Land Quality Index for Paddy (Orzya sativa L.) Cultivation Area based on Deep Learning Approach using Geographical Information System and Geostatistical Techniques

Many fields in Turkey have suitable ecological conditions for rice production and its per hectare yield is usually above the global average. However, because of such as unstable fertilization, lack of nutrients, and irrigation, the production amount cannot meet the consumption amount when soil properties are not considered. This case entails Republic of Turkey to be a rice-importing country. The present study was conducted to on a 1,763-hectare field in the province of C ¸ orum, Osmancık district, 1 Springer Nature 2022 L A T E X template Land Quality Index based on DL using GIS and Geostatistical Techniques which is one of the most significant rice fields in Turkey. The main purpose of this study is to determine land quality classes for rice (Oryza sativa L.) production based on Geographic Information System (GIS) and the deep learning approach. Deep learning is a popular technique for image processing and data analysis with procuring results and great potential. This technique has recently been used extensively in precision agriculture applications. In the study, Feedforward Neural Networks (FNN), a basic deep learning model, was used. It is used 15 different physicochemical properties (pH, EC, lime, OM, depth, slope, HI, HA, clay, silt, sand, N, P, K, and Zn). Using this parameters soil types was classified and regression analysis (index, efficiency, NAI, and SQI) was performed by deep learning. In the study, using the index, efficiency, NAI, and SQI soil parameters as network outputs caused different performance levels in models. Therefore, different models were suggested for each network output. R2 values are at an acceptable level for predicting parameters (index: 91.14%, efficiency: 87.50%, NAI: 87.54%, and SQI: 87.54%). A success rate of 88% is achieved in classifying ”class” information. It has been shown that deep learning can be used successfully in predicting soil parameters and identifying land quality classes. In addition, identified land quality classes have been confirmed by field study. As regards research results, a significant positive relationship was found between land quality classes and yield using Statistical methods. In addition, according to field validation study, rice yield was significantly affected by the land quality classes which was at a p < 0.001 level. The highest product yield was achieved in (7197 kg ha −1 in S1 class), (5032 kg ha −1 in S2 class), and then (3572 kg ha −1 in S3 class).