Detecting Soil pH from Open Source Remote Sensing Data: A Case Study of Angul and Balangir districts, Odisha StatePGogumallaauthorSRupavatharamauthorADattaauthorRKhopadeauthorPChoudhariauthorR KDhulipalaauthorDDixitauthorInternational Crops Research Institute for the Semi-Arid Tropics (ICRISAT) is implementing ‘Odisha Bhoochetana’, an agricultural development project in Angul and Balangir districts in India. Under this project, soil health improvement activity was initiated by collecting soil samples from selected villages of the districts. Soil information before sowing helps farmers not only to choose a crop but also in planning crop nutritional inputs. Soil sampling, collection, and analysis is a costly and laborintensive activity that cannot cover the entire farmlands, hence it was conceived to use high-speed open-source platforms like Google Earth Engine in this research to estimate soil characteristics remotely using high-resolution open-source satellite data. The objective of this research was to estimate soil pH from Sentinel1, Sentinel 2, and Landsat satellite-derived indices; Data from Sentinel 1, Sentinel 2, and Landsat satellite missions were used to generate indices and as proxies in a statistical model to estimate soil pH. Step-wise multiple regression, Artificial Neural networks (ANN) and Random forest (RF) regression, and Class-wise random forest were used to develop predictive models for soil pH. Step-wise multiple regression, ANN, and RF regression are single class models while class-wise RF models are an integration of RF-Acidic, RF-Alkaline, and RF- Neutral models (based on soil pH). The step-wise regression model retained the bands and indices that were highly correlated with soil pH. Spectral regions that were retained in the step-wise regression are B2, B11, Brightness Index, Salinity Index 2, Salinity Index 5 of Sentinel 2 data; VH/VV index of Sentinel 1 and TIR1 (thermal infrared band1) Landsat with p-value <0.001. Amongst the four statistical models developed, the class-wise RF model performed better than other models with a cumulative R 2 and RMSE of 0.78 and 0.35 respectively. The better performance of class-wise RF models over single class models can be attributed to different spectral characteristics of different soil pH groups. Though neural networks performed better than the stepwise multiple regression model, they are limited to a regression while the random forest model was capable of regression and classification. The large tracts of acidic soils (datasets) in the study area contributed to the training of the model accordingly leading to neutral and alkaline soils that were misclassified hindering the single class model performance. However, the class-wise RF model was able to address this issue with different models for different soil pH classes dramatically improving prediction. Our results show that the spectral bands and indices can be used as proxies to soil pH with individual classes of acidic, neutral, and alkaline soils. This study has shown the potential in using big data analytics to predict soil pH leading to the accurate mapping of soils and help in decision support.GIS Techniques/Remote SensingSoil FertilitySoil Science2022-03SpringerArticle