Photo credit: IFPRI

To every crop there is a season with optimal times for planting, sowing, and harvesting. (And you thought we were going to break into Pete Seeger.) Precisely documenting these times in a Crop Calendar through time and space, however, remains a challenge, particularly at small scales such as the pixel. But pixels are what HarvestChoice is all about.

A Crop Calendar reflects the seasonal progression of a particular crop or cropping system in a specific geographical area from planting to harvest (for an example, check out FAO’s Crop Calendars). Traditional Crop Calendars are based on phenology data gathered from household surveys, national censuses, and the extent of cropping systems or agro-ecological zones. This broad-brushed treatment fails to capture regional variation and is time consuming, labor intensive, and a difficult process to replicate. Using non-traditional methods such as remote sensing data, Crop Calendars can be disaggregated at the pixel level and could be particularly useful for developing regions such as sub-Saharan Africa where data are typically scarce and the mosaic of farm parcels are heterogeneous at small scales. More localized information is important to stakeholders along the seed chain: farmers, input and seed suppliers, processors, and traders – all of whom depend on the right timing at the right location.

Global surface images taken from satellites are capturing the lifecycle of vegetation cover at fine spatial resolutions (30 meters to 1-2 kilometers) around the globe. Through the changing shades of greens and browns as plants grow and mature, remotely-sensed imagery can shed light on the timing for key cropping events such as planting, harvesting, and maximum growth, as well as measure cropping intensity (e.g., number of growing seasons per year) and total length of the growing period. Over the decades, vegetative index data from AVHRR, MODIS, and SPOT VEG have been used to generate crop phenology products.

With the help of remote sensing products, GIS, and spatial analysis, Zhe Guo and colleagues from HarvestChoice and the International Food Policy Research Institute (IFPRI) designed a methodology to harmonize and geo-reference crop phenology data, resulting in the first generation of Crop Calendar products at the pixel scale (1 km2) for sub-Saharan Africa1.

About the data and methodology

The methodology can be broken down into five main steps:

  1. Download time-series NDVI (Normalized Difference Vegetation Index2) datasets from NASA’s MODIS 16-Date Composite and create a data mosaic over Africa (1 km resolution) using ArcGIS tools (Fig 1).
  2. Rank each pixel for quality and reliability of the NDVI data using the corresponding ancillary quality data layers. Weighted values are assigned to represent ‘good’ data (clear weather begets clear images and reliable data), marginal data, and unreliable data (snow/ice on the ground or continuous cloud cover limits visibility from space and compromises data quality).
  3. Reclassify the values in the quality layer to harmonize with the next step and reassign weights to each pixel. The most reliable pixels are assigned the value of 1 and the most unreliable (cloudiest) a 0. Mixed pixels are weighted in between and assigned a value of 0.5.
  4. Measure and fit the green area (the vegetative index) under the curve (Fig 2). Using one of the curve-fitting methods provided in the TIMESAT package (in this case, the Adaptive Savitzky-Golay filter) eliminates ‘noise’ and smooths the NDVI curve for better analysis. Pixels assigned a zero (see step 3) bear no weight while good-data pixels carry the most influence in the analysis.
  5. Extract and explore seasonality parameters from the NDVI datasets using TIMESAT software and generate Crop Calendar products at the pixel level (1 km) for sub-Saharan Africa (Fig 3).

figure 1_1.png

figure 2-1.png

figure 3_2.png

The curve fitting procedure in the methodology we describe is a necessary step to reduce noise caused by cloud effects. A look at the above map (Fig 3), however, indicates a data challenge along the coast of West Africa since clouds cover the bird’s eye view during much of the growing season. To overcome this, the next step is to investigate the possibility of using other satellite imagery products, such as SPOT Veg NDVI, to supplement MODIS. A reliable land cover map could also help to identify the extent of cropland in sub-Saharan Africa.

Meanwhile, Crop Calendars can be developed in one of two ways: through coarser, more traditional methods that rely on household surveys, country census data, and ground truthing; or via modern methods such as we describe using remotely-sensed, time-series data. Both methods have their advantages and limitations, depending on the nature of the region and the quality of the information needed. A next step in this study is to evaluate Crop Calendar products derived from both ways and design a strategy to geo-reference and harmonize the best of them by using spatial extensions. By combining phenology products derived through remote sensing and geo-referenced tabulation data, the quality of Crop Calendar products could substantially improve for sub-Saharan Africa and better inform stakeholders from suppliers and growers to marketers and traders.

1Guo, Zhe  (2013). An effort to retrieve crop phenology information from NDVI time series in Africa. Paper presented at the International Geoscience and Remote Sensing Symposium, Melbourne, Australia.

2 From Wikipedia: The Normalized Difference Vegetation Index (NDVI) is a simple graphical indicator that can be used to analyze remote sensing measurements, typically but not necessarily from a space platform, and assess whether the target being observed contains live green vegetation or not.


HarvestChoice, 2013. "A Time to Sow: How HarvestChoice Is Mapping the Life Cycle of Crops ." International Food Policy Research Institute, Washington, DC., and University of Minnesota, St. Paul, MN. Available online at

Sep 27, 2013