Using Remote Sensing Normalised Difference Vegetation Index to Rec-ognise Irrigated Croplands via Agroland Classifier Application

Abstract

Formulation of the problem. Recognition between irrigated and non-irrigated croplands is an important task of modern agri-cultural science in order to ensure efficient management of water resources in agriculture and control the usage of irrigation systems. Remote sensing data could be utilized as a means for the automation of this task through the implementation of machine classifica-tion algorithms. The normalised difference vegetation index, calculated based onaerospace images, could be ofgreat usefulness in this regard to determine the patterns of vegetation cover in different humidification conditions and provide a key to distinguish be-tweenrainfed and irrigated crops.The purposeof this study was to assess the accuracy of croplandmeliorative status recognition using remote sensing normal-ised difference vegetation index through different mathematical algorithms within Agroland Classifier application and to findout whether this application could be applied forautomated cropland recognition. Methods. The study was conducted for the Southern Steppe zone of Ukraine, and included 100 randomly selected fields (50 ir-rigated, and 50 non-irrigated) within the boundaries of Kherson and Mykolaiv regions. The data on the values of the field normalised difference vegetation index were obtained through the calculation of the average monthly index value using free of distortion cloud-less aerospace imagery with aresolution of 250 m from OneSoil remote sensing platform, and then fetched to the application Agro-land Classifier to get a decision on the meliorative status of the field (irrigated or non-irrigated). Agroland Classifier utilises linear canonical discriminant function and logistic regression algorithms to distinguish between the irrigated and rainfed fields. The accura-cy of the application recognition was evaluated through the calculation of general correctness rate, as well as correctness rates for each recognition algorithm separately. Results. The study revealed that Agroland Classifier provides high general correctness rate (92% for the combined algorithms) for the recognition between the irrigated and non-irrigated croplands. Eachalgorithm of the application was established to haveits unique advantages and disadvantages. The linear canonical discriminant function provides more stable results both for the irrigated (88% of correct assumptions) and non-irrigatedlands (84% of correct assumptions). At the same time, logistic regression failed to recognize the irrigated crops (just 78% of correct assumptions), while the accuracy of the non-irrigated lands recognition was signifi-cantly higher (96% of correct assumptions). Scientific novelty and practical significance. The article provides novel insights on the implementation of remote sensing da-ta in the classification betweenirrigated and non-irrigatedcrops in the Southern Steppezone of Ukraine via Agroland Classifier application. The application could be recommendedfor scientific and practical purposes to improve croplandmapping and monitor-ing of the use ofwater resourcesin agriculture.