This paper presents Prithvi-EO-2.0, a new geospatial foundation model that offers significant improvements over its predecessor, Prithvi-EO-1.0. Trained on 4.2 million global time series samples from NASA’s Harmonized Landsat and Sentinel-2 data archive at 30-m resolution, the new model incorporates temporal and location embeddings for enhanced performance across various geospatial tasks. Through extensive benchmarking with GEO-Bench, the model outperforms the previous Prithvi-EO model by 8% across a range of tasks. It also outperforms six other geospatial foundation models when benchmarked on remote sensing tasks from different domains and resolutions (i.e. from 0.1 m to 15 m). The results demonstrate the versatility of the model in both classical Earth observation and high-resolution applications. Early involvement of end-users and subject matter experts (SMEs) allowed constant feedback on model and dataset design, enabling customization across diverse SME-led applications in disaster response, land cover and crop mapping, and ecosystem dynamics monitoring. Prithvi-EO-2.0 is available as an open-source model on Hugging Face and IBM TerraTorch, with additional resources on GitHub. The project exemplifies the Trusted Open Science approach embraced by all involved organizations.

This study developed a rapid rice yield estimation workflow and customized yield prediction model by integrating remote sensing and meteorological data with machine learning (ML). Several issues need to be addressed while developing a crop yield estimation model, including data quality issues, data processing issues, selecting a suitable machine learning model that can learn from few available time-series data, and understanding the non-linear relationship between historical crop yield and remote sensing and meteorological factors. This study applied a series of data processing techniques and a customized ML model to improve the accuracy of crop yield estimation at the district level in Nepal. It was found that remote sensing-derived NDVI product alone was not sufficient for accurate estimation of crop yield. After incorporating other meteorological variables into the ML models, estimation accuracy improved dramatically. Along with NDVI, the meteorological variables of rainfall, soil moisture, and evapotranspiration also exhibited a strong association with rice yield. This study also found that stacking multiple tree-based regression models together could achieve better accuracy than benchmark linear regression or standalone ML models. Due to the unique and distinct physio-geographical setting of each district, a variation in estimation accuracy from district to district could be observed. Our data processing and ML model workflow achieved an average of 92% accuracy of yield estimation with RMSE 328.06 kg/ha and MAE 317.21 kg/ha. This methodological workflow can be replicated in other study areas and the results can help the local authorities and stakeholders understand the factors affecting crop yields as well as estimating crop yield before harvesting season to ensure food security and sustainability.

The large scale quantification of impervious surfaces provides valuable information for urban planning and socioeconomic development. Remote sensing and GIS techniques provide spatial and temporal information of land surfaces and are widely used for modeling impervious surfaces. Traditionally, these surfaces are predicted by computing statistical indices derived from different bands available in remotely sensed data, such as the Landsat and Sentinel series. More recently, researchers have explored classification and regression techniques to model impervious surfaces. However, these modeling efforts are limited due to lack of labeled data for training and evaluation. This in turn requires significant effort for manual labeling of data and visual interpretation of results. In this paper, we train deep learning neural networks using TensorFlow to predict impervious surfaces from Landsat 8 images. We used OpenStreetMap (OSM), a crowd-sourced map of the world with manually interpreted impervious surfaces such as roads and buildings, to programmatically generate large amounts of training and evaluation data, thus overcoming the need for manual labeling. We conducted extensive experimentation to compare the performance of different deep learning neural network architectures, optimization methods, and the set of features used to train the networks. The four model configurations labeled U-Net_SGD_Bands, U-Net_Adam_Bands, U-Net_Adam_Bands+SI, and VGG-19_Adam_Bands+SI resulted in a root mean squared error (RMSE) of 0.1582, 0.1358, 0.1375, and 0.1582 and an accuracy of 90.87%, 92.28%, 92.46%, and 90.11%, respectively, on the test set. The U-Net_Adam_Bands+SI Model, similar to the others mentioned above, is a deep learning neural network that combines Landsat 8 bands with statistical indices. This model performs the best among all four on statistical accuracy and produces qualitatively sharper and brighter predictions of impervious surfaces as compared to the other models.

Mapping rice area is a critical resource planning task in many South Asia countries where rice is the primary crop. Remote sensing-based methods typically rely on domain knowledge, such as crop calendar and crop phenology, and supervised classification with ground truth samples. Applying such methods on Google Earth Engine (GEE) has been proven effective especially at large scale owing to the comprehensive and up-to-date data catalog and massive server-side processing power. However, writing scripts through the code editor requires users to program in JavaScript and understand GEE Application Programming Interface (API), which can be challenging for many researchers. Thus, this paper presents a GEE-based web application that aims to eliminate the programming requirements for data selection, preprocessing, and visualizations so that users can easily produce rice maps and refine ground truth collections through intuitive Graphical User Interfaces (GUI). This software includes 3 sub-module apps, namely the ground truth collection app, threshold-based rice mapping app, and classification-based rice mapping app. Users can customize data processing flow using GUI designed with Bootstrap, and the backend server uses GEE Python API, and a Google service account for authentication to execute the workflow on Google cloud servers. The experiment shows both the overall accuracy (OA) and Kappa scores of the mapping result are higher than 0.9 which suggests RiceMapEngine can significantly reduce the complexity and time costs it takes to produce accurate rice area maps and meet the demands of real-world stakeholders.

Declining populations of Zizania palustris L. (northern wildrice, or wildrice) during the last century drives the demand for new and innovative techniques to support monitoring of this culturally and ecologically significant crop wild relative. We trained three wildrice detection models in R and Google Earth Engine using data from annual aquatic vegetation surveys in northern Minnesota. Three different training datasets, varying in the definition of wildrice presence, were combined with Landsat 8 Operational Land Imager (OLI) and Sentinel-1 C-band synthetic aperture radar (SAR) imagery to map wildrice in 2015 using random forests. Spectral predictors were derived from phenologically important time periods of emergence (June–July) and peak harvest (August–September). The range of the Vertical Vertical (VV) polarization between the two time periods was consistently the top predictor. Model outputs were evaluated using both point and area-based validation (polygon). While all models performed well in the point validation with percent correctly classified ranging from 83.8% to 91.1%, we found polygon validation necessary to comprehensively assess wildrice detection accuracy. Our practical approach highlights a variety of applications that can be applied to guide field excursions and estimate the extent of occurrence at landscape scales. Further testing and validation of the methods we present may support multiyear monitoring which is foundational for the preservation of wildrice for future generations.

Creating annual crop type maps for enabling improved food security decision making has remained a challenge in Bhutan. This is in part due to the level of effort required for data collection, technical model development, and reliability of an on-the-ground application. Through focusing on advancing Science, Technology, Engineering, and Mathematics (STEM) in Bhutan, an effort to co-develop a geospatial application known as the Agricultural Classification and Estimation Service (ACES) was created. This paper focuses on the co-development of an Earth observation informed climate smart crop type framework which incorporates both modeling and training sample collection. The ACES web application and subsequent ACES modeling software package enables stakeholders to more readily use Earth observation into their decision making process. Additionally, this paper offers a transparent and replicable approach for addressing and combating remote sensing limitations due to topography and cloud cover, a common problem in Bhutan. Lastly, this approach resulted in a Random Forest “LTE 555” model, from a set of 3,600 possible models, with an overall test Accuracy of 85% and F-1 Score of .88 for 2020. The model was independently validated resulting in an independent accuracy of 83% and F-1 Score of .45 for 2020. The insight into the model perturbation via hyperparameter tuning and input features is key for future practitioners.