Remote sensing has the power to transform and complement traditional approaches to impact evaluation. With the emergence of new technologies and the deployment of advanced sensors aboard satellites, there is an increase in the use of satellite imagery to measure impact in low- and middle-income countries across the world. Some recent examples of using remote sensing for measuring developmental outcomes include night-time lights as a proxy for measuring economic activity, and day-time satellite images for measuring monthly forest loss and land-usage classification in previously inaccessible areas.

However, only a handful of organizations have applied these methods to their work thus far. Lack of familiarity with the tools and techniques, resource constraints, and lack of practical guidance available for informing that work are some of the contributing factors for that. This post provides guidance on how to get started with remote sensing for those considering using satellite imagery to measure the impact(s) of development interventions.

Variable identification

The first step is to identify a geospatial or proxy variable of interest to inform the evaluation which in turn is informed by the theory of change and evaluation questions. For example, night-time light radiance is highly correlated with the economic activity of a region. Therefore, if you are interested in measuring economic growth changes after an intervention, you may look at the night-time light radiance. However, in some papers researchers have taken day-time and night-time imagery together to give a better estimate of economic activity. One example here by Neal Jean et al combines satellite imagery and machine learning to predict poverty. Another application is measuring agricultural crop yields. In this case, the health of a crop can be measured by the Enhanced Vegetation Index or other measures like the NDVI, GNDVI, or SIF. Since there is no single index that is suitable for all purposes, you can search the Internet and academic databases to see what other researchers have used and find the best one for your research question.

Decide on source, availability, and resolution of satellite image

After identifying the possible variables, you need to decide on which data source to use for those variables. There are many public and commercial satellites available. Decide which one is appropriate for your research question.

Two of the popular publicly available sources are SENTINEL and LANDSAT satellites. LANDSAT offers extensive historical data but has a lower resolution. SENTINEL provides higher-resolution data, but only after 2015. Two of the commercial sources of satellite data are MAXAR and OneAtlas. They have an extremely fine resolution but charge a premium for accessing their services. You need to focus on the research question and decide what the ideal resolution of the satellite image needs to be for answering the question. For example, measuring agricultural yield in a country with a small average land-holding size would require a high-resolution image. The same could be said about the availability of data. Research requiring high-resolution images before 1990 may not be possible because the technology was not mature at that time. At this point, it would be helpful to go through the previous work done by research teams to know the methods, sources of data, and rationale for using a specific method.

Exploratory Data Analysis

Exploratory data analysis serves as the first step in understanding the data and seeing previously-unseen patterns. Explore the datasets by visualizing the data and verifying that you can measure your outcome of interest, check the time period, check the level of consistency using available satellite bands. Additionally, this exercise will help you understand and restate your assumptions, detect outliers, and understand patterns in the data. This process may help reframe or bring further clarity to the research questions. Some of the tools which can be employed at this stage include QGIS, ArcGIS, or any other spatial software you may be using. We recommend QGIS because it’s open-source and supported by a large community. Google Earth Engine is a cloud-based platform that is helpful for exploratory data analysis, as well as calculating some geospatial variables.

Of course, this blog post only discusses the possibilities remote sensing offers for measurement – but planning an impact evaluation requires many more steps: defining a counterfactual, developing an identification strategy, planning data analysis, and so on. At 3ie, we have been incorporating remote sensing into our own projects and working to build capacity in the Global South. With support from FCDO, 3ie is developing a hands-on tutorial with introductory discussions on theory along with real-time coding exercises in Python. Through the tutorial, beginners can learn practical considerations while applying remote sensing, applications of remotely sensed satellite imagery, and calculation of various indices used in agriculture and forestry interventions. Our goal is to empower our partners to learn and apply the most up-to-date remote sensing techniques in their research projects and work settings.