Numbers of Earth Observation (EO) satellites have increased exponentially over the past decade, fuelled by a shift towards constellation models that promise to deliver data at finer spatial, temporal and spectral resolutions compared to the past. The result is the now >1000 EO satellites in orbit, a population that is rapidly increasing because of a booming private-sector interest in space imaging. Flowing from this, EO data volumes have mushroomed in recent years, and data processing has migrated to the cloud, with scientists leveraging tools such as Google Earth Engine for information retrieval. Whilst considerable attention has been given to the launch and in-orbit environmental impacts of satellites (e.g. rocket emissions and space-junk risks), specific environmental impacts from EO missions (data infrastructures and cloud computation); have so far escaped critical scrutiny. It is urgent that the environmental science community address this gap, so that the environmental good of EO can withstand scrutiny.

Data centres consume high quantities of water and energy and they may be situated in sensitive geographical situations far away from both users and launchpads (i.e. a transboundary environmental concern). There are also hidden impacts in the carbon-intensive processes of computer component manufacture, impacting places and communities far from the site of EO information retrieval. We scope the broad suite of transboundary environmental impacts that EO generates. Related to the data aspect of the EO life-cycle, we quantify the current volume of global EO data holdings (> 800 PB currently, increasing by 100 PB / year). Mapping the distribution of datasets across different data centre providers, our work shows high redundancy of datasets, with collections from NASA and ESA replicated across many data centres globally. Storage of this data volume generates annual CO₂ equivalent emissions summing to >4000 tonnes/year. We quantify the environmental cost of performing EO functions on the cloud compared to desktop machines, using Google Earth Engine as an exemplar, scaling emissions using the ‘Earth Engine Compute Unit’. We show how large-scale analyses executed within GEE rapidly scale to produce the equivalent emissions of a single ticket on an economy flight ticket from London-Paris. Executing these processes on the cloud takes seconds, and these estimates do not account for emissions from microprocessor manufacture, nor do they account for users running processes multiple times (e.g. during code development).  A major blind-spot is that the geography of GEE data centres is hidden from users, with no choice given to users about where GEE processes are executed. It is important that EO providers become more transparent about the location-specific impacts of EO work, and provide tools for measuring the environmental cost of cloud computation. Furthermore, the EO community as one which is concerned with the fate of Earth’s environment must now urgently and critically consider the broad suite of EO data life-cycle impacts that lie (a) beyond the launchpad, and (b) on Earth rather than in space; taking action to minimise and mitigate them. This is important particularly because EO data will long outlive the satellites that provided them.

How to cite: Anderson, K., Mleczko, M., Brewin, R., Gaston, K., Mueller, M., Jamie, S., Yan, X., and Wilkinson, R.: Quantifying the transboundary environmental impacts of Earth observation in the 'big data' and constellation era, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11307, https://doi.org/10.5194/egusphere-egu24-11307, 2024.

Secondary forests, forest naturally regrowing on areas of previously deforested, now abandoned lands are crucial to help maintain and increase the carbon sink on land, helping to tackle the climate and ecological emergencies. There is a growing research field on improving our understanding of the carbon dynamics of secondary forests, often using novel remote sensing techniques to map the temporal and spatial patterns of change. However, the full datasets from the research are often not fully accessible to all users, either because the whole dataset is not published, or because it is presented in a specialist format. Given both the scientific and policy interest in forest-carbon dynamics, it is critical to ensure datasets are accessible to a wide range of audiences. 

Here we present “RE:Growth” – a user-friendly toolkit designed in Google Earth Engine, enabling users to view and download the aboveground carbon dynamics spatially and temporally in secondary forests in one of the largest tropical forest regions; the Brazilian Amazon. Designed to be easily updated as more temporal and spatial data are made available, ‘RE:Growth’ provides spatial and aggregated information on carbon flux dynamics in secondary forests through time based on Earth Observation. Uniquely, the user can draw their own region or select a jurisdictional boundary in the Brazilian Amazon to focus the analysis to their region of interest. Such information can be used within the measurement, reporting and verification framework, which is critical for results-based payments at all scales. For example, the toolkit can be used to provide spatial and quantitative data to inform the spatial prioritization of secondary forest conservation and expansion in line with Brazil’s Nationally Determined Contributions (NDC) to the Paris Agreement.

How to cite: Heinrich, V., Sitch, S., Rosan, T., Silva Junior, C., and Aragão, L.: RE:Growth—A Google Earth Engine toolkit analyzing secondary forest carbon dynamics in the Brazilian Amazon, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5123, https://doi.org/10.5194/egusphere-egu24-5123, 2024.

Two innovative and powerful Google Earth Engine (GEE) Apps have recently been developed to identify and map volcanic thermal features and investigate gas flaring sources at global scale in daylight conditions. Both the GEE Apps are based on the Normalized Hotspot Indices (NHI; Marchese et al., 2019), which analyze the Near Infrared (NIR) and Short-Wave Infrared (SWIR) radiances from the Multispectral Instrument (MSI) and the Operational Land Imager (OLI/OLI-2) sensors, respectively aboard the Sentinel-2 and Landsat 8/9 satellites, to detect high-temperature features. The NHI tool enables the analysis of volcanic thermal anomalies through plots of hotspot pixel number, total SWIR radiance and total hotspot area. In addition, an automated module of the tool notifies the active volcanoes over the past 48 hours (https://sites.google.com/view/nhi-tool/home-page). DAFI (Daytime Approach for gas Flaring Investigation) by performing a multitemporal analysis of the NHI identifies the gas flares on annual basis both onshore and offshore, providing information about the gas flaring sources in terms of persistence of thermal activity and through the computation of the radiative power (https://sites.google.com/view/flaringsitesinventory). These systems demonstrate the relevance of the GEE platform in supporting the analysis, monitoring and characterization of hot targets (both natural and industrial ones) thanks to the massive computational resources and the availability of extended datasets of multisource satellite observations.

How to cite: Faruolo, M., Pergola, N., Genzano, N., and Marchese, F.: Two powerful Google Earth Engine (GEE) Apps for the worldwide high-temperature features monitoring and investigation, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-20371, https://doi.org/10.5194/egusphere-egu24-20371, 2024.

Enabling the resilience of local food systems is crucial to ensure a steady supply of nutritious food to people living in fragile and conflict-affected locations. While the majority of interventions often focus on staple crops, there is an increasing tendency by humanitarian organizations to include vegetable production solutions in their programs. However, information on land suitability for vegetable production is usually lacking or available at a coarse spatial resolution, thereby limiting targeted interventions for smallholder farmers.

This study proposes a comprehensive geospatial data-driven framework for mapping suitable areas for vegetable production in Africa using a machine learning algorithm (ML) implemented in Google Earth Engine (GEE) and a Multi-Criteria Decision Analysis (MCA) approach. Mali (West Africa) and Ethiopia (East Africa) are selected as case studies given the current fragility of both countries, and support provided by the USAID's Bureau for Humanitarian Assistance (BHA). Field data of vegetable production locations was collected to train and validate the ML and MCA models. Several publicly available geospatial datasets, including FAO’s WaPOR database, were reviewed to select the predictor variables, which include information on climate, soil, topography, surface water, groundwater, socioeconomics and disaster risks. A suitability map was produced for all vegetables, and separate suitability maps were generated for the top five most cultivated vegetables in Mali and Ethiopia.

Comparison of the ML approach to the MCA approach revealed a lower performance of the former due to the limited availability of field data, thereby highlighting the benefit of expert knowledge in addition to the data-driven approach. The results show that the most suitable areas are found in the region of Segou in Mali (up to 88%), while the region of Oromia has the most suitable areas in Ethiopia (up to 85%). The resulting maps of land suitability for vegetable production serve to develop an irrigation investment targeting tool, which can be used to assist humanitarian organizations in implementing suitable irrigation solutions for vegetables.

How to cite: Dembélé, M., Leh, M., Wickramasinghe, D., Velpuri, N., Sanogo, K., Tegegne, D., Andarcia, M. G., and Schmitter, P.: Geospatial suitability mapping for targeted vegetable production in fragile African regions, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11472, https://doi.org/10.5194/egusphere-egu24-11472, 2024.

Monitoring agricultural water use is essential to ensure water security, especially in regions facing water scarcity. Satellite-acquired multi-spectral images of the Earth’s surface provide crucial data to enable frequent estimations of crop water use. In large data-scarce regions, these estimations represent a key source of information for water management. The emergence of cloud-based platforms, such as Google Earth Engine (GEE), has made it feasible, accessible, and cost-effective to automate crop water use monitoring pipelines. Here, we demonstrate the potential benefits of a cloud-based crop water use estimation and monitoring framework by estimating a decade's worth of agricultural water use over Saudi Arabia. In Saudi Arabia, large-scale agricultural activities account for the majority  (>80%) of water use, water which is sourced primarily from non-renewable groundwater resources from the Arabian Shelf. Saudi Arabia’s large land area (> 2 million  km²) and the long study period (+10 years) forms the basis of a case-study for our cloud-based model. Previous mapping efforts provided annual maps of individual field boundary delineations at, identifying more than 30,000 fields covering a total of more than 10,000 km² of croplands that are distributed across several large-scale agricultural clusters within the Kingdom. As a preprocessing step, we developed an approach to generate large-scale delineations of irrigated agricultural regions over arid areas. This approach helped reduce processing efforts for field delineations, and at the same time reducing the water use estimation computations. Our GEE cloud-based model implements a two-source energy balance model (TSEB) and automatically incorporates all available Landsat collection 2 surface reflectance and surface temperature products from Landsat 7, 8, and 9, along with climate reanalysis data from the ECMWF ERA5-Land hourly product. The model can be readily applied elsewhere by defining just the date range and study geometry, while allowing the flexibility for more advanced users to control parameters within the TSEB model. Total crop water use (evapotranspiration term only; not accounting for irrigation efficiency) was estimated at between 7 and 12 BCM per year of study, with the highest use in 2016 and the lowest in 2020. Both Riyadh and al Jawf administrative regions collectively shared more than half of the total study cropland area, and a similar contribution of water use. This study represents the convergence of a number of efforts towards developing operational crop water use monitoring, while motivating further applications in other regions and providing a rich dataset for further food and water security related studies.

How to cite: Lopez Valencia, O. M., Li, T., Aragon Solorio, B. J. L., and McCabe, M. F.: Cloud-based agricultural crop water use monitoring across Saudi Arabia, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6454, https://doi.org/10.5194/egusphere-egu24-6454, 2024.

The Google Earth Engine platform has transformed access to long term scientific datasets as more than thirty years of image data are readily available for analysis on the cloud. Despite this rapid access to long term data and cloud computing resources of GEE, there is still the requirement to remove noise (e.g., from cloud and haze) and correct for the effects of calibration between image types to process raw data into information about changes in the Earth’s surface. Using the Python Earth Engine API, we developed an algorithm to combine timeseries datasets from high spatial resolution sensors (Landsat-8 and Sentinel-2) and filter noise whilst still retaining high temporal resolution. A second algorithm was then developed to automate the decomposition of pre-processed timeseries data into individual agricultural seasons for the extraction phenological stages for sugarcane fields across India. Our approach was developed and validated on over 800 sugar cane field parcels and overcomes the challenge of previous machine-learning methods for sugarcane monitoring using remote sensing that rely on information on planting and harvesting. Fully automated monitoring of sugarcane is possible over wide areas without the need to download image datasets or process time series data locally. This approach can significantly improve the sustainability of sugarcane production by optimising the harvest to maintain efficiency in the supply of sugarcane to mills during the crushing season and reduce waste by avoiding the harvest of immature cane. The use of GEE means that this approach can be easily modified for use with other crops and in other geographical areas to improve satellite-based monitoring of crops. These tools are essential for realising the goals of sustainable development, and time-series analysis can be used to help producers demonstrate commodities have not come from recently deforested land (for example, the EU regulation on deforestation-free products).

How to cite: Joshi, N., Simms, D., and Burgess, P.: The application of Google Earth Engine to monitor sugarcane development in India., EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19440, https://doi.org/10.5194/egusphere-egu24-19440, 2024.

The Qinghai-Tibetan Plateau (QTP) is one of the most sensitive and vulnerable regions under global climate change. Vegetation is the key component of the QTP ecosystem and is closely related to the ecological vulnerability. Fractional vegetation cover (FVC) is an important parameter to characterize vegetation conditions in the horizontal direction. Therefore, dense time-series of spatially continuous FVC at high spatial resolution is essential for understanding the detailed spatiotemporal dynamic changes in vegetation across QTP. Landsat is an ideal remote sensing data source for high spatial resolution FVC monitoring. However, frequent cloud cover during the growing season on QTP makes it challenging to observe FVC constantly only using Landsat. Spatiotemporal fusion methods integrating the advantages of Landsat and MODIS have been widely developed. Currently, most spatiotemporal fusion methods assume that the relationship between Landsat and MODIS is fixed over the prediction period. For regions with strong heterogeneity and large temporal variations, the relationship between Landsat and MODIS is variable along time. In addition, most methods fuse the reflectance bands separately without considering the interrelationship between bands. Therefore, a method blending Landsat and MODIS reflectance to generate FVC with 30m spatial resolution and 8-day interval based on Google Earth Engine (GEE) is proposed in this study. This method considers the dynamic relationship between MODIS and Landsat by analyzing the time-series data collected from multiple years. And a novel two-band simultaneous smoothing strategy is developed in this method, which can generate smoothed and consistent time-series of red and near-infrared bands simultaneously. Compared with three previous typical methods in two challenging QTP regions with rapid vegetation change, it can be found that the synthesized 30m reflectance data generated by the proposed method can get more accurate FVC. The validation results compared with the field-measured FVC further confirm the validity of the proposed method. The generated FVC products across QTP exhibit spatial continuity and reasonable time-series profiles. The proposed method is thus expected to provide high-quality FVC time-series with high spatiotemporal resolution over multiple years for QTP and other regions with frequent data missing based on GEE.

How to cite: Tao, G. and Jia, K.: Generating dense time-series of spatially continuous 30m fractional vegetation cover for the Qinghai-Tibetan Plateau based on Google Earth Engine, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13730, https://doi.org/10.5194/egusphere-egu24-13730, 2024.

Effective agricultural management policies rely heavily on accurate crop identification across various scales. This is particularly challenging in highly fragmented agricultural landscapes, such as those found in many European regions, particularly for fruit tree orchard identification. While solutions exist, the confidence in individual orchard classification is often overlooked, despite its importance in enhancing classification accuracy (precision, recall and specificity).

Several confidence metrics at pixel level have been proposed by estimating the probabilities of a pixel to belong to each of the possible classes. The higher the probability of class membership for a given class, the greater the confidence associated with that class. In this sense, a measure of confidence can be based in the differences of probability between the first two highest values (sometimes called the distance to the second cluster).

This study introduces an innovative methodological approach to build a classification confidence metric at object (orchard) level. Once segmentation is completed, all pixels whose confidence is not above a certain threshold are masked out. Then, each orchard is initially assigned to a class by computing the mode of the unmasked pixels inside its perimeter. In a subsequent step, a confidence metric at orchard level is estimated, based on the number of mode class pixels, the total number of pixels completely inside the orchard, and the proportion of the mode pixels and unmasked pixels within the orchard. This confidence metric allows for a balance between increased precision and a reduction in the number of classified orchards (those with insufficient confidence in their classification).

The proposed method, fully implemented in Google Earth Engine, was tested in a highly fragmented area in Valencia (Spain). The system’s performance was assessed by using a Random Forest classification algorithm on Fourier coefficients of spectral indexes time-series at pixel-level plus a specific spatial cross-validation procedure. By setting a 70% orchard classification confidence level, the mean overall accuracy increased from 88.74 ± 3.03% to 93.58 ± 2.85%, and the Kappa index from 0.78 ± 0.06% to 0.87 ± 0.05%, albeit at the cost of leaving 12.60 ± 7.18 % of orchards unclassified.

How to cite: Izquierdo Sanz, H., Moltó garcía, E., and Morell Monzó, S.: Enhancing Orchard Classification Accuracy through Object-Level Confidence Metrics: A Case Study in Automatic Orchard Identification in Valencia, Spain, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18012, https://doi.org/10.5194/egusphere-egu24-18012, 2024.

How to cite: Shivamurthy, V., Pawale, S., Chandrashekar Shetty, Y., and Bhat, S. T.: Comprehensive Hybrid Approach for LULC classification using GEE-API and ML, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19017, https://doi.org/10.5194/egusphere-egu24-19017, 2024.