Abstract:Ensuring food security has been the top priority of many regions, particularly in developing countries in Africa. In recent decades, increasing population, together with growing food demands, have put great pressure on the world's food production. Long-term, up-to-date, annual cropland mapping at high resolution (i.e., at tens-of-metre levels) is in urgent demand for tracking spatial and temporal patterns of cropland change. However, because of the difficulty of capturing seasonality and flexible cropping systems, few studies have focused on understanding the dynamics of cropland using Landsat data in Africa. Here, we propose a new method of updating annual cropland mapping using a change-detection approach and post-classification to improve on traditional bi-temporal change vector analysis. Three Landsat footprints in Africa were selected (Egypt, Ethiopia and South Africa) as our study areas based on their different cropping systems and field sizes. The potential annual change areas were detected by employing multiple indices and thresholds in reference and long-term annual composite Landsat images. Next, map updates were conducted in the potential change pixels using random forest-based classification. Different training sample metrics were used (seasonal and annual samples) and compared in the classification step. The long-term cropland mapping accuracies for these three sites ranged from 88.04% to 94.30% (Egypt), 76.28% to 82.88% (Ethiopia) and 56.52% to 67.53% (South Africa). The results showed improvements in the accuracy and consistency of updating the annual cropland information using change-detection approaches, accounting for accuracy increases of 2.40%, 10.62% and 0.55% compared with a yearly cropland mapping approach in our previous research. The best results using annual samples extracted from the same season with the classified images supported the use of annual and growing samples in long-term annual mapping. Overall, a common trend of cropland expansion in all three sites was revealed, with an increase rate of 10.06, 3.73 and 1.35 kha/year in Egypt, Ethiopia and South Africa, respectively. The results indicated a rapid increasing pattern from bare land (desert) to irrigated systems (Egyptian site) but smaller and stable cropland changes in smallholder and farming-pastoral ecotones (Ethiopian and South African site), where limited land was still available for an expansion of agricultural area. This study highlights the potential application of time-series Landsat data in documenting and contributing missing cropland distribution information required for assessing and solving food security in Africa.

Tracking spatiotemporal dynamics of crop-specific areas through machine learning and statistics disaggregating

CROPUP: Historical Products Are All You Need? an End-to-end Cross-Year Crop Map Updating Framework Without the Need for in Situ Samples

Mapping corn dynamics using limited but representative samples with adaptive strategies

Croplayer-China: A 2-Meter Resolution Cropland Map of China Based on Active Learning of Segmentation with Mapbox and Google Satellite Imagery

Multi-year mapping of cropping systems in regions with smallholder farms from Sentinel-2 images in Google Earth engine

Generating plausible crop distribution maps for Sub-Saharan Africa using a spatially disaggregated data fusion and optimization approach

Development of a 10-m resolution maize and soybean map over China: Matching satellite-based crop classification with sample-based area estimation

Incorporating Multi-Temporal Remote Sensing and a Pixel-Based Deep Learning Classification Algorithm to Map Multiple-Crop Cultivated Areas

Pre-Constrained Machine Learning Method for Multi-Year Mapping of Three Major Crops in a Large Irrigation District

Large-scale crop mapping from multi-source optical satellite imageries using machine learning with discrete grids

A New Method for Crop Type Mapping at the Regional Scale Using Multi-Source and Multi-Temporal Sentinel Imagery

Tracking annual cropland changes from 1984 to 2016 using time-series Landsat images with a change-detection and post-classification approach: Experiments from three sites in Africa

Mapping Crop Cycles in China Using MODIS-EVI Time Series

Large-Scale Crop Mapping Based on Multisource Remote Sensing Intelligent Interpretation: A Spatiotemporal Data Cubes Approach

Mapping Maize Area in Heterogeneous Agricultural Landscape with Multi-Temporal Sentinel-1 and Sentinel-2 Images Based on Random Forest

Crop Mapping Using the Historical Crop Data Layer and Deep Neural Networks: A Case Study in Jilin Province, China

Early- and in-season crop type mapping without current-year ground truth: Generating labels from historical information via a topology-based approach

Improving Parcel-Level Mapping of Smallholder Crops from VHSR Imagery: An Ensemble Machine-Learning-Based Framework

Validation and refinement of cropland data layer using a spatial-temporal decision tree algorithm

Mapping annual 10-m soybean cropland with spatiotemporal sample migration

High Resolution Mapping of Cropping Cycles by Fusion of Landsat and MODIS Data