Abstract

Given the budgetary constraints that prevented the implementation of a national agricultural census during the 1980s, Colombia adopted an approach based on area sample surveys as an alternative for producing agricultural statistics. In this context, the construction of an area sampling frame based on the photointerpretation of aerial imagery began, which served as the basis for the National Agricultural Survey (ENA) from 1995 to 2016.
Subsequently, with the availability of updated and georeferenced information from the Third National Agricultural Census (CNA 2014), a dual sampling frame was developed, composed of a list frame and an optimized area frame. This redesign included the redefinition of strata, the adjustment of the size of the primary sampling units (PSUs), and the precise delineation of study domains. The 2017 ENA sample design incorporated these improvements, enabling a shift from a two-stage to a single-stage sampling scheme, which reduced sampling errors and improved estimator efficiency.
In implementing the dual frame, differentiated methodological criteria were established: the list frame targeted large-scale agricultural holding, mainly located in flat areas where agro-industrial farms and intensive livestock operations predominate, while the area frame focused on small APUs, mostly located in mountainous regions where land fragmentation and limited accessibility hinder the updating of administrative records.
During the design and construction of the dual frame, remote sensing tools, geographic information systems (GIS), and high-resolution satellite imagery were integrated for the delineation of geographic units, land-use identification, and cartographic validation. Likewise, statistical and operational criteria were applied for stratification, determining the optimal cluster size within each stratum, and defining inclusion/exclusion thresholds in the list frame, thus ensuring the statistical coverage and representativeness of the sample.
Regarding the sample design, different selection strategies were analyzed—from Simple Random Sampling (SRS) to Probability Proportional to Size (PPS) schemes—assessing their operational feasibility and their impact on the precision of the estimates. Mechanisms implemented to address field contingencies (such as absence of respondents, replacements, and non-response) were also documented, as well as the adjustment and calibration methods used during the estimation stage.
Finally, the adopted methodology for the maintenance and periodic updating of the sampling frame and sample will be presented, including strategies for field verification, cartographic updating, and cross-validation with administrative records and satellite sources.