithms to non-stand-replacing events, as both achieved comparable PA. However, BEAST struggled to balance UA and PA when using a single parameter set, achieving a maximum F1 score of 0.717. Moreover, the computational efficiency of BEAST in processing high-dimensional time series was very limited due to its univariate nature based on the Bayesian approach. The adaptability of HILANDYN to detect a wide range of disturbance severities using a single parameter set and its computational efficiency in handling high-dimensional time series promotes its scalability to large study areas characterized by heterogeneous ecological conditions.

Mapping Forest Stock Volume Using Phenological Features Derived from Time-Serial Sentinel-2 Imagery in Planted Larch

Mapping Forest Stock Volume Using Phenological Features Derived from Time-Serial Sentinel-2 Imagery in Planted Larch