Measuring, tracking and monitoring agricultural production in developing countries.

Challenges in measuring agricultural land

The challenge poised to development agencies is not the acceptance of how critical land utilization indicators are but rather how to measure them accurately. A lack of accurate and current maps coupled with poor land ownership frameworks, remote areas and the small size of agricultural holdings makes the task difficult. Upwards of several thousand, geographically spread and often of less than a hectare are areas not easy to measure, track and monitor over intervention life.

Typically these constraints lead to an approach in the field of estimating rather than measuring. Approaches vary from organization to organization and from field worker to field worker. A project field worker counting out the paces of the land under production , relying on the land owners estimate or, in perhaps the most sophisticated cases, using a surveyor's wheel are examples of the general approach.

Accuracy of measuring agricultural land

All of these methods are inherently inaccurate. Typically for development projects farm sizes are small at between 0.5 and 1.25 hectares, often irregular in shape and so not only are they difficult to estimate but also, because of the small size, sensitive to error. FAO agricultural census put African farm size at a mean of 1.28 Ha with 78% of all farms below 2 hectares and with further studies demonstrating a link between farm size and economic development agencies often deal with even smaller holdings.

In response to these challenges some development agencies have turned to the adoption of GPS technology. Even the most basic of hand held GPS units now have accurate area measurement capacity and the land in question can easily and accurately measured by walking its perimeter. The one significant draw back of this approach over previous measurement estimates is time. It takes about 20 – 30 minutes to measure a perimeter of about 1 ha and so for an average program size of over 1000 beneficiaries this would be 500 man hours excluding organization, transport and logistics. And it is with these figures, whilst not exact, that we serve to illustrate the scale of the task.

So whilst historically the issue of land measurement has been a question of accuracy with the adoption of new technologies to address this, the issue is now one of the time resources necessary to use the technology properly. This problem is compounded by the consideration that often land usage indicators are both seasonal and compressed within a certain time scale. With land cultivation, planting, and yield indicators being temporal to a month or less it is hard to see how a hard pressed monitoring and evaluation department can allocate resource here.

History of statistical approach

This is not a unique problem. Indeed since the early 18th century states have been confronted with the very problem of land measurement and the adoption of statistical sampling was a response to this. Manufacturers rapidly followed suit and by measuring only a sample of their production of hundreds of thousands they found they could confidently predict the characteristics of the rest.

Since the early 1800 hundreds when confronted with large population or sample size and a time consuming or difficult measurement task the most common approach is to adopt statistical sampling. Originally involving only measurements of country states by the 19th century Francis Galton had applied the first concept of statistical correlation to human heights. Thereafter manufacturers rapidly followed and the statistical sampling of production became both widespread in use and conventional as a way to accurately measure the characteristics of hundreds of items. By measuring only a sample of their production they found they could reasonably and with confidence predict the characteristics of the rest. And it was the cost effectiveness of this that made statistical sampling both widespread and accepted as the normal way to measure large groups.

Steps for applying statistics to land measurement

Applying this to land use indicators in the development space is fairly straight forward and requires the answer to the question- within a certain margin of error, what is the average farm size and how confident can be be that this is an accurate measure of all the farms in the project. This is a relatively straight foward four step process and in this example we will look to accurately find our projects's average farm size, with a margin of error of 0.05 ha and with a confidence that this is 98% accurate.

  • Step One: Divide the confidence interval by 2 and find the nearest value in the Z-table (normal distribution table). So this would be .98/2 =.49 which corresponds to 2.33 in the Z-table.
  • Step Two: Accurately measure 20 – 30 farms with gps and ensure that these are relatively typical of the types of farms you project will deal with and establish the standard deviation of this group. And so for this example we will assume our farm sample gave a standard deviation of 0.356.
  • Step Three: Multiple step one with step two which would be 2.33 * 0.356= 0.8295.
  • Step Four: Now divide by the margin of error which is our case is 0.05 ha giving 0.8295/0.05 = 16.58.
  • Step Five: Square step four to get the number of farms you need to measure to get an accurate sample size. So this is 16.58 * 16.58 = 275.
  • Step Six: As the number of farms we need to measure to be 98% confidence we have an accurate average is greater than our original sample of 20 – 30 farms steps two five need to be repeated every time a new farm is added to the sample.


The above approach has been tested recently in Northern Mozambique on a soya production initiative with over 4000 farmers. And the figures above were taken from this case study. So in oder to accurately measure 4000 small, irregular and remote farms accurately we can see we simply need to measure around 300.