Recent studies have shown that Convolutional Neural Networks (CNNs) achieve impressive results in crop segmentation of Satellite Image Time-Series (SITS). However, the emergence of transformer networks in various vision tasks raises the question of whether they can outperform CNNs in crop segmentation of SITS. This paper presents a revised version of the Transformer-based Swin UNETR model adapted specifically for crop segmentation of SITS. The proposed model demonstrates significant advancements, achieving a validation accuracy of 96.14% and a test accuracy of 95.26% on the Munich dataset, surpassing the previous best results of 93.55% for validation and 92.94% for the test. Additionally, the model’s performance on the Lombardia dataset is comparable to UNet3D and superior to FPN and DeepLabV3. Experiments of this study indicate that the model will likely achieve comparable or superior accuracy to CNNs while requiring significantly less training time. These findings highlight the potential of transformer-based architectures for crop segmentation in SITS, opening new avenues for remote sensing applications.
Mediterranean forests are every year affected by wildfires which have a significant effect on the ecosystem. Mapping
burned areas is an important field of application for optical remote sensing techniques and several methodologies have
been developed in order to improve mapping accuracy. We developed an automated procedure based on spectral indices
and fuzzy theory for mapping burned areas from atmospherically corrected Landsat TM images. The algorithm proved to
provide consistent accuracy over Mediterranean areas. We further tested algorithm’s performance to assess the influence
of the atmospheric correction on the accuracy of burned areas. In particular, we ran the Second Simulation of a Satellite
Signal in the Solar Spectrum (6S) code with different Atmospheric Optical Thickness (AOT) levels and two aerosol
models (continental and maritime) on one TM image acquired over Portugal (12/08/2003). Burned area maps derived
from atmospherically corrected images and from the non corrected image (Top Of Atmosphere, TOA) have been
analyzed. In the output burned areas maps the omission error varies in the range 4.6-6.5% and the commission error
fluctuates between 11.9 and 22.2%; the highest omission (commission) errors occur with the continental (maritime)
model. The accuracy of burned area maps derived from non corrected image is very low, with omission error greater than
90%. These results show that, although atmospheric correction is needed for the application of the algorithm, the AOT
value does not significantly affect the performance.
Large wildfires in forests of southern European countries such as Portugal, Spain, Greece, France and Italy are one key
ecological disturbance of the Mediterranean environment. Optical data have been largely used for burned area mapping and literature provides an extensive reference for the typical spectral signal of burns and the methodologies applied to extract burn perimeters. However, optical remote sensing techniques have the major limitation of a reduced frequency of clear images due to cloud cover; moreover, for the specific application of burned area mapping, unburned targets such as shadows, can be spectrally confused and misclassified as burns. For this reason radar images could be integrated as an additional source of information. We developed an approach for mapping burned areas in Mediterranean regions based on Landsat TM/ETM+ data and vegetation indices that provided satisfactory results. However, we are currently working for further improving our approach by exploiting the synergy between optical and radar data. In this paper we present the first results of the analysis of the SAR backscatter over burned areas for future integration into the formal framework previously developed. Although results are preliminary, they encourage us to test the approach over different regions of the Mediterranean environment to evaluate its robustness.
Rice farming, one of the most important agricultural activities in the world producing staple food for nearly one-fifth of the global population, covers 153 MHa every year corresponding to a production of more than 670 Mton. Retrieve updated information on actual rice cultivated areas and on key phenological stages occurrence is fundamental to support policy makers, rice farmers and consumers providing the necessary information to increase food security and control market prices. In particular, remote sensing is very important to retrieve spatial distributed information on large scale fundamental to set up operational agro-ecosystem monitoring tool. The present work wants to assess the reliability of automatic image processing algorithm for the identification of rice cultivated areas. A method, originally tested for Asian tropical rice areas, was applied on temperate European Mediterranean environment. Modifications of the method have been evaluated to adapt the original algorithm to the different experimental conditions. Finally, a novel approach based on phenological detection analysis has been tested on Northern Italy rice district. Rice detection was conducted using times series of Vegetation Indices derived by MODIS MOD09A1 products for the year 2006 and the accuracy of the maps was assessed using available thematic cartography. Error matrix analysis shows that the new proposed method, applied in a fully automatic way, is comparable to the results of the original approach when it is customized and adapted for the specific study area. The new algorithm minimizes the use of external data and provides also spatial distributed information on crop phenological stages.
Leaf area index (LAI) is a key variable for modeling the interaction between vegetation and the atmosphere. We collected field LAI measurements over 12 sites in 2005 in the Lys valley, Northern Italy, to calibrate regressive models using normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), and wide dynamic range vegetation index (WDRVI) products derived from 250-m moderate resolution imaging spectroradiomete (MODIS) imagery. Field data were compared to the 1 km MODIS leaf area index-fraction of photosynthetically active radiation (LAI/fAPAR) product to show that regressive techniques are better suited for local applications. We investigated these LAI-vegetation index (VI) regressive models to 1. test the sensitivity of the model to forest type and phenology, 2. identify the most suitable VI for LAI retrieval, and 3. verify the feasibility of using a linear model. Results show that in our experimental conditions the LAI-VI relationship is primarily influenced by phenology and that the leaf constant period (maximum LAI) is significantly different compared to the other phenological phases. Among the indices, EVI yielded the poorest performance (R2<0.41) and the linear regressive models for NDVI and WDRVI derived by pooling together data from different phenological phases show a good correlation with field data (R2>0.65); the use of a logarithmic model does not improve the performance. The LAI-WDRVI and LAI-NDVI models were inverted and applied to 2004 MODIS data and model performance was assessed by comparing predicted and measured LAI. Results show that WDRVI performs best in a linear regressive model, yielding a relative root mean square error <23%.
The Sahelian belt of West Africa is a region characterized by wide climate variations, which can in turn affect the
survival of local populations especially in rangeland, as happened during the dramatic food crisis in the 70-80s caused by
severe drought. This work has been carried out in the framework of the EU FP7 Geoland2 project as a contribution to the
ECOWAS component (Economic Community Of West African States) of the AMESD (African Monitoring of the
Environment for Sustainable Development) programme with the purpose of establishing the reliability of Dry Matter
Productivity (DMP) developed by Flemish Institute for Technological Research (VITO), a spatial estimation of dry
matter (DM) obtained from remotely sensed data. DMP can be of great help in monitoring savanna pasturelands in a
region characterized by food insecurity and a significant variability of biomass production, linked to climate variations,
which can in turn affect the survival of local populations. The evaluation of DMP was carried out thanks to the Centre de
Suivi Ecologique (CSE) and Action Contre la Fame (ACF), the partners who provided the field biomass measurements.
The paper shows the correlation of DMP with field measurements of herbaceous biomass, and discusses the differences
among the different sites where ground data were collected. The analysis of other environmental variables (land cover,
rainfall), which can be influential on rangeland biomass production, is presented in order to better explain the variance of
field measurements among the different years.
Natural Resource Monitoring in Africa (NARMA) is one of the Core Information Services of EU-FP7 project Geoland2
addressing important sectoral policies that concern with the development of an environmental monitoring capacity over
African countries for the needs of the European Commission (EC) services and for regional and continental EC partners
in African countries. Congo basin is one of the target area where NARMA has to contribute to the development of
AMESD/CICOS services in support to management of water resources focusing on environmental aspects of watersheds.
In this contest and to better understand dynamics that occur in the watershed, an analysis has been conducted on the
relation between precipitation, river discharge and vegetation dynamics by exploiting available time series of Earth
Observation data. Rainfall dynamics has been described using FEWS-NET RFE estimations, river discharge has been
monitored using ENVISAT radar altimeter data provided by LEGOS laboratory and vegetation dynamics have been
examined through vegetation indices available from long term series of SPOT-VGT data. The comparison between river
discharge measured at Bangui (Central African Republic), gauging station and radar altimeter virtual station data
demonstrated that these data can be used to estimate river discharge. This result allowed to focus a preliminary analysis
on the Uele watershed, Ubangi sub basin, using radar data as a proxy of river discharge, comparing these trends to
seasonal rainfall estimates and trying to disentangling the effect of vegetation on discharge-rain relation. Results showed
that a strong positive correlation is obtained between rain data and river discharge only at the end of the vegetation
season when plants have reduced water demand for evapotranspiration and less intercept rain. Trend analysis on the
considered time windows are provided and the contribution of these finding for river water alert monitoring system is discussed.
Studies of impact of human activity on the vegetation dynamics in the Sahel belt of Africa are recently re-invigorated
due to a new scientific findings that highlighted the primary role of climate in the drought crises of the 70s-80s. Time
series of satellite observations allowed identifying re-greening of the Sahel belt that indicates no sensible human effect
on vegetation dynamics at sub continental scale from 80s to late 90s. However, several regional/local crises related to
natural resources occurred in the last decades underling that more detailed studies are needed. This study contribute to
the understanding of climate/human impact on pasture vegetation status in the Sahel region in the last decade (1999-
2008). The use of a time-series of SPOT-VGT NDVI and FEWS-RFE rainfall estimates allowed to analyze vegetation
and rainfall trends and identify local anomalous situation in the region. Trend analysis has been conducted to map a)
areas where vegetation has been significantly decreased or increased due to rainfall pattern and b) anomalous zones
where vegetation dynamics could not be fully explained by rainfall pattern by. The identified hot-spots areas have been
compared with spatial information on the reported humanitarian-food crisis events in order to understand chronic
situation where ecosystems carrying capacity is endangered. The results of this study show that even if a general positive
re-greening situation is evident for the entire Sahel, some serious hot spots exist in areas where cropping system and
pasture activity are conflicting.
Monitoring crop conditions and assessing nutrition requirements is fundamental for implementing sustainable agriculture. Rational nitrogen fertilization is of particular importance in rice crops in order to guarantee high production levels while minimising the impact on the environment. In fact, the typical flooded condition of rice fields can be a significant source of greenhouse gasses. Information on plant nitrogen concentration can be used, coupled with information about the phenological stage, to plan strategies for a rational and spatially differentiated fertilization schedule. A field experiment was carried out in a rice field Northern Italy, in order to evaluate the potential of field radiometric measurements for the prediction of rice nitrogen concentration. The results indicate that rice reflectance is influenced by nitrogen supply at certain wavelengths although N concentration cannot be accurately predicted based on the reflectance measured at a given wavelength. Regression analysis highlighted that the visible region of the spectrum is most sensitive to plant nitrogen concentration when reflectance measures are combined into a spectral index. An automated procedure allowed the analysis of all the possible combinations into a Normalized Difference Index (NDI) of the narrow spectral bands derived by spectral resampling of field measurements. The derived index appeared to be least influenced by plant biomass and Leaf Area Index (LAI) providing a useful approach to detect rice nutritional status. The validation of the regressive model showed that the model is able to predict rice N concentration (R2=0.55 [p<0.01]; RRMSE=29.4; modelling efficiency close to the optimum value).
Crop growth and production can be simulated by models for the whole canopy as a function of intercepted radiation, water availability, air temperature and nitrogen availability. Simulation models supply quantitative outputs starting from quantitative inputs and they need quite complex databases to run simulations. In practice, the more complex and physically based these tools are, the more inputs are required for their application. In most cases such data are not available. This is the reason why, for large scale evaluations, simplified models are often applied and satellite data are used as input. In particular, multi-temporal Earth Observation data represent a valid tool to define crop phenological stages and derive temporal and spatial variability of vegetation biophysical parameters, such as the Leaf Area Index (LAI). In 2003 and 2004 two intensive field campaigns were conducted over different areas of the Italian Rice Belt, Northern Italy, with the objective of collecting data for growth model calibration. Field spectroradiometer measurements and LAI estimation, retrieved by LAI2000, have been used to study the best Vegetation Index (VI) for rice growth monitoring. VI vs LAI relationship has been scaled up to MODIS data to produce LAI map for the entire growing season and the key phenological rice events have been detected by multitemporal MODIS analysis. Preliminary results of rice production estimation using a Light Use efficiency model that ingests spatially distributed phenological information are presented. Comparison with CropSyst model phenological parameters are provided and the contribution of multi-temporal EO data for regional crop monitoring is discussed.
Hyperspectral imaging is becoming an important analytical tool for generating land-use map. High dimensionality in hyperspectral remote sensing data, on one hand, provides us with more potential discrimination power for classification tasks. On the other hand, the classification performance improves up to a limited point as additional features are added, and then deteriorates due to the limited number of training samples. Proceeding from these considerations, the present work is aimed to systematically evaluate the robustness of novel classification techniques in classifying hyperspectral data under the twofold condition of high dimensionality and minimal training. We consider in the study a neural adaptive model based on Multi Layer Perceptron (MLP). Accuracy has been evaluated experimentally, classifying MIVIS Hyperspectral data to identify different typology of vegetation in Ticino Regional Park. A performance analysis has been conducted comparing the novel approach with Support Vector Machine and conventional statistical and neural techniques. The adaptive model shows advantages especially when mixed data are presented to the classifiers in combination with minimal training conditions.
Rice is one of the most important crops in the whole world, providing staple food for more than 3000 million people. For this reason FAO declared the year 2004 as The International Year of Rice promoting initiatives and researches on this valuable crop. Assessing the Net Primary Production (NPP) is fundamental to support a sustainable development and to give crop yield forecast essential to food security policy. Crop growth models can be useful tools for estimating growth, development and yield but require complex spatial distributed input parameters to produce valuable map. Light use efficiency (LUE) models, using satellite-borne data to achieve daily surface parameters, represent an alternative approach able to monitor differences in vegetation compound providing spatial distributed NPP maps. An experiment aimed at testing the capability of a LUE model using daily MODIS data to estimate rice crop production was conducted in a rice area of Northern Italy. Direct LAI measurements and indirect LAI2000 estimation were collected on different fields during the growing season to define a relationship with MODIS data. An hyperspectral MIVIS image was acquired in early July on the experimental site to provide high spatial resolution information on land cover distribution. LUE-NPP estimations on several fields were compared with CropSyst model outputs and field biomass measurements. A comparison of different methods performance is presented and relative advantages and drawbacks in spatialization are discussed.
In this paper, we propose a method able to fuse spectral information with spatial contextual information in order to solve “operationally” classification problem. The salient aspect of the method is the integration of heterogeneous data within a Multi-Layer Perceptron model. Spatial and spectral relationships are not explicitly formalized in an attempt to limit design and computational complexity; raw data are instead presented directly as input to the neural network classifier. The method in particular addresses new open problems in processing hyperspectral and high resolution data finding solution for multisource analysis. Experimental results in real domain show this fusing approach is able to produce accurate classification. The method in fact is able to handle the problem of a volumetric mixture typical of natural forest ecosystems identifying the different surfaces present under the tree canopy. The understory map, produced by the neural classification method, was used as input to the inversion of radiative transfer models that show a significant increase in the retrieval of important biophysical vegetation parameter.
Knowledge of the characteristics of the vegetation cover is of great interest due to its role in the mass and energy exchanges at the surface/atmosphere interface (e.g. water and carbon cycles). This study is part of DARFEM experiments, EU-funded HySens project (DLR), designed to provide a better understanding of the capability of airborne hyperspectral and directional observations to retrieve biophysical vegetation parameters. Different airborne hyperspectral data were acquired in late June 2001 on the experimental site, a poplar plantation belonging to CARBOEUROFLUX network, located in Northern Italy. An intensive field campaign was accomplished during the aerial survey to collect vegetation parameters and radiometric measurements. Leaf area index (LAI) and vegetation fractional cover (Fc), were retrieved from remote sensing data by statistical relationships with ground measurements. A radiative transfer model was used in direct mode to simulate and analyse the canopy spectral signature changes for varying overstory LAI and different understory conditions. In order to minimize the influence of the extensive understory vegetation on the relationship between spectral Vegetation Index (VI) and LAI, an optical index exploiting short wave infrared (SWIR) was evaluated. A comparison of different VIs performance is presented and relative advantages and drawbacks of SWIR exploitation are discussed.
In this study a relationship between water surface area and river discharge was derived by using multitemporal radar images in the central braided part of the Ticino River (Italy). Ticino River is the outlet of the Lake Maggiore (north-west Italy) and it streams for about 100 km before flowing into the Po River, the biggest river in Italy. The braided part of Ticino River is about 50% of the total length and its active flow increases or decreases in responding to river discharge. Ground measurement of discharge (Q) were related to satellite-derived effective width (We), where We is the water surface area within a braided reach divided by the reach length, showing a good correlation. Power functions were fitted through plots of We and Q and they represent satellite derived rating curves that can be helpful for the estimation of mean daily discharge by using satellite radar data. Those satellite-derived rating curves can be apply where the gauging station are either enough distributed, like in Ticino River, or impossible to set in order to predict the river discharge or in remote areas, where river discharge is not easily available.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.