Prototyping a global algorithm for systematic fire-affected area mapping using MODIS time series data

The remote sensing of Earth surface changes is an active research field aimed at the development of methods and data products needed by scientists, resource managers, and policymakers. Fire is a major cause of surface change and occurs in most vegetation zones across the world. The identification and delineation of fire-affected areas, also known as burned areas or fire scars, may be considered a change detection problem. Remote sensing algorithms developed to map fire-affected areas are difficult to implement reliably over large areas because of variations in both the surface state and those imposed by the sensing system. The availability of robustly calibrated, atmospherically corrected, cloud-screened, geolocated data provided by the latest generation of moderate resolution remote sensing systems allows for major advances in satellite mapping of fire-affected area. This paper describes an algorithm developed to map fire-affected areas at a global scale using Moderate Resolution Imaging Spectroradiometer (MODIS) surface reflectance time series data. The algorithm is developed from the recently published Bi-Directional Reflectance Model-Based Expectation change detection approach and maps at 500 m the location and approximate day of burning. Improvements made to the algorithm for systematic global implementation are presented and the algorithm performance is demonstrated for southern African, Australian, South American, and Boreal fire regimes. The algorithm does not use training data but rather applies a wavelength independent threshold and spectral constraints defined by the noise characteristics of the reflectance data and knowledge of the spectral behavior of burned vegetation and spectrally confusing changes that are not associated with burning. Temporal constraints are applied capitalizing on the spectral persistence of fire-affected areas. Differences between mapped fire-affected areas and cumulative MODIS active fire detections are illustrated and discussed for each fire regime. The results reveal a coherent spatio-temporal mapping of fire-affected area and indicate that the algorithm shows potential for global application.     

Modeling MODIS LAI time series using three statistical methods

Leaf Area Index (LAI) is one of the most important variables characterizing land surface vegetation and dynamics. Many satellite data, such as the Moderate Resolution Imaging Spectroradiometer (MODIS), have been used to generate LAI products. It is important to characterize their spatial and temporal variations by developing mathematical models from these products. In this study, we aim to model MODIS LAI time series and further predict its future values by decomposing the LAI time series of each pixel into several components: trend, intra-annual variations, seasonal cycle, and stochastic stationary or irregular parts. Three such models that can characterize the non-stationary time series data and predict the future values are explored, including Dynamic Harmonics Regression (DHR), STL (Seasonal-Trend Decomposition Procedure based on Loess), and Seasonal ARIMA (AutoRegressive Intergrated Moving Average) (SARIMA). The preliminary results using six years (2001-2006) of the MODIS LAI product indicate that all these methods are effective to model LAI time series and predict 2007 LAI values reasonably well. The SARIMA model gives the best prediction, DHR produces the smoothest curve, and STL is more sensitive to noise in the data. These methods work best for land cover types with pronounced seasonal variations.     

Real-time retrieval of Leaf Area Index from MODIS time series data

Real-time/near real-time inversion of land surface biogeophysical variables from satellite observations is required to monitor rapid land surface changes, and provide the necessary input for numerical weather forecasting models and decision support systems. This paper develops a new inversion method for the real-time estimation of the Leaf Area Index (LAI) of land surfaces from MODIS time series reflectance data (MOD09A1). It consists of a series of procedures, including time series data smoothing, data quality control and real-time estimation of LAI. After the historical LAI time series is smoothed by a multi-step Savitzky-Golay filter to determine the upper LAI envelope, a Seasonal Auto-Regressive Integrated Moving Average (SARIMA) model is used to derive the LAI climatology. Based on the climatology from the SARIMA model to evolve LAI in time, a dynamic model is then constructed and used to provide the short-range forecast of LAI. Predictions from this model are used with Ensemble Kalman Filter (EnKF) techniques to recursively update biophysical variables as new observations arrive. The validation results produced using MODIS surface reflectance data and field-measured LAI data at eight BELMANIP sites show that the real-time inversion method is able to efficiently produce a relatively smooth LAI product. In addition, the accuracy is significantly improved over the MODIS LAI product.     

Detecting trend and seasonal changes in satellite image time series

A wealth of remotely sensed image time series covering large areas is now available to the earth science community. Change detection methods are often not capable of detecting land cover changes within time series that are heavily influenced by seasonal climatic variations. Detecting change within the trend and seasonal components of time series enables the classification of different types of changes. Changes occurring in the trend component often indicate disturbances (e.g. fires, insect attacks), while changes occurring in the seasonal component indicate phenological changes (e.g. change in land cover type). A generic change detection approach is proposed for time series by detecting and characterizing Breaks For Additive Seasonal and Trend (BFAST). BFAST integrates the decomposition of time series into trend, seasonal, and remainder components with methods for detecting change within time series. BFAST iteratively estimates the time and number of changes, and characterizes change by its magnitude and direction. We tested BFAST by simulating 16-day Normalized Difference Vegetation Index (NDVI) time series with varying amounts of seasonality and noise, and by adding abrupt changes at different times and magnitudes. This revealed that BFAST can robustly detect change with different magnitudes (> 0.1 NDVI) within time series with different noise levels (0.01-0.07 σ) and seasonal amplitudes (0.1-0.5 NDVI). Additionally, BFAST was applied to 16-day NDVI Moderate Resolution Imaging Spectroradiometer (MODIS) composites for a forested study area in south eastern Australia. This showed that BFAST is able to detect and characterize spatial and temporal changes in a forested landscape. BFAST is not specific to a particular data type and can be applied to time series without the need to normalize for land cover types, select a reference period, or change trajectory. The method can be integrated within monitoring frameworks and used as an alarm system to flag when and where changes occur.     

基于MODIS时间序列森林扰动监测指数比较研究

森林扰动是影响陆地生态系统的重要因素之一,遥感可定期地获得大面积森林覆盖数据,成为定期和连续森林扰动监测的重要手段,基于时间序列数据的森林监测成为主要方式。研究利用2001~2013年MODIS时间序列遥感影像,以福建省为例,利用NDVI、NBRI、NDMI、IFZ和DI 5种森林扰动监测指数,结合植被变化追踪算法提取森林扰动区域,并从光谱响应特征和对不同扰动类型的响应能力等方面,分析了对我国南方森林扰动的监测能力。结果表明:DI对森林砍伐、森林病虫害和植树造林3种扰动类型的响应能力较强,NBR对森林火灾最为敏感,NDVI对4种扰动类型的响应能力相对较弱;5种指数中DI对森林扰动的响应能力较强,森林扰动提取精度最高,IFZ次之,NDMI和NBR监测精度相当,且优于NDVI。     

AVHRR、VEGETATION和MODIS时间系列遥感数据产品现状与应用研究进展

全球时间系列卫星遥感产品自产生之日起就得到了高度关注,被广泛地应用于全球、洲际和区域的地表动态监测,并与气温、降水等气候变化表征参数结合起来,应用于全球变化分析。随着时间系列的逐渐延长和新兴传感器的不断涌现,时间系列遥感产品的内容和应用领域更是得到了极大扩展。主要介绍了:①当前国际上流行的可见光/近红外、短波红外和热红外时间系列卫星数据产品的发展现状,传感器主要包括AVHRR、VEGETATION和MODIS。早期以开发波段信息和植被指数等基础数据为主,当前大量专题产品的生产得到广泛开展;② 在数据产品的进一步处理和分析方面,重点介绍了时间系列重建、比较和延长、产品真实性检验的研究进展和发展趋势;③在数据产品应用方面,重点介绍了地表覆被特征的动态监测、物候和种植结构等信息提取、遥感产品在模型中的应用等方面的研究进展和发展趋势。     

Automatic outlier detection for time series: an application to sensor data

In this article we consider the problem of detecting unusual values or outliers from time series data where the process by which the data are created is difficult to model. The main consideration is the fact that data closer in time are more correlated to each other than those farther apart. We propose two variations of a method that uses the median from a neighborhood of a data point and a threshold value to compare the difference between the median and the observed data value. Both variations of the method are fast and can be used for data streams that occur in quick succession such as sensor data on an airplane. Martin Meckesheimer has been a member of the Applied Statistics Group at Phantom Works, Boeing since 2001. He received a Bachelor of Science Degree in Industrial Engineering from the University of Pittsburgh in 1997, and a Master's Degree in Industrial and Systems Engineering from Ecole Centrale Paris in 1999. Martin earned a Doctorate in Industrial Engineering from The Pennsylvania State University in August 2001, as a student of Professor Russell R. Barton and Dr. Timothy W. Simpson. His primary research interests are in the areas of design of experiments and surrogate modeling. Sabyasachi Basu received his Ph.D. is Statistics from the University of Wisconsin at Madison in 1990. Since his Ph.D., he has worked in both academia and in industry. He has taught and guided Ph.D. students in the Department of Statistics at the Southern Methodist University. He has also worked as a senior marketing statistician at the J. C. Penney Company. Dr. Basu is also an American Society of Quality certified Six Sigma Black Belt. He is currently an Associate Technical Fellow in Statistics and Data Mining at the Boeing Company. In this capacity, he works as a researcher and a technical consultant within Boeing for data mining, statistics and process improvements. He has published more than 20 papers and technical reports. He has also served as journal referee for several journals, organized conferences and been invited to present at conferences.     

Assessment of MODIS NDVI time series data products for detecting forest defoliation by gypsy moth outbreaks

This paper discusses an assessment of Moderate Resolution Imaging Spectroradiometer (MODIS) time-series data products for detecting forest defoliation from European gypsy moth (Lymantria dispar). This paper describes an effort to aid the United States Department of Agriculture (USDA) Forest Service in developing and assessing MODIS-based gypsy moth defoliation detection products and methods that could be applied in near real time without intensive field survey data collection as a precursor. In our study, MODIS data for 2000-2006 were processed for the mid-Appalachian highland region of the United States. Gypsy moth defoliation maps showing defoliated forests versus non-defoliated areas were produced from temporally filtered and composited MOD02 and MOD13 data using unsupervised classification and image thresholding of maximum value normalized difference vegetation index (NDVI) datasets computed for the defoliation period (June 10-July 27) of 2001 and of the entire time series. These products were validated by comparing stratified random sample locations to relevant Landsat and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) reference data sets. Composites of 250 m daily MOD02 outperformed 16-day MOD13 data in terms of classifying forest defoliation, showing a lower omission error rate (0.09 versus 0.56), a similar Kappa (0.67 versus 0.79), a comparable commission error rate (0.22 versus 0.14), and higher overall classification agreement (88 versus 79%). Results suggest that temporally processed MODIS time-series data can detect with good agreement to available reference data the extent and location of historical regional gypsy moth defoliation patches of 0.25 km² or more for 250-meter products. The temporal processing techniques used in this study enabled effective broad regional, “wall to wall” gypsy moth defoliation detection products for a 6.2 million ha region that were not produced previously with either MODIS or other satellite data. This study provides new, previously unavailable information on the relative agreement of temporally processed, gypsy moth defoliation detection products from MODIS NDVI time series data with respect to higher spatial resolution Landsat and ASTER data. These results also provided needed timely information on the potential of MODIS data for contributing near real time defoliation products to a USDA Forest Service Forest Threat Early Warning System.     

Phenological change detection while accounting for abrupt and gradual trends in satellite image time series

A challenge in phenology studies is understanding what constitutes phenological change amidst background variation. The majority of phenological studies have focused on extracting critical points in the seasonal growth cycle, without exploiting the full temporal detail. The high degree of phenological variability between years demonstrates the necessity of distinguishing long-term phenological change from temporal variability. Here, we demonstrate the phenological change detection ability of a method for detecting change within time series. BFAST, Breaks For Additive Seasonal and Trend, integrates the decomposition of time series into trend, seasonal, and remainder components with methods for detecting change. We tested BFAST by simulating 16-day NDVI time series with varying amounts of seasonal amplitude and noise, containing abrupt disturbances (e.g. fires) and long-term phenological changes. This revealed that the method is able to detect the timing of phenological changes within time series while accounting for abrupt disturbances and noise. Results showed that the phenological change detection is influenced by the signal-to-noise ratio of the time series. Between different land cover types the seasonal amplitude varies and determines the signal-to-noise ratio, and as such the capacity to differentiate phenological changes from noise. Application of the method on 16-day NDVI MODIS images from 2000 until 2009 for a forested study area in south eastern Australia confirmed these results. It was shown that a minimum seasonal amplitude of 0.1 NDVI is required to detect phenological change within cleaned MODIS NDVI time series using the quality flags. BFAST identifies phenological change independent of phenological metrics by exploiting the full time series. The method is globally applicable since it analyzes each pixel individually without the setting of thresholds to detect change within a time series. Long-term phenological changes can be detected within NDVI time series of a large range of land cover types (e.g. grassland, woodlands and deciduous forests) having a seasonal amplitude larger than the noise level. The method can be applied to any time series data and it is not necessarily limited to NDVI.     

Optimized extraction of daily bio-optical time series derived from MODIS/Aqua imagery for Lake Tanganyika, Africa

Lake Tanganyika is one of the world's great freshwater ecosystems. In recent decades its hydrodynamic characteristics have undergone important changes that have had consequences on the lake's primary productivity. The establishment of a long-term Ocean Color dataset for Lake Tanganyika is a fundamental tool for understanding and monitoring these changes. We developed an approach to create a regionally calibrated dataset of chlorophyll-a concentrations (CHL) and attenuation coefficients at 490 nm (K490) for the period from July 2002 to December 2006 using daily calibrated radiances retrieved from the MODIS-Aqua sensor. Standard MODIS Aqua Ocean Color products were found to not provide a suitable calibration for high altitude lakes such as the Lake Tanganyika. An optimization of the extraction process and the validation of the dataset were performed with independent sets of in situ measurements. Our results show that for the geographical, atmospheric and optical conditions of Lake Tanganyika: (i) a coastal aerosol model set with high relative humidity (90%) provides a suitable atmospheric correction; (ii) a significant correlation between in situ data and CHL estimates using the MODIS specific OC3 algorithm is possible; and (iii) K490 estimates provide a good level of significance. The resulting validated time series of bio-optical properties provides a fundamental information base for the study of phytoplankton and primary production dynamics and interannual trends. A comparison between surface chlorophyll-a concentrations estimated from field monitoring and from the MODIS based dataset shows that remote sensing allows improved detection of surface blooms in Lake Tanganyika.     

An integrated temporal data model incorporating time series concept

Most temporal data models have concentrated on describing temporal data based on versioning of objects, tuples or attributes. The concept of time series, which is often needed in temporal applications, does not fit well within these models. The goal of this paper is to propose a generalized temporal database model that integrates the modeling of both version-based and time-series based temporal data into a single conceptual framework. The concept of calendar is also integrated into our proposed model. We also discuss how a conceptual Extended-ER design in our model can be mapped to an object-oriented or relational database implementation.     

Clustering of time series data—a survey

Time series clustering has been shown effective in providing useful information in various domains. There seems to be an increased interest in time series clustering as part of the effort in temporal data mining research. To provide an overview, this paper surveys and summarizes previous works that investigated the clustering of time series data in various application domains. The basics of time series clustering are presented, including general-purpose clustering algorithms commonly used in time series clustering studies, the criteria for evaluating the performance of the clustering results, and the measures to determine the similarity/dissimilarity between two time series being compared, either in the forms of raw data, extracted features, or some model parameters. The past researchs are organized into three groups depending upon whether they work directly with the raw data either in the time or frequency domain, indirectly with features extracted from the raw data, or indirectly with models built from the raw data. The uniqueness and limitation of previous research are discussed and several possible topics for future research are identified. Moreover, the areas that time series clustering have been applied to are also summarized, including the sources of data used. It is hoped that this review will serve as the steppingstone for those interested in advancing this area of research.     

Pairwise likelihood inference for ordinal categorical time series

Ordinal categorical time series may be analyzed as censored observations from a suitable latent stochastic process, which describes the underlying evolution of the system. This approach may be considered as an alternative to Markov chain models or to regression methods for categorical time series data. The problem of parameter estimation is solved through a simple pseudolikelihood, called pairwise likelihood. This inferential methodology is successfully applied to the class of autoregressive ordered probit models. Potential usefulness for inference and model selection within more general classes of models are also emphasized. Illustrations include simulation studies and two simple real data applications.     

Combining medium and coarse spatial resolution satellite data to improve the estimation of sub-pixel NDVI time series

The potential of multitemporal coarse spatial resolution remotely sensed images for vegetation monitoring is reduced in fragmented landscapes, where most of the pixels are composed of a mixture of different surfaces. Several approaches have been proposed for the estimation of reflectance or NDVI values of the different land-cover classes included in a low resolution mixed pixel. In this paper, we propose a novel approach for the estimation of sub-pixel NDVI values from multitemporal coarse resolution satellite data. Sub-pixel NDVIs for the different land-cover classes are calculated by solving a weighted linear system of equations for each pixel of a coarse resolution image, exploiting information about within-pixel fractional cover derived from a high resolution land-use map. The weights assigned to the different pixels of the image for the estimation of sub-pixel NDVIs of a target pixel i are calculated taking into account both the spatial distance between each pixel and the target and their spectral dissimilarity estimated on medium-resolution remote-sensing images acquired in different periods of the year. The algorithm was applied to daily and 16-day composite MODIS NDVI images, using Landsat-5 TM images for calculation of weights and accuracy evaluation.Results showed that application of the algorithm provided good estimates of sub-pixel NDVIs even for poorly represented land-cover classes (i.e., with a low total cover in the test area). No significant accuracy differences were found between results obtained on daily and composite MODIS images. The main advantage of the proposed technique with respect to others is that the inclusion of the spectral term in weight calculation allows an accurate estimate of sub-pixel NDVI time series even for land-cover classes characterized by large and rapid spatial variations in their spectral properties.     

Finding the most unusual time series subsequence: algorithms and applications

In this work we introduce the new problem of finding time seriesdiscords. Time series discords are subsequences of longer time series that are maximally different to all the rest of the time series subsequences. They thus capture the sense of the most unusual subsequence within a time series. While discords have many uses for data mining, they are particularly attractive as anomaly detectors because they only require one intuitive parameter (the length of the subsequence) unlike most anomaly detection algorithms that typically require many parameters. While the brute force algorithm to discover time series discords is quadratic in the length of the time series, we show a simple algorithm that is three to four orders of magnitude faster than brute force, while guaranteed to produce identical results. We evaluate our work with a comprehensive set of experiments on diverse data sources including electrocardiograms, space telemetry, respiration physiology, anthropological and video datasets. Eamonn Keogh is an Assistant Professor of computer science at the University of California, Riverside. His research interests include data mining, machine learning and information retrieval. Several of his papers have won best paper awards, including papers at SIGKDD and SIGMOD. Dr. Keogh is the recipient of a 5-year NSF Career Award for “Efficient discovery of previously unknown patterns and relationships in massive time series databases.” Jessica Lin is an Assistant Professor of information and software engineering at George Mason University. She received her Ph.D. from the University of California, Riverside. Her research interests include data mining and informational retrieval. Sang-Hee Lee is a paleoanthropologist at the University of California, Riverside. Her research interests include the evolution of human morphological variation and how different mechanisms (such as taxonomy, sex, age, and time) explain what is observed in fossil data. Dr. Lee obtained her Ph.D. in anthropology from the University of Michigan in 1999. Helga Van Herle is an Assistant Clinical Professor of medicine at the Division of Cardiology of the Geffen School of Medicine at UCLA. She received her M.D. from UCLA in 1993; completed her residency in internal medicine at the New York Hospital (Cornell University, 1993–1996) and her cardiology fellowship at UCLA (1997–2001). Dr. Van Herle holds a M.Sc. in bioengineering from Columbia University (1987) and a B.Sc. in Chemical Engineering from UCLA (1985)     

Traffic conduction analysis model with time series rule mining

The traffic density situation in a traffic network, especially traffic congestion, exhibits characteristics similar to thermodynamic heat conduction, e.g., the traffic congestion in one section can be conducted to other adjacent sections of the traffic network sequentially. Analyzing this conduction facilitates the forecasting of future traffic situation; therefore, a navigation system can reduce traffic congestion and improve transportation mobility. This study describes a methodology for traffic conduction analysis modeling based on extracting important time-related conduction rules using a type of evolutionary algorithm named Genetic Network Programming (GNP). The extracted rules construct a useful model for forecasting future traffic situations and analyzing traffic conduction. The proposed methodology was implemented and experimentally evaluated using a large scale real-time traffic simulator, SOUND/4U.     

A spatio-temporal decomposition based deep neural network for time series forecasting

Spatio-temporal problems arise in a broad range of applications, such as climate science and transportation systems. These problems are challenging because of unique spatial, short-term and long-term patterns, as well as the curse of dimensionality. In this paper, we propose a deep learning framework for spatio-temporal forecasting problems. We explicitly design the neural network architecture for capturing various types of spatial and temporal patterns, and the model is robust to missing data. In a preprocessing step, a time series decomposition method is applied to separately feed short-term, long-term and spatial patterns into different components of the neural network. A fuzzy clustering method finds clusters of neighboring time series residuals, as these contain short-term spatial patterns. The first component of the neural network consists of multi-kernel convolutional layers which are designed to extract short-term features from clusters of time series data. Each convolutional kernel receives a single cluster of input time series. The output of convolutional layers is concatenated by trends and followed by convolutional-LSTM layers to capture long-term spatial patterns. To have a robust forecasting model when faced with missing data, a pretrained denoising autoencoder reconstructs the model’s output in a fine-tuning step. In experimental results, we evaluate the performance of the proposed model for the traffic flow prediction. The results show that the proposed model outperforms baseline and state-of-the-art neural network models.     

Assessment of biases in MODIS surface reflectance due to Lambertian approximation

Using MODIS data and the AERONET-based Surface Reflectance Validation Network (ASRVN), this work studies errors of MODIS atmospheric correction caused by the Lambertian approximation. On one hand, this approximation greatly simplifies the radiative transfer model, reduces the size of the look-up tables, and makes operational algorithm faster. On the other hand, uncompensated atmospheric scattering caused by Lambertian model systematically biases the results. For example, for a typical bowl-shaped bidirectional reflectance distribution function (BRDF), the derived reflectance is underestimated at high solar or view zenith angles, where BRDF is high, and is overestimated at low zenith angles where BRDF is low. The magnitude of biases grows with the amount of scattering in the atmosphere, i.e., at shorter wavelengths and at higher aerosol concentration. The slope of regression of Lambertian surface reflectance vs. ASRVN bidirectional reflectance factor (BRF) is about 0.85 in the red and 0.6 in the green bands. This error propagates into the MODIS BRDF/albedo algorithm, slightly reducing the magnitude of overall reflectance and anisotropy of BRDF. This results in a small negative bias of spectral surface albedo. An assessment for the GSFC (Greenbelt, USA) validation site shows the albedo reduction by 0.004 in the near infrared, 0.005 in the red, and 0.008 in the green MODIS bands.     

Reconstruction of missing data in multidimensional time series by fuzzy similarity

The present work addresses the problem of missing data in multidimensional time series such as those collected during operational transients in industrial plants. We propose a novel method for missing data reconstruction based on three main steps: (1) computing a fuzzy similarity measure between a segment of the time series containing the missing data and segments of reference time series; (2) assigning a weight to each reference segment; (3) reconstructing the missing values as a weighted average of the reference segments. The performance of the proposed method is compared with that of an Auto Associative Kernel Regression (AAKR) method on an artificial case study and a real industrial application regarding shut-down transients of a Nuclear Power Plant (NPP) turbine.     

Mapping canopy damage from understory fires in Amazon forests using annual time series of Landsat and MODIS data

Understory fires in Amazon forests alter forest structure, species composition, and the likelihood of future disturbance. The annual extent of fire-damaged forest in Amazonia remains uncertain due to difficulties in separating burning from other types of forest damage in satellite data. We developed a new approach, the Burn Damage and Recovery (BDR) algorithm, to identify fire-related canopy damages using spatial and spectral information from multi-year time series of satellite data. The BDR approach identifies understory fires in intact and logged Amazon forests based on the reduction and recovery of live canopy cover in the years following fire damages and the size and shape of individual understory burn scars. The BDR algorithm was applied to time series of Landsat (1997-2004) and MODIS (2000-2005) data covering one Landsat scene (path/row 226/068) in southern Amazonia and the results were compared to field observations, image-derived burn scars, and independent data on selective logging and deforestation. Landsat resolution was essential for detection of burn scars < 50 ha, yet these small burns contributed only 12% of all burned forest detected during 1997-2002. MODIS data were suitable for mapping medium (50-500 ha) and large (> 500 ha) burn scars that accounted for the majority of all fire-damaged forests in this study. Therefore, moderate resolution satellite data may be suitable to provide estimates of the extent of fire-damaged Amazon forest at a regional scale. In the study region, Landsat-based understory fire damages in 1999 (1508 km²) were an order of magnitude higher than during the 1997-1998 El Niño event (124 km² and 39 km², respectively), suggesting a different link between climate and understory fires than previously reported for other Amazon regions. The results in this study illustrate the potential to address critical questions concerning climate and fire risk in Amazon forests by applying the BDR algorithm over larger areas and longer image time series.