Fairness related to locations (i.e., “where”) is critical for the use of machine learning in a variety of societal domains involving spatial datasets (e.g., agriculture, disaster response, urban planning). Spatial biases incurred by learning, if left unattended, may cause or exacerbate unfair distribution of resources, social division, spatial disparity, etc. The goal of this work is to develop statistically-robust formulations and model-agnostic learning strategies to understand and promote spatial fairness. The problem is challenging as locations are often from continuous spaces with no well-defined categories (e.g., gender), and statistical conclusions from spatial data are fragile to changes in spatial partitionings and scales. Existing studies in fairness-driven learning have generated valuable insights related to non-spatial factors including race, gender, education level, etc., but research to mitigate location-related biases still remains in its infancy, leaving the main challenges unaddressed. To bridge the gap, we first propose a robust space-as-distribution (SPAD) representation of spatial fairness to reduce statistical sensitivity related to partitionings and scales in continuous space. Furthermore, we propose a new SPAD-based stochastic strategy to efficiently optimize over an extensive distribution of fairness criteria, and a bi-level training framework to enforce fairness via adaptive adjustment of priorities among locations. Finally, we extend this framework with a similarity-based training strategy to improve the computational efficiency. Experiments conducted on two real-world problems, crop monitoring in the US and palm oil plantation mapping in Indonesia, show that SPAD can effectively reduce sensitivity in fairness evaluation and the stochastic bi-level training framework can greatly improve the fairness. Controlled experiments also show that similarity-based acceleration can greatly reduce the training time while keeping the prediction performance and fairness results at the same level.

Despite improvements in safe water and sanitation services in low-income countries, a substantial proportion of the population in Africa still does not have access to these essential services. Up-to-date fine-scale maps of low-income settlements are urgently needed by authorities to improve service provision. We aim to develop a cost-effective solution to generate fine-scale maps of these vulnerable populations using multi-source public information. The problem is challenging as ground-truth maps are available at only a limited number of cities, and the patterns are heterogeneous across cities. Recent attempts tackling the spatial heterogeneity issue focus on scenarios where true labels partially exist for each input region, which are unavailable for the present problem. We propose a dynamic point-to-region co-learning framework to learn heterogeneity patterns that cannot be reflected by point-level information and generalize deep learners to new areas with no labels. We also propose an attention-based correction layer to remove spurious signatures, and a region-gate to capture both region-invariant and variant patterns. Experiment results on real-world fine-scale data in three cities of Kenya show that the proposed approach can largely improve model performance on various base network architectures.

Cloud masking is both a fundamental and a critical task in the vast majority of Earth observation problems across social sectors, including agriculture, energy, water, etc. The sheer volume of satellite imagery to be processed has fast-climbed to a scale (e.g., >10 PBs/year) that is prohibitive for manual processing. Meanwhile, generating reliable cloud masks and image composite is increasingly challenging due to the continued distribution-shifts in the imagery collected by existing sensors and the ever-growing variety of sensors and platforms. Moreover, labeled samples are scarce and geographically limited compared to the needs in real large-scale applications. In related work, traditional remote sensing methods are often physics-based and rely on special spectral signatures from multi- or hyper-spectral bands, which are often not available in data collected by many -- and especially more recent -- high-resolution platforms. Machine learning and deep learning based methods, on the other hand, often require large volumes of up-to-date training data to be reliable and generalizable over space. We propose an autonomous image composition and masking (Auto-CM) framework to learn to solve the fundamental tasks in a label-free manner, by leveraging different dynamics of events in both geographic domains and time-series. Our experiments show that Auto-CM outperforms existing methods on a wide-range of data with different satellite platforms, geographic regions and bands.

Since 2000, Russia experienced large-scale increases (149%) in wheat production and starting in 2018 accounted for almost 25% of all global wheat exports. This growth indicates significant land cover and land use change (LCLUC) and is driven primarily by winter wheat growth adding 9 million hectares of cropland area, a 117% increase. Here we show that 40% of southwestern Russia experienced changes in land cover and land use including a 29% growth in winter wheat cropland. Of this growth, 66% is attributed to winter wheat cropland expansion (planting in new areas) and 34% to intensification (increased planting rate). The observed growth in winter cropland was latitudinally dichotomous where northernmost regions experienced areal expansion and southernmost regions intensification. Based on rates of winter cropland use, we conclude that there remains significant capacity for winter crop intensification and provide probable trajectories of continued growth.

Ukraine plays an important role in global food security. Ukraine produces about half of the global sunflower oil production, and Ukraine-produced barley, corn, wheat and rapeseed are exported to Europe, China, India, North Africa and the Middle East countries. However, after the invasion of Russia in February 2022 and blocking the ports in Black Sea, the global food prices increased. Uncertainties over crop production over the Russia-occupied territories in Ukraine also impacted the food prices. In this paper, we developed approaches to detect and map crop planted areas over Ukraine using Synthetic Aperture Radar (SAR). This information is very important to assess the agricultural activities, especially on occupied territories, and potentially reduce global food market volatility.

Climate velocity estimation is an important task for climate change study. At the same time, analysis of biological response to climate change requires climate-analog approach for climate velocity estimation. This approach uses a nearest neighborhood technique for the climate cells matching between climate data with temporal distance. Despite benefits in the quality, usability and accuracy this approach has serious limitations related to the distance measurements, climate analog search, and parameters of algorithm that necessary to tune for each region, scale, spatial resolution and type of climate data. These limitations are making this algorithm difficult to use, less reliable and not usable on the global scale. In this paper we introducing the new method for climate-analog velocity estimation based on the optical flow iterative Lucas-Kanade (iLK) approach. This algorithm is using phase-correlation as a matching cost function and allows to avoid tuning of dissemination threshold parameters and provide robust and accurate result on any scale or resolution.

Fairness related to locations (i.e., "where") is critical for the use of machine learning in a variety of societal domains involving spatial datasets (e.g., agriculture, disaster response, urban planning). Spatial biases incurred by learning, if left unattended, may cause or exacerbate unfair distribution of resources, social division, spatial disparity, etc. The goal of this work is to develop statistically-robust formulations and model-agnostic learning strategies to understand and promote spatial fairness. The problem is challenging as locations are often from continuous spaces with no well-defined categories (e.g., gender), and statistical conclusions from spatial data are fragile to changes in spatial partitionings and scales. Existing studies in fairness-driven learning have generated valuable insights related to non-spatial factors including race, gender, education level, etc., but research to mitigate location-related biases still remains in its infancy, leaving the main challenges unaddressed. To bridge the gap, we first propose a robust space-as-distribution (SPAD) representation of spatial fairness to reduce statistical sensitivity related to partitioning and scales in continuous space. Furthermore, we propose a new SPAD-based stochastic strategy to efficiently optimize over an extensive distribution of fairness criteria, and a bi-level training framework to enforce fairness via adaptive adjustment of priorities among locations. Experiments on real-world crop monitoring show that SPAD can effectively reduce sensitivity in fairness evaluation and the stochastic bi-level training framework can greatly improve the fairness.

Aerosols play a critical role in radiative transfer within the atmosphere and in climate change. As part of the validation of atmospheric correction of remote sensing data affected by the atmosphere, it is critical to utilize appropriate aerosol models as aerosols are a main source of error. Here, we define the aerosol model by recalculating the aerosol microphysical properties based on the optical thickness at 440 nm and the Ångström coefficient obtained from numerous AERONET sites. The associated uncertainties are up to 23%, except for the imaginary part of the refractive index (about 38%). Uncertainties of the retrieved aerosol microphysical properties were incorporated in the framework for validating surface reflectance derived from space-borne Earth observation sensors. It yields an overall uncertainty of approximately of 1 to 3% of the retrieved surface reflectance in the MODIS red spectral band, well below the specification used for atmospheric correction.

We present the validation of surface reflectance from Sentinel-2 (S2) produced by LaSRC (Land Surface Reflectance Code) using an automated camera system (CAMSIS). The system is composed of four cameras (470, 550, 650, and 850nm wavelengths), and a motorized reference (50% reflectance) calibration target. CAMSIS is installed 120m above ground on a TV tower (WLEF) near Park Falls, Wisconsin, and captures data every 15 minutes. Surface reflectance and NDVI computed from CAMSIS calibrated data show good agreement when compared to observations from Sentinel-2. These results show good performance of LaSRC atmospheric correction.

The long-term data record (LTDR) has the goal of developing a quality and consistent Advanced Very High Resolution Radiometer (AVHRR) surface reflectance and albedo products dating back to 1982 at 0.05° spatial resolution. Distinguishing between cloud and snow is of critical importance when analyzing global albedo trends, for they influence the Earth’s energy balance. However, this task is specially challenging when working with AVHRR given its limited spectral bands. Therefore, the current version of the LTDR does not distinguish between snow and clouds. To this end, we propose the Moderate Resolution Imaging Spectroradiometer (MODIS)-based AVHRR Class Separation Algorithm (MACSSA), whose goal is to identify clear land and snow pixels using AVHRR data. We make use of a combination of optical and thermal information from satellite and reanalysis data, along with monthly climatology information. These are used as inputs for two different support vector machine (SVM) models, which are then applied to AVHRR data to retrieve the MACSSA predicted tags. These are compared first against reference tags retrieved from the MYD10C1 product over pixels with less than 2-min overpass time difference between MODIS Aqua and NOAA16–19, distributed all around the world, and second against the Climate Change Initiative Cloud (Cloud_cci AVHRR) project. We found the product to be highly accurate in identifying clear land pixels, with a probability of detection of clear pixels (PODclear) of 97%. The discrimination of snow and clouds shows a PODsnow of 89%, which is encouraging given the spectral limitations of the AVHRR sensor.

In this study we present a model to forecast wheat yield based on the evolution of the Difference Vegetation Index (DVI) and the Growing Degree Days (GDD), presented in Franch et al. (2015), but adapted to Franch et al. (2019) model. Additionally, we explore how the Land Surface Temperature (LST) can be included into the model and if this parameter adds any value to the model when combined with the optical information. This study is applied to MODIS data at 1km resolution to monitor the national and state level yield of winter wheat in the United States and Ukraine from 2001 to 2019.

Crop yield information at field scale is important for farmers, crop insurance companies and agricultural communities in general. In this study, a wide range of ground-collected yield data was used to develop crop yield forecast models for the two internationally important crops: wheat and soybeans. A deep neural network (NN), a long short-term memory (LSTM), was trained for both crops individually. For each crop, the LSTM model was trained for two different scenarios including using Synthetic Aperture Radar (SAR)-only data as first scenario and using integration of SAR and optical satellite data as a second scenario. The root mean square error (RMSE) and coefficient of determination$(R^{2})$were estimated for each scenario. The results demonstrated that the accuracies improved from RMSE of 516.7 kg/ha and$R^{2}$of 0.79 (scenario 1) to RMSE of 433.77 kg/ha and$R^{2}$of 0.87 (scenario 2) for soybeans. For wheat, the accuracies improved from RMSE of 617.14 kg/ha and$R^{2}$of 0.83 (scenario 1) to RMSE of 423.04 kg/ha and$R^{2}$of 0.87 (scenario 2). These results show that using SAR data and their integration with optical satellite data is a promising approach for crop yield forecast at field scale.

In this study we present a methodology to develop a global winter wheat crop calendar based on the existing crop calendar products from FAO and GEOGLAM Crop Monitor in the framework of the WorldCereal project. It is based on integrating both datasets by building on the accuracy from Crop Monitor and the spatial resolution from the Food and Agriculture Organization of the United Nations (FAO). Additionally, given the global extent of WorldCereal and the gaps that both products present at global scale, we simulated the crop calendars in those areas not covered by any of the products. To do so, we integrated a Regression-Kriging model considering as training data the calendars derived from both products and based on the latitude, height and distance to the coast (DTC).

Timely monitoring of agricultural production and early yield predictions are essential for food security. Crop growth conditions and yield are related to climate variability and are impacted by extreme events. Remotely sensed time-series could be used to study the variability in crop growth and agricultural production. However, the choice of remotely sensed data and methods is still an issue, as different datasets have different spatiotemporal characteristics. Our primary goal was to test different algorithms and several remotely sensed time-series datasets for yield estimation in U.S. at county and field scale. For a county-level analysis, MODIS-based surface reflectance, Land Surface Temperature, and Evapotranspiration time series were used as input datasets. Field-level analysis was carried out using NASA's Harmonized Landsat Sentinel-2 (HLS) product. For this purpose, 3D convolutional neural network (CNN) and CNN followed by long-short term memory (LSTM) were used. For county-level analysis, the CNN-LSTM model had the highest accuracy, with a mean percentage error of 10.3% for maize and 9.6% for soybean. This model presented robust results for the year 2012, which is considered a drought year. In the case of field-level analysis, all models achieved accurate results with R2exceeding 0.8 when data from mid growing season were used. The results highlight the potential of using satellite data for yield estimation at different management scales.

This study aims at mapping Land Use and Land Cover (LULC) in the region of Roraima, Brazil, using time-series of Sentinel-1 Synthetic Aperture Radar (SAR) data. All available Sentinel-1 images covering the study area were used and classified using two machine learning algorithms, namely random forest and multilayer perceptron. LULC heterogeneity with the SAR process complexity makes the process challenging in distinguishing certain classes. Results show that SAR data could be used for LULC mapping, as rainforest, savannas, water, and sandbank/outcrop classes. But cannot provide accurate separation for all classes, mainly for those with similar geometrical structures, such as regeneration areas, perennial crops, and buritizais.

In this work, we focus on exploring very high spatial resolution (1-3 m) satellite imagery for capturing crop yield variability at field scale. In-field yields of soybean and corn were collected in Iowa, USA, and were correlated with multi-spectral satellite data acquired by WorldView-3 (at 1.25 m) and PlanetScope (Dove-Classic) (at 3 m). Results show that the most important spectral bands explaining corn and soybean yield variability are green/yellow, red edge and NIR. High temporal frequency of Planet data allowed identification of best suitable date for yield assessment: PlanetScope's spectral bands at 3 m explained 10% to 75% of in-field corn and soybean yield variability.

Detecting changes in impervious surface cover is one of the most important topics in land cover and land use (LCLU) change. This study focuses on detecting infrastructure constructions, such as residential areas, commercial building, and roads, in the State of Maryland (US) from 2018 to 2019 by utilizing Sentinel-2 images at 10 m spatial resolution. We use a time-series of Sentinel-2 images to derive land cover maps in 2018 and 2019 and derive the change detection map. The multi-layer perceptron (MLP) neural network is used to classify satellite images into general land cover classes (impervious surface, forest/tree cover, grassland/cropland, water). The derived change detection map allows one to identify areas of changes with new constructions.

This study focuses on winter wheat yield assessment from NASA's Harmonized Landsat Sentinel-2 (HLS) product and meteorological observations through phenological fitting. Vegetation indices (VIs), namely difference vegetation index (DVI), normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI2), extracted from satellite optical data, are fitted per pixel against accumulated growing degree days (AGDD) using a quadratic function. Accumulated VIs are correlated against winter wheat yields. Results show a better performance from DVI compared to NDVI and EVI2.

Accurate and timely crop yield forecasts are critical for making informed agricultural policies and investments, as well as increasing market efficiency and stability. In Becker-Reshef et al. (2010) and Franch et al. (2015) we developed an empirical generalized model for forecasting winter wheat yield. In this study we present a new model based on the extrapolation of the pure wheat signal (100% of wheat within the pixel) from MODIS data at 1 km resolution and using the Difference Vegetation Index (DVI). The model has been applied to monitor the national and state level yield of winter wheat in the United States from 2001 to 2016.

The Harmonized Landsat-8 and Sentinel-2 (HLS) project is a NASA initiative aiming to produce a seamless, harmonized surface reflectance record from the Operational Land Imager (OLI) and Multi-Spectral Instrument (MSI) aboard Landsat-8 and Sentinel-2 remote sensing satellites, respectively. The HLS products are based on a set of algorithms to obtain seamless products from both sensors (OLI and MSI): atmospheric correction, cloud and cloud-shadow masking, geographic co-registration and common gridding, bidirectional reflectance distribution function normalization and bandpass adjustment. As of version 1.3, the HLS v1.3 data set covers 9.12 million km2 and spans from first Landsat-8 data (2013) to present. HLS products provide near-daily surface reflectance information with a common geometric framework, and are suitable for a variety of agricultural and vegetation monitoring tasks, including analysis of crop type, condition, and phenology.

Cross-calibration between sensors is necessary to bring measurements to a common radiometric scale; it allows a more complete monitoring of land surface processes and enhances data continuity and harmonization. However, differences in the Relative Spectral Response (RSR) of sensors generate uncertainties in the process [1]. For this reason, compensating for these differences is of great importance and can be achieved by using a spectral band adjustment factor (SBAF), which establishes a relationship between two spectrally adjusted bands. Nonetheless, this relationship has been shown to depend greatly on the surface type [2] and therefore needs to be corrected. In this work, we compute the SBAF between the historical Landsat and Sentinel 2 sensors by using the RSRs of different passive optical sensors in the Green, Red and NIR bands and the surface reflectance spectral libraries (ASTER, AVIRIS, IGCP) with a wide variety of classes. We produce a quadratic fit of the SBAF vs the surface's NDVI (ρnir- ρred)/(ρnir+ ρred) and propose an exponential correction equation dependent on the NDVI value for both bands. A comparison between Landsat 8 and Sentinel 2 images using the HLS product shows that this method improves the red band and NDVI accuracy by 46.4% and 63.9% respectively when the difference between the Relative Spectral Responses (RSR) is significant, but is inaccurate for the green band, where the atmospheric correction is likely to introduce same order errors.

With availability of images acquired by NASA/USGS Landsat 8 and European Copernicus Sentinel-2 remote sensing satellites, it becomes possible to provide a global coverage of Earth's surface every 3-5 days. Such high temporal resolution is a prerequisite for developing next generation products at moderate spatial resolution (10-30 m). This is especially important for applications, involving agricultural monitoring. This paper explores a combined use of Landsat 8 and Sentinel-2 data to winter wheat yield assessment at regional scale. We take advantage of the NASA's Harmonized Landsat and Sentinel-2 (HLS) product, which provides a seamless unified product from different sensors aboard both satellites. Multiple features are evaluated through correlation with winter wheat yield values with normalized difference vegetation index (NDVI) serving as a benchmark. We show that, when using Landsat 8 and Sentinel-2 data together, the error of winter wheat yield estimates can be reduced up to 1.8 times, compared to using a single satellite.

This paper presents a generic approach developed to derive surface reflectance over land from a variety of sensors. This technique builds on the extensive dataset acquired by the Terra platform by combining MODIS and MISR to derive an explicit and dynamic map of band ratio's between blue and red channels and is a refinement of the operational approach used for MODIS and LANDSAT over the past 15 years. We will present the generic approach and the application to MODIS VIIRS, LANDSAT and Sentinel 2 data's and its validation using the AERONET data [1].

Ukraine is a large agricultural country situated in Eastern Europe (603,500 km2). Nowadays in Ukraine, there is no any land market due to the moratorium on land sales. Nevertheless, in all areas preparation for land market is undergoing. Cropland productivity assessment based on satellite data is a challenging task for Ukraine because of a large territory and big diversity of agricultural crops. Cropland productivity is one of the major factors for forming the land price. In this paper, we aim to provide land productivity maps based on analysis of MODIS and Landsat-8 data due to availability long term time-series of Normalized Difference Vegetation index (NDVI) from sensors aboard those remote sensing satellites. Taking into account the huge amount of satellite products to be analyzed, in the study we propose to exploit the Google Earth Engine (GEE) cloud platform. It was found that land productivity maps provided from MODIS data for different time periods are strongly correlated. The experiment shows that land productivity maps should have high resolution. That is why, Landsat-8 data is more appropriate for land market purpose, despite of some bias in values comparing to results based on MODIS data. Comparing crop mask from ESA Sen2Agri project and obtained results it was found the dependence of land productivity value and crop/non-crop cover. It was found that irrigated fields from the south part of the study area are the most productive lands in Ukraine.

The land surface reflectance is a fundamental climate data record at the basis of the derivation of other climate data records (Albedo, LAI/Fpar, Vegetation indices) and has been recognized as a key parameter in the understanding of the land-surface-climate processes. In this presentation, we present the validation of the Land surface reflectance used for MODIS, VIIRS, Landsat 8 and Sentinel 2 data. This methodology uses the 6SV Code and data from the AERONET network. The overall accuracy clearly reaches the satellite specifications. To understand how to improve the validation, we developed an exhaustive error budget. Results show an impact of the absorption of aerosol and of the fine mode volume concentration.

For many applied problems in agricultural monitoring and food security it is important to provide reliable crop classification maps in national or global scale. Large amount of satellite data for large scale crop mapping generate a “Big Data” problem. The main idea of this paper was comparison of pixel-based approaches to crop mapping in Ukraine and exploring efficiency of the Google Earth Engine (GEE) cloud platform for solving “Big Data” problem and providing high resolution crop classification map for large territory. The study is carried out for the Joint Experiment of Crop Assessment and Monitoring (JECAM) test site in Ukraine covering the Kyiv region (North of Ukraine) in 2013. We found that Google Earth Engine (GEE) provided very good performance in enabling access to remote sensing products through the cloud platform, but our own approach based on ensemble of neural networks outperformed SVM, decision tree and random forest classifiers that are available in GEE.

This study aims at addressing misregistration issues between Landsat-8/OLI and Sentinel-2A/MSI at 30 m resolution using a phase correlation approach and multiple transformation functions. Phase correlation proved to be a robust approach that allowed us to identify hundreds and thousands of control points on images acquired more than 100 days apart. Overall, misregistration of up to 1.6 pixels at 30 m resolution between Landsat-8 and Sentinel-2A images were observed. The Random Forest regression used for constructing the mapping function showed best results, yielding an average RMSE error of 0.07 pixels at 30 m resolution for multiple tiles and multiple conditions.

Deep learning (DL) is a powerful state-of-the-art technique for image processing including remote sensing (RS) images. This letter describes a multilevel DL architecture that targets land cover and crop type classification from multitemporal multisource satellite imagery. The pillars of the architecture are unsupervised neural network (NN) that is used for optical imagery segmentation and missing data restoration due to clouds and shadows, and an ensemble of supervised NNs. As basic supervised NN architecture, we use a traditional fully connected multilayer perceptron (MLP) and the most commonly used approach in RS community random forest, and compare them with convolutional NNs (CNNs). Experiments are carried out for the joint experiment of crop assessment and monitoring test site in Ukraine for classification of crops in a heterogeneous environment using nineteen multitemporal scenes acquired by Landsat-8 and Sentinel-1A RS satellites. The architecture with an ensemble of CNNs outperforms the one with MLPs allowing us to better discriminate certain summer crop types, in particular maize and soybeans, and yielding the target accuracies more than 85% for all major crops (wheat, maize, sunflower, soybeans, and sugar beet).

This study aims at analyzing sub-pixel misregistration between multi-spectral images acquired by the Multi-Spectral Instrument (MSI) aboard Sentinel-2A remote sensing satellite, and exploring its potential for moving target and cloud detection. By virtue of its hardware design, MSI's detectors exhibit a parallax angle that leads to sub-pixel shifts that are corrected with special pre-processing routines. However, these routines do not correct shifts for moving and/or high altitude objects. In this letter, we apply a phase correlation approach to detect sub-pixel shifts between B2 (blue), B3 (green) and B4 (red) Sentinel-2A/MSI images. We show that shifts of more than 1.1 pixels can be observed for moving targets, such as airplanes and clouds, and can be used for cloud detection. We demonstrate that the proposed approach can detect clouds that are not identified in the built-in cloud mask provided within the Sentinel-2A Level-1C (L1C) product.

In the paper we propose the methodology for solving the large scale classification and area estimation problems in the remote sensing domain on the basis of deep learning paradigm. It is based on a hierarchical model that includes self-organizing maps (SOM) for data preprocessing and segmentation (clustering), ensemble of multi-layer perceptrons (MLP) for data classification and heterogeneous data fusion and geospatial analysis for post-processing. The proposed methodology is applied for generation of high resolution land cover and land use maps for the territory of Ukraine from 1990 to 2010 and 2015.

Wheat is one of the most important cereal crops in the world. Timely and accurate forecast of wheat yield and production at global scale is vital in implementing food security policy. Becker-Reshef et al. (2010) developed a generalized empirical model for forecasting winter wheat production using remote sensing data and official statistics. This model was implemented using static wheat maps. In this paper, we analyze the impact of incorporating yearly wheat masks into the forecasting model. We propose a new approach of producing in season winter wheat maps exploiting satellite data and official statistics on crop area only. Validation on independent data showed that the proposed approach reached 6% to 23% of omission error and 10% to 16% of commission error when mapping winter wheat 2-3 months before harvest. In general, we found a limited impact of using yearly winter wheat masks over a static mask for the study regions.

In this paper, we propose a new approach to pixel and parcel-based classification of multi-temporal optical satellite imagery. We first restore missing data due to clouds and shadows based on vector and raster data fusion in different phases of classification methodology. Pixel-based classification maps are derived from an ensemble of neural networks, in particular multilayer perceptrons (MLPs). The proposed approach is applied for regional scale crop classification using multi-temporal Landsat-8 images for the JECAM site in the Kyivska oblast of Ukraine in 2013. The obtained results on crop area estimates are also compared to official statistics.

In this paper we propose a new methodology to automatically generate retrospective high resolution land cover maps on a regular basis for the whole territory of Ukraine. An ensemble of neural networks, in particular multilayer perceptrons (MLPs), is used for multi-temporal Landsat-4/5/7 satellites imagery classification with previously restored missing data due to clouds, shadows and non-regular coverage. This methodology was used to obtain land cover maps for the territory of Ukraine for three decades, namely 1990s, 2000s and 2010s, with overall accuracy more than 97%.

In this paper, we propose an approach for estimation biophysical parameters, namely LAI effective, FAPAR, and FCOVER, based on in-situ and satellite measurements. In-situ data were collected during 2013-2014 within several field campaigns at the JECAM test site in Ukraine. We have built 30-meter resolution crop specific maps of biophysical parameters based on regression dependencies between ground measurements and NDVI derived from high resolution imagery (Landsat, SPOT) and Proba-V (100 m). In this paper, we discuss the best model selection for LAI effective, FAPAR and FCOVER mapping as well as selection of optimal source of satellite images. Obtained results are compared to available coarse resolution global biophysical products such as MODIS and SPOT-Vegetation.

In this paper, we focus on assessing efficiency of using different satellite-derived parameters (biophysical variables), such as NDVI, FAPAPR, and VHI, for crop yield forecasting for Ukraine. The main objective of this study is to select the optimal parameters using a rigorous feature selection procedure based on random forest and cross validation. Empirical models incorporating different parameters are built, and used in the operational framework. The forecasts that are available 2-3 months prior harvest are compared to official statistics for the years 2011-2013.

In this paper, we focus on the application of satellite synthetic-aperture radar (SAR) images for discriminating summer crops in Ukraine within the JECAM project. Both optical (EO-1/ALI) and SAR (RADARSAT-2) images are used in order to assess impact adding SAR images for classification purposes. Three different classifiers, in particular neural networks, support vector machine and decision trees, are applied with neural networks giving the best overall accuracy. It is found that major impact of using SAR images is for sunflower and sugar beet classes while there was no gain for other crops (maize and soybeans).

In this paper, we focus on quantitative drought risk assessment using satellite data. Methods of the extreme value theory (EVT) are applied for a time-series of vegetation health index (VHI) derived from NOAA satellites in order to provide drought hazard mapping. For this, a Poisson-GP (Generalized Pareto) model is applied for modelling VHI extreme values. The model allows estimation and mapping of return periods of different categories of drought severity. An approach to economical risk assessment due to droughts is presented. The derived drought hazard map is integrated with high resolution crop map to provide final estimates of risk. The proposed approach is implemented for quantitative assessment of drought risk for the Kyiv region in Ukraine.

Wheat is one of the most important and grown crops in Ukraine. Enabling reliable and accurate winter wheat yield forecasts several months in advance of the harvest is an important problem. In this paper we assess relative efficiency of using MODIS data to winter wheat yield forecasting in Ukraine at oblast level. Relative efficiency is defined as the ratio between the variance of the sample (in our case the winter wheat yield official statistics) and the variance of the estimate that has been made with the aid of satellite data. Performance of the forecasting models in terms of relative efficiency was dependant on the agroclimatic zone being on average 1.2 for Plane-Polissya, 1.5 for Forest-Steppe, and 1.9 for Steppe.

In this paper we discuss advantages and benefits of Sensor Web approach to flood monitoring and risk assessment. A general framework of using Sensor Web based services is discussed that incorporates heterogeneous data sources to provide disaster hazard mapping. Probability density function of the disaster is estimated based on the analysis of heterogeneous geospatial data. We use risk functional minimization theory that is developed within the theoretical framework known as a statistical learning theory. A particular case-study, the Namibia SensorWeb Pilot Project, of exploiting this framework is described.

In this paper we discuss results of a pilot study conducted by Ukrainian Space Research Institute of NASU-NSAU, in collaboration with the MARS team of the JRC, to explore the feasibility, cost-efficiency and specific difficulties of crop area estimation assisted by satellite remote sensing in Ukraine. The study compares the cost efficiency of several image types (MODIS, Landsat TM, AWiFS, LISS-III and RapidEye) combined with a field survey on a stratified sample of square segments. Additionally, field data were collected “along the road” as training data for image classification algorithms. The study shows that TM images from Landsat 5 yielded the best results, in spite of the old age of this sensor. Among the sensors that were tested, only MODIS and Landsat TM reach cost-efficiency thresholds.

A system is proposed for user activity monitoring in computer networks. The system is based on the use of neural networks and is implemented using agent approach. The monitoring system allows to detect anomalies in user activity, and consists of two components-on-line and off-line. On-line monitoring is carried out in real time and is used to predict the processes started by an user on the basis of previous ones. Off-line monitoring is carried out at the end of the day and is based on the analysis of statistical parameters of user behavior (user signature). Both on-line and off-line monitoring use neural network approach to detect anomalies in user behavior. Proposed system was verified on real data obtained in Intranet of Space Research Institute of NASU-NSAU and Institute of Physics and Technologies of National Technical University of Ukraine "Kiev Polytechnic Institute".