The 21st century has witnessed an increase in the availability and use of satellite images to capture changes in landscape patterns through time. You may have already been exposed to classical change detection analysis, which is a type of monitoring in which changes in landscape patterns are quantified from satellite imagery between few snapshots in time. Change detection analysis in this way is insufficient however when the processes under investigation are highly dynamic, e.g., crop rotation and ecosystem disturbances/recovery. Such cases require continuous monitoring of satellite images at frequent intervals with time series analysis (TSA). Continuous satellite image data, referred to as Satellite Image Time Series (SITS) in this course, are used to monitor dynamic processes. Ecological indicators derived from SITS capture landscape patterns consistently at frequent intervals, which enable researchers and practioners alike to detect both abrupt or seasonal changes and gradual trends over time. In addition, SITS spanning long periods of time, provide insights into the “drivers of change” and underlying mechanisms governing change. Several satellite image archives are now publicly available with the emergence of relatively inexpensive high-performance cloud computing platforms. Each archive presents unique challenges in terms of acquisition and processing. At the same time, TSA encompasses an array of quantitative approaches to monitor and forecast ecological indicators derived from SITS. These include among others, autoregressive (AR), moving average (MA) and autoregressive moving average (ARMA) models.
The number of SITS and methods for TSA can make environmental monitoring with Earth observation data a daunting task. The overall goal of this course therefore is to provide participants with sufficient knowledge and tools to acquire and process SITS, perform TSA on ecological indicators derived from SITS and design a successful environmental monitoring solution.
We begin the course with a review of key terms and concepts in environmental monitoring with Earth observation. These include: landscape patterns, pattern-generating processes and process interactions. The course continues with the exploitation of SITS to identify eco-physiological traits (ecological indicators) that can be used to monitor landscape patterns through time. With this foundation, we enter the nuts and bolts of the course: how to acquire, process, analyze and evaluate SITS for environmental monitoring. We use the Google Earth Engine cloud computing platform, Breaks For Additive Season and Trend (BFAST) algorithm and Box–Jenkins method for TSA at these stages. Google Earth Engine is a freely-available, convenient and widely used platform to acquire and process SITS. BFAST is an intuitive and widely used algorithm to decompose ecological indicators derived from SITS based on trend, seasonality, cyclical irregularity and structural changes. Box–Jenkins is a classical and systematic method for constructing ARMA models for retrospective time series analysis and forecasting. The ARMA process consists of five stages: (i) model identification; (ii) model estimation; (iii) model validation; (iv) forecasting; and (v) forecasting evaluation. You will then apply your new knowledge and skills to two case studies. The first case study deals with ecosystem detecting tipping points with the time series segmentation tool Landtrendr. The second case study involves modelling and forecasting crop rotations with AR models. Each case study links a problem to an ecological indicator, SITS and method for TSA. For the remainder of the course, participants will form groups to design and execute their own small environmental monitoring solution. Each group will present their findings to the entire class at the end of the course.