Assessing the potential of DESIS hyperspectral data to discriminate cropping patterns

Research output: Contribution to conferenceAbstractAcademic

Abstract

Cropping patterns; the annual sequence and spatial arrangement of crops on a piece of land, are rarely reported and information on the location, extent and type remains unknown. Such information is important for measuring crop yields and land use intensity, also it required for food security studies. The study is located in the sub-Saharan region, where local farmers have small field sizes, highly fragmented landscapes, and diverse cropping practices. The use of remote sensing technology such as optical sensors have enabled the mapping of cropping patterns, however, there are no studies that have explored hyperspectral sensors for discriminating cropping patterns. The study assesses the potential discrimination of maize cropping patterns using DESIS hyperspectral satellite data to characterize maize and intercropped maize (imaize) fields. We extracted reflectance of the fields of both cropping patterns based on the data collected from the field. A Random Forests classifier was used for feature selection to identify the best subset of features (bands) that would further be used for classification. Ten bands dominated in the red edge and NIR (752.2nm, 767.5nm, 775.2nm, 783nm, 814.2nm, 849.8nm, 870,5nm, 898,3nm, 903,7, 934,4nm). These bands were used in a Random Forest classifier obtaining an overall accuracy (OA) of 78% with a producer accuracy (PA) of 76% for maize and 84% for intercropped maize and user accuracy (UA) of 89% for maize and 54% for intercropped maize. An F1 score of 0.82 for maize and 0.65 for intercropped maize was obtained. A kappa coefficient was 0.57, showing the complexity of discriminating and classifying maize cropping patterns. The date of the image was obtained at the later growth phase of maize, when the maize canopy cover has obscured the intercrops (beans). An earlier date can improve the results, as intercrops will still be visible and can be detected with less obscurity by the sensor. This research opens more opportunities for future research into cropping patterns discrimination in small-holder agriculture using hyperspectral sensors.
Original languageEnglish
Publication statusPublished - 2 Nov 2023
Event13th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing, WHISPERS 2023 - The Eugenides Foundation’s technologically advanced Conference Centre, Athens, Greece
Duration: 31 Oct 20232 Nov 2023
Conference number: 13
https://www.ieee-whispers.com/

Conference

Conference13th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing, WHISPERS 2023
Abbreviated titleWHISPERS 2023
Country/TerritoryGreece
CityAthens
Period31/10/232/11/23
Internet address

Fingerprint

Dive into the research topics of 'Assessing the potential of DESIS hyperspectral data to discriminate cropping patterns'. Together they form a unique fingerprint.

Cite this