Evan B. Brooks

On-the-Fly Massively Multitemporal Change Detection Using Statistical Quality Control Charts and Landsat Data

One challenge to implementing spectral change detection algorithms using multitemporal Landsat data is that key dates and periods are often missing from the record due to weather disturbances and lapses in continuous coverage. This paper presents a method that utilizes residuals from harmonic regression over years of Landsat data, in conjunction with statistical quality control charts, to signal subtle disturbances in vegetative cover. These charts are able to detect changes from both deforestation and subtler forest degradation and thinning. First, harmonic regression residuals are computed after fitting models to interannual training data. These residual time series are then subjected to Shewhart X-bar control charts and exponentially weighted moving average charts. The Shewhart X-bar charts are also utilized in the algorithm to generate a data-driven cloud filter, effectively removing clouds and cloud shadows on a location-specific basis. Disturbed pixels are indicated when the charts signal a deviation from data-driven control limits. The methods are applied to a collection of loblolly pine ( Pinus taeda) stands in Alabama, USA. The results are compared with stands for which known thinning has occurred at known times. The method yielded an overall accuracy of 85%, with the particular result that it provided afforestation/deforestation maps on a per-image basis, producing new maps with each successive incorporated image. These maps matched very well with observed changes in aerial photography over the test period. Accordingly, the method is highly recommended for on-the-fly change detection, for changes in both land use and land management within a given land use.

Fitting the Multitemporal Curve: A Fourier Series Approach to the Missing Data Problem in Remote Sensing Analysis

With the advent of free Landsat data stretching back decades, there has been a surge of interest in utilizing remotely sensed data in multitemporal analysis for estimation of biophysical parameters. Such analysis is confounded by cloud cover and other image-specific problems, which result in missing data at various aperiodic times of the year. While there is a wealth of information contained in remotely sensed time series, the analysis of such time series is severely limited due to the missing data. This paper illustrates a technique which can greatly expand the possibilities of such analyses, a Fourier regression algorithm, here on time series of normalized difference vegetation indices (NDVIs) for Landsat pixels with 30-m resolution. It compares the results with those using the spatial and temporal adaptive reflectance fusion model (STAR-FM), a popular approach that depends on having MODIS pixels with resolutions of 250 m or coarser. STAR-FM uses changes in the MODIS pixels as a template for predicting changes in the Landsat pixels. Fourier regression had an R2 of at least 90% over three quarters of all pixels, and it had the highest RPredicted2 values (compared to STAR-FM) on two thirds of the pixels. The typical root-mean-square error for Fourier regression fitting was about 0.05 for NDVI, ranging from 0 to 1. This indicates that Fourier regression may be used to interpolate missing data for multitemporal analysis at the Landsat scale, especially for annual or longer studies.

Using Window Regression to Gap-Fill Landsat ETM+ Post SLC-Off Data

The continued development of algorithms using multitemporal Landsat data creates opportunities to develop and adapt imputation algorithms to improve the quality of that data as part of preprocessing. One example is de-striping Enhanced Thematic Mapper Plus (ETM+, Landsat 7) images acquired after the Scan Line Corrector failure in 2003. In this study, we apply window regression, an algorithm that was originally designed to impute low-quality Moderate Resolution Imaging Spectroradiometer (MODIS) data, to Landsat Analysis Ready Data from 2014–2016. We mask Operational Land Imager (OLI; Landsat 8) image stacks from five study areas with corresponding ETM+ missing data layers, using these modified OLI stacks as inputs. We explored the algorithm’s parameter space, particularly window size in the spatial and temporal dimensions. Window regression yielded the best accuracy (and moderately long computation time) with a large spatial radius (a 7 × 7 pixel window) and a moderate temporal radius (here, five layers). In this case, root mean square error for deviations from the observed reflectance ranged from 3.7–7.6% over all study areas, depending on the band. Second-order response surface analysis suggested that a 15× 15 pixel window, in conjunction with a 9-layer temporal window, may produce the best accuracy. Compared to the neighborhood similar pixel interpolator gap-filling algorithm, window regression yielded slightly better accuracy on average. Because it relies on no ancillary data, window regression may be used to conveniently preprocess stacks for other data-intensive algorithms.

A mid‐century ecological forecast with partitioned uncertainty predicts increases in loblolly pine forest productivity

Ecological forecasting of forest productivity involves integrating observations into a process-based model and propagating the dominant components of uncertainty to generate probability distributions for future states and fluxes. Here, we develop a forecast for the biomass change in loblolly pine (Pinus taeda) forests of the southeastern United States and evaluate the relative contribution of different forms of uncertainty to the total forecast uncertainty. Specifically, we assimilated observations of carbon and flux stocks and fluxes from sites across the region, including global change experiments, into a forest ecosystem model to calibrate the parameter distributions and estimate the process uncertainty (i.e., model structure uncertainty revealed in the residuals of the calibration). Using this calibration, we forecasted the change in biomass within each 12-digit Hydrologic (H12) unit across the native range of loblolly pine between 2010 and 2055 under the Representative Concentration Pathway 8.5 scenario. Averaged across the region, productivity is predicted to increase by a mean of 31% between 2010 and 2055 with an average forecast 95% quantile interval of ±15 percentage units. The largest increases were predicted in cooler locations, corresponding to the largest projected changes in temperature. The forecasted mean change varied considerably among the H12 units (3-80% productivity increase), but only units in the warmest and driest extents of the loblolly pine range had forecast distributions with probabilities of a decline in productivity that exceeded 5%. By isolating the individual components of the forecast uncertainty, we found that ecosystem model process uncertainty made the largest individual contribution. Ecosystem model parameter and climate model uncertainty had similar contributions to the overall forecast uncertainty, but with differing spatial patterns across the study region. The probabilistic framework developed here could be modified to include additional sources of uncertainty, including changes due to fire, insects, and pests: processes that would result in lower productivity changes than forecasted here. Overall, this study presents an ecological forecast at the ecosystem management scale so that land managers can explicitly account for uncertainty in decision analysis. Furthermore, it highlights that future work should focus on quantifying, propagating, and reducing ecosystem model process uncertainty.