Potato Fields Mapping Based on the Phenology Feature and Support Vector Machine Utilizing Google Earth Engine Platform

Potato is the fourth most cultivated crop worldwide. In terms of its strategic role in food security, accurate potato mapping provides essential information for national crop censuses and potato yield estimation and prediction at any scale. Although remote sensing (RS) approaches based on optical and/or microwave sensors have been widely employed to monitor cultivated lands (including crop area, conditions, and yield forecasting), the identification of potato planting areas using RS data and machine learning has not been much addressed. As a result, the present research addresses the literature gap by suggesting an effective potato mapping approach in Iran's main production center and tries to provide accurate information on the cultivated areas of this crop for the field of agricultural management.

Since most crops have specific spectral and temporal characteristics during their cultivation period, this research has presented a method to discriminate potato fields from other crops using time series images without explicit thresholding. Is. This method identified this product by using layers based on potato phenology and machine learning. We employed the ground truth data of the crop types from the studied site, which included a total of 1648 samples of potato fields and other crops, to optimize the internal parameters of the algorithm, train, and evaluate the model. A handheld GPS receiver was used to collect this data. This research employed Sentinel-2 satellite images and the Support Vector Machine (SVM) algorithm to map potato fields. To accurately identify potato fields, we prepared appropriate input layers, including the phenological index of the potato crop and the median statistical index of NDVI (time series of Sentinel-2 satellite images) at specific intervals. We used these layers as inputs to the SVM. We optimized the gamma and C values using the 5-fold cross-validation method to train the optimal model for SVM using the RBF kernel. We then used these values in the algorithm implementation process under the Google Earth Engine cloud computing platform. We assessed the efficacy of the suggested approach in the Iranian cities of Hamedan and Bahar, key sites for the cultivation of this particular crop.

Based on the results, the optimal values for the internal parameters of the model (C = 70 and γ = 0.3) were calculated. We included these values in the RBF function to identify the cultivated areas of the potato crop. By implementing the classification algorithm and then applying the majority filter, a map of the areas under potato cultivation was prepared for the study area. This map showed the highest density of potato cultivation in the border area of two cities (northwest of Hamedan city and east of Bahar city). The calculated total area for potato farming was 4527.1 hectares in Hamedan city and 6088.3 hectares in Bahar city. The estimated overall accuracy and Kappa coefficient are 90.9% and 0.82 for Hamedan and 93.3% and 0.87 for Bahar, respectively. The present research's results demonstrate the effectiveness of the SVM algorithm in detecting potato cultivation areas, highlighting the potential of using indicators corresponding to potato phenology as distinguishing features for improved identification.

By employing the SVM method, we effectively identified potato fields by utilizing layers of indicators that correspond to crop phenology. At the trial stage, it was demonstrated that this method can improve the potato acreage mapping process. Therefore, a similar approach can be evaluated for identifying other important crops in other regions. It is also suggested that the efficiency of microwave data and other machine learning algorithms be considered in future research.