Forest losses are most evident for the Northern and Trans-Baikal regions and the greatest discrepancy between RS and the SFR is in the Sakha Republic (Table 2). Although these are differences among the estimates, all of the most accurate RS products report much smaller forestarea than indicated in the SFR. The highest discrepancy are between estimates produced by MODIS VCF and the SFR data was -88.7 million hectares. This difference can be explained partly by nature of larch forest (light, sparse, deciduous crown) and by the instrument (accounted canopy cover, which recognizes not only the gaps between the crowns projection, but also gaps within the crown). Therefore, the universal threshold of 25% canopy cover is too high for the larch forests, which make up about 75% of the forests of the Republic of Sakha. Our GWR model uses a geographically variable threshold to approximate the training dataset. As a result, the hybrid dataset estimates forestarea in the Sakha Republic to be 107.9 million hectares (the deviation from SFR is -31.0 million hectares). This is close to the estimates of the high-resolution products – Hansen et al. (2013), Sexton et al. (2013), JAXA Palsar (102-118 million hectares, with the SFR deviation from – 30.2 to -36.4 million hectares). Northern sparse and low productive (V-Vb site indexes) larch forest typically can be classified as forest by the Russian definition starting with 11% tree cover. Incidentally, application of this threshold with MODIS VCF produces an estimate of the forestarea of the Sakha Republic, which almost identical to that of the hybrid map.
The presented study quanti ﬁes the proportion of stand-replacement ﬁres in Russian forests through the integrated analysis of Landsat and Moderate Resolution Imaging Spectroradiometer (MODIS) data products. We employed 30 m Landsat Enhanced Thematic Mapper Plus derived tree canopy cover and decadal (2001 –2012) forest cover loss (Hansen et al 2013 High-resolution global maps of 21st-century forest cover change Science 342 850–53 ) to identify forest extent and disturbance. These data were overlaid with 1 km MODIS active ﬁre ( earthdata.nasa.gov/ data/near-real-time-data/ ﬁrms ) and 500 m regional burned area data (Loboda et al 2007 Regionally adaptable dNBR-based algorithm for burned area mapping from MODIS data Remote Sens. Environ. 109 429–42 and Loboda et al 2011 Mapping burned area in Alaska using MODIS data: a data limitations-driven modi ﬁcation to the regional burned area algorithm Int. J. Wildl. Fire 20 487–96 ) to differentiate stand-replacement disturbances due to ﬁre versus other causes. Total stand replacement forest ﬁre area within the Russian Federation from 2002 to 2011 was estimated to be 17.6 million ha (Mha). The smallest stand-replacement ﬁre loss occurred in 2004 (0.4 Mha) and the largest annual loss in 2003 (3.3 Mha). Of total burned area within forests, 33.6% resulted in stand-replacement. Light conifer stands comprised 65% of all non- stand-replacement and 79% of all stand-replacement ﬁre in Russia. Stand-replacement area for the study period is estimated to be two times higher than the reported logging area. Results of this analysis can be used with historical ﬁre regime estimations to develop effective ﬁre management policy, increase accuracy of carbon calculations, and improve ﬁre behavior and climate change modeling efforts.
A number of new satellite sensors designed more speciﬁ cally for observing land cover and land cover changes have been launched recently. Th ese provide improved data in terms of spatial, spectral and angular resolutions, and atmospheric, radiometric and geometric correction. However, land cover mapping is most often based on the spectral information although, for example, the angular sampling of the sensors has improved considerably (Asner et al. 1998; Diner et al. 1999). Furthermore, new medium spatial resolution sensors have good temporal resolution, which increases the potential ap- plications of temporal information in cloud-prone northern latitudes. For example, the NASA’s Earth Observing System (EOS) sensors ASTER, MODIS and MISR are used to make available a range of preprocessed data products in support of a variety of applications. Higher level data products include, among others, global land cover maps (Friedl et al. 2002) and retrievals of biophysical parameters, such as leaf area index (Myneni et al. 2002). Th ese data are distributed together with extensive meta- data over the Internet free of charge or at low price. Furthermore, the temporal continuation of the satellite observations is important for monitoring long-term land cover changes. Th e threat of a possible data gap in the very popular Landsat program has motivated the search for substitutive data sources (Goetz 2007). EOS sensor ASTER, for example, could provide supplementary data, which has been used so far only rarely to study land cover and forests.
Compared with NOAA data and TM data, MODIS data has higher time and middle spatial resolution. The data sharing service has made MODIS data be used to remotesensing monitoring winter wheat more and more [10-13]. Yigang Jing set models and extracted winter wheat area with accuracy of over 91% using the NDVI changes of March, May and June . Jinqiu Zou extracted area using the EVI (Enhanced Vegetation Index) difference between May and October, and the error rate was -0.04% . Wenpeng Lin used NIR, RED, BLUE and ESWIR 4 bands MODIS data of October and December to class 6 kinds of surface features with the method of fuzzy ARTMAP, and the accuracy reached 80.3% . The main method of comparing spectral changes in the key growth periods after sowing for extracting winter wheat area was used frequently using MODIS data . Over time, the amount of information increased, and the accuracy enhanced [18-20]. The paper studied one method with high accuracy to extract winter wheat area at early periods, combining with the former stubble crops based on cropping system in Shandong province.
Out of the 3885.24 ha of closed forest canopy area in 1991, 2803.23 ha remained closed forest in 2002, but 617.04 ha was con- verted to built-up areas and 157.14 ha also changed to opened for- ests in 2002. This negative decrease in closed forestarea from 1991 to 2002 could be due to human induced pressures ( FAO, 2010 ) more specifically, illegal logging and clearing of forest lands for farming in the Bosomtwe Range Forest Reserve ( Boakye et al., 2008 ) coupled with high prevalence of wildfires which had altered the vegetation of most forests in Ghana to Panicum maximum ( Hawthorne and Abu-Juan, 1995 ). Moreover, opened forest canopy retained an area of 929.34 ha in 2002 out of the total 1621.99 ha in 1991 with larger areas being converted to bare landscapes (305.64 ha). Simultaneously, the increase in built-up areas from 1991 to 2002, was principally from conversion of opened forest areas (275.97 ha). Changing conditions due to increasing forest fragmentation make land cover and change detection analysis an extremely important consideration for management, planning and inventory mapping ( Lu et al., 2004 ).
Bosomtwe Range Forest Reserve (BRFR) was considered for change detection because of the forest’s proximity to the ‘Bosomtwe’ Lake situated in the Ashanti Region of Ghana. It is one of six meteoritic global lakes. Also, the overlapping southern- most section of the Lake with northern part of the reserve creates a combination of forests, wetland and mountain ecosystem. Recently, the World Network of Biosphere Reserves under the Man and Biosphere Programme designated the Lake as a globally significant hotspot to reconcile the conservation of biodiversity with sustainable use of natural resources ( Abreu et al., 2016 ). Fur- thermore, Lake ‘Bosomtwe’ is rich in aquatic biodiversity of national and global significance, but has become particularly vul- nerable due to intense human pressures. Existing traditional norms and ethics have contributed significantly to ensuring sustainable fish production of the lake. However, confined contamination and the degradation of the vegetation in the catchment areas are desta- bilizing the veracity of the lake. Increasing demand for resources due to rising population as the situation currently is envisaged to have detrimental effects on the available resources ( Abreu et al., 2016 ) as well. Nevertheless, it is highly envisioned that sustainable management of the reserve could secondarily offer protection for the Lake and increase forest cover in the catchment areas.
The management of water resources in this study refers to the control of catchment water in the state of Kelantan. The quantity of water yield at a given time is important to identify the amount of water in a catchment area. The objective of this study is to obtain the quantities of water in the watershed of Kelantan. This study focuses on identifying 21 new water catchments based on Soil and Water Assessment Tool (SWAT) model and GIS location of Water Treatment Plant intake. The catchment area features the data relating to each catchment area such as the main river, land lot and area. Subsequently, a simple water balance model was used to obtain the total amount of water in each catchment area in June 2010. This model uses the rainfall and actual evapotranspiration derived from the Tropical Rain Measuring Mission (TRMM 2A25) satellite and the Landsat-5 TM satellite, respectively. The actual evapotranspiration was extracted using the False Color-Composite Model (FCC). This study shows the remotesensing-based water yield model is able to measure the amount of water in June 2010 for 20 catchment areas.
identify dynamic vegetation information . Compared with the traditional two-phase image classification, the image time-series analysis can improve the accuracy of information extraction . Huang et al.  used Landsat historical data to express the specific spectral index of a disturbance state to construct a time series and pro- posed the Landsat time-series stacks-vegetation change tracker algorithm (LTSS-VCT) for automatic mapping of forest disturbances. Kennedy et al.  proposed that the LandTrends time segmentation method was considered to be able to identify the mutation points and trends in the time series of the interannual time series. Verbesselt et al.  proposed the breaks for additive season and trend algorithm (BFAST) for real-time remotesensing of ecological disturbance information. For the highly dis- turbed coal mining area, Lei et al.  analyzed, with the MODIS-NDVI time series, the temporal and spatial evolution characteristics of vegetation under mining ac- tion in Shendong Mining area. With the deepening of research, many machine learning methods, such as C5.0 decision tree , random forest algorithm (RF) , support vector machine (SVM) [13, 14], and fuzzy c-means clustering , are used for extracting dynamic information of the time-series images. Due to big amount of image time-series data, the time spectrum in- formation is susceptible to noise interference, so that the data is subject to high uncertainty and is characterized in seasonality, space-time autocorrelation, etc. [16, 17], which limits the application of many time-series math- ematical models, and the effective mining and classifica- tion methods of image time-series information is still faced with many challenges [18, 19]. The extraction of existing time-series information is mostly based on the pixel time series , but further studies are required for the disturbance of vegetation phenological difference in different years, the statistical dependence of neighbor- hood pixel and category marking probability, the sensi- tivity of similarity measurement, and how to improve generalization ability of fuzzy classification.
Modelling approaches that have determined a parametric association between response and predictor variables have been successfully applied to the attribution of forest structure (Hudak et al., 2002; Wulder and Seemann, 2003; Pascual et al., 2010). However, in more recent years machine learning techniques have been utilised for remotesensing applications, where the complex statistical associations of multi-source datasets require more advanced approaches to characterise forests over large areas (McInerney et al., 2010; Mellor et al., 2013; Mora et al., 2013). A machine learning technique that has gained in popularity is random forest, an ensemble regression tree technique from the Classification And Regression Tree or CART family (Breiman, 2001). Random forest works by constructing “weak” regression trees (usually in the order of hundreds) from bootstrapped samples of input variables. The “weak” regression trees are then aggregated in an ensemble to produce a robust model that is insensitive to collinear predictor variables and a non-normal distributed response variable (Breiman, 1996). Furthermore, the ease of application (e.g. only two model parameters, see Section 2.3.1) and the ability to run efficiently over large datasets makes random forest an ideal choice for large area attribution (Rodriguez-Galiano et al., 2012b). A number of studies have utilised random forest for mapping forest attributes with remotely sensed data, including biomass (Baccini et al., 2008; Mascaro et al., 2014), species extent (Evans and Cushman, 2009), forest extent (Ørka et al., 2010; Mellor et al., 2012, 2015), canopy cover (Armston et al., 2009; Johansen et al., 2010; Ahmed et al., 2015) and canopy height (Kellndorfer et al., 2010; Stojanova et al., 2010; Simard et al., 2011; Cartus et al., 2012; Peterson and Nelson, 2014; Ahmed et al., 2015).
To account for the impacts of natural disturbances caused by insects and drought, the Canadian Forest Service of Natural Resources Canada has developed the National Forest Carbon Monitoring, Accounting and Reporting System (NFCMARS) (Kurz and Apps 2006). Outputs from NFCMARS will inform national policy makers and resource managers on the impacts of resource management, land-use change and disturbances on forest carbon stocks (Kurz et al. 2008). The system is providing data for annual reporting on greenhouse gas emissions provided to Environment Canada as part of Canada's report to the United Nations Framework Convention on Climate Change (Environment Canada 2007), Criteria and Indicators reporting, and to provide a framework for national level forest monitoring (Wulder et al. 2004).
Chapter 4 concentrates on the analysis of FINSAR airborne SAR cam- paign data. It demonstrates that tree height estimationbased on polari- metric interferometry using L-band, a technique which was developed for tropical and temperate forest, also works well in the boreal zone. Addi- tional analysis shows that X-band SAR interferometry can also be used for forest height retrieval. A careful comparison with LIDAR measured tree height reveals that X-band backscattering is arriving from the top third of the forest and the phase height correlates well with the canopy height. It is demonstrated that with the Random Volume over Ground model, the penetration depth can be calculated accurately and thus, when the ground model is available from other sources, forest height can be retrieved from X-band interferometric SAR images with a high accuracy. The same tech- nique can be applied to L-band SAR measurements in order to reduce the need for fully polarimetric measurement or to improve RVoG inversion performance. The forest height estimation scheme proposed in this work potentially enables tree height measurement with TanDEM-X, the only available existing spaceborne interferometric X-band SAR system. Forest height is related to forest biomass through allometry equations and there- fore tree height measurement from space would enable to produce more accurate global forest biomass maps. The method can give signiﬁcant im- provements to estimates, for example, in Finland, where a highly accurate ground model is available for the entire country free of charge. The work for validating the method with satellite data is already in progress.
We compared remotesensing daily evapotranspiration esti- mates with sap flow measurements upscaled to stand tran- spiration. Sap flow density in the outer xylem was mea- sured with 20 mm long heat dissipation probes constructed according to Granier (1985); 15-min averages of data col- lected every 10 seconds were stored in a datalogger (DT 500, DataTaker, Australia). Heat dissipation gauges were installed at breast height on the north-facing side of 12 Scots pine trees and were covered with reflective insulation to avoid the influ- ence of natural temperature gradients in the trunk. Sap flow density measured in the outer xylem was corrected for radial variability in sap flow using correction coefficients derived from radial patterns of sap flow within the xylem measured with a multi-point heat field deformation sensor (Nadezhd- ina et al., 2002). A gravimetric analysis of wood cores was carried out to estimate sapwood depths in a sample of Scots pine trees, and a linear regression was obtained between the basal area and sapwood area of individual trees. Stand tran- spiration was then calculated by multiplying the average sap flow density within a diametric class by the total sapwood area of trees in that class. Instantaneous values (15-min av- erages) were then summed to compute daily stand transpira- tion. Further details on the methodology and results of sap flow measurements used in this study can be found in Poy- atos et al. (2005, 2008).
Remotesensing hyperspectral sensors are important and powerful instruments for addressing classifica- tion problems in complex forest scenarios, as they allow one a detailed characterization of the spectral behavior of the considered information classes. However, the processing of hyperspectral data is particu- larly complex both from a theoretical viewpoint (e.g. problems related to the Hughes phenomenon ) and from a computational perspective. Despite many previous investigations have been presented in the literature on feature reduction and feature extraction in hyperspectral data, only a few studies have ana- lyzed the role of spectral resolution on the classification accuracy in different application domains. In this chapter, we present an empirical study aimed at understanding the relationship among spectral reso- lution, classifier complexity, and classification accuracy obtained with hyperspectral sensors for the clas- sification of forest areas. We considered two different test sets characterized by images acquired by an AISA Eagle sensor over 126 bands with a spectral resolution of 4.6 nm, and we subsequently degraded its spectral resolution to 9.2, 13.8, 18.4, 23, 27.6, 32.2 and 36.8 nm. A series of classification experiments were carried out with bands at each of the degraded spectral resolutions, and bands selected with a fea- ture selection algorithm at the highest spectral resolution (4.6 nm). The classification experiments were carried out with three different classifiers: Support Vector Machine, Gaussian Maximum Likelihood with Leave-One-Out-Covariance estimator, and Linear Discriminant Analysis. From the experimental results, important conclusions can be made about the choice of the spectral resolution of hyperspectral sensors as applied to forest areas, also in relation to the complexity of the adopted classification methodology. The outcome of these experiments are also applicable in terms of directing the user towards a more efficient use of the current instruments (e.g. programming of the spectral channels to be acquired) and classifica- tion techniques in forest applications, as well as in the design of future hyperspectral sensors.
While landcover may be observed directly in the field or by remotesensing, observations of its changes generally require the integration of natural and social scientific methods. This aids in determining which human activities are occurring in different parts of the landscape, even when landcover appears to be the same (Angelsen and Kaimowitz, 2001, Lambin et al., 2003). Also, the distribution of natural landcover types is related to the heterogeneity of environmental conditions such as temperature and moisture. The actual distribution is however, only occasionally a result of physical limitations and it is mainly due to competition between individual plants or between ecosystems as a whole. This is also due to anthropogenic alteration of land which disrupts the structure and functioning of ecosystems. Agriculture, forestry, and other land-management practices have modified entire landscapes and altered plant and animal communities of many ecosystems throughout the world. The most spatially and economically important human uses of land globally include cultivation in various forms, livestock grazing, settlement and construction, reserves and protected lands, and timber extraction (Turner et al., 1994; Verburg, 2000).
The collected information of these subcatchments was carefully investigated to select a suitable subcatchment for this study. Least or no streamflow regulation, appropriate catchment size (> 1000 km 2 ) and the availability of key ground measured data were considered in this selection process. These key data are streamflow (to be used for accuracy assessment) and air temperature (which is needed to estimate potential evapotranspiration). After considering the information mentioned above, both the Jemma and the Lake Tana subcatchments were selected as potential catchments for the investigation. According to the field officers of the Ministry of Water Resources (Ethiopia), the Jemma subcatchment is subject to flash floods during the rainy season and often many of the installed gauges have been washed away during those flash floods. It also appears that the available data are not accurate due to serious issues (absence of periodic maintenance and calibration) with the calibration of the streamflow gauges (Personal Communication, MoWR, 2011). Furthermore, the ground condition of the Jemma subcatchment is extremely rough, and accessibility is difficult because of its terrain conditions and poor infrastructure, which means that it would be difficult to collect ground-truth data for LULC classification. Therefore, the Jemma option was quickly eliminated because of the unfavorable issues involved with stream gauge data, which are needed to assess the accuracy of the estimated streamflow data with RS data.
The results of applying two different weighting functions for forest AGB estimation are shown in Figure 2. The simple 1 minus the spectral distance estimator worked best. In order to take advantage of weighting functions such as inversely proportional to the distance or 1 minus the distance, it is necessary to have an identifiable close neighbor and some distance between this and the next one. It was apparent that instead of having one spectrally close neighbor, in most cases, there were several close neighbors. Thus, for subsequent trials, Eq. (3) was used to allocate equal weights to neighbors.
Several applications of airborne LiDAR data in the monitoring and evaluation of existing forest roads have been demonstrated. Craven and Wing (2014) considered the influence of 4 different canopy condi- tions on the accuracy of estimation of road geometry based on LiDAR data, and showed mean vertical error of 0.28 m and horizontal error of 1.21 m, when consid- ered against existing road centrelines. Road slopes were estimated to within 1% and error in horizontal curve radii was estimated with an absolute error of 3.17 m. The follow up work by Beck et al. (2015) used varying intensity values and return densities in clas- sifying roads and demonstrated a high level of accu- racy in doing so. In other applications, LiDAR has been used in detecting, monitoring or extracting the geometry of existing roads to evaluate whether they meet certain specifications. For instance White et al. (2010) extracted alignment and gradient data from a mountain forest road, showing deviations of 1.5 m in position, 0.5% in slope and 0.2% in terms of length when compared with field survey data. A similar ap- proach applied by Azizi et al. (2014) resulted in more than 95% of the road length being classified within 1.3 m of the field surveyed normal. These develop- ments represent considerable time and effort savings in providing detailed road geometry, providing essen- tial complementary data to conventional field surveys. However, Krogstad and Scheiss (2004) list pitfalls of a blind adoption of these models including inconsistent data returns depending on canopy density and a re- sultant data smoothing that can provide a false basis for road design, as well as subsurface issues not re- flected in the topography.
available regarding the nature of the components before actual processing.
Fung (1990) reported on a comparative analysis of a multidate standardized PCA and a multidate tasseled cap transformation of a multitemporal Landsat TM data set. The PCA rotation was based on the merged 12-band TM image. Three components were found associated with change: PC3 with changes in soil brightness, PC5 with changes in NIR reflectance and thus vegetative vigor, and PC6 with changes in the contrast between the middle infrared bands and the PAR (photosynthetically active radiation) bands, thus changes in wetness. The usefulness of these lower-order multidate PC's to highlight localized change was confirmed by Lee et al. (1989). They ran a principal factor analysis on the transformed image and found the lower-order PC's to contain significantly more unique variance. The tasseled cap transformation was carried out as follows: The spectral bands of the two dates were assigned the tasseled cap coefficients as derived by Crist & Cicone (1984) with positive coefficients for the first date and negative coefficients for the second. The derived vectors were not orthogonal and consequently were subjected to a Gram-Schmidt transformation. The process generated output vectors that were effectively orthogonal to each other. Three change tasseled cap images were produced detailing differences in greenness, brightness, and wetness. The greenness change image gave the highest classification accuracy among all resulting PCA and tasseled cap change-related images. Fung (1990) clearly advocated the use of the tasseled cap transformation. Under this algorithm, the inherent data structure could be clearly depicted and the derived variables were physically based and independent of scene content. Twelve TM input bands could moreover be reduced to two, maximum three, significant change bands. Collins & Woodcock (1994) similarly applied the Gramm-Schmidt orthogonalization process to map forest mortality, however, defining their own transformation coefficients.
An ever-increasing demand for weather prediction and high climate modelling ac- curacy drives the need for better atmospheric data collection. These demands in- clude better spatial and temporal coverage of mainly humidity and temperature distributions in the atmosphere. A new type of remotesensing satellite technol- ogy is emerging, originating in the field of radio astronomy where telescope aper- ture upscaling could not keep up with the increasing demand for higher resolution. Aperture synthesis imaging takes an array of receivers and emulates apertures ex- tending way beyond what is possible with any single antenna. In the field of Earth remotesensing, the same idea could be used to construct satellites observing in the microwave region at a high resolution with foldable antenna arrays. If placed in a geostationary orbit, these could produce images with high temporal resolution, however, such altitudes make the resolution requirement and, hence, signal process- ing very demanding. The relentless development in miniaturization of integrated circuits has in recent years made the concept of high resolution aperture synthesis imaging aboard a satellite platform viable.
Prior forest composition commonly influences subsequent, post-fire, forest species composition and abundance (Frelich and Reich, 1995; Heinselman, 1996; Johnson et al., 2003). Post- fire tree seedling recruitment (especially quaking aspen, black spruce, and jack pine) is directly related to pre-fire abundance of the respective tree species (Frelich and Reich, 1995; Hein- selman, 1996; Johnson et al., 2003), though specific recovery patterns may be modified by competition with regenerating herbaceous cover (Frey et al., 2003). For the PCF area, pre-burn forest structure and composition data (Wolter and Townsend, 2011) indicated that mature aspen cover was prominent in the central, east, and southern regions prior to the PCF , while mature jack pine and black spruce were more dominant within the western region. This arrangement of pre-burn aspen abun- dance generally mirrored our modeled post-burn mapped esti- mates of aspen abundance and structure (Plate 2), which were produced both with and without the pre-burn aspen BA data as a predictive variable. Furthermore, gradients of abundance ob- served among modeled aspen structure results are consistent with a land use history (i.e., logging) that promoted deciduous species including aspen (Heinselman, 1996). Hence, it seems reasonable to suggest that aspen legacy governs aspen response and recovery following wildfire in this region. However, aspen legacy had no effect on vegetative richness.