Historical tropical successional forest cover mapped with Landsat MSS imagery

Maps of tropical successional forest cover of the 1970s and 1980s are needed for long-term modelling of tropical forest-cover change, carbon flux and habitat change. Landsat Multispectral Scanner System (MSS) imagery may provide a basis for such maps, but its capability in this respect is poorly unexplored if not discounted. This article examines how reliably single-date MSS imagery may distinguish tropical successional forest. Statistical and graphical analyses of 2043 MSS pixels of successional forest cover, pasture and mature forest cover of Central Panama indicate that successional forest may be accurately mapped, with a maximum-likelihood classification accuracy of 86–90%. Detectable successional cover is unlikely to be older than 10 years approximately. These findings indicate that MSS imagery may provide a new baseline for historical mapping and long-term modelling of tropical forest-cover change that, unlike that of Advanced Very High Resolution Radiometer (AVHRR) imagery used for this purpose, is amenable to fine-scale spatial analysis and reliable accuracy assessment.     

Regional-scale boreal forest cover and change mapping using Landsat data composites for European Russia

Boreal forests are a critical component of the global carbon cycle, and timely monitoring allows for assessing forest cover change and its impacts on carbon dynamics. Earth observation data sets are an important source of information that allow for systematic monitoring of the entire biome. Landsat imagery, provided free of charge by the USGS Center for Earth Resources Observation and Science (EROS) enable consistent and timely forest cover updates. However, irregular image acquisition within parts of the boreal biome coupled with an absence of atmospherically corrected data hamper regional-scale monitoring efforts using Landsat imagery. A method of boreal forest cover and change mapping using Landsat imagery has been developed and tested within European Russia between circa year 2000 and 2005. The approach employs a multi-year compositing methodology adapted for incomplete annual data availability, within-region variation in growing season length and frequent cloud cover. Relative radiometric normalization and cloud/shadow data screening algorithms were employed to create seamless image composites with remaining cloud/shadow contamination of less than 0.5% of the total composite area. Supervised classification tree algorithms were applied to the time-sequential image composites to characterize forest cover and gross forest loss over the study period. Forest cover results when compared to independently-derived samples of Landsat data have high agreement (overall accuracy of 89%, Kappa of 0.78), and conform with official forest cover statistics of the Russian government. Gross forest cover loss regional-scale mapping results are comparable with individual Landsat image pair change detection (overall accuracy of 98%, Kappa of 0.71). The gross forest cover loss within European Russia 2000-2005 is estimated to be 2210 thousand hectares, and constitutes a 1.5% reduction of year 2000 forest cover. At the regional scale, the highest proportional forest cover loss is estimated for the most populated regions (Leningradskaya and Moskovskaya Oblast). Our results highlight the forest cover depletion around large industrial cities as the hotspot of forest cover change in European Russia.     

Analysis of MODIS imagery for detection of clear cuts in the boreal forest in north-west Russia

This study examines the feasibility of using MODIS images (MOD02 products) for the detection and monitoring of forest clear cuts in the boreal forest in north-west Russia. The proposed approach combines three change detection methods, including Change Vector Analysis, Textural Analysis using the coefficient of variation, and Constrained Energy Minimization analysis. For each individual method a series of thresholds was tested in order to obtain an optimal identification of clear cuts. A clear cut detection was only accepted if the change was detected by each individual method. All input parameters needed were derived from a set of reference clear cuts, mapped from 30 m resolution Landsat ETM+ imagery and used also for accuracy assessment. Change assessment was tested with MODIS images of two and of three acquisition dates. Referring to two test sites (Karelia, Komi) the detection omission and commission errors, assessed within a 3 × 3 pixels moving kernel, were at 23% and 8%, and at 21% and 17%, respectively. In terms of detectable clear cut size, a detection accuracy of about 90% can be expected for clear cuts in the size category above 15 ha, which contains the majority of cuts in the region. MODIS therefore provides good capabilities for large scale monitoring of major clear cut activities in the boreal forests of north-western Russia.     

Mapping crop cover using multi-temporal Landsat 8 OLI imagery

Crop classification maps are useful for estimating amounts of crops harvested, which could help address challenges in food security. Remote-sensing techniques are useful tools for generating crop maps. Optical remote sensing is one of the most attractive options because it offers vegetation indices (VIs) with frequent revisits and has adequate spatial and spectral resolution and some data has been distributed free of charge. However, sufficient consideration has not been given to the potential of VIs calculated from Landsat 8 Operational Land Imager (OLI) data. This article describes the use of Landsat 8 OLI data for the classification of crops in Hokkaido, Japan. In addition to reflectance, VIs calculated from simple formulas that consisted of combinations of two or more reflectance wavebands were evaluated, as well as the six components of the Kauth–Thomas transform. The VIs based on shortwave infrared bands (bands 6 or 7) improved classification accuracy, and using a combination of all derived data from Landsat 8 OLI data resulted in an overall accuracy of 94.5% (allocation disagreement = 4.492 and quantity disagreement = 1.017).     

Estimating stand structural details using nearest neighbor analyses to link ground data, forest cover maps, and Landsat imagery

Forest information over a landscape is often represented as a spatial mosaic of polygons, separated by differences in species composition, height, age, crown closure, productivity, and other variables. These polygons are commonly delineated on medium-scale photography (e.g., 1:15,000) by a photo-interpreter familiar with the inventory area, and displayed and stored in a Geographic Information System (GIS) layer as a forest cover map. Forest cover maps are used for multiple purposes including timber and habitat supply analyses, and carbon inventories, at a regional or management unit level, and for parks planning, operational planning, and selection of stands for many purposes at a local level. Attribute data for each polygon commonly include the variables used to delineate the polygon, and other variables that can be measured or estimated using these medium-scale photographs. Additional measures that can only be obtained via expensive ground measures or possibly on high resolution photographs (e.g., volume per unit area, biomass components per unit area, tree-list of species and diameters) are available only for a sample of polygons, or may have been gathered independently using a sample survey over the land area. Improved linkages over a variety of data sources may help to support landscape level analyses. This study presents an approach to combine information from a systematic (grid) ground survey, forest cover (polygon) data, and Landsat Thematic Mapper (TM) imagery using variable-space nearest neighbor methods to estimate (i) mean ground-measured attributes for each polygon, in particular, volume per ha (m³/ha), stems per ha, and quadratic mean diameter for each polygon; and (ii) variation of these ground attributes within polygons. The approach was initially evaluated using Monte Carlo simulations with known measures of these attributes. Nearest neighbor methods were then applied to an approximate 5000 ha area (about 1000 polygons) of high productivity, mountainous forests located near the Pacific Coast of British Columbia, Canada. Based on the simulation results, the use of Landsat pixel reflectances to estimate volume per ha, average tree size (i.e., quadratic mean diameter), and stems per ha did not show great promise in improving estimates for each polygon over using forest cover data alone. However, in application, the use of remotely sensed data provided estimates of within-polygon variability. At the same time, the estimated means of these three imputed variables over the entire study area were very similar to the representative sample estimates using the ground data only. Extensions to other variables such as ranges of diameters and numbers of snags may also be possible providing useful data for habitat and forest growth analysis.     

Characterizing boreal forest wildfire with multi-temporal Landsat and LIDAR data

Wildfire is an important disturbance agent in Canada's boreal forest. Optical remotely sensed imagery (e.g., Landsat TM/ETM+), is well suited for capturing horizontally distributed forest conditions, structure, and change, while Light Detection and Ranging (LIDAR) data are more appropriate for capturing vertically distributed elements of forest structure and change. The integration of optical remotely sensed imagery and LIDAR data provides improved opportunities to characterize post-fire conditions. The objective of this study is to compare changes in forest structure, as measured with a discrete return profiling LIDAR, to post-fire conditions, as measured with remotely sensed data. Our research is focused on a boreal forest fire that occurred in May 2002 in Alberta, Canada. The Normalized Burn Ratio (NBR), the differenced NBR (dNBR), and the relative dNBR (RdNBR) were calculated from two dates of Landsat data (August 2001 and September 2002). Forest structural attributes were derived from two spatially coincident discrete return LIDAR profiles acquired in September 1997 and 2002 respectively. Image segmentation was used to produce homogeneous spatial patches analogous to forest stands, with analysis conducted at this patch level.In this study area, which was relatively homogenous and dominated by open forest, no statistically significant relationships were found between pre-fire forest structure and post-fire conditions (r < 0.5; p > 0.05). Post-fire forest structure and absolute and relative changes in forest structure were strongly correlated to post-fire conditions (r ranging from − 0.507 to 0.712; p < 0.0001). Measures of vegetation fill (VF) (LIDAR capture of cross-sectional vegetation amount), post-fire and absolute change in crown closure (CC), and relative change in average canopy height, were most useful for characterizing post-fire conditions. Forest structural attributes generated from the post-fire LIDAR data were most strongly correlated to post-fire NBR, while dNBR and RdNBR had stronger correlations with absolute and relative changes in the forest structural attributes. Absolute and relative changes in VF and changes in CC had the strongest positive correlations with respect to dNBR and RdNBR, ranging from 0.514 to 0.715 (p < 0.05). Measures of average inter-tree distance and volume were not strongly correlated to post-fire NBR, dNBR, or RdNBR. No marked differences were found in the strength or significance of correlations between post-fire structure and the post-fire NBR, dNBR, RdNBR, indicating that for the conditions present in this study area all three burn severity indices captured post-fire conditions in a similar manner. Finally, the relationship between post-fire forest structure and post-fire condition was strongest for dense forests (> 60% crown closure) compared to open (26-60%) and sparse forests (10-25%). Forest structure information provided by LIDAR is useful for characterizing post-fire conditions and burn induced structural change, and will complement other attributes such as vegetation type and moisture, topography, and long-term weather patterns, all of which will also influence variations in post-fire conditions.     

Crop and land cover classification in Iran using Landsat 7 imagery

Remote sensing provides one way of obtaining more accurate information on total cropped area and crop types in irrigated areas. The technique is particularly well suited to arid and semi‐arid areas where almost all vegetative growth is associated with irrigation. In order to obtain more information with regard to crop patterns in the irrigated areas in the Zayandeh Rud basin, a classification analysis was made of the Landsat 7 image of 2 July 2000. The target of the classification was to primarily focus on the agricultural land use. The date of the image fell in the transition period where the first crops were harvested and many fields were being prepared for the second crop. The image has therefore captured an instantaneous picture of a system generally in transition from the first to the second crop, but with significant differences from system to system, both with respect to crop types and agricultural cycles. The overall accuracy of image registration was about 30 m (one pixel). Fieldwork was conducted on various occasions in August–October 2000 and May–October 2001. Farmers were interviewed to determine the situation on 2 July 2000. Fields were mapped in detail with the GPS instruments, and data compiled for 112 fields. Using a supervised classification system, training areas were selected and initial classifications were made to determine the validity of the classes. After merging several classes and testing several new classes a final classification system was made. All seven Landsat bands were used in the determination of the feature statistics. The final classification was made with the minimum distance algorithm. The statistics with respect to areas and crop type for the districts was obtained by crossing the raster map with the irrigation district raster map. The results with respect to crop type and total irrigated area per district were compared with those of previous studies. This included both NOAA/AVHRR and conventional agricultural district statistics.     

An improved high spatial and temporal data fusion approach for combining Landsat and MODIS data to generate daily synthetic Landsat imagery

Because of low temporal resolution and cloud influence, many remote-sensing applications lack high spatial resolution remote-sensing data. To address this problem, this study introduced an improved spatial and temporal data fusion approach (ISTDFA) to generate daily synthetic Landsat imagery. This algorithm was designed to avoid the weaknesses of the spatial and temporal data fusion approach (STDFA) method, including the sensor difference and spatial variability. A weighted linear mixed model was used to adjust the spatial variability of surface reflectance. A linear-regression method was used to remove the influence of differences in sensor systems. This method was tested and validated in three study areas located in Xinjiang and Anhui province, China. The other two methods, the STDFA and the Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), were also applied and compared in those three study areas. The results showed that the ISTDFA algorithm can generate daily synthetic Landsat imagery accurately, with correlation coefficient r equal to 0.9857 and root mean square error (RMSE) equal to 0.0195, which is superior to the STDFA method. The ISTDFA method had higher accuracy than ESTARFM in areas greater than 200 × 200 MODIS pixels while the ESTARFM method had higher accuracy than the ISTDFA method in small areas. The correlation coefficient r had a negative power relation with ratio of land-cover change pixels. A land-cover change of 20.25% pixels can lead to a reduced correlation coefficient r of 0.295 in the blue band. The accuracy of the ISTDFA method indicated a logarithmic relationship with the size of the applied area, so it is recommended for use in large-scale areas.     

Combining Landsat images with historic records to estimate the live coral cover of Luhuitou fringing reef in Northern South China Sea

As an efficient indicator of coral reef health, live coral cover (LCC) is regularly surveyed and recorded by many coral reef documents. However, there usually exist some blanks for the historic records, while current in-field surveys are impossible to fill the blanks. To overcome such difficulties, we focus on exploiting the potential of optical satellite images. The purpose is to fill the blanks of the records over the past and further estimate the LCC in future. As historic records were usually lack of accurate geographical locations to match to the satellite images, a spectral index was defined based on the mean of the subsurface remote sensing reflectance. The index was then used to link the LCC with the satellite images by a cubic polynomial function. Thereafter, the LCC and the coefficients of the polynomial function were finally estimated by simultaneously combining the mean subsurface remote sensing reflectance, the historic LCC records, and the constraints among LCC in adjacent years. Experiments on a series of Landsat images of Luhuitou fringing reef (1973 to 2018) demonstrated that the proposed method is effective and feasible, where the introduction of the satellite images can greatly improve the accuracy. The Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Relative Errors (MRE) of the LCC were able to reach 5.4%, 4.0%, and 15.9% respectively. This is regarded as the first test on LCC estimation by combining such a long-term LCC records with a series of satellite images.     

Contextual classification of Landsat TM images to forest inventory cover types

In the context of Landsat TM images forest stands are a cluster of homogeneous pixels. Contextual classification of forest cover types exploits relationships between neighbouring pixels in the pursuit of an increase in classification accuracy. Results with six contextual classifiers from two sites in Canada were compared to results with a maximum likelihood (ML) classifier. The comparisons were done at three levels of spectral class separation. Training and validation data were obtained from single-stage cluster sampling of 2?km×2?km primary sampling units (PSU) located on a 20?km×20?km grid. A strong relationship between contextual and ML classification accuracy was explored with logistic regression analysis. Effects of contextual classification were predicted for given levels of ML accuracy. Estimates of the spatial autocorrelation of reflectance values within a PSU were deemed consistent with a first-order autoregressive process. Iterative Conditional Modes (ICM) was the best contextual method; it improved the overall accuracy by four to six percentage points (statistically significant) when ML accuracy was between 50% and 80%. A relaxed ICM and a smoothing algorithm were second and third best. Contextual classification is most promising when an ML accuracy is around 70%. ICM results were sensitive to the level of spatial autocorrelation of ML classification errors and to the homogeneity of a PSU.     

A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin

In this paper we demonstrate a new approach that uses regional/continental MODIS (MODerate Resolution Imaging Spectroradiometer) derived forest cover products to calibrate Landsat data for exhaustive high spatial resolution mapping of forest cover and clearing in the Congo River Basin. The approach employs multi-temporal Landsat acquisitions to account for cloud cover, a primary limiting factor in humid tropical forest mapping. A Basin-wide MODIS 250 m Vegetation Continuous Field (VCF) percent tree cover product is used as a regionally consistent reference data set to train Landsat imagery. The approach is automated and greatly shortens mapping time. Results for approximately one third of the Congo Basin are shown. Derived high spatial resolution forest change estimates indicate that less than 1% of the forests were cleared from 1990 to 2000. However, forest clearing is spatially pervasive and fragmented in the landscapes studied to date, with implications for sustaining the region's biodiversity. The forest cover and change data are being used by the Central African Regional Program for the Environment (CARPE) program to study deforestation and biodiversity loss in the Congo Basin forest zone. Data from this study are available at http://carpe.umd.edu.     

Comparing annual MODIS and PRODES forest cover change data for advancing monitoring of Brazilian forest cover

Annual forest cover loss indicator maps for the humid tropics from 2000 to 2005 derived from time-series 500 m data from the MODerate Resolution Imaging Spectroradiometer (MODIS) were compared with annual deforestation data from the PRODES (Amazon Deforestation Monitoring Project) data set produced by the Brazilian National Institute for Space Research (INPE). The annual PRODES data were used to calibrate the MODIS annual change indicator data in estimating forest loss for Brazil. Results indicate that MODIS data may be useful in providing a first estimate of national forest cover change on an annual basis for Brazil. When directly compared with PRODES change at the MODIS grid scale for all years of the analysis, MODIS change indicator maps accounted for 75% of the PRODES change. This ratio was used to scale the MODIS change indicators to the PRODES area estimates. A sliding threshold of percent PRODES forest and 2000 to 2005 deforestation classes per MODIS grid cell was used to match the scaled MODIS to the official PRODES change estimates, and then to differentiate MODIS change within various sub-areas of the PRODES analysis. Results indicate significant change outside of the PRODES-defined intact forest class. Total scaled MODIS change area within the PRODES historical deforestation and forest area of study is 120% of the official PRODES estimate. Results emphasize the importance of synoptic monitoring of all forest change dynamics, including the cover dynamics of intact humid forest, regrowth, plantations, and cerrado tree cover assemblages. Results also indicate that operational MODIS-only forest cover loss algorithms may be useful in providing near-real time areal estimates of annual change within the Amazon Basin.     

Classification of boreal forest cover types using SAR images

Mapping forest cover types in the boreal ecosystem is important for understanding the processes governing the interaction of the surface with the atmosphere. In this paper, we report the results of the land-cover classification of the SAR (synthetic aperture radar) data acquired during the Boreal Ecosystem Atmospheric Study's intensive field campaigns over the southern study area near Prince Albert, Canada. A Bayesian maximum a posteriori classifier was applied on the national Aeronautics and Space Administration/Jet Propulsion Laboratory airborne SAR images covering the region during the peak of the growing season in July 1994. The approach is supervised in the sense that a combination of field data and existing land-cover maps are used to develop training areas for the desired classes. The images acquired were first radiometrically and absolutely calibrated, the incidence angle effect in airborne images was corrected to an acceptable accuracy, and the images were used in a mosaic form and geocoded and georeferenced with an existing land-cover map for validation purposes. The results show that SAR images can be classified into dominant forest types such as jack pine, black spruce, trembling aspen, clearing, open water, and three categories of mixed strands with better than 90% accuracy. The unispecies stands such as jack pine and black spruce are separated with 98% accuracy, but the accuracy of mixed coniferous and deciduous stands suffers from confusing factors such as varying species composition, surface moisture, and understory effects. To satisfy the requirements of process models, the number of cover types was reduced from eight to five general classes of conifer wet, conifer dry, mixed deciduous, disturbed, and open water. Reduction of classes improved the overall accuracy of the classification over the entire region from 77% to 92%.     

Predicting forest successional stages using multitemporal Landsat imagery with forest inventory and analysis data

Forest succession is an important ecological process that has profound biophysical, biological and biogeochemical implications in terrestrial ecosystems. Therefore, information on forest successional stages over an extensive forested landscape is crucial for us to understand ecosystem processes, such as carbon assimilation and energy interception. This study explored the potential of using Forest Inventory and Analysis (FIA) plot data to extract forest successional stage information from remotely sensed imagery with three widely used predictive models, linear regression (LR), decision trees (DTs) and neural networks (NNs). The predictive results in this study agree with previous findings that multitemporal Landsat Thematic Mapper (TM) imagery can improve the accuracy of forest successional stage prediction compared to models using a single image. Because of the overlap of spectral signatures of forests in different successional stages, it is difficult to accurately separate forest successional stages into more than three broad age classes (young, mature and old) with reasonable accuracy based on the age information of FIA plots and the spectral data of the plots from Landsat TM imagery. Given the mixed spectral response of forest age classes, new approaches need to be explored to improve the prediction of forest successional stages using FIA data.     

Land cover assessment with MODIS imagery in southern African Miombo ecosystems

Global land use and land cover products in highly dynamic tropical ecosystems lack the detail needed for natural resource management and monitoring at the national and provincial level. The MODIS sensor provides improved opportunities to combine multispectral and multitemporal data for land use and land cover mapping. In this paper we compare the MODIS Global Land Cover Classification Product with recent land use and land cover maps at the national level over a characteristic location of Miombo woodlands in the province of Zambezia, Mozambique. The performances of three land cover-mapping approaches were assessed: single-date supervised classification, principal component analysis of band-pair difference images, and multitemporal NDVI analysis. Extensive recent field data were used for the definition of the test sites and accuracy assessment. Encouraging results were achieved with the three approaches. The classification results were refined with the help of a digital elevation model. The most consistent results were achieved using principal component analysis of band-pair difference images. This method provided the most accurate classifications for agriculture, wetlands, grasslands, thicket and open forest. The overall classification accuracy reached 90%. The multitemporal NDVI provided a more accurate classification for the dense forest cover class. The selection of the right image dates proved to be critical for all the cases evaluated. The flexibility of these alternatives makes them promising options for rapid and inexpensive land cover mapping in regions of high environmental variability such as tropical developing countries.     

Mapping forest vegetation using Landsat TM imagery and a canopy reflectance model

Estimates of mean tree size and cover for each forest stand from an invertible forest canopy reflectance model are part of a new forest vegetation mapping system. Image segmentation defines stands which are sorted into general growth forms using per-pixel image classifications. Ecological models based on terrain relations predict species associations for the conifer, hardwood, and brush growth forms. The combination of the model-based estimates of tree size and cover with species associations yields general-purpose vegetation maps useful for a variety of land management needs. Results of timber inventories in the Tahoe and Stanislaus National Forests indicate the vegetation maps form a useful basis for stratification. Patterns in timber volumes for the strata reveal that the cover estimates are more reliable than the tree size estimates. A map accuracy assessment of the Stanislaus National Forest shows high overall map accuracy and also illustrates the problems in estimating tree size.     

Partial blind model inversion of mountain forest structure from MODIS imagery

Blind model inversion of forest structure allows the user to run powerful physically based canopy reflectance models (CRMs) without having to specify any input model parameters as these are instead automatically derived. This is particularly important for large areas and regional scales where obtaining these model parameters may be costly, impractical, not representative or impossible. This is especially challenging in high-relief mountainous terrain. This article presents the multiple-forward mode-partial blind (MFM-PB) inversion capability as an important advancement from MFM-User and MFM adaptive full-blind (AFB) processing in that MFM-PB permits all available user input data to be utilized, while facilitating PB analyses for model parameters that are missing, a more typical operational-level requirement. MFM-PB was compared with MFM-User analysis of Moderate Resolution Imaging Spectroradiometer (MODIS) data in the Canadian Rocky Mountains and was shown to be comparable in terms of both generated inputs and all biophysical structural outputs, with differences for stand density of ±42 stems/ha, crown radii ±0.08 m, height to crown centre (HCC) ±0.10 m and tree height (HGT) ±0.37 m. These mountain results were further compared with MFM results from flat, boreal forest terrain and were found to be comparable. MFM-PB provides full flexibility for CRM inversion, and is particularly important for (but not limited to) larger area, regional-scale studies for which user input data are typically constrained.     

Assessment of tropical forest degradation by selective logging and fire using Landsat imagery

Many studies have assessed the process of forest degradation in the Brazilian Amazon using remote sensing approaches to estimate the extent and impact by selective logging and forest fires on tropical rain forest. However, only a few have estimated the combined impacts of those anthropogenic activities. We conducted a detailed analysis of selective logging and forest fire impacts on natural forests in the southern Brazilian Amazon state of Mato Grosso, one of the key logging centers in the country. To achieve this goal a 13-year series of annual Landsat images (1992-2004) was used to test different remote sensing techniques for measuring the extent of selective logging and forest fires, and to estimate their impact and interaction with other land use types occurring in the study region. Forest canopy regeneration following these disturbances was also assessed. Field measurements and visual observations were conducted to validate remote sensing techniques. Our results indicated that the Modified Soil Adjusted Vegetation Index aerosol free (MSAVI_af) is a reliable estimator of fractional coverage under both clear sky and under smoky conditions in this study region. During the period of analysis, selective logging was responsible for disturbing the largest proportion (31%) of natural forest in the study area, immediately followed by deforestation (29%). Altogether, forest disturbances by selective logging and forest fires affected approximately 40% of the study site area. Once disturbed by selective logging activities, forests became more susceptible to fire in the study site. However, our results showed that fires may also occur in undisturbed forests. This indicates that there are further factors that may increase forest fire susceptibility in the study area. Those factors need to be better understood. Although selective logging affected the largest amount of natural forest in the study period, 35% and 28% of the observed losses of forest canopy cover were due to forest fire and selective logging combined and to forest fire only, respectively. Moreover, forest areas degraded by selective logging and forest fire is an addition to outright deforestation estimates and has yet to be accounted for by land use and land cover change assessments in tropical regions. Assuming that this observed trend of land use and land cover conversion continues, we predict that there will be no undisturbed forests remaining by 2011 in this study site. Finally, we estimated that 70% of the total forest area disturbed by logging and fire had sufficiently recovered to become undetectable using satellite data in 2004.     

Predicting categorical forest variables using an improved k-Nearest Neighbour estimator and Landsat imagery

The k-Nearest Neighbour (k-NN) estimation and prediction technique is widely used to produce pixel-level predictions and areal estimates of continuous forest variables such as area and volume, often by sub-categories such as species. An advantage of k-NN is that the same parameters (e.g., k-value, distance metric, weight vector for the feature space variables) can be used for all variables, whether continuous or categorical. An obvious question is the degree to which accuracy can be improved if the k-NN estimation parameters are tailored for specific variable groups such as volumes by tree species or categorical variables. We investigated prediction of categorical forest attribute variables from satellite image spectral data using k-NN with optimisation of the weight vector for the ancillary variables obtained using a genetic algorithm. We tested several genetic algorithm fitness functions, all derived from well-known accuracy measures. For a Finnish test site, the categorical forest attribute variables were site fertility and tree species dominance, and for an Italian test site, the variables were forest type and conifer/broad-leaved dominance. The results for both test sites were validated using independent data sets. Our results indicate that use of the genetic algorithm to optimize the weight vector for prediction of a single forest attribute variable had a slight positive effect on the prediction accuracies for other variables. Errors can be further decreased if the optimisation is done by variable groups.