The LandTrendr (LT) algorithm has been used widely for analysis of change in Landsat spectral time series data, but requires significant pre-processing, data management, and computational resources, and is only accessible to the community in a proprietary programming language (IDL). Here, we introduce LT for the Google Earth Engine (GEE) platform. The GEE platform simplifies pre-processing steps, allowing focus on the translation of the core temporal segmentation algorithm. Temporal segmentation involved a series of repeated random access calls to each pixel’s time series, resulting in a set of breakpoints (“vertices”) that bound straight-line segments. The translation of the algorithm into GEE included both transliteration and code analysis, resulting in improvement and logic error fixes. At six study areas representing diverse land cover types across the U.S., we conducted a direct comparison of the new LT-GEE code against the heritage code (LT-IDL). The algorithms agreed in most cases, and where disagreements occurred, they were largely attributable to logic error fixes in the code translation process. The practical impact of these changes is minimal, as shown by an example of forest disturbance mapping. We conclude that the LT-GEE algorithm represents a faithful translation of the LT code into a platform easily accessible by the broader user community.

Disturbance is a critical ecological process in forested systems, and disturbance maps are important for understanding forest dynamics. Landsat data are a key remote sensing dataset for monitoring forest disturbance and there recently has been major growth in the development of disturbance mapping algorithms. Many of these algorithms take advantage of the high temporal data volume to mine subtle signals in Landsat time series, but as those signals become subtler, they are more likely to be mixed with noise in Landsat data. This study examines the similarity among seven different algorithms in their ability to map the full range of magnitudes of forest disturbance over six different Landsat scenes distributed across the conterminous US. The maps agreed very well in terms of the amount of undisturbed forest over time; however, for the ~30% of forest mapped as disturbed in a given year by at least one algorithm, there was little agreement about which pixels were affected. Algorithms that targeted higher-magnitude disturbances exhibited higher omission errors but lower commission errors than those targeting a broader range of disturbance magnitudes. These results suggest that a user of any given forest disturbance map should understand the map’s strengths and weaknesses (in terms of omission and commission error rates), with respect to the disturbance targets of interest.

In November 2016, after 52 years of armed conflict, the Colombian government and the primary rebel group, the FARC (Fuerzas Armadas Revolucionarias de Colombia) reached a peace agreement. The agreement incorporated three changes to institutions governing forest land occupation and use: (a) the demobilization of FARC from forested places, (2) the future distribution of legal land titles and new road construction into forests, and (3) the eradication of illicit crops. However, we document unprecedented rates of forest disturbance in the months following the peace agreement in biodiversity hotspots across the country. Are the declaration of peace and the increased rates of forest disturbance related? Here, we present the first systematic assessment of the impact of the Colombian peace agreement on forest disturbance. Focusing on the Andes-Amazon Transition Belt (AATB), we used automated satellite image disturbance detection methods and ethnographic data to quantify and interpret forest cover change from 2010 to 2018 that span wartime, peace negotiation, and post-peace agreement stages. Our findings indicate that during the post-peace agreement period (2017–2018), the area of forest disturbance increased by 50% (about 238 000 ha) across the AATB in comparison with the four-year peace negotiation stage (2013–2016); these changes reflect the end of FARC-led gunpoint conservation in the region. Forest disturbance also spread deeper into the Amazon watershed and increased in area by 187% within the AATB’s protected areas. We find that following the peace agreement and the withdrawal of FARC, key actors (viz. drug cartels, large landowners, campesinos and dissidents) with expectations of favorable land tenure policies swept into the region; this led to increases in large-scale cattle ranching, coca cultivation dispersal, and speculative illegal land markets each of which contributed to the widespread forest disturbance that we mapped. The rapid increase in forest disturbance occurred despite the interest of the international community in promoting forest conservation initiatives in the AATB and Colombia’s existing conservation and land titling frameworks for public lands. Our findings underscore the need for conservation strategies sensitive to rapid institutional and demographic changes in the course of the peace agreement to prevent forests from becoming an unexpected casualty of premature and unstable peace.

This paper presents a prototype Carbon Monitoring System (CMS) developed to produce regionally unbiased annual estimates of aboveground biomass (AGB). Our CMS employed a bottom-up, two-step modeling strategy beginning with a spatially and temporally biased sample: project datasets collected and contributed by US Forest Service (USFS) and other forestry stakeholders in 29 different project areas in the northwestern USA. Plot-level AGB estimates collected in the project areas served as the response variable for predicting AGB primarily from lidar metrics of canopy height and density (R² = 0.8, RMSE = 115 Mg ha⁻¹, Bias = 2 Mg ha⁻¹). This landscape model was used to map AGB estimates at 30 m resolution where lidar data were available. A stratified random sample of AGB pixels from these landscape-level AGB maps then served as training data for predicting AGB regionally from Landsat image time series variables processed through LandTrendr. In addition, climate metrics calculated from downscaled 30 year climate normals were considered as predictors in both models, as were topographic metrics calculated from elevation data; these environmental predictors allowed AGB estimation over the full range of observations with the regional model (R² = 0.8, RMSE = 152 Mg ha⁻¹, Bias = 9 Mg ha⁻¹), including higher AGB values (>400 Mg ha⁻¹) where spectral predictors alone saturate. For both the landscape and regional models, the machine-learning algorithm Random Forests (RF) was consistently applied to select predictor variables and estimate AGB. We then calibrated the regional AGB maps using field plot data systematically collected without bias by the national Forest Inventory and Analysis (FIA) Program. We found both our project landscape and regional, annual AGB estimates to be unbiased with respect to FIA estimates (Biases of 1% and 0.7%, respectively) and conclude that they are well suited to inform forest management and planning decisions by our contributing stakeholders. Social media abstract Lidar-based biomass estimates can be upscaled with Landsat data to regionally unbiased annual maps.

The Landsat record represents an amazing resource for discovering land-cover changes and monitoring the Earth’s surface. However, making the most use of the available data, especially for automated applications ingesting thousands of images without human intervention, requires a robust screening of cloud and cloud-shadow, which contaminate clear views of the land surface. We constructed a deep convolutional neural network (CNN) model to semantically segment Landsat 8 images into regions labeled clear-sky, clouds, cloud-shadow, water, and snow/ice. For training, we constructed a global, hand-labeled dataset of Landsat 8 imagery; this labor-intensive process resulted in the uniquely high-quality dataset needed for the creation of a high-quality model. The CNN model achieves results on par with the ability of human interpreters, with a total accuracy of 97.1%, omitting only 3.5% of cloud pixels and 4.8% of cloud shadow pixels, which is seven to eight times fewer missed pixels than the masks distributed with the imagery. By harnessing the power of advanced tensor processing units, the classification of full images is I/O bound, making this approach a feasible method to generate masks for the entire Landsat 8 archive.