Improving the Accuracy of Remotely Sensed Irrigated Areas Using Post-Classification Enhancement Through UAV Capability

Abstract

Although advances in remote sensing have enhanced mapping and monitoring of irrigated areas, producing accurate cropping information through satellite image classification remains elusive due to the complexity of landscapes, changes in reflectance of different land-covers, the remote sensing data selected, and image processing methods used, among others. This study extracted agricultural fields in the former homelands of Venda and Gazankulu in Limpopo Province, South Africa. Landsat 8 imageries for 2015 were used, applying the maximum likelihood supervised classifier to delineate the agricultural fields. The normalized difference vegetation index (NDVI) applied on Landsat imageries on the mapped fields during the dry season (July to August) was used to identify irrigated areas, because years of satellite data analysis suggest that healthy crop conditions during dry seasons are only possible with irrigation. Ground truth points totaling 137 were collected during fieldwork for pre-processing and accuracy assessment. An accuracy of 96% was achieved on the mapped agricultural fields, yet the irrigated area map produced an initial accuracy of only 71%. This study explains and improves the 29% error margin from the irrigated areas. Accuracy was enhanced through post-classification correction (PCC) using 74 post-classification points randomly selected from the 2015 irrigated area map. High resolution aerial photographs of the 74 sample fields were acquired by an unmanned aerial vehicle (UAV) to give a clearer picture of the irrigated fields. The analysis shows that mapped irrigated fields that presented anomalies included abandoned croplands that had green invasive alien species or abandoned fruit plantations that had high NDVI values. The PCC analysis improved irrigated area mapping accuracy from 71% to 95%.

Prospects of Improving Agricultural and Water Productivity through Unmanned Aerial Vehicles

“…Most of the freely available space-borne satellite images have been used to produce coarse landuse/cover maps but, the advent of UAVs has improved mapping accuracy because of their high resolution, although at a smaller coverage [34]. The use of UAVs in mapping landuse makes it potentially possible to monitor smallholder farming fields, which are generally too small to be detected by readily available moderate to low resolution satellite images [9]. Smallholder farming plots measure about two hectares in area per farmer [35].…”

“…The NDVI is important for providing information on the variability of the health of crops, as well as in large-scale monitoring of plantations, assess changes in the field, quantifying crop acreage and analyse crop loss [48]. It is also used to estimate plant attributes such as physiological status, yield production, crop distribution and irrigation mapping [9,49,50]. Another modification of the NDVI used to assess changes in plant health is the SAVI (Soil Adjusted Vegetation Index) (Equation ( 2)) [51].…”

“…Improved agricultural water management requires innovative and evidence-based solutions applied throughout the whole agriculture value chain [7,8]. The use of Unmanned Aerial Vehicles (UAVs), also known as drones, offers advanced crop image data analytics at high spatial and temporal resolution and crop monitoring in near real-time, important elements in agricultural water management [9].…”

“…Satellite and UAV imagery serve the same purpose but differ significantly in spatial and temporal resolution. The choice of the two is determined by the scale and accuracy the user requires to achieve [9,21]. The advantage of UAVs over space-borne sensors is that UAVs offer low cost imaging at high spatial resolution and at user determined revisit periods [22].…”

IoT enabled Plant Sensing Systems for Small and Large Scale Automated Horticultural Monitoring

“…Current techniques use drones equipped with highresolution cameras to obtain a heat map of crops from the air. However, such methods have been shown to be inconclusive as false readings and failure rates are predominant due to complex topographical landscapes, and limitations of the image processing methods used [2]. Moreover, such methods are unable to detect the humidity and light levels in different areas of the field from such a distance.…”