Three-phase hierarchical model-based and hybrid inference

Global commitments to mitigating climate change and halting biodiversity loss require reliable information about Earth's ecosystems. Increasingly, such information is obtained from multiple sources of remotely sensed data combined with data acquired in the field. This new wealth of data poses challenges regarding the combination of different data sources to derive the required information and assess uncertainties. In this article, we show how predictors and their variances can be derived when hierarchically nested models are applied. Previous studies have developed methods for cases involving two modeling steps, such as biomass prediction relying on tree-level allometric models and models linking plot-level field data with remotely sensed data. This study extends the analysis to cases involving three modeling steps to cover new important applications. The additional step might involve an intermediate model, linking field and remotely sensed data available from a small sample, for making predictions that are subsequently used for training a final prediction model based on remotely sensed data:•In cases where the data in the final step are available wall-to-wall, we denote the approach three-phase hierarchical model-based inference (3pHMB),•In cases where the data in the final step are available as a probability sample, we denote the approach three-phase hierarchical hybrid inference (3pHHY).

Estimating biomass and soil carbon change at the level of forest stands using repeated forest surveys assisted by airborne laser scanner data

BACKGROUND

Under the growing pressure to implement mitigation actions, the focus of forest management is shifting from a traditional resource centric view to incorporate more forest ecosystem services objectives such as carbon sequestration. Estimating the above-ground biomass in forests using airborne laser scanning (ALS) is now an operational practice in Northern Europe and is being adopted in many parts of the world. In the boreal forests, however, most of the carbon (85%) is stored in the soil organic (SO) matter. While this very important carbon pool is "invisible" to ALS, it is closely connected and feeds from the growing forest stocks. We propose an integrated methodology to estimate the changes in forest carbon pools at the level of forest stands by combining field measurements and ALS data.

RESULTS

ALS-based models of dominant height, mean diameter, and biomass were fitted using the field observations and were used to predict mean tree biophysical properties across the entire study area (50 km²) which was in turn used to estimate the biomass carbon stocks and the litter production that feeds into the soil. For the soil carbon pool estimation, we used the Yasso15 model. The methodology was based on (1) approximating the initial soil carbon stocks using simulations; (2) predicting the annual litter input based on the predicted growing stocks in each cell; (3) predicting the soil carbon dynamics of the annual litter using the Yasso15 soil carbon model. The estimated total carbon change (standard errors in parenthesis) for the entire area was 0.741 (0.14) Mg ha^-1 yr^-1. The biomass carbon change was 0.405 (0.13) Mg ha^-1 yr^-1, the litter carbon change (e.g., deadwood and leaves) was 0.346 (0.027) Mg ha^-1 yr^-1, and the change in SO carbon was - 0.01 (0.003) Mg ha^-1 yr^-1.

CONCLUSIONS

Our results show that ALS data can be used indirectly through a chain of models to estimate soil carbon changes in addition to changes in biomass at the primary level of forest management, namely the forest stands. Having control of the errors contributed by each model, the stand-level uncertainty can be estimated under a model-based inferential approach.

Analysis-ready satellite data mosaics from Landsat and Sentinel-2 imagery

Today's enormous amounts of freely available high-resolution satellite imagery provide the demand for effective preprocessing methods. One such preprocessing method needed in many applications utilizing optical satellite imagery from the Landsat and Sentinel-2 archives is mosaicking. Merging hundreds of single scenes into a single satellite data mosaic before conducting analysis such as land cover classification, change detection, or modelling is often a prerequisite. Maintaining the original data structure and preserving metadata for further modelling or classification would be advantageous for many applications. Furthermore, in other applications, e.g., connected to land cover classification creating the mosaic for a specific period matching the phenological state of the phenomena in nature would be beneficial. In addition, supporting in-house and computing centers not directly connected to a specific cloud provider could be a requirement for some institutions or companies. In the current work, we present a method called Geomosaic that meets these criteria and produces analysis-ready satellite data mosaics from Landsat and Sentinel-2 imagery.•The method described produces analysis-ready satellite data mosaics.•The satellite data mosaics contain pixel metadata usable for further analysis.•The algorithm is available as an open-source tool coded in Python and can be used on multiple platforms.

Monitoring small pioneer trees in the forest-tundra ecotone: using multi-temporal airborne laser scanning data to model height growth

Monitoring of forest resources through national forest inventory programmes is carried out in many countries. The expected climate changes will affect trees and forests and might cause an expansion of trees into presently treeless areas, such as above the current alpine tree line. It is therefore a need to develop methods that enable the inclusion of also these areas into monitoring programmes. Airborne laser scanning (ALS) is an established tool in operational forest inventories, and could be a viable option for monitoring tasks. In the present study, we used multi-temporal ALS data with point density of 8-15 points per m², together with field measurements from single trees in the forest-tundra ecotone along a 1500-km-long transect in Norway. The material comprised 262 small trees with an average height of 1.78 m. The field-measured height growth was derived from height measurements at two points in time. The elapsed time between the two measurements was 4 years. Regression models were then used to model the relationship between ALS-derived variables and tree heights as well as the height growth. Strong relationships between ALS-derived variables and tree heights were found, with R ² values of 0.93 and 0.97 for the two points in time. The relationship between the ALS data and the field-derived height growth was weaker, with R ² values of 0.36-0.42. A cross-validation gave corresponding results, with root mean square errors of 19 and 11% for the ALS height models and 60% for the model relating ALS data to single-tree height growth.

Combining airborne laser scanning and Landsat data for statistical modeling of soil carbon and tree biomass in Tanzanian Miombo woodlands

BACKGROUND

Soil carbon and biomass depletion can be used to identify and quantify degraded soils, and by using remote sensing, there is potential to map soil conditions over large areas. Landsat 8 Operational Land Imager satellite data and airborne laser scanning data were evaluated separately and in combination for modeling soil organic carbon, above ground tree biomass and below ground tree biomass. The test site is situated in the Liwale district in southeastern Tanzania and is dominated by Miombo woodlands. Tree data from 15 m radius field-surveyed plots and samples of soil carbon down to a depth of 30 cm were used as reference data for tree biomass and soil carbon estimations.

RESULTS

Cross-validated plot level error (RMSE) for predicting soil organic carbon was 28% using only Landsat 8, 26% using laser only, and 23% for the combination of the two. The plot level error for above ground tree biomass was 66% when using only Landsat 8, 50% for laser and 49% for the combination of Landsat 8 and laser data. Results for below ground tree biomass were similar to above ground biomass. Additionally it was found that an early dry season satellite image was preferable for modelling biomass while images from later in the dry season were better for modelling soil carbon.

CONCLUSION

The results show that laser data is superior to Landsat 8 when predicting both soil carbon and biomass above and below ground in landscapes dominated by Miombo woodlands. Furthermore, the combination of laser data and Landsat data were marginally better than using laser data only.

Mapping and estimating the total living biomass and carbon in low-biomass woodlands using Landsat 8 CDR data

BACKGROUND

A functional forest carbon measuring, reporting and verification (MRV) system to support climate change mitigation policies, such as REDD+, requires estimates of forest biomass carbon, as an input to estimate emissions. A combination of field inventory and remote sensing is expected to provide those data. By linking Landsat 8 and forest inventory data, we (1) developed linear mixed effects models for total living biomass (TLB) estimation as a function of spectral variables, (2) developed a 30 m resolution map of the total living carbon (TLC), and (3) estimated the total TLB stock of the study area. Inventory data consisted of tree measurements from 500 plots in 63 clusters in a 15,700 km² study area, in miombo woodlands of Tanzania. The Landsat 8 data comprised two climate data record images covering the inventory area.

RESULTS

We found a linear relationship between TLB and Landsat 8 derived spectral variables, and there was no clear evidence of spectral data saturation at higher biomass values. The root-mean-square error of the values predicted by the linear model linking the TLB and the normalized difference vegetation index (NDVI) is equal to 44 t/ha (49 % of the mean value). The estimated TLB for the study area was 140 Mt, with a mean TLB density of 81 t/ha, and a 95 % confidence interval of 74-88 t/ha. We mapped the distribution of TLC of the study area using the TLB model, where TLC was estimated at 47 % of TLB.

CONCLUSION

The low biomass in the miombo woodlands, and the absence of a spectral data saturation problem suggested that Landsat 8 derived NDVI is suitable auxiliary information for carbon monitoring in the context of REDD+, for low-biomass, open-canopy woodlands.

Monitoring forest carbon in a Tanzanian woodland using interferometric SAR: a novel methodology for REDD

BACKGROUND

REDD+ implementation requires establishment of a system for measuring, reporting and verification (MRV) of forest carbon changes. A challenge for MRV is the lack of satellite based methods that can track not only deforestation, but also degradation and forest growth, as well as a lack of historical data that can serve as a basis for a reference emission level. Working in a miombo woodland in Tanzania, we here aim at demonstrating a novel 3D satellite approach based on interferometric processing of radar imagery (InSAR).

RESULTS

Forest carbon changes are derived from changes in the forest canopy height obtained from InSAR, i.e. decreases represent carbon loss from logging and increases represent carbon sequestration through forest growth. We fitted a model of above-ground biomass (AGB) against InSAR height, and used this to convert height changes to biomass and carbon changes. The relationship between AGB and InSAR height was weak, as the individual plots were widely scattered around the model fit. However, we consider the approach to be unique and feasible for large-scale MRV efforts in REDD+ because the low accuracy was attributable partly to small plots and other limitations in the data set, and partly to a random pixel-to-pixel variation in trunk forms. Further processing of the InSAR data provides data on the categories of forest change. The combination of InSAR data from the Shuttle RADAR Topography Mission (SRTM) and the TanDEM-X satellite mission provided both historic baseline of change for the period 2000-2011, as well as annual change 2011-2012.

CONCLUSIONS

A 3D data set from InSAR is a promising tool for MRV in REDD+. The temporal changes seen by InSAR data corresponded well with, but largely supplemented, the changes derived from Landsat data.

Spatial distribution of temporal dynamics in anthropogenic fires in miombo savanna woodlands of Tanzania

BACKGROUND

Anthropogenic uses of fire play a key role in regulating fire regimes in African savannas. These fires contribute the highest proportion of the globally burned area, substantial biomass burning emissions and threaten maintenance and enhancement of carbon stocks. An understanding of fire regimes at local scales is required for the estimation and prediction of the contribution of these fires to the global carbon cycle and for fire management. We assessed the spatio-temporal distribution of fires in miombo woodlands of Tanzania, utilizing the MODIS active fire product and Landsat satellite images for the past ~40 years.

RESULTS

Our results show that up to 50.6% of the woodland area is affected by fire each year. An early and a late dry season peak in wetter and drier miombo, respectively, characterize the annual fire season. Wetter miombo areas have higher fire activity within a shorter annual fire season and have shorter return intervals. The fire regime is characterized by small-sized fires, with a higher ratio of small than large burned areas in the frequency-size distribution (β = 2.16 ± 0.04). Large-sized fires are rare, and occur more frequently in drier than in wetter miombo. Both fire prevalence and burned extents have decreased in the past decade. At a large scale, more than half of the woodland area has less than 2 years of fire return intervals, which prevent the occurrence of large intense fires.

CONCLUSION

The sizes of fires, season of burning and spatial extent of occurrence are generally consistent across time, at the scale of the current analysis. Where traditional use of fire is restricted, a reassessment of fire management strategies may be required, if sustainability of tree cover is a priority. In such cases, there is a need to combine traditional and contemporary fire management practices.

Effects of field plot size on prediction accuracy of aboveground biomass in airborne laser scanning-assisted inventories in tropical rain forests of Tanzania

BACKGROUND

Airborne laser scanning (ALS) has recently emerged as a promising tool to acquire auxiliary information for improving aboveground biomass (AGB) estimation in sample-based forest inventories. Under design-based and model-assisted inferential frameworks, the estimation relies on a model that relates the auxiliary ALS metrics to AGB estimated on ground plots. The size of the field plots has been identified as one source of model uncertainty because of the so-called boundary effects which increases with decreasing plot size. Recent research in tropical forests has aimed to quantify the boundary effects on model prediction accuracy, but evidence of the consequences for the final AGB estimates is lacking. In this study we analyzed the effect of field plot size on model prediction accuracy and its implication when used in a model-assisted inferential framework.

RESULTS

The results showed that the prediction accuracy of the model improved as the plot size increased. The adjusted R² increased from 0.35 to 0.74 while the relative root mean square error decreased from 63.6 to 29.2%. Indicators of boundary effects were identified and confirmed to have significant effects on the model residuals. Variance estimates of model-assisted mean AGB relative to corresponding variance estimates of pure field-based AGB, decreased with increasing plot size in the range from 200 to 3000 m². The variance ratio of field-based estimates relative to model-assisted variance ranged from 1.7 to 7.7.

CONCLUSIONS

This study showed that the relative improvement in precision of AGB estimation when increasing field-plot size, was greater for an ALS-assisted inventory compared to that of a pure field-based inventory.

Modelling aboveground forest biomass using airborne laser scanner data in the miombo woodlands of Tanzania

BACKGROUND

Airborne laser scanning (ALS) has emerged as one of the most promising remote sensing technologies for estimating aboveground biomass (AGB) in forests. Use of ALS data in area-based forest inventories relies on the development of statistical models that relate AGB and metrics derived from ALS. Such models are firstly calibrated on a sample of corresponding field- and ALS observations, and then used to predict AGB over the entire area covered by ALS data. Several statistical methods, both parametric and non-parametric, have been applied in ALS-based forest inventories, but studies that compare different methods in tropical forests in particular are few in number and less frequent than studies reported in temperate and boreal forests. We compared parametric and non-parametric methods, specifically linear mixed effects model (LMM) and k-nearest neighbor (k-NN).

RESULTS

The results showed that the prediction accuracy obtained when using LMM was slightly better than when using the k-NN approach. Relative root mean square errors from the cross validation was 46.8 % for the LMM and 58.1 % for the k-NN. Post-stratification according to vegetation types improved the prediction accuracy of LMM more as compared to post-stratification by using land use types.

CONCLUSION

Although there were differences in prediction accuracy between the two methods, their accuracies indicated that both of methods have potentials to be used for estimation of AGB using ALS data in the miombo woodlands. Future studies on effects of field plot size and the errors due to allometric models on the prediction accuracy are recommended.

Forest biomass change estimated from height change in interferometric SAR height models

BACKGROUND

There is a need for new satellite remote sensing methods for monitoring tropical forest carbon stocks. Advanced RADAR instruments on board satellites can contribute with novel methods. RADARs can see through clouds, and furthermore, by applying stereo RADAR imaging we can measure forest height and its changes. Such height changes are related to carbon stock changes in the biomass. We here apply data from the current Tandem-X satellite mission, where two RADAR equipped satellites go in close formation providing stereo imaging. We combine that with similar data acquired with one of the space shuttles in the year 2000, i.e. the so-called SRTM mission. We derive height information from a RADAR image pair using a method called interferometry.

RESULTS

We demonstrate an approach for REDD based on interferometry data from a boreal forest in Norway. We fitted a model to the data where above-ground biomass in the forest increases with 15 t/ha for every m increase of the height of the RADAR echo. When the RADAR echo is at the ground the estimated biomass is zero, and when it is 20 m above the ground the estimated above-ground biomass is 300 t/ha. Using this model we obtained fairly accurate estimates of biomass changes from 2000 to 2011. For 200 m² plots we obtained an accuracy of 65 t/ha, which corresponds to 50% of the mean above-ground biomass value. We also demonstrate that this method can be applied without having accurate terrain heights and without having former in-situ biomass data, both of which are generally lacking in tropical countries. The gain in accuracy was marginal when we included such data in the estimation. Finally, we demonstrate that logging and other biomass changes can be accurately mapped. A biomass change map based on interferometry corresponded well to a very accurate map derived from repeated scanning with airborne laser.

CONCLUSIONS

Satellite based, stereo imaging with advanced RADAR instruments appears to be a promising method for REDD. Interferometric processing of the RADAR data provides maps of forest height changes from which we can estimate temporal changes in biomass and carbon.