Janne Heiskanen

Estimating aboveground tree biomass and leaf area index in a mountain birch forest using ASTER satellite data

Biomass and leaf area index (LAI) are important variables in many ecological and environmental applications. In this study, the suitability of visible to shortwave infrared advanced spaceborne thermal emission and reflection radiometer (ASTER) data for estimating aboveground tree and LAI in the treeline mountain birch forests was tested in northernmost Finland. The biomass and LAI of the 128 plots were surveyed, and the empirical relationships between forest variables and ASTER data were studied using correlation analysis and linear and non‐linear regression analysis. The studied spectral features also included several spectral vegetation indices (SVI) and canonical correlation analysis (CCA) transformed reflectances. The results indicate significant relationships between the biomass, LAI and ASTER data. The variables were predicted most accurately by CCA transformed reflectances, the approach corresponding to the multiple regression analysis. The lowest RMSEs were 3.45 t ha−1 (41.0%) and 0.28 m2m−2 (37.0%) for biomass and LAI respectively. The red band was the band with the strongest correlation against the biomass and LAI. SR and NDVI were the SVIs with the strongest linear and non‐linear relationships. Although the best models explained about 85% of the variation in biomass and LAI, the undergrowth vegetation and background reflectance are likely to affect the observed relationships.

Retrieval of boreal forest LAI using a forest reflectance model and empirical regressions

Abstract Spectral invariants provide a novel approach for characterizing canopy structure in forest reflectance models and for mapping biophysical variables using satellite images. We applied a photon recollision probability (p) based forest reflectance model (PARAS) to retrieve leaf area index (LAI) from fine resolution SPOT HRVIR and Landsat ETM+ satellite data. First, PARAS was parameterized using an extensive database of LAI-2000 measurements from five conifer-dominated boreal forest sites in Finland, and mixtures of field-measured forest understory spectra. The selected vegetation indices (e.g. reduced simple ratio, RSR), neural networks and kNN method were used to retrieve effective LAI (Le) based on reflectance model simulations. For comparison, we established empirical vegetation index-LAI regression models for our study sites. The empirical RSR–Le regression performed best when applied to an independent test site in southern Finland [RMSE 0.57 (24.2%)]. However, the difference to the best reflectance model based retrievals produced by neural networks was only marginal [RMSE 0.59 (25.1%)]. According to this study, the PARAS model provides a simple and flexible modelling tool for calibrating algorithms for LAI retrieval in conifer-dominated boreal forests. The advantage of PARAS is that it directly uses field measurements to parameterize canopy structure (LAI-2000, hemispherical photographs) and optical properties of foliage and understory.

Evaluation of global land cover data sets over the tundra-taiga transition zone in northernmost Finland

The remote sensing‐based continental to global scale land cover data sets provide several land cover depictions over the circumpolar tundra–taiga transition zone. The aim of this study was to evaluate three data sets in northernmost Finland: the Global Land Cover 2000 Northern Eurasia map (GLC2000‐NE), the MODIS global land cover map (MODIS‐IGBP) and the tree cover layer of the MODIS vegetation continuous fields product (MODIS‐VCF). The data sets were first compared both visually and statistically to biotope inventory data including tree cover, height, species composition and shrub cover information as continuous variables. The agreement with reference data was poor because the classifications do not correspond to the class descriptions. The MODIS‐VCF tree cover overestimates the tree cover in the low values and underestimates it in the high values. The agreement was relatively good when the global data sets were aggregated to a forest–non‐forest level and compared to the Finnish CORINE Land Cover 2000 map over a larger area. However, the inaccurate mapping of the deciduous broadleaf forests and mires reduced the agreement at the forest–non‐forest level. The vegetation transitions are difficult to map using low‐resolution satellite data and further improvements to the land cover characterization over the tundra–taiga transition zone are required.

Can MODIS EVI monitor ecosystem productivity in the Amazon rainforest

The enhanced vegetation index (EVI) obtained from satellite imagery has often been used as a proxy of vegetation functioning and productivity in the Amazon rainforest. However, recent studies indicate that EVI patterns are strongly affected by satellite data artifacts. Hence, it is unclear if EVI is sensitive to subtle seasonal variations in evergreen Amazon forest productivity. This study analyzes 12 years of Moderate Resolution Imaging Spectroradiometer (MODIS) EVI in order to evaluate its response to factors driving productivity in the Amazon. We show that, after removing cloud and aerosol contamination, and correcting bidirectional reflectance distribution function effects, radiation and rainfall extremes show no influence on EVI anomalies. However, EVI seasonal patterns are still evident after accounting for Sun-sensor geometry effects. This remaining pattern cannot be explained by solar radiation or rainfall, but it is significantly correlated to gross primary production (GPP). Nevertheless, we argue that the causality between GPP and EVI should be interpreted with caution.

Classification of crops across heterogeneous agricultural landscape in Kenya using AisaEAGLE imaging spectroscopy data

Abstract Land use practices are changing at a fast pace in the tropics. In sub-Saharan Africa forests, woodlands and bushlands are being transformed for agricultural use to produce food for the rapidly growing population. The objective of this study was to assess the prospects of mapping the common agricultural crops in highly heterogeneous study area in south-eastern Kenya using high spatial and spectral resolution AisaEAGLE imaging spectroscopy data. Minimum noise fraction transformation was used to pack the coherent information in smaller set of bands and the data was classified with support vector machine (SVM) algorithm. A total of 35 plant species were mapped in the field and seven most dominant ones were used as classification targets. Five of the targets were agricultural crops. The overall accuracy (OA) for the classification was 90.8%. To assess the possibility of excluding the remaining 28 plant species from the classification results, 10 different probability thresholds (PT) were tried with SVM. The impact of PT was assessed with validation polygons of all 35 mapped plant species. The results showed that while PT was increased more pixels were excluded from non-target polygons than from the polygons of the seven classification targets. This increased the OA and reduced salt-and-pepper effects in the classification results. Very high spatial resolution imagery and pixel-based classification approach worked well with small targets such as maize while there was mixing of classes on the sides of the tree crowns.

Ecological applications of physically based remote sensing methods

Abstract Global monitoring of vegetation using optical remote sensing has undergone rapid technological and methodological development during the past decade. Physically based methods generally apply reflectance models for interpreting remotely sensed data sets. These methods have become increasingly important in the assessment of terrestrial variables from satellite-borne and airborne images. Products based on satellite images currently include various ecological variables that are needed for monitoring changes in forest cover, structure and functioning, including biophysical variables such as the amount of photosynthesizing leaf area. This paper reviews variables and global products estimated from optical satellite sensors describing, for example, the amount and functioning of green biomass and forest carbon exchange. Continuous validation work as new vegetation products are released continues to be important. More emphasis is needed on the collection of field data equivalent to satellite retrievals, data harmonization and continuous measurements of seasonal forest dynamics.

Seasonal Contribution of Understory Vegetation to the Reflectance of a Boreal Landscape at Different Spatial Scales

The composition of understory vegetation plays a central role in satellite-based estimation of biophysical properties of the tree layer in managed boreal forests. In this letter, we assess the contribution of understory vegetation to the reflectance of a managed boreal forest area throughout a growing season both at stand and landscape levels. We use a time series of SPOT and Hyperion, and MODIS satellite images, and a concurrent set of ground reference. Our results show that the contribution of understory to stand or landscape reflectance depends on tree canopy gap fractions and understory reflectance spectra, and their seasonal development. The reflected signal from the understory vegetation can account for more than 40% of boreal forest reflectance at stand level and for 20% at landscape level.

Modelling lidar-derived boreal forest canopy cover with SPOT 4 HRVIR data

Forest canopy cover (C) is needed in forest area monitoring and for many ecological models. Airborne scanning lidar sensors can produce fairly accurate C estimates even without field training data. However, optical satellite images are more cost-efficient for large area inventories. Our objective was to use airborne lidar data to obtain accurate estimates of C for a set of sample plots in a boreal forest and to generalize C for a large area using a satellite image. The normalized difference vegetation index (NDVI) and reduced simple ratio (RSR) were calculated from the satellite image and used as predictors in the regressions. RSR, which combines information from the red, near-infrared, and shortwave infrared bands, provided the best performance in terms of absolute root mean square error (RMSE) (7.3%) in the training data. NDVI produced a markedly larger RMSE (10.0%). However, in an independent validation data set, RMSE increased (13.0–17.1%) because the systematic sample of validation plots contained more variation than the training plots. Our results are better than those reported earlier, which is probably explained by more consistent C estimates derived from the lidar. Our approach provides an efficient method for creating C maps for large areas.

Land Cover Characterization in West Sudanian Savannas Using Seasonal Features from Annual Landsat Time Series

With the increasing temporal resolution of medium spatial resolution data, seasonal features are becoming more readily available for land cover characterization. However, in the tropical regions, images can be severely contaminated by clouds during the rainy season and fires during the dry season, with possible effects to seasonal features. In this study, we evaluated the performance of seasonal features based on an annual Landsat time series (LTS) of 35 images for land cover characterization in West Sudanian savanna woodlands. First, the burnt areas were detected and removed. Second, the reflectance seasonality was modelled using a harmonic model, and model parameters were used as inputs for land cover classification and tree crown cover prediction using the random forest algorithm. Furthermore, to study the sensitivity of the approach to the burnt areas, we repeated the analyses without the first step. Our results showed that seasonal features improved classification accuracy significantly from 68.7% and 66.1% to 76.2%, and decreased root mean square error (RMSE) of tree crown cover predictions from 11.7% and 11.4% to 10.4%, in comparison to the dry and rainy season single date images, respectively. The burnt areas biased the seasonal parameters in near-infrared and shortwave infrared bands, and decreased the accuracy of classification and tree crown cover prediction, suggesting that burnt areas should be removed before fitting the harmonic model. We conclude that seasonal features from annual LTS improved land cover characterization performance, and the harmonic model, provided a simple method for computing annual seasonal features with burnt area removal.

Use of airborne lidar for estimating canopy gap fraction and leaf area index of tropical montane forests

Leaf area index (LAI) is one descriptor of forest canopy structure and can be linked to vegetation productivity, carbon cycling, and several other ecosystem services. Airborne lidar (light detection and ranging) provides proxies of canopy gap fraction (GF) in the near-vertical direction, which can be related to LAI using a logarithmic model derived from Beer’s Law. The approach has been successful in LAI mapping in boreal and temperate forests. In this study, we evaluated the logarithmic model and several GF proxies in tropical montane forests in southeastern Kenya. We used two discrete-return lidar datasets (max. scan angle ~16°) with different flying heights and pulse densities (5.4 and 2.6 pulses m–2). GF for the 0–15° zenith angle range (GF15) and effective LAI (Le) were estimated for 29 sample plots using digital hemispherical photography. Twenty-one plots were located in indigenous forests and eight plots in plantation forests. According to the results, GF15 was best approximated by the proxies that included all canopy and ground return types (all echo cover index, ACI, root mean square error, RMSE = 0.050, bias = –0.003; Solberg’s cover index, SCI, RMSE = 0.057, bias = 0.002) although some saturation occurred when using data from the higher flight altitude. The results of the Le modelling propose that the logarithmic model needs to be fit separately for indigenous forest and plantations. Furthermore, the slope parameters of the models based on SCI suggest planophile (β ≈ 1.6) and spherical (β ≈ 2) leaf angle distribution for indigenous forests and plantations, respectively. We conclude that lidar cover indices based on all returns can estimate GF15 in closed-canopy tropical forests but the detection of the smallest gaps can be limited by the scanner or scanning parameters. The application of the logarithmic model requires stratification in the structurally heterogeneous and multi-species forest areas as β should be estimated separately for the different forest types.