The native vegetationof Italy reflects the heterogeneity of the physical environments. At least three macro-areas of dif- fering vegetation can be detected: the Alps, the Po Valley, and the Mediterranean-Apennine area. Native vegetation is in practice closeted in high altitude areas. In the strictly Mediterranean part of the Apennines, where the forests have been destroyed, typical scrubland vegetation, composed pri- marily of leathery, broad-leaved evergreen shrubs or small trees, which is called maquis, has grown up. It can be mainly found on the lower slopes of mountains bordering the Mediterranean Sea. Altogether, vegetation typologies are very heterogeneous. They range from plants typical of warm climates, such the papyrus plant that is present in Sicily, to plants that are endemic in northern Europe, such as the Nor- way spruce and the Scotch pine. However, the main part of the territory is man managed; agriculture is quite diffused even in non level areas. It is mainly divided into field crops, fruit tree plantations, and agro-forestry areas. Land use and land cover patterns in anthropized areas are rather erratic since urban and industrial areas coexist with cultivated and densely vegetated areas. AVHRR data at 1 km proved to be useful for characterizing such heterogeneity (Bonfiglio et al., 2002; Maselli, 2004; Simoniello et al., 2004). In this work, for the first time, the potential use of 8 km data for the same purpose has been tested.
In this study, two sources of data were considered. The 300m resolution NDVI data, consisting of 10 days aggregates of NDVI, provided by the SPOT-Vegetation mission and available for the whole world since 2014 were downloaded from Copernicus global land service portal (CGLSP, 2017). Landsat 8 Operational Land Imager (OLI) images (30m resolution) were downloaded from the portal of the United States Geological Survey (USGS, 2018) for the growing season (March to October) from 2014 to 2017. Landsat 8 images have a temporal resolution of 16 days. However, cloud cover rendered most of them useless (Saranya, 2014; Sun et al., 2017) in our study. Therefore, selected images included all available images with no or insignificant cloud cover in the footprint of the tower. Contrarily to SPOT-Vegetation NDVI data, Landsat images represent single date images. The Table 1 presents the different bands of Landsat 8 and those of SPOT-Vegetationsatellite.
ABSTRACT. Arabica Coffee ( Coffea arabica L.) demonstrates a two-year phenological cycle, this knowledge is important for crop forecast in Brazil. This work aimed to describe the coffee crop phenology from MODIS vegetation index timeseries. The study area is located in the western Bahia State, Brazil, due to its remark- able agribusiness development. MODIS timeseries data comprehended 10-year Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI). However, these times series are usually contaminated by noise caused by atmospheric variations that are harmful to the surface discrimination. Median filter and the Minimum Noise Fraction (MNF) were used together to smooth the original dataset. NDVI and EVI temporal profiles showed differences of amplitude and gradient. The results evidenced the Arabica Coffee phenological stages, as described in previous fieldworks. These results showed potential application for large-area land cover monitoring.
Not only are satellite sensors different in resolution, coverage and cost but there are also different methods and techniques for measuring vegetation carbon stock and sequestration. These differences between them may vary in time labor, techniques, need of special software and especially in investment. Although remote sensing looks promising, some complications are naturally observed. Cloud cover for instance is sometimes presents year-round and impossible to overcome for optical remote sensing. Flooding also imposes difficulties when trying to measure vegetation carbon stock. Selective logging and forest diversity is another crucial discussion when a land classification is considered. The removal of specific species may imply differences in total carbon stored but may not imply on a distinct change in an image pixel value. As for forest diversity, methods for quantifying carbon in a forest with single species is significantly different from a mixed species forest (Vincent, Saatchi 1999).
The aim of this study to yield information about vegeta- tion and LC to differ between human-driven change, climate- driven change, vegetation trends and inter-annual ecosystem variability, even if the ground truth knowledge is scarce, is a challenge to the definition of parameters of the de- cision tree classifier. It was not possible to use varying thresholds throughout the years since knowledge on inter- annual ecosystem variability in the study region was not known before. The adaptation of classification thresholds to precipitation sums was also impossible because we had no knowledge on the vegetation-precipitation interaction. This way the use of fix classification thresholds ensures to extract the core of each LCT as stably classified areas while the re- maining areas vary normally between two LCT as described. By examining the stably classified areas it becomes possible to extract information about the LCT and vegetation (dynam- ics). There is though, a big difference of LCC every year due to inter-annual fluctuations caused by fix thresholds. In terms of LCC detection we had to find a solution to delimit varying areas from areas with actual LCC. Hence we used the requirements that there is exactly one LCC between two LCT within the years to delimit actual LCC and to receive a sufficient likelihood to call it actual LCC. Apart from the many documented LCC due to inter-annual ecosystem vari- ability documenting e.g. dryer and wetter years it was pos- sible to extract 1.6 % (∼374 km 2 ) of the study region which have actually changed LCT. In total, change detection of LC for areas with actual LCC has raised value in comparison to commonly change detection techniques because it does not compare two different points in time but a gapless time se- ries of data in 16 d steps. Thus, misleading results due to inter-annual variability (as possible by comparing e.g. Fig. 5: 2007 and 2009) are minimised.
The study represents an initial attempt to use a suite of new flux tower data in sub-Saharan Africa to improve ET estima- tion using remote sensing and surface climate reanalysis data for regular near real-time continental scale monitoring. The integration of the evaporation components from GNOAH with the PT-JPL transpiration component further improves the correlations and reduces the RMSE for both humid and semi-arid sites compared to the models before hybridization. The fractional total vegetationcoverof GNOAH is the sin- gle most important variable controlling transpiration. This is consistent with other findings that showed LAI was a strong control on the ratio of ET/PET at savanna sites (Williams et al., 2009). The use of climatology to derive this component often suffers in semi-arid climates where variability in veg- etation can be high. This is most apparent at the drier sites, where GNOAH shows low interannual variability. The intro- duction of a time-varying component dramatically improves the correlation between observed and modeled LE. The PT- JPL and GNOAH model use two different satellite sensors (AVHRR and MODIS) to determine vegetation indices, and this undoubtedly plays a role in the results, which should be considered along with other products in the future.
Figure 7 illustrates how each individual instrument con- tributes to the all instrument average (green line) for the 25– 35 km and 30 ◦ S–30 ◦ N bin once the respective anomalies are calculated. From October 1991 until January 1995 we only consider HALOE data as the SAGE II data are removed due to aerosol contamination. The main feature from 1991–1996 is a general increase in values until the middle of 1996 where values level off until 2002. HALOE and SAGE II values are by and large consistent with each other from 1995 un- til 2005 when each instrument ceased operation. After 2002 mean water vapour anomaly values drop, which is more pro- nounced in the HALOE data than in the SAGE II data. This result is in agreement with reports of a sudden decrease in water vapour values in the lower most stratosphere in 2001, which are coupled with a decreased tropical tropopause tem- perature (Randel et al., 2004, 2006). Perhaps most inter- estingly of all is the excellent agreement of SMR and MLS anomalies, which show a similar pronounced structure as the HALOE timeseries during their respective overlapping pe- riods (2001–2005, and 2004–2005 respectively), but are also less noisy due to their better spatial and temporal sampling of the emission sensors (SMR measures with global cover- age in approximately one day per week, while MLS mea- sures a global coverage daily). Not only does this confirm the drop in water vapour values seen previously using HALOE and other measuring techniques (such as the balloon sonde frost point hygrometer measurements at Boulder, Colorado), it more importantly shows how well the different measure- ments agree considering the different techniques used. Since
In this study, we address some of these challenges. We start by establishing a methodology to develop a multitemporal analysis of the LULC changes and thereby provide essential information regarding the extent of the changes. The goal of this study was twofold: (i) access free LULC data to be used as training samples for a long-term satellite image time-series classification; and (ii) identify the LULC changes that occurred from 1995 to 2015 in a rural region characterized by a mixed agro-silvo-pastoral environment in the municipality of Beja, Portugal. This region is characterized by a set of LULC complex patterns typical of the Iberian Peninsula. It is a Mediterranean agro-forestry system [33,34] that includes a vast landscape of intermingling cultures, such as wheat, olive groves, vineyards, cork oak forests and pastures, which have a high economic importance in the Portuguese agricultural industry . Therefore, we first sought to integrate the free official Portuguese LULC maps (Carta de Uso e Ocupação do Solo, COS) to produce a data source for training purposes. Second, we sought to investigate the potential of a k-means clustering technique [25,36,37] to refine the broad range of spectral signatures for each LULC class of the training data. Third, we tested the TWDTW algorithm on a Landsat imagery time-series classification, using the refined sample source. Finally, we evaluated the extent of the changes over 21 years.
missing values, the least squares estimator for a m−AR(p) model is consistent and asymptotically normal (L¨utkepohl, 1993) and is recommended, provided that the model to be fit- ted is stable (Schneider and Griffies, 1999). Difficulties in the estimationof a m−AR(p) are often associated with the large number of parameters involved. For large fields the number of spatial degrees of freedom can be reduced by consider- ing the leading components from a PCA analysis thereby re- ducing dimensionality while retaining most of the variance in the original field. Although PCA is a multivariate anal- ysis method for independent observations, and thus should not be used for timeseries, non-independence does not have a serious effect when the main objective of the analysis is descriptive rather than inferential (Jollife, 2002). A positive by-product of estimating a multivariate autoregressive model in PCA space is the exclusion of noisy components from the analysis and diagonalisation of the error covariance matrix.
We used a timeseriesof 37 Landsat satellite images that were acquired between 1984 and 2009, path/row 228/061. Thirty-three of these images were recorded by Landsat 5/TM, while the remaining four images were recorded by Landsat 7/ ETM+ (Table 1). We obtained up to two images per year for the months of September–November, corresponding to the low water period of the Amazon River and the season with lowest cloud incidence. The LandTrendr algorithm requires that all images correspond to the same season of each year to minimize variations caused by phenology, flooding, or changes in solar geometry, which could be detected as false cover changes (Kennedy et al. 2010). All images were acquired from the Landsat Surface Reflectance Climate Data Record (CDR, http://earthexplorer.usgs.gov/). CDR images are corrected for atmospheric interference (surface reflectance) and geometric distortions. They also include cloud and cloud shadow masks, generated using the fmask algorithm (Schmidt et al. 2013). Registration and geolocation errors for CDR products have an expected Root Mean Square Error (RMSE) of 50 meters (Loveland and Dwyer 2012).
Remote sensing has long been used as a means of detecting and classifying changes on the land. Analysis of multi-year timeseriesof land surface attributes and their seasonal change indicates a complexity of land use land cover change (LULCC). This paper explores the temporal complexity of land change considering temporal vegetation dynamics, in other words, distinguishing the changes regarding to their properties in long-term image analysis. This study is based on the hypothesis that land cover might be dynamics; however, consistent land use has a typical, distinct and repeated temporal pattern ofvegetation index inter-annually. Therefore, pixels represent a change when the inter-annual temporal dynamics is changed. We analysed the dynamics pattern of long-term image data of wavelet-filtered MODIS EVI from 2001 to 2007. The change of temporal vegetation dynamics was detected by differentiating distance between two successive annual EVI patterns. Moreover, we defined the type of changes using the clustering method, which were then validated by ground check points and secondary data sets.
The goal of our study was mainly to highlight the individual and combinatorial influence of the spectral and of the temporal components of remotely sensed images reflectances in land cover classification. As our aim was not to propose an operational classifier directed at thematic mapping based on the most efficient combination of reflectance inputs, we intentionally restricted our experimental framework to continental Portugal. In the course, we were led to define a distance-based topography to rank these features in terms of their relevant contribution to a discriminatory representation space. The chosen Mahalanobis median distance, although it was not shown to strictly optimize any classification criterion, deemed among all the possible arbitrary choices a reasonable indicator of the classes’ dispersion and of the clusters compactness. Then, following this resulting arrangement of features inputs to sequentially train a Support Vector Machine classifier, we were able to demonstrate that the Enhanced Vegetation Index (EVI) calculated in August was the most informative combination of one spectral band with one date to characterize the land cover classes that we retained to describe the Portuguese mainland. Continuing to gradually include the remaining bands and dates, we also exposed the context dependent advantages of each new component to the classification performances, and thus proved the multitemporality assets and limitations. In this way, we showed that spectral diversity is a richer source of information than time variety. In fact, the multitemporal factor has a significant effect when coupled with combinations of few spectral bands, but it turns negligible as soon as the full spectral information is exploited. In contrast, even with a full year measurement, there is always substantial interest in considering no less than three spectral channels. As a by-product of our study, we evidenced the poor adequacy of spectral and temporal recourses at differentiating certain land cover classes. A situation often pointed out by previous investigations in diverse bioclimatic study areas, and due to baffling temporal and spectral similarities between distinct classes’ phenologies.
The estimationof sugarcane (Saccharum officinarum L.) yield can be conducted at local (e.g. sugar mills) and regional (e.g. government) scales. The yield estimation methods for sugarcane adopted by the Brazilian govern- ment are considered subjective because they are based on information gathered from direct inquiries to the pro- duction sector, such as field research using question- naires, surveys on information about demands on agri- culture raw materials, use of yield historical data and field observations on plant behavior (IBGE, 2002; CONAB, 2007). The possibility of determining sugar- cane development by spectral data such as the Normal-
Agriculture is the backbone of Gojjam economy as it depends on seasonal characteristics of rainfall. This study analyses the components of regional climate variability, especially La Niña or El Niño Southern Oscillation (ENSO) events and their impact on rainfall variability and the growing season normalized difference vegetation index. The temporal and spatial distribution of temperature, precipitation and vegetationcover have been investigated statistically in two agricultural productive seasons for a period of 9 years (2000–2008), using data from 11 meteorological station and MODIS satellite data in Gojam, Ethiopia. The normalized difference vegetation index (NDVI) is widely accepted as a good indicator for providing vegetation properties and associated changes for large scale geographic regions. Investigations indicate that climate variability is persistent particularly in the small rainy season Belg and continues to affect vegetation condition and thus Belg crop production. Statistical correlation analyses shows strong positive correlation between NDVI and rainfall in most years, and negative relationship between temperature and NDVI in both seasons. Although El Niño and La Niña events vary in magnitude in time, ENSO analyses shows that two strong La Niña years and one strong El Niño years. ENSO analyses result shows that its impact to the region rainfall variability is mostly noticeable but it is inconsistent and difficult to predict all the time. The NDVI anomaly patterns approximately agree with the main documented precipitation and temperature anomaly patterns associated with ENSO, but also show additional patterns not related to ENSO. The spatial and temporal analyses of climate elements and NDVI values for the growing season shows that NDVI and rainfall are very unstable and variable during the 9 years period. ENSO /El Niño and La Niña events analyses shows an increase ofvegetation coverage during El Niño episodes contrasting to La Niña episodes. Moreover, El Niño years are good for Belg crop production.
Mongolia is a land-locked country located in Central Asia between Russia and China. The climate is continental with harsh winters and hot, short summers. The total population of Mongolia is 2.3 million with an area of 1.565 million sq. km, making it one of the lowest population density areas (1.5 persons per sq. km) in the world. The main economy is nomadic animal husbandry with 33.4% GDP. Of the work force, 48.6% is in the livestock- breeding sector. Until 1990, Mongolia had a central economy with state owned farms of cattle, sheep, goat, camels and horses. There were also state- supported monitoring systems for livestock production, water supply and pasture quality. After 1990 Mongolia shifted towards a market-economy, privatization took place for much of the state owned properties and former structures for monitoring failed to function effectively. Privatization has stimulated the livestock industry, which has reached its maximum and pastureland has approached its maximum carrying capacity (Erdenebaatar et al., 2001). Recent extreme climate variations and pastureland deterioration brought waves of problems in this sector: high livestock mortality; decrease in pasture productivity; increase in livestock disease and as a result, a dramatic decrease in lifestyle quality for herders’ families.
NDVI S10 images from SPOT VGT from 1999 to 2011 were processed for Rondonia. As precipitation has consider- able influence on the vegetation development, the most relevant period for this region is from April to September. Furthermore, the presence of clouds limits the use ofsatellite data in the rainy season. Maximum Value Compo- site images of January and AprilSeptember using NDVI VGT data from 2000 are presented in Figure 4. Due to the frequent cloud cover, the NDVI values from January might not be clearly related to the vegetation type and condition. In the remaining images, it was noticeable that the central areas of Rondonia have lower NDVI values. This is because this part of the state has low height herbaceous or semi- herbaceous vegetation, as the soil is mostly used for agriculture and pasture. Moreover, Rondonia is a remark- able example of land cover change in the past decades as a result of deforestation induced by human and natural causes. The converted forest cover in a fishbone pattern, mainly due to forestation caused by agricultural and urban expansion as reported by Eva et al. (2002), is noticeable in Figure 4. The least influence of clouds occurred between June and August. During this period a large increase in NDVI occurred, especially in forests. The better transpar- ency of the atmosphere at this timeof the year promotes a net balance of radiation greater than the other parts of the year (Da Rocha et al., 2004; Malhi et al., 2002). This high availability of energy and the ability of forestry trees to capture water from deep soil explain the trend of the NDVI increase in forest areas between June and August.
cells in the central United States, eastern China and Argentina. In 10 % of the land area the correlation coefficient of all LPJmL versions was smaller than 0.2 compared to the GIMMS3g dataset. These grid cells were mostly located in the Amazon, the Kongo Basis and the Sunda Islands where FAPAR timeseriesof tropical forests do not ex- hibit seasonal cycles and where optical satellite observations are often distorted from
Woodland caribou (Ranifer Tarandus Caribou) is an important wildlife species ecologically and culturally in Canada. This amazing creature is now at risk and listed as a threatened species provincially and nationally (Ontario Woodland Caribou Recovery Team, 2008). The decline in woodland caribou populations is primarily the result of habitat loss and forest fragmentation which is caused by human interference including spread of agriculture, oil and gas exploration, and mining (McLoughlin, 2003).To protect them from the threat of extinction, it is critical to conserve its habitat conservation. To map the wildlife habitat, the basic idea is to establish a link between the organism’s characteristics and behaviors to the physical habitat (Guisan and Zimmermann 2000). Conventionally, the observation was conducted by field work which has the promising accuracy and generally been used on the local scale. For the past decades, remote sensing techniques played a key role in wildlife habitats mapping, when large spatial information becomes more favourable for broader area analysis while conventional field work is time and labour intensity. (Kerr and Dostoevsky, 2003). The forest inventory map and land use map which generated from lower resolution satellite imagery has been employed as the major data source to map the woodland caribou habitat (Hansen, 2001). In addition to land cover/use maps, other remote sensing products, such as NDVI (Normalized Difference Vegetation Index) images calculated fromsatellite data with medium to low spatial resolution have been used in other wildlife habitat mapping. However, the map was not usually up to date, and classes from products usually generated for other or generic applications are not best suited for woodland caribou habitat mapping. These
It should be noted that just by compiling estimatetrend software using MAGMA libraries without changing anything in the code, the software execution time automatically increased by about 3 seconds. This is one of the reasons why GPU cannot be faster than CPU using very small datasets. Another explanation to the poor performance, and the most relevant regarding to GPU program- ming is about GPU insignificance for small data. For a human, counting from one until 10000 is a very lengthy task. An array of 10000 size also seems pretty big, however, that array, with double precision values, takes 78KiB which is so small that can perfectly be stored in CPU cache. For such a small data, even the best GPU can be useless. This explains exactly why GPU version of estimatetrend starts to be faster only after using significant higher datasets.
The experiment was conducted in a farm located near the Açu National Forest, Rio Grande do Norte state, Brazil, 5°32'23.01"S and 36°57'18.66"O coordinates. The climate is semi-arid with average annual rainfall between 400 and 600mm (Alvares et al. 2013). After clear-cutting, areas are generally used for agriculture and pasture for livestock (goats, cattle and horses). The experimental area suffered clear-cutting about 50 years ago for implementation of cotton plantations (widely practiced in the Brazilian semi-arid region in the last century). However, the past 20 years agricultural activities have not been developed and the area was abandoned. In the rainy season (March-June) the study area consisted of a few isolated trees (mainly Zizyphus joazeiro and Combretum leprosum) surrounded by a dominant herbaceous cover. In the dry season the herbaceous cover disappears, forming a scenario of isolated trees in a bare soil matrix. In the study area, cattle and horse grazing is a frequent activity. The herbaceous cover is an important source of food for these animals in Caatinga, especially during rainy season. In addition to livestock, the movement of abandoned donkeys by local communities is common. About 30 animals among cattle, horses and donkeys can be seen circulating in the study area daily.