Search

Accurate estimation of crown fire potential (CFP) can improve guidance on crown fire control and management. However, robust simulations of crown fire behavior are still challenging, limiting the accuracy of regional-scale CFP mapping. This study aims to incorporate fire spread simulation and machine learning algorithms to improve CFP mapping at a regional scale. First, we built a crown fire dataset using the fire simulations from the FARSITE model, as well as multi-source data, including fuel, weather, and topography variables. Fuel model parameters were optimized with four metaheuristic algorithms for robust fire simulations. Then, the hybrid models of CFP estimation (TBA-ML) were established by coupling with the transfer AdaBoost (TrAdaBoost) algorithm and three machine learning (ML) algorithms, i.e., Bayesian Network (BN), Random Forest (RF), and Support Vector Machine (SVM), to estimate CFP for crown fire danger assessment spatially. Results showed that the TBA-BN model performed best in estimating CFP with higher accuracy (AUC>0.9 and F1 score > 0.8) than the RF- and SVM-based CFP models. The variable importance and causal analysis showed that fuel and topography variables have major contributions to crown fire occurrence. Finally, we mapped monthly average passive and active CFP at regional scales and qualitatively demonstrated that our CFP time-series products successfully captured the dynamic change of crown fire danger. The above results suggest the potential of integrating fire spread simulation and machine learning algorithms to accurately estimate CFP to improve crown fire management.

The upcoming BIOMASS mission will provide P-band repeat-pass PolInSAR data from space for the improved mapping of global biomass. PolInSAR technique has been widely validated with the potential to invert forest height and estimate forest aboveground biomass (AGB). However, the robustness of PolInSAR-based AGB estimation across different sites still lacks full evaluation, especially for those with a varied forest type, heterogeneity (varied growth ratio between cover and height), and topographic relief. In this study, we concentrated on backscatter decomposition and forest height inversion, and developed a robust AGB estimation method that can be applied to different sites. Two dense and closed tropical forest sites (Paracou and Nouragues) and one open and heterogeneous boreal forest site (Krycklan) were selected as the study areas, and the corresponding airborne PolInSAR, LiDAR, and ground measured AGB data were used for validation and analysis. Results show that ground backscatter has the strongest correlation with AGB in boreal forests, but this correlation cannot be transferred to the tropical forests. Only canopy volume backscatter is almost free from topographic influence, and its relationship with AGB across three sites can be formulated using one exponential equation, producing the best estimation accuracy, with R2 of 0.79 and RMSE of 61.5 tons/ha (relative RMSE of 20.0 %). Multi-baseline PolInSAR retrieved forest height with little bias in spite of the presence of temporal decorrelation. One power equation can be used to correlate PolInSAR forest height with AGB across three different sites, and LOO (leave-one-out) validation shows the R2 of 0.85 and RMSE of 51.8 tons/ha (relative RMSE of 16.9 %). However, the RVoG-inverted PolInSAR FH was found to mainly represent the top forest height for open and heterogeneous forests, which means PolInSAR FH (forest height) lacks consideration for forest horizontal structure (e.g. forest density). In contrast, volume backscatter better captured forest density, and the proposed AGB model that combines PolInSAR FH and volume backscatter further improved the AGB estimation accuracy, especially for open forests: the plot-scale validation from all three sites shows R2 was improved from 0.79 (volume backscatter) and 0.85 (PolInSAR FH) to 0.89, and RMSE decreased from 61.5 and 51.8 to 45.2 (relative RMSE of 14.7 %) tons/ha; for region-scale validation, R2 was improved from 0.77 and 0.83 to 0.89, and RMSE decreased from 64.2 (relative RMSE of 39.0 %) and 54.5 (34.5 %) to 48.1 (29.4 %) tons/ha.

Optical data records have been widely used for monitoring forest aboveground biomass (AGB) dynamics given their freely available long-term archives and high temporal and spatial resolution. However, most of the methods utilized only the spatial information of reflectance (e.g. annual mean reflectance) to estimate AGB, lack of consideration for dynamic characteristics of reflectance. These methods are susceptible to reflectance saturation and temporal noise and may cause uncertainties in their estimation. In this study, we exploited dynamic time-series information of reflectance to enhance the accuracy and robustness of AGB estimation. More specifically, we utilized the continuous change detection and classification (CCDC) algorithm for temporal modeling of Landsat reflectance, followed by estimation of forest AGB with a random forest (RF) algorithm using the temporal information extracted by the CCDC model as input. The AGB estimation method was developed using in situ AGB measurements in Australia (10304 plots, range: [0.05, 936] tons/ha, mean: 153 tons/ha) and Landsat 5/7/8 data. Plot-scale results show that the overall reflectance derived by the CCDC model outperformed the annual mean reflectance when used in RF to estimate AGB because the overall reflectance is more robust to noisy or missing data. Introducing the CCDC model parameters into the RF model further improved estimation accuracy. The use of overall reflectance plus temporal parameters in AGB estimation increased R2 to 0.60 and reduced RMSE to 72.9 tons/ha, compared to using annual mean reflectance (R2 of 0.45 and RMSE of 85.7 tons/ha). This improvement may be attributed to CCDC model better capturing temporal changes in reflectance as the solar angle changes between seasons. Those changes relate to forest canopy coverage, width, and height, which are linked to AGB. Regional analysis for a 19,000-km2 area in Western Australia showed that this approach could characterize different types of forest AGB change over a 30 years period (1990–2019), including both gradual and sudden AGB losses and gains from, e.g., forest disturbance and management activities. This AGB estimation method has the potential to mitigate noisy or missing data and could be applied to large-scale forest biomass monitoring using the rich Landsat data archive.

P-band PolInSAR technique has been validated to be a useful tool for forest height inversion. Aiming at the residual ground scattering contribution and temporal decorrelation in repeat-pass interferometry that affect the inversion accuracy, we proposed a dual-baseline method based on the RVoG + VTD model. This method is firstly applied to E-SAR P-band data acquired over the Krycklan Catchment, Sweden with mixed forests, and the influence of ground scattering contribution and temporal decorrelation related to vertical wavenumber kz and height hv are theoretically and experimentally explored by comparing the traditional three-stage inversion method, the fixed extinction method and the proposed method. The experimental results show that the traditional three-stage inversion method has a overall serious overestimation for the whole study area, resulting in a R2 of ca. 0.4 and RMSE of ca. 7 m. Compared with the three-stage inversion method, the fixed extinction method improves the overestimation in the near range but has little effect in the far range. After introducing the RVoG + VTD model, both the overestimation in the near and far range are improved. In general, the study area is affected by both the influence of ground scattering contribution and temporal decorrelation. The overestimation in the near range (with kz larger than ca. 0.1 rad/m) is mainly caused by ground scattering contribution while for the far range (with kz smaller than ca. 0.06 rad/m) temporal decorrelation is the dominated reason. The simulation experiment further reveals the relationship between the influence of ground scattering contribution, temporal decorrelation and kz. Simulative results also show that ground scattering contribution has a greater influence for a large kz while temporal decorrelation generally brings overestimation, and has a greater impact for a small kz inversely. The proposed method simultaneously reduces the effects of ground scattering contribution and temporal decorrelation, and produces the highest accuracy with R2 reaching to ca. 0.6 and RMSE reducing to ca. 3.48 m, further demonstrating its validity for forest height inversion.

Detailed spatial information on the presence and properties of woody vegetation serves many purposes, including carbon accounting, environmental reporting and land management. Here, we investigated whether machine learning can be used to combine multiple spatial observations and training data to estimate woody vegetation canopy cover fraction (‘cover’), vegetation height (‘height’) and woody above-ground biomass dry matter (‘biomass’) at 25-m resolution across the Australian continent, where possible on an annual basis. We trained a Random Forest algorithm on cover and height estimates derived from airborne LiDAR over 11 regions and inventory-based biomass estimates for many thousands of plots across Australia. As predictors, we used annual geomedian Landsat surface reflectance, ALOS/PALSAR L-band radar backscatter mosaics, spatial vegetation structure data derived primarily from ICESat/GLAS satellite altimetry, and spatial climate data. Cross-validation experiments were undertaken to optimize the selection of predictors and the configuration of the algorithm. The resulting estimation errors were 0.07 for cover, 3.4 m for height, and 80 t dry matter ha-1 for biomass. A large fraction (89–94 %) of the observed variance was explained in each case. Priorities for future research include validation of the LiDAR-derived cover training data and the use of new satellite vegetation height data from the GEDI mission. Annual cover mapping for 2000–2018 provided detailed insight in woody vegetation dynamics. Continentally, woody vegetation change was primarily driven by water availability and its effect on bushfire and mortality, particularly in the drier interior. Changes in woody vegetation made a substantial contribution to Australia’s total carbon emissions since 2000. Whether these ecosystems will recover biomass in future remains to be seen, given the persistent pressures of climate change and land use.

Synthetic Aperture Radar (SAR) texture has been demonstrated to have the potential to improve forest biomass estimation using backscatter. However, forests are 3D objects with a vertical structure. The strong penetration of SAR signals means that each pixel contains the contributions of all the scatterers inside the forest canopy, especially for the P-band. Consequently, the traditional texture derived from SAR images is affected by forest vertical heterogeneity, although the influence on texture-based biomass estimation has not yet been explicitly explored. To separate and explore the influence of forest vertical heterogeneity, we introduced the SAR tomography technique into the traditional texture analysis, aiming to explore whether TomoSAR could improve the performance of texture-based aboveground biomass (AGB) estimation and whether texture plus tomographic backscatter could further improve the TomoSAR-based AGB estimation. Based on the P-band TomoSAR dataset from TropiSAR 2009 at two different sites, the results show that ground backscatter variance dominated the texture features of the original SAR image and reduced the biomass estimation accuracy. The texture from upper vegetation layers presented a stronger correlation with forest biomass. Texture successfully improved tomographic backscatter-based biomass estimation, and the texture from upper vegetation layers made AGB models much more transferable between different sites. In addition, the correlation between texture indices varied greatly among different tomographic heights. The texture from the 10 to 30 m layers was able to provide more independent information than the other layers and the original images, which helped to improve the backscatter-based AGB estimation.

Fuel load is the key factor driving fire ignition, spread and intensity. The current literature reports the light detection and ranging (LiDAR), optical and airborne synthetic aperture radar (SAR) data for fuel load estimation, but the optical and SAR data are generally individually explored. Optical and SAR data are expected to be sensitive to different types of fuel loads because of their different imaging mechanisms. Optical data mainly captures the characteristics of leaf and forest canopy, while the latter is more sensitive to forest vertical structures due to its strong penetrability. This study aims to explore the performance of Landsat Enhanced Thematic Mapper Plus (ETM+) and Advanced Land Observing Satellite (ALOS) Phased Arrayed L-band Synthetic Aperture Radar (PALSAR) data as well as their combination on estimating three different types of fuel load—stem fuel load (SFL), branch fuel load (BFL) and foliage fuel load (FFL). We first analyzed the correlation between the three types of fuel load and optical and SAR data. Then, the partial least squares regression (PLSR) was used to build the fuel load estimation models based on the fuel load measurements from Vindeln, Sweden, and variables derived from optical and SAR data. Based on the leave-one-out cross-validation (LOOCV) method, results show that L-band SAR data performed well on all three types of fuel load (R2 = 0.72, 0.70, 0.72). The optical data performed best for FFL estimation (R2 = 0.66), followed by BFL (R2 = 0.56) and SFL (R2 = 0.37). Further improvements were found for the SFL, BFL and FFL estimation when integrating optical and SAR data (R2 = 0.76, 0.81, 0.82), highlighting the importance of data selection and combination for fuel load estimation.

Existing drought indices have been widely used to monitor meteorological drought and agricultural drought; however, few of them are focus on drought monitoring for grassland regions. This study presented a new drought index, the Grassland Drought Index (GDI), for monitoring drought conditions in global grassland regions. These regions are vital for the environment and human society but susceptible to drought. The GDI was constructed based on three measures of water content: precipitation, soil moisture (SM), and canopy water content (CWC). The precipitation information was extracted from the available precipitation datasets, and SM was estimated by downscaling exiting soil moisture data to a 1 km resolution, and CWC was retrieved based on the PROSAIL (PROSPECT + SAIL) model. Each variable was scaled from 0 to 1 for each pixel based on absolute minimum and maximum values over time, and these scaled variables were combined with the selected weights to construct the GDI. According to validation at the regional scale, the GDI was correlated with the Standardized Precipitation Index (SPI) to some extent, and captured most of the drought area identified by the United States Drought Monitor (USDM) maps. In addition, the global GDI product at a 1 km spatial resolution substantially agreed with the global Standardized Precipitation Evapotranspiration Index (SPEI) product throughout the period 2005–2010, and it provided detailed and accurate information about the location and the duration of drought based on the evaluation using the known drought events.

In this study, the Standardized Precipitation Evaporation Index (SPEI) was applied to characterize the drought conditions in Southwest China from 1982–2012. The SPEI was calculated by precipitation and temperature data for various accumulation periods. Based on the SPEI, the multi-scale patterns, the trend, and the spatio-temporal extent of drought were evaluated, respectively. The results explicitly showed a drying trend of Southwest China. The mean SPEI values at five time scales all decreased significantly. Some moderate and severe droughts were captured after 2005 and the droughts were even getting aggravated. By examining the spatio-temporal extent, the aggravating condition of drought was further revealed. To investigate the performance of SPEI, correlation analysis was conducted between SPEI and two remotely sensed drought indices: Soil Moisture Condition Index (SMCI) and Vegetation Condition Index (VCI). The comparison was also conducted with the Standardized Precipitation Index (SPI). The results showed that for both SMCI and VCI, the SPI and SPEI had approximate correlations with them. The SPEI could better monitor the soil moisture than the SPI in months with significant increase of temperature. The correlations between the VCI and SPI/SPEI were lower; nevertheless, the SPEI was slightly superior to the SPI.

Fire Spread Rate (FSR) can indicate how fast a fire is spreading, which is especially helpful for wildfire rescue and management. Historically, images obtained from sun-orbiting satellites such as Moderate Resolution Imaging Spectroradiometer (MODIS) were used to detect active fire and burned area at the large spatial scale. However, the daily revisit cycles make them inherently unable to extract FSR in near real­-time (hourly or less). We argue that the Himawari-8, a next generation geostationary satellite with a 10-min temporal resolution and 0.5–2 km spatial resolution, may have the potential for near real-time FSR extraction. To that end, we propose a novel method (named H8-FSR) for near real-time FSR extraction based on the Himawari-8 data. The method first defines the centroid of the burned area as the fire center and then the near real-time FSR is extracted by timely computing the movement rate of the fire center. As a case study, the method was applied to the Esperance bushfire that broke out on 17 November, 2015, in Western Australia. Compared with the estimated FSR using the Commonwealth Scientific and Industrial Research Organization (CSIRO) Grassland Fire Spread (GFS) model, H8-FSR achieved favorable performance with a coefficient of determination (R2) of 0.54, mean bias error of –0.75 m/s, mean absolute percent error of 33.20% and root mean square error of 1.17 m/s, respectively. These results demonstrated that the Himawari-8 data are valuable for near real-time FSR extraction, and also suggested that the proposed method could be potentially applicable to other next generation geostationary satellite data.