Date of Award

Spring 2019

Project Type


Program or Major

Natural Resources and Environmental Studies

Degree Name

Doctor of Philosophy

First Advisor

Russell G Congalton

Second Advisor

Mark J. Ducey

Third Advisor

Richard Smith





Kamini Yadav

University of New Hampshire, May 2019

Global cropland products are continuously being produced at different spatial resolutions using remotely sensed satellite imagery. Recently, with our increased accessibility to higher computing processing, three different cropland extent maps have been developed as a part of Global Food Security-Support Analysis Data (GFSAD) project at three spatial resolutions (i.e., GFSAD1km, GFSAD250m, and GFSAD30m). All cropland maps should be assessed for their accuracy, errors, and uncertainty for various agriculture monitoring applications. However, in previous assessment efforts appropriate assessment strategies have not always been applied and many have reported only a single accuracy measure for the entire world. This research was divided into four components to provide more attention and focus on the accuracy assessment of large area cropland products.

First, a valid assessment of cropland extent maps was performed addressing different strategies, issues, and constraints depending upon various conditions related to the cropland distribution, proportion, and pattern present in each continent. This research focused on dealing with some specific issues encountered when assessing the cropland extent of North America (confined to the United States), Africa and Australia. Continent-specific sampling strategies and accuracy assessments were performed within homogenous regions (i.e., strata) of different continents to ensure that an appropriate reference data set was collected to generate rigorous and valid accuracy results indicative of the actual cropland proportion.

Second, all the three different GFSAD cropland extent maps were assessed using appropriate sampling and collection of a cropland reference data based on the cropland distribution and proportion for different regions in the entire world. In addition to the accuracy assessment, the cropland extent maps developed at the three spatial resolutions were compared to investigate the differences among them and provide guidance for users to select the appropriate resolution given different agriculture field sizes. The comparison of three different GFSAD cropland extent maps was performed based on the similarity of the cropland area proportion (CAP) and landscape clumping at different spatial resolutions to provide specific recommendations for when to apply these maps in different agriculture field sizes.

Third, an issue was discovered with the accuracy assessment of 30m global cropland extent map (i.e., GFSAD30m) in that insufficient samples were collected resulting in an ineffective assessment when the cropland map class was rare as occurred in some regions around the world. This research evaluated the sampling designs for different cropland regions to achieve sufficient samples and effective accuracy of rare cropland map class by comparing the distribution, allocation of samples and accuracy measures. The evaluation of sampling designs demonstrated that the cropland regions of <15% CAP must be sampled with an appropriate stratified sampling combined with a predetermined minimum sample size for each map class.

Finally, the accuracy assessment of all thematic maps (e.g., crop type maps) needs sufficient reference data to conduct a valid assessment. The availability of reference data is a severely limiting factor over large geographic region because of the time, effort, cost, and accessibility in different parts of the world. The objectives of this research were to augment and extend the limited availability of crop type reference data using non-ground-based sources of crop type information for creating and assessing large area crop type maps. There is the potential to either interpret the photographs available from Google Street View (GSV) or classify High Resolution Imagery (HRI) using a phenology-based classification approach to generate additional reference data within similar agriculture ecological zones (AEZs) based on the crop characteristics, their types, and growing season. These two methods of augmenting and extending crop type reference data were developed for the United States (US) where high-quality crop type reference data already exist so that the methods could be effectively and efficiently tested.

This research described a tale of three continents providing recommendations to adapt accuracy assessment strategies and methodologies for assessing global cropland extent maps. Based on these results, the assessment and comparison of different resolution GFSAD cropland extent maps were performed to provide specific recommendations for when to apply each of the maps for agriculture monitoring based on the agriculture field sizes. When assessing the cropland extent maps, different sampling strategies perform differently in the various cropland proportion regions and therefore, must be selected according to the cropland extent maps to be assessed. Finally, this research concluded that the limited crop type reference data can be effectively extended using a phenology-based classification approach and is more efficient than the interpretation of photographs collected from GSV.