Search

Copyright of Agronomy Journal is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

The global potential distribution of biomes (natural vegetation) was modelled using 8,959 training points from the BIOME 6000 dataset and a stack of 72 environmental covariates representing terrain and the current climatic conditions based on historical long term averages (1979-2013). An ensemble machine learning model based on stacked regularization was used, with multinomial logistic regression as the meta-learner and spatial blocking (100 km) to deal with spatial autocorrelation of the training points. Results of spatial cross-validation for the BIOME 6000 classes show an overall accuracy of 0.67 and R²logloss of 0.61, with "tropical evergreen broadleaf forest" being the class with highest gain in predictive performances (R²logloss = 0.74) and "prostrate dwarf shrub tundra" the class with the lowest (R²logloss = -0.09) compared to the baseline. Temperature-related covariates were the most important predictors, with the mean diurnal range (BIO2) being shared by all the base-learners (i.e.,random forest, gradient boosted trees and generalized linear models). The model was next used to predict the distribution of future biomes for the periods 2040-2060 and 2061-2080 under three climate change scenarios (RCP 2.6, 4.5 and 8.5). Comparisons of predictions for the three epochs (present, 2040-2060 and 2061-2080) show that increasing aridity and higher temperatures will likely result in significant shifts in natural vegetation in the tropical area (shifts from tropical forests to savannas up to 1.7 x105 km² by 2080) and around the Arctic Circle (shifts from tundra to boreal forests up to 2.4 x 105 km² by 2080). Projected global maps at 1 km spatial resolution are provided as probability and hard classes maps for BIOME 6000 classes and as hard classes maps for the IUCN classes (six aggregated classes). Uncertainty maps (prediction error) are also provided and should be used for careful interpretation of the future projections. [ABSTRACT FROM AUTHOR]

Location‐specific information is required to support decision making in crop variety management, especially under increasingly challenging climate conditions. Data synthesis can aggregate data from individual trials to produce information that supports decision making in plant breeding programs, extension services, and of farmers. Data from on‐farm trials using the novel approach of triadic comparison of technologies (tricot) are increasingly available, from which more insights could be gained using a data synthesis approach. The objective of our study was to present the applicability of a rank‐based data synthesis approach to several datasets from tricot trials to generate location‐specific information supporting decision making in crop variety management. Our study focuses on tricot data from 14 trials of common bean (Phaseolus vulgaris L.) performed between 2015 and 2018 across four countries in Central America (Costa Rica, El Salvador, Honduras, and Nicaragua). The combined data of 17 common bean genotypes were rank aggregated and analyzed with the Plackett–Luce model. Model‐based recursive partitioning was used to assess the influence of spatially explicit environmental covariates on the performance of common bean genotypes. Location‐specific performance was predicted for the three main growing seasons in Central America. We demonstrate how the rank‐based data synthesis methodology allows integrating tricot trial data from heterogenous sources to provide location‐specific information to support decision making in crop variety management. Maps of genotype performance can support decision making in crop variety evaluation such as variety recommendations to farmers and variety release processes. Core Ideas: We aggregate data from trials established by different organizations across different seasons and locations.We generate location‐specific insights on genotype performance and environmental interaction.We characterize uncertainty of model predictions using Shannon's entropy and area of applicability assessment. [ABSTRACT FROM AUTHOR]

This article describes a data-driven framework based on spatiotemporal machine learning to produce distribution maps for 16 tree species (Abies alba Mill., Castanea sativa Mill., Corylus avellana L., Fagus sylvatica L., Olea europaea L., Picea abies L. H. Karst., Pinus halepensis Mill., Pinus nigra J. F. Arnold, Pinus pinea L., Pinus sylvestris L., Prunus avium L., Quercus cerris L., Quercus ilex L., Quercus robur L., Quercus suber L. and Salix caprea L.) at high spatial resolution (30 m). Tree occurrence data for a total of three million of points was used to train different algorithms: random forest, gradient-boosted trees, generalized linear models, k-nearest neighbors, CART and an artificial neural network. A stack of 305 coarse and high resolution covariates representing spectral reflectance, different biophysical conditions and biotic competition was used as predictors for realized distributions, while potential distribution was modelled with environmental predictors only. Logloss and computing time were used to select the three best algorithms to tune and train an ensemble model based on stacking with a logistic regressor as a meta-learner. An ensemble model was trained for each species: probability and model uncertainty maps of realized distribution were produced for each species using a time window of 4 years for a total of six distribution maps per species, while for potential distributions only one map per species was produced. Results of spatial cross validation show that the ensemble model consistently outperformed or performed as good as the best individual model in both potential and realized distribution tasks, with potential distribution models achieving higher predictive performances (TSS = 0.898, R²logloss = 0.857) than realized distribution ones on average (TSS = 0.874, R²logloss = 0.839). Ensemble models for Q. suber achieved the best performances in both potential (TSS = 0.968, R²logloss = 0.952) and realized (TSS = 0.959, R²logloss = 0.949) distribution, while P. sylvestris (TSS = 0.731, 0.785, R²logloss = 0.585, 0.670, respectively, for potential and realized distribution) and P. nigra (TSS = 0.658, 0.686, R²logloss = 0.623, 0.664) achieved the worst. Importance of predictor variables differed across species and models, with the green band for summer and the Normalized Difference Vegetation Index (NDVI) for fall for realized distribution and the diffuse irradiation and precipitation of the driest quarter (BIO17) being the most frequent and important for potential distribution. On average, fine-resolution models outperformed coarse resolution models (250 m) for realized distribution (TSS = +6.5%, R²logloss = +7.5%). The framework shows how combining continuous and consistent Earth Observation time series data with state of the art machine learning can be used to derive dynamic distribution maps. The produced predictions can be used to quantify temporal trends of potential forest degradation and species composition change. [ABSTRACT FROM AUTHOR]

Estimates of the area of land cover classes or land change are frequently calculated from land cover classification maps by counting the pixels labeled as each class in the map. This procedure is known to produce biased estimates of area for many widely used classification algorithms, including random forests. Poststratification estimation using the mapped classes as strata has been proposed to obtain unbiased estimates of the class areas. Still, the method requires additional sampling units, which may not be available or be the most efficient method depending on the application. Alternatively, consistent estimates of class areas can be obtained using class membership probabilities estimates from a random forest classification. This article demonstrates that, for a large sample and proper set of explanatory variables, the error of the predicted class membership probabilities obtained from a random forest classification converges to zero. Therefore, the expected class areas calculated from these probabilities converge to the population class areas. On average, the relative error of the expected class proportions computed by class membership probabilities from a random forests model was 40% points lower than the proportions estimated by pixel counting. Our proposed approach is also comparable to the area-adjusted method, which is currently considered the best practice by the remote sensing community. We recommend that class probability estimates area always retained and used for calculating expected class areas or area proportions based on our results. Our method reduces bias compared to statistics calculated by pixel counting and circumvents the need for poststratification area estimates under certain conditions. [ABSTRACT FROM AUTHOR]

The terrestrial forest carbon pool is poorly quantified, in particular in regions with low forest inventory capacity. By combining multiple satellite observations of synthetic aperture radar (SAR) backscatter around the year 2010, we generated a global, spatially explicit dataset of above-ground live biomass (AGB; dry mass) stored in forests with a spatial resolution of 1 ha. Using an extensive database of 110 897 AGB measurements from field inventory plots, we show that the spatial patterns and magnitude of AGB are well captured in our map with the exception of regional uncertainties in high-carbon-stock forests with AGB >250 Mgha-1 , where the retrieval was effectively based on a single radar observation. With a total global AGB of 522 Pg , our estimate of the terrestrial biomass pool in forests is lower than most estimates published in the literature (426–571 Pg). Nonetheless, our dataset increases knowledge on the spatial distribution of AGB compared to the Global Forest Resources Assessment (FRA) by the Food and Agriculture Organization (FAO) and highlights the impact of a country's national inventory capacity on the accuracy of the biomass statistics reported to the FRA. We also reassessed previous remote sensing AGB maps and identified major biases compared to inventory data, up to 120 % of the inventory value in dry tropical forests, in the subtropics and temperate zone. Because of the high level of detail and the overall reliability of the AGB spatial patterns, our global dataset of AGB is likely to have significant impacts on climate, carbon, and socio-economic modelling schemes and provides a crucial baseline in future carbon stock change estimates. The dataset is available at 10.1594/PANGAEA.894711 (Santoro, 2018). [ABSTRACT FROM AUTHOR]

Reference data for large-scale land cover map are commonly acquired by visual interpretation of remotely sensed data. To assure consistency, multiple images are used, interpreters are trained, sites are interpreted by several individuals, or the procedure includes a review. But little is known about important factors influencing the quality of visually interpreted data. We assessed the effect of multiple variables on land cover class agreement between interpreters and reviewers. Our analyses concerned data collected for validation of a global land cover map within the Copernicus Global Land Service project. Four cycles of visual interpretation were conducted, each was followed by review and feedback. Each interpreted site element was labelled according to dominant land cover type. We assessed relationships between the number of interpretation updates following feedback and the variables grouped in personal, training, and environmental categories. Variable importance was assessed using random forest regression. Personal variable interpreter identifier and training variable timestamp were found the strongest predictors of update counts, while the environmental variables complexity and image availability had least impact. Feedback loops reduced updating and hence improved consistency of the interpretations. Implementing feedback loops into the visually interpreted data collection increases the consistency of acquired land cover reference data. [ABSTRACT FROM AUTHOR]

River discharges are often predicted based on a calibrated rainfall-runoff model. The major sources of uncertainty, namely input, parameter and model structural uncertainty must all be taken into account to obtain realistic estimates of the accuracy of discharge predictions. Over the past years, Bayesian calibration has emerged as a suitable method for quantifying uncertainty in model parameters and model structure, where the latter is usually modelled by an additive or multiplicative stochastic term. Recently, much work has also been done to include input uncertainty in the Bayesian framework. However, the use of geostatistical methods for characterizing the prior distribution of the catchment rainfall is underexplored, particularly in combination with assessments of the influence of increasing or decreasing rain gauge network density on discharge prediction accuracy. In this article we integrate geostatistics and Bayesian calibration to analyze the effect of rain gauge density on river discharge prediction accuracy. We calibrated the HBV hydrological model while accounting for input, initial state, model parameter and model structural uncertainty, and also taking uncertainties in the discharge measurements into account. Results for the Thur basin in Switzerland showed that model parameter uncertainty was the main contributor to the joint posterior uncertainty. We also showed that a low rain gauge density is enough for the Bayesian calibration, and that increasing the number of rain gauges improved model prediction until reaching a density of one gauge per 340 km². While the optimal rain gauge density is case-study specific, we make recommendations on how to handle input uncertainty in Bayesian calibration for river discharge prediction and present the methodology that may be used to carry out such experiments. [ABSTRACT FROM AUTHOR]

Rain gauge networks are crucial for enhancing the spatio‐temporal characterization of precipitation. In tropical regions, scarcity of rain gauge data, climatic variability, and variable spatial accessibility make conventional approaches to design rain gauge networks inadequate and impractical. In this study, we propose the use of conditioned Latin hypercube sampling (cLHS) method with multi‐temporal layers of remotely sensed precipitation measurements for capturing the spatio‐temporal precipitation patterns in ungauged areas. The study was conducted in the Amazon region of Ecuador, for which monthly precipitation averages were derived based on a 16‐year period of Tropical Rainfall Measuring Mission (TRMM 3B43 V7) data which were used as prior information to select representative sampling points through cLHS. Two scenarios for the sampling design were considered and evaluated, one without and one with restrictions on accessible sites according to the proximity to roads and settlements. Results showed that both optimized networks captured the variability of precipitation according to the TRMM climatology. Furthermore, evaluation against an independent satellite precipitation data set showed that the optimized networks support mapping precipitation based on ordinary kriging (OK). Comparison with regular and random sampling methods showed that particularly when a practical scenario is considered, the optimized network provided more reliable results over time, highlighting the suitability of the network to capture temporal changes and map precipitation with high accuracy. The proposed approach could be easily adopted in other ungauged and poorly accessible regions for rain gauge network design as well as to the design of multi‐objective monitoring networks. A sampling scheme to monitor precipitation distribution along the year is proposed thought the use of conditioned Latin hypercube sampling and satellite precipitation images. It was applied on monthly TRMM images in the relatively ungauged region of the Ecuadorian Amazon, in which accessibility restrictions were considered. The proposed method captured effectively the spatial and temporal changes of precipitation and supports mapping precipitation in the region. [ABSTRACT FROM AUTHOR]

To target food security interventions for smallholder households, decision makers need large-scale information, such as maps on poverty, food security and key livelihood activities. Such information is often based on expert knowledge or aggregated data, despite the fact that food security and poverty are driven largely by processes at the household level. At present, it is unclear if and how household level information can contribute to the spatial prediction of such welfare indicators or to what extent local variability is ignored by current mapping efforts. A combination of geo-referenced household level information with spatially continuous information is an underused approach to quantify local and large-scale variation, while it can provide a direct estimate of the variability of welfare indicators at the most relevant scale. We applied a stepwise regression kriging procedure to translate point information to spatially explicit patterns and create country-wide predictions with associated uncertainty estimates for indicators on food availability and related livelihood activities using household survey data from Uganda. With few exceptions, predictions of the indicators were weak, highlighting the difficulty in capturing variability at larger scale. Household explanatory variables identified little additional variation compared to environmental explanatory variables alone. Spatial predictability was strongest for indicators whose distribution was determined by environmental gradients. In contrast, indicators of crops that were more ubiquitously present across agroecological zones showed large local variation, which often overruled large-scale patterns. Our procedure adds to existing approaches that often only show large-scale patterns by revealing that local variation in welfare is large. Interventions that aim to target the poor must recognise that diversity in livelihood activities for income generation within any given area often overrides the variability of livelihood activities between distant regions in the country. [ABSTRACT FROM AUTHOR]

Global scale land cover (LC) mapping has interested many researchers over the last two decades as it is an input data source for various applications. Current global land cover (GLC) maps often do not meet the accuracy and thematic requirements of specific users. This study aimed to create an improved GLC map by integrating available GLC maps and reference datasets. We also address the thematic requirements of multiple users by demonstrating a concept of producing GLC maps with user-specific legends. We used a regression kriging method to integrate Globcover-2009, LC-CCI-2010, MODIS-2010 and Globeland30 maps and several publicly available GLC reference datasets. Overall correspondence of the integrated GLC map with reference LC was 80% based on 10-fold cross-validation using 24,681 sample sites. This is globally 10% and regionally 6–13% higher than the input map correspondences. Based on LC class presence probability maps, expected LC proportion maps at coarser resolution were created and used for characterizing mosaic classes for land system modelling and biodiversity assessments. Since more reference datasets are becoming freely accessible, GLC mapping can be further improved by using the pool of all available reference datasets. LC proportion information allow tuning LC products to specific user needs. [ABSTRACT FROM PUBLISHER]

The Agriculture, Forestry and Other Land Use (AFOLU) sector contributes with ca. 20-25% of global anthropogenic emissions (2010), making it a key component of any climate change mitigation strategy. AFOLU estimates, however, remain highly uncertain, jeopardizing the mitigation effectiveness of this sector. Comparisons of global AFOLU emissions have shown divergences of up to 25%, urging for improved understanding of the reasons behind these differences. Here we compare a variety of AFOLU emission datasets and estimates given in the Fifth Assessment Report for the tropics (2000-2005) to identify plausible explanations for the differences in (i) aggregated gross AFOLU emissions, and (ii) disaggregated emissions by sources and gases (CO2, CH4, N2O). We also aim to (iii) identify countries with low agreement among AFOLU datasets to navigate research efforts. The datasets are FAOSTAT (Food and Agriculture Organization of the United Nations, Statistics Division), EDGAR (Emissions Database for Global Atmospheric Research), the newly developed AFOLU "Hotspots", "Houghton", "Baccini", and EPA (US Environmental Protection Agency) datasets. Aggregated gross emissions were similar for all databases for the AFOLU sector: 8.2 (5.5-12.2), 8.4, and 8.0 PgCO2 eq. yr-1 (for Hotspots, FAOSTAT, and EDGAR respectively), forests reached 6.0 (3.8-10), 5.9, 5.9, and 5.4 PgCO2 eq. yr-1 (Hotspots, FAOSTAT, EDGAR, and Houghton), and agricultural sectors were with 1.9 (1.5-2.5), 2.5, 2.1, and 2.0 PgCO2 eq. yr-1 (Hotspots, FAOSTAT, EDGAR, and EPA). However, this agreement was lost when disaggregating the emissions by sources, continents, and gases, particularly for the forest sector, with fire leading the differences. Agricultural emissions were more homogeneous, especially from livestock, while those from croplands were the most diverse. CO2 showed the largest differences among the datasets. Cropland soils and enteric fermentation led to the smaller N2O and CH4 differences. Disagreements are explained by differences in conceptual frameworks (carbon-only vs. multi-gas assessments, definitions, land use vs. land cover, etc.), in methods (tiers, scales, compliance with Intergovernmental Panel on Climate Change (IPCC) guidelines, legacies, etc.) and in assumptions (carbon neutrality of certain emissions, instantaneous emissions release, etc.) which call for more complete and transparent documentation for all the available datasets. An enhanced dialogue between the carbon (CO2/ and the AFOLU (multi-gas) communities is needed to reduce discrepancies of land use estimates. [ABSTRACT FROM AUTHOR]

The carbon emissions and removals due to land cover changes between 2001 and 2010 in the Vu Gia Thu Bon River Basin, Central Vietnam, were estimated using Landsat satellite images and 3083 forest inventory plots. The net emissions from above- and belowground vegetation biomass were equal to 1.76 ± 0.12 Tg CO2, about 1.1% of the existing stocks. The vast majority of carbon emissions were due to forest loss, with the conversion of forest to cropland accounting for 67% of net emissions. Forest regrowth had a substantial impact on net carbon changes, removing 22% of emissions from deforestation. Most deforestation occurred in regrowth forest (60%) and plantations (29%), characterized by low carbon stock density. Thus identifying the type of forest where deforestation occurred and using local field data were critical with net emissions being 4 times larger when considering only one forest class with average carbon stock, and 5–7 times higher when using literature default values or global emission maps. Carbon emissions from soil (up to 30 cm) were estimated for the main land change class. Due to the low emission factors from biomass, soils proved a key emission category, accounting for 30% of total land emissions that occurred during the monitoring period. [ABSTRACT FROM AUTHOR]

Current methods for monitoring deforestation from satellite data at sub-annual scales require pixel time series to have many historical observations in the reference period to model normal forest dynamics before detecting deforestation. However, in some areas, pixel time series often do not have many historical observations. Detecting deforestation at a pixel with scarce historical observations can be improved by complementing the pixel time series with spatial context information. In this work, we propose a data-driven space-time change detection method that detects deforestation events at sub-annual scales in data cubes of satellite image time series. First we spatially normalised observations in the local space-time data cube to reduce seasonality. Subsequently, we detected deforestation by assessing whether a newly acquired observation in the monitoring period is an extreme when compared against spatially normalised values in a local space-time data cube defined over reference period. We demonstrated our method at two sites, a dry tropical Bolivian forest and a humid tropical Brazilian forest, by varying the spatial and temporal extent of data cube. We emulated a "near real-time" monitoring scenario, implying that observations in the monitoring period were sequentially rather than simultaneously assessed for deforestation. Using Landsat normalised difference vegetation index (NDVI) time series, we achieved a median temporal detection delay of less than three observations, a producer's accuracy above 70%, a user's accuracy above 65%, and an overall accuracy above 80% at both sites, even when the reference period of the data cube only contained one year of data. Our results also show that large percentile thresholds (e.g., 5th percentile) achieve higher producer's accuracy and shorter temporal detection delay, whereas smaller percentiles (e.g., 0.1 percentile) achieve higher user's accuracy, but longer temporal detection delay. The method is data-driven, not based on statistical assumption on the data distribution, and can be applied on different forest types. However, it may face challenges in mixed forests where, for example, deciduous and evergreen forests coexist within short distances. A pixel to be assessed for deforestation should have a minimum of three temporal observations, the first of which must be known to represent forest. Such short time series allow rapid deployment of newly launched sensors (e.g., Sentinel-2) for detecting deforestation events at sub-annual scales. [ABSTRACT FROM AUTHOR]

This paper describes an interactive web-based near real-time (NRT) forest monitoring system using four levels of geographic information services: 1) the acquisition of continuous data streams from satellite and community-based monitoring using mobile devices, 2) NRT forest disturbance detection based on satellite time-series, 3) presentation of forest disturbance data through a web-based application and social media and 4) interaction of the satellite based disturbance alerts with the end-user communities to enhance the collection of ground data. The system is developed using open source technologies and has been implemented together with local experts in the UNESCO Kafa Biosphere Reserve, Ethiopia. The results show that the system is able to provide easy access to information on forest change and considerably improves the collection and storage of ground observation by local experts. Social media leads to higher levels of user interaction and noticeably improves communication among stakeholders. Finally, an evaluation of the system confirms the usability of the system in Ethiopia. The implemented system can provide a foundation for an operational forest monitoring system at the national level for REDD+ MRV applications. [ABSTRACT FROM AUTHOR]

According to the latest report of the Intergovernmental Panel on Climate Change (IPCC), emissions must be cut by 41-72 % below 2010 levels by 2050 for a likely chance of containing the global mean temperature increase to 2°C. The AFOLU sector (Agriculture, Forestry and Other Land Use) roughly contributes with a quarter (~ 10-12 PgCO2e.yr-1) of the net anthropogenic GHG emissions mainly from deforestation, fire, wood harvesting, and agricultural emissions including croplands, paddy rice and livestock. In spite of the importance of this sector, it is unclear where are the regions in the planet with AFOLU emissions hotspots, and how uncertain these emissions are. Here we present a novel spatially comparable dataset containing annual mean estimates of gross AFOLU emissions (CO2, CH4, N2O), associated uncertainties, and leading emission sources, in a spatially disaggregated manner (0.5°), for the tropics, for the period 2000-2005. Our data highlight: i) the existence of AFOLU emissions hotspots on all continents, with particular importance of evergreen rainforest deforestation in Central and South America, fire in dry forests in Africa, and both peatland emissions and agriculture in Asia; ii) a predominant contribution of forests and CO2 to the total AFOLU emissions (75 %) and to their uncertainties (98 %), iii) higher gross fluxes from forests coincide with higher uncertainties, making agricultural hotspots more appealing for effective mitigation action, and iv) a lower contribution of non-CO2 agricultural emissions to the total gross budget (ca. 25 %) with livestock (15.5 %) and rice (7 %) leading the emissions. Gross AFOLU tropical emissions 8.2 (5.5-12.2) were in the range of other databases 8.4 and 8.0 PgCO2e.yr-1 (FAOSTAT and EDGAR respectively), but we offer a spatially detailed benchmark for monitoring progress on reducing emissions from the land sector in the tropics. The location of the AFOLU hotspots of emissions and data on their associated uncertainties, will assist national policy makers, investors and other decision-makers who seek to understand the mitigation potential of the AFOLU sector. [ABSTRACT FROM AUTHOR]

Along with the creation of new maps, current efforts for improving global land cover (GLC) maps focus on integrating maps by accounting for their relative merits, e.g., agreement amongst maps or map accuracy. Such integration efforts may benefit from the use of multiple GLC reference datasets. Using available reference datasets, this study assesses spatial accuracy of recent GLC maps and compares methods for creating an improved land cover (LC) map. Spatial correspondence with reference dataset was modeled for Globcover-2009, Land Cover-CCI-2010, MODIS-2010 and Globeland30 maps for Africa. Using different scenarios concerning the used input data, five integration methods for an improved LC map were tested and cross-validated. Comparison of the spatial correspondences showed that the preferences for GLC maps varied spatially. Integration methods using both the GLC maps and reference data at their locations resulted in 4.5%-13% higher correspondence with the reference LC than any of the input GLC maps. An integrated LC map and LC class probability maps were computed using regression kriging, which produced the highest correspondence (76%). Our results demonstrate the added value of using reference datasets and geostatistics for improving GLC maps. This approach is useful asmore GLC reference datasets are becoming publicly available and their reuse is being encouraged. [ABSTRACT FROM AUTHOR]

Mapping of environmental variables often relies on map accuracy assessment through cross-validation with the data used for calibrating the underlying mapping model. When the data points are spatially clustered, conventional cross-validation leads to optimistically biased estimates of map accuracy. Several papers have promoted spatial cross-validation as a means to tackle this over-optimism. Many of these papers blame spatial autocorrelation as the cause of the bias and propagate the widespread misconception that spatial proximity of calibration points to validation points invalidates classical statistical validation of maps. We present and evaluate alternative cross-validation approaches for assessing map accuracy from clustered sample data. The first method uses inverse sampling-intensity weighting to correct for selection bias. Sampling-intensity is estimated by a two-dimensional kernel approach. The two other approaches are model-based methods rooted in geostatistics, where the first assumes homogeneity of residual variance over the study area whilst the second accounts for heteroscedasticity as a function of the sampling intensity. The methods were tested and compared against conventional k -fold cross-validation and blocked spatial cross-validation to estimate map accuracy metrics of above-ground biomass and soil organic carbon stock maps covering western Europe. Results acquired over 100 realizations of five sampling designs ranging from non-clustered to strongly clustered confirmed that inverse sampling-intensity weighting and the heteroscedastic model-based method had smaller bias than conventional and spatial cross-validation for all but the most strongly clustered design. For the strongly clustered design where large portions of the maps were predicted by extrapolation, blocked spatial cross-validation was closest to the reference map accuracy metrics, but still biased. For such cases, extrapolation is best avoided by additional sampling or limitation of the prediction area. Weighted cross-validation is recommended for moderately clustered samples, while conventional random cross-validation suits fairly regularly spread samples. • Cross-validation with clustered data produces too optimistic map accuracy estimates. • In contrast, blocked spatial cross-validation is pessimistically biased. • Sampling-intensity weighted cross-validation is recommended for clustered samples. [ABSTRACT FROM AUTHOR]

To address the need for timely information on newly deforested areas at medium resolution scale, we introduce a Bayesian approach to combine SAR and optical time series for near real-time deforestation detection. Once a new image of either of the input time series is available, the conditional probability of deforestation is computed using Bayesian updating, and deforestation events are indicated. Future observations are used to update the conditional probability of deforestation and, thus, to confirm or reject an indicated deforestation event. A proof of concept was demonstrated using Landsat NDVI and ALOS PALSAR time series acquired at an evergreen forest plantation in Fiji. We emulated a near real-time scenario and assessed the deforestation detection accuracies using three-monthly reference data covering the entire study site. Spatial and temporal accuracies for the fused Landsat-PALSAR case (overall accuracy = 87.4%; mean time lag of detected deforestation = 1.3 months) were consistently higher than those of the Landsat- and PALSAR-only cases. The improvement maintained even for increasing missing data in the Landsat time series. [ABSTRACT FROM AUTHOR]

Land cover identification and area quantification are key aspects of implementing the European Common Agriculture Policy. Legitimacy of support provided to farmers is monitored using the Land Parcel Identification System (LPIS), with land cover identification performed by visual image interpretation. While the geometric orthoimage quality required for correct interpretation is well understood, little is known about the photometric quality needed for LPIS applications. This paper analyzes the orthoimage quality characteristics chosen by authors as being most suitable for visual identification of agricultural fields. We designed a survey to assess users' preferred brightness and contrast ranges for orthoimages used for LPIS purposes. Survey questions also tested the influence of a background color on the preferred orthoimage brightness and contrast, the preferred orthoimage format and color composite, assessments of orthoimages with shadowed areas, appreciation of image enhancements and, finally, consistency of individuals' preferred brightness and contrast settings across multiple sample images. We find that image appreciation is stable at the individual level, but preferences vary across respondents. We therefore recommend that LPIS operators be enabled to personalize photometric settings, such as brightness and contrast values, and to choose the displayed band combination from at least four spectral bands. [ABSTRACT FROM AUTHOR]

Within the Reducing Emissions from Deforestation and Degradation (REDD+) framework, the involvement of local communities in national forest monitoring activities has the potential to enhance monitoring efficiency at lower costs while simultaneously promoting transparency and better forest management. We assessed the consistency of forest monitoring data (mostly activity data related to forest change) collected by local experts in the UNESCO Kafa Biosphere Reserve, Ethiopia. Professional ground measurements and high resolution satellite images were used as validation data to assess over 700 forest change observations collected by the local experts. Furthermore, we examined the complementary use of local datasets and remote sensing by assessing spatial, temporal and thematic data quality factors. Based on this complementarity, we propose a framework to integrate local expert monitoring data with satellite-based monitoring data into a National Forest Monitoring System (NFMS) in support of REDD+ Measuring, Reporting and Verifying (MRV) and near real-time forest change monitoring. [ABSTRACT FROM AUTHOR]

Monitoring tropical deforestation and forest degradation is one of the central elements for the Reduced Emissions from Deforestation and Forest Degradation in developing countries (REDD+) scheme. Current arrangements for monitoring are based on remote sensing and field measurements. Since monitoring is the periodic process of assessing forest stands properties with respect to reference data, adopting the current REDD+ requirements for implementing monitoring at national levels is a challenging task. Recently, the advancement in Information and Communications Technologies (ICT) and mobile devices has enabled local communities to monitor their forest in a basic resource setting such as no or slow internet connection link, limited power supply, etc. Despite the potential, the use of mobile device system for community based monitoring (CBM) is still exceptional and faces implementation challenges. This paper presents an integrated data collection system based on mobile devices that streamlines the community-based forest monitoring data collection, transmission and visualization process. This paper also assesses the accuracy and reliability of CBM data and proposes a way to fit them into national REDD+ Monitoring, Reporting and Verification (MRV) scheme. The system performance is evaluated at Tra Bui commune, Quang Nam province, Central Vietnam, where forest carbon and change activities were tracked. The results show that the local community is able to provide data with accuracy comparable to expert measurements (index of agreement greater than 0.88), but against lower costs. Furthermore, the results confirm that communities are more effective to monitor small scale forest degradation due to subsistence fuel wood collection and selective logging, than high resolution remote sensing SPOT imagery. [ABSTRACT FROM AUTHOR]

In case of an environmental accident, initially available data are often insufficient for properly managing the situation. In this paper, new sensor observations are iteratively added to an initial sample by maximising the global expected value of information of the points for decision making. This is equivalent to minimizing the aggregated expected misclassification costs over the study area. The method considers measurement error and different costs for class omissions and false class commissions. Constraints imposed by a mobile sensor web are accounted for using cost distances to decide which sensor should move to the next sample location. The method is demonstrated using synthetic examples of static and dynamic phenomena. This allowed computation of the true misclassification costs and comparison with other sampling approaches. The probability of local contamination levels being above a given critical threshold were computed by indicator kriging. In the case of multiple sensors being relocated simultaneously, a genetic algorithm was used to find sets of suitable new measurement locations. Otherwise, all grid nodes were searched exhaustively, which is computationally demanding. In terms of true misclassification costs, the method outperformed random sampling and sampling based on minimisation of the kriging variance. [ABSTRACT FROM AUTHOR]