Challenges of mapping agricultural soils in an arid Colorado River alluvial area using remote sensing, Southwestern U.S

Authors: MANCINI, MARCELO - University Of Arkansas; Winzeler, Hans; Blackstock, Joshua; Owens, Phillip; Libohova, Zamir; MOORE, JOSHUA - Colorado Department Of Agriculture; MILLER, DAVID - University Of Arkansas; SILVA, SERGIO - Federal University Of Lavras; Ashworth, Amanda

Submitted to: Remote Sensing Applications: Society and Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/20/2026
Publication Date: N/A
Citation: N/A

Interpretive Summary: Many tribal nations lack information about soil properties and soil function related to tribal lands. Accurate detailed soil maps are key to for optimizing agriculture to help optimize use of resources and achieve more sustainable production systems. The agricultural systems of tribes in Southwest US rely on water from the Colorado River and are conducted mainly on alluvial soils that are often highly saline and challenging to manage Detailed high resolution maps of soil fertility properties were produced for the Colorado River Indian Tribes (CRIT) that are needed for precision agriculture. High resolution soil maps of nutrients that are essential for soil fertility were made by combining advanced methods for mapping soils like Digital Soil Mapping (DSM) with satellite images and field observations This approach creates functional soil map units that will help Indian Tribes in Colorado River better manage their limited soil and water resources in a sustainable manner while saving cost and the environment.

Technical Abstract: The challenge of mapping alluvial soils lies in the spatial heterogeneity of alluvial sediments and is compounded by agricultural activities in fertile floodplains. Here, we used Google Earth Engine (GEE) to calculate statistics of time series of remotely sensed data (555 scenes) to map the spatial variability of macronutrients (N, Ca, K, Mg, P, and S) in soils within the alluvial plains in the Colorado River Indian Tribes (CRIT). Vegetation was masked using GEE to mitigate the effects of intense management. Gradient boosting models were trained with masked and unmasked Sentinel-2 bands, and prediction performance was compared to that of models trained with a Visible and Near-infrared spectroscope (Vis-NIR) proximal sensor. The objective was to investigate if statistics of masked spatiotemporal remotely sensed data could improve the prediction capabilities of publicly available satellite data by comparing their predictive power to that of a proximal sensor (Vis-NIR). Most models did not perform well due to the high spatial complexity of the study area. Models trained with Vis-NIR data presented promising results for most macronutrients (R2=0.38). Models trained with Sentinel-2 data for which vegetation was not masked had the poorest results. The best prediction results were achieved by training models with statistics of masked bands. Potassium spatial estimations were the most accurate (R2=0.45). Pixel-based statistics unveiled persistent spatial patterns relevant for macronutrient estimations following fluvial landforms, albeit agricultural activities have significantly altered their spatial appearance. GEE facilitates the processing of large spatial datasets and statistics of masked Sentinel-2 bands and produced prediction performance comparable to that of proximal sensors. GEE and Sentinel-2 can be used to support agricultural management in the CRIT and have potential to improve the digital soil mapping of key soil properties in alluvial plains.