Geography
Permanent URI for this communityhttp://hdl.handle.net/1903/2242
Browse
53 results
Search Results
Item Russian Winter Cropland Mapping and Impact on Land Use(2024) Abys, Christian Joseph; Skakun, Sergii Dr.; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This dissertation provides an in-depth analysis of the transformation in Russia’s winter wheat industry over the past two decades, focusing on production growth, land use changes, and advancements in monitoring techniques. The study reveals a substantial 149% increase in wheat production and a 35% rise in farmland area from 2000 to 2020, driven predominantly by winter wheat, which now represents a significant portion of global exports. Despite this growth, there is notable yearly volatility in production, with USDA Foreign Agriculture Service forecasts exhibiting considerable uncertainty, particularly in area estimations which has substantial impacts on the global wheat export market. To address these challenges, the research utilizes long-term MODIS satellite data to analyze cropland expansion and intensification in southwestern Russia, identifying a 29% increase in winter wheat cropland with distinct patterns of expansion and intensification across different latitudes. The study highlights the ongoing capacity for further cropland intensification. Furthermore, this research introduces Sentinel-1 SAR imagery as an effective solution to the issue of cloud coverage, which hampers optical data accuracy. By employing various machine learning models, including multi-layer perceptron, long short-term memory, and random forest, the study demonstrates that Sentinel-1 SAR enhances the accuracy of in-season cropland mapping. The results show that Sentinel-1 SAR data reduces uncertainty in area estimations by two-thirds compared to MODIS data, offering improved monitoring capabilities. Collectively, this research provides valuable insights into Russia’s agricultural dynamics, addresses key uncertainties in forecasting, and proposes advanced methodologies for more accurate and reliable agricultural assessmentsItem DEEP LEARNING APPROACHES FOR ESTIMATING AND FORECASTING SURFACE DOWNWARD SHORTWAVE RADIATION FROM SATELLITE DATA(2024) Li, Ruohan; Wang, Dongdong; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Surface downward shortwave radiation (DSR) designates solar radiation with a wavelength from 300 to 4000 nm received at the Earth’s surface. DSR plays a pivotal role in the surface energy and radiation budget, serving as the primary driver for hydrological, ecological, and biogeochemical cycles (Liang et al., 2010, 2019), and the important input for various earth models (Huang et al., 2019; Liang et al., 2010; Stephens et al., 2012). Given the rising demand for renewable energy, as well as accelerated advancements in solar energy technologies on both utility-scale and residential scale, the precision and resolution in estimating and forecasting DSR have become indispensable for planning and administering solar power plants (Gueymard, 2014; Jiang et al., 2019). This dissertation delves into the potential of integrating deep learning with satellite observations to address the deficiencies in current DSR estimation and forecasting methods, aiming to cater to the evolving needs of solar radiation estimation. The research begins by examining current DSR satellite products, emphasizing their limitations, particularly concerning spatial resolution and performance in snowy, cloudy, and high-latitude areas. In such regions, challenges arise from the degradation of radiative transfer models, band saturation, the pronounced effects of 3D cloud dynamics, and temporal resolution constraints (Li et al., 2021). Identifying these gaps, the study introduces the concept of transfer learning to tackle cases where physical methods degrade and limited training data is available. By combining data from physical simulations and ground observations, the proposed models enhance both the accuracy and adaptability of DSR predictions on a global scale. The investigation further reveals the influence of training data volume on model performance, illustrating how transfer learning can ameliorate these effects (Li et al., 2022). Moreover, the dissertation compares the application of DenseNet, Gated Recurrent Unit (GRU), and a hybrid of Convolutional Neural Network (CNN) and GRU (CNNGRU) to geostationary satellite data, achieving precise and timely DSR estimates. These models underscore their prowess in tackling 3D cloud effects and reducing dependency on additional data sources by the spatial and temporal structure of DL (Li et al., 2023b). Finally, the dissertation introduces the SolarFormer, a space-time transformer neural network adept at forecasting solar radiation up to three hours in advance at 15-minute intervals. By harnessing solely geostationary satellite imagery without the need for ground measurements, this model facilitates expansive DSR predictions, which are crucial for optimizing solar energy distribution at both utility and micro scales. This chapter also highlights the Transformer model's potential for extended forecasting due to its computational and memory efficiency.Item INTEGRATED MONITORING OF DISTURBANCE AND FOREST ATTRIBUTES(2024) Lu, Jiaming; Huang, Chengquan; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Forests provide numerous ecosystem services and are shaped by historical disturbance events. The intensity of disturbance significantly influences the post-disturbance forest structure, species composition, and subsequent forest regrowth. Under the influence of anthropogenic activities and climate change, disturbance regime has undergone unprecedent changes and is subsequently affecting a suite of interrelated forest attributes that are critical in understanding forests dynamics. Historical large-scale disturbance intensity information is needed for understanding the change in disturbance regimes and create linkage to forest dynamics, but such dataset was not available. Forest attributes can be estimated from the spectral information of remote sensing imagery; however, inconsistency exists among the developed product, and the usage of the dataset is limited by accuracy. To fill the research gaps, this dissertation aims to develop a framework that integrates the historical disturbance and the inter-relationship between forest attributes to provide more consistent, and likely more accurate forest attribute estimations. Age, a key attribute that can be the determinant of many ecosystem processes and tree/forest stand develop stage, was selected as the prototype attribute to study. The dissertation started by producing the first set of annual forest disturbance intensity map products quantifying thepercentage of basal area removal (PBAR) at the 30-m resolution for the CONUS from 1986 to 2015, by integrating field plot measurements collected by the Forest Inventory and Analysis program with time series Landsat observations. Compared to other published disturbance products, the maps derived through this study can provide the unique thematic (intensity) information on forest disturbances, precise details critical for understanding forest dynamics across CONUS over multiple decades. The dissertation then proceeded to quantify individual tree age. The tree age was estimated for every tree in the FIA database (over 10 million trees) across the US from our modeling approach that had higher accuracies than existing studies. The developed tree age dataset allows better characterization of tree age distribution, which is important for understanding the disturbance history, functioning, and growth vigor of forest ecosystems. With the disturbance intensity and tree age dataset, the dissertation was able to develop an integrated modeling approach for the forest age mapping. The forest age and complexity maps were produced for 2015 and 2005. The combination of the two metrics should provide a more comprehensive characterization of the forest development condition. The maps provide valuable information for knowing forest conditions, estimating forest growth and carbon sequestration potential, understanding the relationship between age and other forest attributes, evaluating forest health, and planning sustainable forest management practices. This modeling framework developed by the dissertation will enhance the ability to retrieve forest attributes in a broader scale so that with the remote sensing observation, we can not only provide spatially explicit forest structure information, but also review forest status over the decades. Furthermore, when combined with the ecosystem models, these estimations will provide a better prediction for future vegetation and climate dynamics.Item Advances in Mapping Forest Biomass and Old-Growth Conditions Using Waveform Lidar(2023) Bruening, Jamis; Dubayah, Ralph; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The Global Ecosystem Dynamics Investigation (GEDI) is a spaceborne waveform lidar sys- tem that has transformed scientific understanding of the world’s forests through billions of pre- cise measurements of ecosystem structure. Relative to forest processes that operate on decadal to millennial timescales, the four year period during which GEDI collected these measurements is short, and GEDI’s ability to analyze how forest structure changes over time is mostly unproven. However, fusion efforts that integrate GEDI data with forest inventory measurements and ecosys- tem models hold immense potential for discovery. In this dissertation, I explore the limitations and capabilities of GEDI data for inference into structural and successional dynamics within east- ern US forests. First, I used a forest gap model to quantify uncertainty in biomass predictions for individual GEDI waveforms, and discovered a relationship between biomass uncertainty and successional stage. Next, I investigated uncertainties and errors in large-scale GEDI biomass estimates relative to unbiased estimates from the US forest inventory. I developed a novel mod- eling framework based on fusion between GEDI and the US forest inventory data that corrected these errors, and I produced unbiased and precise maps of forest biomass for the continental US. Lastly, I assessed GEDI’s ability to identify and map different types of old-growth forests, and discovered that GEDI can detect some old forests more effectively than others. This research identified key limitations associated with using GEDI to study forest dynamics, and I leveraged these discoveries to develop new ways of using GEDI data for ecological and successional in- ference. These discoveries will inform new uses of GEDI data and its integration with inventory data and ecosystem modeling to better characterize changes within forest ecosystems.Item A SYNTHETIC APERTURE RADAR (SAR)-BASED GENERALIZED APPROACH FOR SUNFLOWER MAPPING AND AREA ESTIMATION(2023) KHAN, MOHAMMAD ABDUL QADIR; Skakun, Sergii; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The effectiveness of remote sensing-based supervised classification models in crop type mapping and area estimation is contingent upon the availability of sufficient and high-quality calibration or training data. The current challenge lies in the absence of field-level crop labels, impeding the advancement of training supervised classification models. To address the needs of operational crop monitoring there is a pressing demand for the development of generalized classification models applicable for various agricultural areas and across different years, even in the absence of calibration data. This dissertation aims to explore the potential of the C-band Sentinel-1 Synthetic Aperture Radar (SAR) capabilities for building generalized crop type models with a specific focus on identifying and monitoring sunflower crop in Eastern Europe. Globally, the sunflower ranks as the fourth most important oilseed crop and stands out as the most profitable and economically significant oilseed crop. It is extensively cultivated for the production of vegetable oil, biodiesel, and animal feed with Ukraine and Russia as the largest producer and exporter in the world. In the first step, this study explores the interaction of Sentinel-1 (S1) SAR signal with sunflower to identify and monitor phenological stages of sunflower. The analysis examines SAR backscattering coefficients and polarizations in Vertical-Horizontal (VH), Vertical-Vertical (VV) and VH/VV ratio, highlighting differences between ascending and descending orbits due to sunflower directional behavior caused by heliotropism. Based on the unique SAR-based signature of sunflower the study introduces a generalized model for sunflower identification and mapping which is applicable across time and space. It was observed that the model based on features acquired from S1-based descending orbits outperforms the one based on ascending orbit because of the sunflower’s directional behavior: user’s accuracy (UA) of 96%, producer’s accuracy (PA) of 97% and F-score of 97% (descending) compared to UA of 90%, PA of 89% and F-score of 90% (ascending). This model was generalized and validated for selected sites in Ukraine, France, Hungary, Russia and USA. When the model is generalized to other years and regions it yields an F-score of > 77% for all cases, with F-score being the highest (>91%) for Mykolaiv region in Ukraine. The generalized approach to map sunflower was applied to assess the impact of the Russian full-scale invasion of Ukraine on national sunflower planted areas. The sunflower planted areas and corresponding changes in 2021 and 2022 were estimated using a sample-based approach for area estimation. Sunflower area was estimated at 7.10±0.45 million hectares (Mha) in 2021 which was further reduced to 6.75±0.45 Mha in 2022 representing a 5% decrease. The findings suggest spatial shifts in sunflower cultivation after the Russian invasion from southern/south-eastern Ukraine under Russian controlled to south-central region under Ukrainian control. The first objective of this dissertation highlights the difference of ascending and descending S1 orbits for sunflower monitoring due to its directional behavior, an aspect not fully researched and documented previously. The implemented generalized approach based on sunflower phenology proves to be an accurate and space-time generalized classifier, reducing time, cost and resources for operational sunflower mapping for large areas. Also, the disparity between sample-based area estimates and SAR-based mapped areas caused due to speckle were substantially reduced emphasizing the role of S1/SAR in global food security monitoring.Item MULTISCALE, MULTITEMPORAL ASSESSMENT OF CHIMPANZEE (Pan troglodytes) HABITAT USING REMOTELY SENSED DATASETS(2023) Jantz, Samuel M; Hansen, Matthew C; Geography/Library & Information Systems; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)All four sub-species of our closest living relative, the chimpanzee, are listed as endangered by the International Union for the Conservation of Nature (IUCN), and their populations continue to decline due to human activities. Effective conservation efforts require information on their population status and distribution. Traditional field surveys are expensive and impractical for covering large areas at regular time intervals, making it difficult to track population trends. Given that chimpanzees occupy a large range (2.3 x 106 km2), new cost-effective methods and data are needed to provide relevant information on population status and trends across large geographic and time scales. The objective of this dissertation is to help fill this gap by leveraging freely available and regularly updated remotely sensed datasets to map and monitor chimpanzee habitat across their range. This research begins by first producing annual forest cover and change maps for the Greater Gombe (GGE) and Greater Mahale ecosystems (GME) in western Tanzania using Landsat phenological metrics and machine learning methods. Canopy cover was predicted at 30-meter resolution and the Cumulative Sums (CuSum) algorithm was applied to the canopy cover time series to detect forest loss and gain events between 2000-2020. An accuracy assessment showed the CuSum algorithm was able to detect forest loss well but had more difficulty detecting gradual forest gain events. A total of 276,000 ha (+/- 27,000 ha) of gross forest loss was detected between 2000 and 2020 in the GGE and GME; however, loss was not spread equally among the two ecosystems. The results show widespread forest loss in the GME, contrasted with net forest cover gain in the GGE. Next, the annual forest cover maps, and additional derived variables, were used to train an ensemble model to predict the relative encounter rate of chimpanzee nest sightings in the GGE and GME. Model output exhibited a strong linear relationship to chimpanzee abundances and population density estimated from a recent ground survey, enabling model output to be linearly transformed into population estimates. The model predicted the two ecosystems harbor just over 3,000 individuals, which agrees with the upper limit of population estimates from ground surveys. Most importantly, the model can be applied to annually updated variables enabling the detection of potential population shifts caused by changes in landscape condition. Model output indicates a possible population reduction in portions of the GME, while the GGE is predicted to have increased its ability to sustain a larger population. Finally, Random Forests regression was used to relate predictor variables, primarily derived from Landsat data to a coarse resolution, range-wide habitat suitability map enabling the prediction of habitat suitability at 30 meter resolution. The model showed good agreement with the calibration data; however, there was considerable variation in predictive capability among the four chimpanzee sub-species. Elevation, Landsat ETM+ band 5 and Landsat derived canopy cover were the strongest predictors; highly suitable areas were associated with dense tree canopy cover for all but the Nigeria-Cameroon and Central Chimpanzee sub-species. The model can detect changes in suitability to support monitoring and conservation planning across the chimpanzee range. Results from this dissertation highlight the promise of integrating continuously updated satellite data into habitat suitability models to detect changes through time and inform conservation efforts for chimpanzees at multiple scales.Item Multispectral satellite remote sensing approaches for estimating cover crop performance in Maryland and Delaware(2022) THIEME, ALISON; Justice, Chris; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Winter cover crops encompass a range of species planted in late summer and fall for a variety of reasons relating to soil health, nutrient retention, soil compaction, biotic diversity, and erosion prevention. As agricultural intensification continues, the practice of winter cover cropping remains a crucial practice to reduce leaching from agricultural fields. Maryland and Delaware both incentivize cover cropping to meet water quality objectives in the Chesapeake Bay Watershed. These large-scale programs necessitate methods to evaluate cover crop performance over the landscape. Cover crop quantity and quality was measured at 2,700 locations between 2006-2021 with a focus on fields planted to four cereal species: wheat, rye, barley, and triticale. Samples were GPS located and timed with satellite remote sensing observations from SPOT 4, SPOT 5, Landsat 5, Landsat 7, Landsat 8, or Sentinel-2. When paired imagery at 10-30 m spatial resolution , there is a strong relationship between the normalized difference vegetation index (NDVI) and percent ground cover (R2=0.72) as well as NDVI and biomass (as high as R2=0.77). There is also a strong relationship between Δ Red Edge (a combination of 740 nm and 783 nm bands) and nitrogen content (R2=0.75). These equations were applied to Harmonized Landsat Sentinel-2 products and used to estimate cover crop aboveground biomass in ~300,000 ha of Maryland Department of Agricultures and ~60,000 ha of Delaware Association of Conservation Districts enrolled fields from 2019-2021 and grouped by agronomic method. Wintertime and springtime cover crop biomass varied based on planting date, planting method, species, termination date, and termination method. Early planted fields had higher wintertime biomass while fields that delayed termination had higher springtime biomass. Triticale had consistently higher biomass while wheat had the lowest biomass. Fields planted using a drill followed by light tillage or no-till drill had higher biomass, likely due to the better seed-to-soil contact. Fields that were taken to harvest or terminated for on farm use (roller crimped, green chopped) also had higher springtime biomass than other termination methods. Incentives can be used to encourage specific agronomic methods and these findings can be used to inform adaptive management in the Mid-Atlantic Region.Item FINE RESOLUTION ASSESSMENT OF THE CARBON FLUXES FROM CONTEMPORARY FOREST DYNAMICS(2021) Gong, Weishu; Huang, Chengquan; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Current estimation of the Earth’s carbon budget contains large uncertainties, with the largest ones in its terrestrial components. With an overarching goal to improve the understanding of carbon budget at regional to global scales, this study aimed to: 1. Develop a grid-based carbon accounting (GCA) model for estimating carbon fluxes from forest disturbance, tested over North Carolina; 2. Develop a consistent timber product output (TPO) record for a globally important timber production region, including seven states in the southeast U.S.; and 3. Further improve the GCA model based on results from objectives 1 and 2, and use it to derive carbon source/sink estimates for all forest land in North Carolina.The results show that several inputs/parameters such as pre-disturbance carbon density, disturbance intensity, allocation of removed carbon among slash and different wood product pools, and forest growth rates could have large impact on carbon estimates. The total emission between 1986 and 2010 from logging over North Carolina was reduced by one third and two thirds, respectively, when remote sensing-based disturbance intensity and biomass data were used to replace parameter values inherited from the original bookkeeping carbon accounting (BCA) model, and was reduced by over 70% when both were used. Use of the TPO data derived in Chapter 3 to partition the removed carbon among slash and different wood product pools resulted in noticeably higher emission estimates than those derived using the partitioning ratios provided by the original BCA model. In addition, without considering legacy effect from wood products harvested before 1986, the emission value derived using the prompt release method was 50% higher than that derived using the delayed release method. This study addresses multiple sources of uncertainties related to the terrestrial carbon budget. The TPO study demonstrated an approach for deriving consistent TPO records for large timber production regions. The GCA model produced state level carbon estimates comparable to those reported by the U.S. Forest Service while providing critical spatial details needed to support carbon management and advance forest-driven climate change mitigation initiatives.Item Towards An Improved Long-term Data Record From The Advanced Very-high Resolution Radiometer: Evaluation, Atmospheric Correction, And Intercalibration(2021) Santamaria Artigas, Andres Eduardo; Justice, Christopher O; Franch, Belen; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Long-term data records from satellite observations are crucial for the study of land surface properties and their long-term dynamics. The AVHRR long term data record (LTDR) is an ongoing effort to generate a consistent climate record of daily atmospherically corrected observations with global coverage that is suitable for long term studies of the Earth surface. In this dissertation, I identified three areas for the improvement of the LTDR: (1) The comprehensive evaluation of the LTDR performance and characterization if its uncertainties. (2) The retrieval of water vapor information from AVHRR data for a more accurate atmospheric correction. (3) The recalibration of the record to address inconsistency issues. The first study consisted on a global long-term evaluation of the LTDR with matched observations from the Landat-5 Thematic Mapper instrument. Results from this evaluation showed that the record performance was close to the proposed specification. The second study proposed a method for the retrieval of water vapor from AVHRR data, which provides a crucial input for the atmospheric correction process. Evaluation of the retrieved values with reference datasets showed excellent results, with a water vapor error lower than 0.45g/cm2. Finally, the last chapter proposed a novel method for the selection of stable areas suitable for satellite intercalibration and for the derivation of recalibration coefficients. The evaluation of the original and recalibrated record showed that for most cases the recalibrated record performed better.Item DYNAMICS OF GLOBAL SURFACE WATER 1999 - PRESENT(2021) Pickens, Amy; Hansen, Matthew C; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Inland surface waters are critical to life, supplying fresh water and habitat, but are constantly in flux. There have been considerable advances in surface water monitoring over the last decade, though the extent of surface water has not been well-quantified per international reporting standards. Global characterizations of change have been primarily bi-temporal. This is problematic due to significant areas with multi-year cycles of wet and dry periods or anomalous high water or drought years. Many areas also exhibit strong seasonal fluctuations, such as floodplains and other natural wetlands. This dissertation aims to characterize open surface water extent dynamics by employing all of the Landsat archive 1999-present, and to report area estimates with associated uncertainty measures as required by policy guidelines. From 1999 to 2018, the extent of permanent water (in liquid or ice state) was 2.93 (standard error ±0.09) million km2, representing only 60.82 (±1.93)% of the total area that had water for some duration of the period. The unidirectional loss and gain areas were relatively small, accounting for only 1.10 (±0.23)% and 2.87 (±0.58)% of total water area, respectively. The area that transitioned multiple times between water and land states on an annual scale was over four times larger (19.74 (±2.16)%), totaling 0.95 (±0.10) million km2, establishing the need to evaluate the time-series from the entire period to assess change dynamics. From a seasonal perspective, June has over double the amount of open surface water as January, with 3.91 (±0.19) million km2 and 1.59 (±0.21) million km2, respectively. This is due to the vast network of lakes and rivers across the high-latitudes of the northern hemisphere that freeze over during the winter, with a maximum extent of ice over areas of permanent and seasonal water in February, totaling 2.49 (±0.25) million km2. This is the first global study to estimate the areas of extent and change with associated uncertainty measures and evaluate the seasonal dynamics of surface water and ice in a combined analysis. The methods developed here provide a framework for continuing to evaluate past trends and monitoring current dynamics of surface water and ice.