Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at a Landscape Scale

Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at a Landscape Scale | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at a Landscape Scale Guillermo E. Ponce-Campos, Philip Heilman, Cynthia L. Norton, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6857473/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 18 Nov, 2025 Read the published version in Landscape Ecology → Version 1 posted 11 You are reading this latest preprint version Abstract Integrating fine-scale measurements with broad-scale monitoring presents a persistent challenge in rangeland ecology, particularly when scaling detailed Unoccupied Aerial System (UAS) observations to satellite-based landscape assessments. This challenge is especially critical as rangelands face increasing climate variability, requiring reliable methods to detect and monitor ecological changes. We investigated how the Largest Patch Index (LPI) of bare ground patches, derived from 3-dimensional UAS observations, can be scaled to landscape levels for mapping bare ground patchiness across a 100 km² semi-arid rangeland in southern Arizona. Our findings reveal three key advances in landscape monitoring. First, LPI effectively captured vegetation responses to extreme climate events during 2019–2023, showing clear sensitivity to both severe drought (SPEI − 2.47) and exceptional wet periods (SPEI + 1.95). Second, LPI values were consistently 30–60% higher in lower elevations, validating the ability to detect known ecological gradients. Third, and most notably, that LPI is positively scale dependent between the 3-m and 30-m grid sizes, and that the magnitude of that difference varies with the density of data from the satellite sensors. This previously unrecognized role of data density challenges fundamental assumptions about scale effects in landscape pattern analysis. Our approach demonstrates a practical solution for integrating UAS and satellite observations, providing a new approach for supporting the detection and monitoring of ecological changes across landscapes, a critical need given increasing climate uncertainty. Multi-scale analysis Largest Patch Index Remote Sensing Ecological Sites Ecological States Rangeland Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction As climate change reshapes our world, understanding the subtle shifts in landscape patterns becomes not just a scientific pursuit, but a critical tool for environmental stewardship. The integration of Unoccupied Aerial Systems (UAS) data with satellite-based remote sensing presents a promising frontier in landscape ecology because it offers unprecedented opportunities to bridge the gap between high-resolution, localized UAS observations with satellites’ broad-scale and dedicated long-term collection from satellites (Steele et al. 2012 ; Karl et al. 2014 ). Previous studies have highlighted the potential of combining these data sources to improve our understanding of complex ecosystems (Anderson and Gaston 2013 ; Solazzo et al. 2018 ; Tmušić et al. 2020 ; Alvarez-Vanhard et al. 2021 ; Wu et al. 2021 ; Villarreal et al. 2025 ). While UAS technology provides extremely detailed spatial information at fine scales, its application to landscape-level analysis has been limited by small spatial coverage and infrequent temporal sampling (Koh and Wich 2012 ; Klosterman et al. 2018 ). Conversely, satellite-based remote sensing offers extensive spatial coverage and high temporal frequency but often lacks the spatial resolution necessary to represent fine-scale ecological processes (Turner and Gardner 2015 ; Gamon et al. 2019 ). We performed the assessment across of ~ 100 km 2 rangeland landscape of mixed-shrub and grass that is representative of a much larger regional landscape in the southwestern US and northern Mexico (Brown et al. 2007 ). Our research addresses the critical topic of scaling up the UAS-derived information and integrating it with satellite information representing the landscape scale (see review by Villarreal et al. 2025 ; and Dash et al. 2018 ; Gillan et al. 2021 ). Specifically, we report the results of developing and evaluating a novel landscape-scale predictive model of bare ground patchiness based on the relationship between fine-scale 3-dimensional (3-D) UAS data with the landscape-scale spectral data from satellite sources. Previous efforts to integrate these sources have integrated fine-scale near-surface cameras and aircraft with landscape-scale satellite data to assess ecosystem functioning (Myers et al. 2024 ), identify areas of ecological concern (van der Leeuw et al. 2024 ) and the spread of invasive species (Villarreal et al. 2019 ). Recently (Ponce-Campos et al. 2023 ), we presented promising results for a UAS-trained landscape-scale predictive model of the patchiness and found that the Largest Patch Index (LPI) was a better predictor of ecological conditions than the percent bare ground or mean fetch (Kuehl et al. 2001 ) distance between plants. ​​LPI is a common metric in landscape ecology to represent spatial arrangement of features by calculating the percentage of a total landscape unit that is covered by the largest contiguous patch of a feature class (McGarigal 2014 ). The present study expands that analysis to a five-year time series and addresses the scale-sensitive behavior of LPI by comparing LPI values calculated at 3-m and 30-m grid cells for the 5-cm UAS data, and by decoupling satellite spatial resolution and data density by resampling 3-m to a coarser 30-m spatial resolution when applying UAS-based training to satellite spectral data. Scale-sensitivity requires deliberate selection of grid size to match the ecological process of interest and the consistent use of a specific spatial scale when comparing LPI values across space or time. Our attention to scale-sensitivity is based on our interest in detecting changes in the erodibility of hillslopes because of an increase in the patch size of bare ground. The attention to bare ground patchiness is related to the basic hillslope erosion process where the erosion probability increases with the length of bare ground flow path because the resulting absence of vegetation obstructions on the hillslope allows a gain in energy that leads to an increased dislodging and transport of soil (Ludwig et al. 2005 ; Poesen 2018 ). In this context, the larger grid size may better represent greater lengths of unobstructed over land flow and increased erosion potential. On the other hand, larger grid size may limit the precision of the spectra-LPI relationship. Finally, the density of data within a grid cell will influence the potential for detecting fine-scale patterning of bare ground patchiness. This attention to bare ground patchiness is an element in distinguishing of erosion-prone ecological conditions in the State-and-Transition framework that represents how vegetation composition (States) responds to climate and management practices (Transitions) (Westoby et al. 1989 ; Briske et al. 2005 ; Steele et al. 2012 ). Specifically, rangeland professionals have identified bare ground patchiness as a distinguishing characteristic of the “Eroded” state within the landscape unit of similar climate, soil, and geomorphology called the Sandy Loam Upland and Deep ecological site (Bestelmeyer et al. 2003 , 2017 ; EDIT, NRCS 2022). Given that preventing and reversing soil erosion is a fundamental goal of rangeland management, we focus our attention on combining fine-scale and large-scale data to detect and monitor the patchiness of bare ground (Olsoy et al. 2018 ). A natural experiment of extraordinary climate variability occurred during the 5-year study (2019-23) and it provided a set of baseline expectations (LPI increases with drier conditions because plant growth is limited) to evaluate the model’s performance for representing LPI. There was a “whiplash” (Scott et al. 2023 ; Swain et al. 2025 ) of precipitation from the driest 12 months on record (July 2019 - June 2020) followed by one of the wettest summer monsoon (2020), and later sequences of unusually dry and wet periods. The large extent of our study area (~ 100 km 2 ) presented the opportunity to evaluate the performance of UAS-based model for satellite representation of LPI. The ground-truth UAS training polygons used to inform the model were located at a higher and wetter elevation, than other portions of the landscape. Therefore, based on earlier work (Ponce et al. 2023) we expected the model to predict greater LPI values for the lower elevations that are 15–25% drier than the higher portion of the landscape where the Training Area was located. Our main objective is to assess the validity of integrating 3-D UAS with satellite data for mapping the patchiness of bare ground at the landscape scale. To this end, we used three performance assessments, and we evaluated a conceptual hypothesis. The performance assessments are 1) the correspondence between UAS-trained satellite estimates of bare ground LPI and the expected responses of LPI to the extreme interannual and inter-seasonal dynamics of precipitation over 5 years, 2) the consistency between the 2019 and the 2023 stand-alone UAS estimates of LPI, and 3) confirming a lower UAS-trained satellite estimate of LPI for the higher elevation Training Area than the portion of the landscape at lower elevations. The conceptual hypothesis is that the scale-sensitive UAS-trained LPI estimates will increase with grid cell size, but the magnitude of that increase will depend on the data density in the grid cell from the satellite sensor. This research advances methodologies for integrating UAS and satellite data to map bare ground patchiness at the landscape scale that is used by land managers to assess soil erosion vulnerability. We also evaluate the scale-sensitive behavior of LPI to inform discussions of best practices regarding spatial resolution for measuring LPI. Methods Study Area This study area is the ~ 200 km 2 Santa Rita Experimental Range (SRER, 31.817° N, 110.851° W, ~ 1,200 m elevation) in southern Arizona (Fig. 1 ). Established in 1902, the SRER is well-studied, and long-term datasets are easily accessed at https://santarita.arizona.edu/ . We focus on the ~ 100 km 2 Sandy Loam, Upland, and Deep (SLUD) ecological site because features of soil, elevation, climate, and vegetation are common throughout the arid southwest of the US (Notaro et al. 2012 ; Bestelmeyer et al. 2018 ) SLUD Ecological Site and Ecological States Our portion of the SLUD is part of the larger Major Land Resource Area (MLRA) 41 − 3 (12–16-inch precipitation zone) (USDA, NRCS 2022 ). Baboquivari and Combate soil series define the Sandy Loam Upland (R041XC319AZ) and Deep (R041XC318AZ) sub-types, respectively, and the former has a weakly developed argillic horizon at 30 cm depth (Breckenfeld and Robinett 2003 ). The area experiences a semi-arid climate with a mean annual temperature of 18.4°C and an average annual precipitation of 358 mm. Monsoonal influences concentrate most rainfall during the summer months (June-September), with the remaining precipitation occurring in winter (October-May) (McClaran and Wei 2014 ). Vegetation is dominated (10–30% canopy cover) by the large shrub Mesquite ( Prosopis velutina ) and various types of grass (1–5% basal cover) including Lehmann lovegrass ( Eragrostis lehmanniana ) and Arizona cottontop ( Digitaria californica ), the small shrub burroweed (1–2% canopy cover; Isocoma tenuisecta ) and cacti (1–2% canopy cover; Cylindroopuntia and Opuntia species (McClaran et al. 2010 ). There are five ecological states within the SLUD ecological site, and we focus on the Eroded State and seek to distinguish it from the other four. The Eroded State is characterized by large, contiguous patches of bare ground, and not simply the extent of bare ground. The species composition is not a defining characteristic because large bare ground patches are possible with different compositions of mesquite, shrubs and grasses. The term Eroded is applied because the large patches of bare ground indicate on-going or potential for accelerated soil erosion given the increased opportunity of surface water flow to gain erosive energy in the absence of vegetation that would otherwise disrupt and slow the overland flow (Michaelides et al. 2009 ; Okin et al. 2009 ; Urgeghe et al. 2021 ). Study Period Precipitation The study period (2019–2023) included some extreme precipitation variability (SPI Explorer Tool, University of Arizona 2024), that has been referred to as “climate whiplash” (Scott et al. 2023 ). Using the standardized precipitation and evapotranspiration index (SPEI; Vicente-Serrano et al. 2013b ), July-September 2020 was the driest summer on record (starting 1895); (SPEI − 2.47) combined with the next 9 months through June 2021, was the driest 12-months on record (SPEI − 2.38). The whiplash occurred when July-September 2021 experienced the 4th wettest summer on record (SPEI + 1.95). The “whiplash” continued through September 2023, where a dry winter (SPEI 1.0). The whiplash subsided when a slightly wet winter 2022-23 followed the wet summer of 2022, but the next summer (2023) was very dry (SPEI − 2.2). UAS polygons and Model Training Area In winter 2019, Robinett Rangeland Resources mapped 10 polygons (1–4 ha) that contained a distinct ecological state (available at https://bit.ly/srer_es_polys ). Half of the polygons were in the Eroded State and half were not Eroded. The 10 polygons are at the higher elevation portion of the SLUD extent, and we designated a best-fit rectangle to serve as a Training Area to build the UAS-trained satellite model of bare ground LPI (Fig. 1 and Fig. 2 Box 1). The dispersion of pure Eroded and Non-eroded ecological states illustrates the diversity of conditions across this landscape. UAS data We collected UAS high-resolution RGB imagery across the polygons in May and September of both 2019 and 2023 using a DJI Phantom 4 RTK equipped with a 20-megapixel global shutter camera, flying at 38 meters above ground with 80% image overlap to achieve 1-cm ground resolution, capturing approximately 200 images per hectare using both nadir (single grid) and 30° oblique (double grid) acquisition patterns (Fig. 2 Box 3; Gillan et al. 2021 ). The May surveys captured the time of lowest herbaceous vegetation productivity, and presumably the easiest time to detection bare ground. Conversely, September surveys aligned with peak herbaceous productivity, offering optimal conditions for capturing spectral signatures of herbs at their maximum photosynthetic activity, but possibly a more difficult to detect bare ground. The 1-cm resolution imagery was processed to generate 5 cm land cover classifications (Fig. 2 Box 3) using a supervised Random Forest classification model implemented in Google Earth Engine (GEE). The model combined spectral data (red, green, and blue bands) with three-dimensional structural information derived from point-cloud data using AgiSoft Metashape (Gillan et al. 2021 ). This classification process resulted in four distinct land cover classes: grass, shrub/tree, bare ground, and shadows (Fig. 3 ). Model inputs included canopy height and spectral data for each pixel, achieving an overall accuracy of 0.92 as documented in the supplemental material of Gillan et al. ( 2021 ). Largest Patch Index and Scale-sensitivity Largest Patch Index (LPI), which represents the percentage of a grid cell occupied by the largest contiguous patch of a feature, in this case bare ground, was calculated using the 5-cm resolution UAS-derived 3-D classification data and two grid cell sizes (3-m and 30-m) in the 10 polygons using Google Earth Engine (GEE) (Google 2015) following the methodology described in Ponce-Campos et al. ( 2023 ). Grid cells of 3-m match PlanetScope's native resolution, while 30-m cells correspond to both the Landsat 8 native resolution and the PlanetScope resampled data. LPI values are expressed as percentages (0-100), with higher values indicating larger contiguous patches of bare ground (Fig. 2 Box 4). The output consists of a point-based shapefile with LPI values for each grid cell. We applied an inner buffer to clip the LPI points where the 5-cm classification image did not fully overlap the satellite grids, thus mitigating edge effects and avoiding incomplete values. Clearly, bare ground LPI differs but is related to percent bare ground (Fig. 4 ), and LPI is scale-sensitive with greater LPI values at 3-m than 30-m grid cell size. These plots represent the outcome of overlaying the grid cells on the UAS 5-cm land cover classification in May and September 2019. LPI is regularly less than percent bare ground, but is equal to bare ground when the largest bare ground patch covers 100% of the grid cell. The chances of a 100% LPI value are greater for the 3-m than 30-m grid cells because the smaller grids are more likely to fit between plants than the larger 30-m grids. Satellite data Spatial resolution (pixel size) and data density (amount of information per area) are key factors of the satellite imagery for this analysis. Higher spatial resolution (smaller pixel size) typically provides more detailed ground information and results in higher data density. We used data from Landsat 8 and PlanetScope satellite platforms because Landsat provides a longer time-series of collection starting in 1984 at 30-m native resolution and 0.02Bytes/m 2 data density (Crawford et al. 2023 ), whereas PlanetScope, became available in 2019 at a 3-m native resolution and data density of 0.9 Bytes/m2 (Planet Labs 2018 ). offers (Table 1 and Fig. 2 Box 5). Data density (B/m 2 ) was calculated as the product of number of spectral bands multiplied by the Bytes per band divided by the pixel area (Table 1 and Fig. 2 Box 5). Specifically, the 0.02 B/m 2 data density for Landsat 8 is based on 7 spectral bands (R, G, B, NIR, SWIR1, SWIR2, and Thermal) multiplied by 2 Bytes per band, and the product is divided by the 900 m2 (30 x 30 m) pixel size. The PlanetScope data density (0.9 B/m²) was derived using the same formula. To decouple the spatial resolution and data density effects, we resampled PlanetScope data to 30-m resolution in Google Earth Engine using a resolution reducer during reprojection to match Landsat's coordinate reference system. This approach creates a statistically meaningful aggregation of the original high-resolution data, as the resampling process employs bilinear interpolation that generates a weighted average of the surrounding pixels, effectively incorporating information from all original 3-m pixels within each 30-m cell (approximately 100 original pixels per 30-m cell). This approach preserves spatial patterns while making the data directly comparable to Landsat's 30-m resolution. The resampled PlanetScope data maintained the spectral characteristics of 4 bands with 2 Bytes per band and the original 3-m pixel size. Table 1 Characteristics of the surface reflectance and data density for the satellite platforms. Platform Bands Data Density Res. Dates Landsat 8 R, G, B, NIR, SWIR1, SWIR2, Thermal. 0.02 (B/m 2 ) 30 m May and September, annually 2019–2023 PlanetScope R, G, B, NIR 0.9 (B/m 2 ) 3 m May and September, annually 2019–2023 Google Earth Engine (GEE; Gorelick et al. 2017 ) served as our primary geospatial processing platform for satellite data analysis. Our workflow commenced with the identification and acquisition of relevant Landsat 8 and PlanetScope imagery corresponding to our study dates, clipping them to the SLUD ecological site within the Santa Rita Experimental Range (SRER), encompassing an area of ~ 100 km² (Fig. 2 Box 5). For each satellite platform, we selected two images per year, late May and late September, to match our UAS survey dates. We applied this approach from 2019 through 2023. Image selection criteria focused on minimal cloud coverage and proximity to the UAS survey dates. The full set of predictive variables (Fig. 2 Box 5) include the original spectral bands and derived indices (Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index 2 (EVI2). Upscaling Largest Patch Index using Random Forest Building upon our previous work (Ponce-Campos et al. 2023 ), we used Random Forest (RF) modeling to build the predictive model for scaling up the UAS-polygon LPI to a landscape-scale metric for each satellite cell in the SLUD across the SRER (Fig. 2 , box 6). Random Forest, an ensemble learning method based on decision trees, offers several advantages for remote sensing applications, including robustness to outliers, ability to handle non-linear relationships, and resistance to overfitting (Breiman 2001 ). The RF model, implemented in GEE was trained using UAS-derived LPI data from 2019 (Fig. 2 , box 3). Surface reflectance spectral bands and vegetation indices in both satellite platforms, Landsat 8 and PlanetScope, served as predictors (Fig. 2 , box 5). The RF model training used a K-fold cross validation to assess model performance, which demonstrated that models including all three indices consistently achieved lower RMSE values compared to models with fewer indices. This empirical validation confirmed that the additional information, despite correlation, contributed positively to model performance. The 2019-trained RF model was then applied to predict LPI values across the SLUD for May and September of each year from 2019 to 2023. This created a time series of 5 years of LPI maps across the SLUD at a landscape-scale (Fig. 2 Box 7). Assigning Largest Patch Index (LPI) to the Eroded Ecological State We assigned the UAS-trained satellite estimates of LPI to a binomial classification of Eroded or Not Eroded ecological state across the entire SLUD ecological site for May and September 2019 through 2023 (Fig. 2 , Box 8). Following Ponce-Campos et al. ( 2023 ), we assigned the pixels as Eroded using the mean value of LPI for all SLUD pixels in May and September 2019, respectively. This approach aligns with commonly used initial thresholding techniques in image analysis (Otsu 1979 ; Glasbey 1993 ), particularly when the distribution does not meet specific requirements for more complex thresholding methods. The mean LPI value from 2019 serves as a threshold, with grid cells equal to or exceeding this value were assigned as Eroded state, and those below as non-eroded. The reasoning behind this approach is illustrated in Fig. 5 , where overlaid histograms demonstrate reasonable separation between ecological states, supporting the utility of this threshold approach for distinguishing between Eroded and non-Eroded conditions across the landscape. Mapping Eroded State and Largest Patch Index (LPI) Mapping the Eroded state extent started with masking out pixels with LPI values below the established threshold and then converting the remaining contiguous pixels into vector format (polygons). The filtered polygons were rasterized, converting the vector data back into a gridded format consistent with our initial satellite imagery (Fig. 2 , box 9). We applied a 1-ha minimum connectivity criterion (Lloyd 2010 ) to represent the coverage of LPI and Eroded state at a spatial scale that is more representative of features large enough to justify land management attention. Performance and Conceptual Hypotheses Assessments To judge the performance of the UAS-trained satellite-based LPI we compared the estimates to a set of expected behaviors related to 1) the response to extreme precipitation patterns over the 2019-23 period, 2) the consistency in LPI between 2019 and 2023 stand-alone UAS estimates of LPI collected for the 10 training polygons given inter-annual trends between those years, and 3) greater LPI values at elevations lower than the Training Area used to build the UAS-based satellite estimates of LPI (Fig. 2 Box 10). We evaluated the conceptual hypothesis that UAS-trained LPI estimates will increase with grid cell size, but that increase will depend on the data density in the cell by comparing the LPI values between 3-m and 30-m grid cell sizes and two data sources that differ by 50 times (Fig. 2 Box 10). Given the strong positive relationship between precipitation and herbaceous ground cover (Ponce-Campos et al. 2013 ; Knapp et al. 2017 ), we expect a strong negative relationship between LPI and precipitation from 2019 through 2023. A poor result would suggest a weak sensitivity to temporal changes in bare ground extent, especially given the large inter-annual and inter-seasonal swings in precipitation and growing conditions for herbaceous vegetation. To judge the validity of the UAS-based estimates of LPI in 2023, we compared them to the UAS-based estimates of LPI in 2019. We expected the estimates to be very similar because the UAS-based estimates of LPI in 2023 were very similar to 2019, even though the LPI varied greatly between 2019 and 2023. A poor result would suggest a systemic drift in the RF model parameters over time. Because the Training Area is at a higher elevation (~ 1200 m) than the majority of the SLUD (~ 950–1050 m) it receives more precipitation, therefore, we expect the LPI values in the Training Area to be consistently less than that estimated for a large portion of the entire SLUD landscape (Ponce-Campos et al. scale 2023). A poor performance would suggest that the UAS-based satellite estimate of LPI had a weak sensitivity to the different conditions of vegetation abundance across the diverse landscape. For the conceptual hypotheses, we first compare the LPI values calculated at 3-m and 30-m grid cells for the 5-cm UAS data from the training polygons, and second we compare LPI among three sources that vary in grid cell size and data density: 30-m at 0.02 B/m 2 (Landsat) and 3-m and 30-m at 0.9B/m 2 (PlanetScope). We reject the scale sensitivity hypothesis if the LPI values estimated using the 3-m resolution are not different than when using the 30-m resolution, and we will reject the amending role of data density if the LPI is not different for the greater data density source (PlanetScope) than the smaller density source (Landsat 8) at 30-m resolution. Results Inter-annual and Inter-seasonal Trends in Largest Patch Index (LPI) and Precipitation The September LPI values are positively related to the extraordinarily dry summers in 2020 and 2023 and negatively related to the consecutive wet summers in 2021 and 2022 (Fig. 6 ). The May LPI values are also positively related to the consecutive dry winters in 2020-21 and 2021-22 and negatively related to the wet winter of 2022-23. The September LPI values had a greater range of values than May LPI over the 5 years. Similarly, the inter-annual range of summer SPEI values were greater across the 5 summers than across the 5 winters. Landsat 8 estimates of September LPI were consistently greater than the native 3-m PlanetScope across the 5 years, but there was not a large difference in May LPI values among the satellites and resolutions. There is not an obvious lag or trans-seasonal trend in September LPI values (Fig. 6 ). For example, the decline in September LPI following wet summers in 2021 and 2022, appears unaffected by the dry winter before those wet summers (2020-21) or the dry winter between those summers (2021-22). However, there was a decline in May LPI in 2023 following a wet 12-month period including the summer of 2022 and the winter of 2022-23. There was no other sequence of consecutive wet summer-wet seasons to judge the consistency of the trans-seasonal trend in the LPI-SPEI relationship. 2019 and 2023 Largest Patch Index (LPI) in Training Area Polygons In both May and September, and for both grid sizes (3-m and 30-m), the UAS-based measures of LPI in the 10 Training Area polygons were greater in 2019 than 2023 (Fig. 7 ). This pattern of greater LPI values in 2019 than 2023 was repeated in 5 of 6 cases for the UAS-trained satellite estimates of LPI (Fig. 7 ). May LPI estimates were always greater in 2019 (~ 20 LPI) than 2023 (~ 8 LPI) for the entire Training Area at all three of the satellite spatial resolution-data density combinations (Fig. 8 ). The one exception to greater LPI in 2019 than 2023 occurred in September using the UAS-trained Landsat 8 data for the entire Training Area (Fig. 8 ). LPI increased from 35 to 48 using the low resolution and low data density Landsat 8 data but declined from 31 to 24 at 30-m resolution and 20 to 17 at 3-m when using the high data density 0.9 B/m 2 PlanetScope data. Largest Patch Index (LPI) and Eroded Ecological State in the Training Area and SLUD As expected, in both May and September, and across all years the LPI was 30–60% greater for the entire SLUD landscape than the Training Area that was located at the higher elevations of the SLUD (Table 2, Fig. 9 ). The difference between the Training Area and entire SLUD was greater in September than May. However, there was a consistent trend in the LPI response to wet and dry periods between the Training Area and the entire SLUD landscape. The extreme dry and wet events between 2020 and 2023 were associated with the maximum and minimum LPI values for both the Training and SLUD areas. The maximum LPI in May followed dry winters in 2021 and 2022 and the minimum value in 2023 followed a wet winter. The maximum LPI in September followed the dry summer in 2020, and the minimum followed the wet summer in 2022. As expected, the coverage of the Eroded ecological state followed the same general pattern as the LPI because it is based on a fixed LPI threshold set in 2019 (Table 2 and Fig. 9 ). The influence of data density on the Eroded state coverage was greater in September than in May, and this was especially obvious in September 2022 when both the Landsat 8 estimate of the Eroded state was 47% but it was only 3–7% for the PlanetScope at 3-m and 30-m resolutions. At that time, most of the Eroded state coverage using Landsat 8 data was at the lower elevation (Fig. 9 ), whereas the location of the Eroded state using the PlanetScope sources was limited to larger dirt roads. Largest Patch Index (LPI), Grid Size, and Data Density As expected, LPI was sensitive to grid cell size in the comparison of 3-m and 30-m for the UAS-based LPI estimates in the 10 Training Area polygons in May and September, 2019 and 2023 (Fig. 8 ). Sensitivity to grid cell size occurred using the UAS-trained satellite estimates of LPI for the Training Area over 5 years (2019-23) and 2 seasons (May and September) in the Training Area, but there was also a relationship with data density at the 30-m grid (Table 2). Lowest mean LPI was estimated using the smallest grid (3-m) and the greatest data density (0.9 B/m 2 ) with the native PlanetScope (Table 2). In contrast, the May estimated LPI was largest in 4 of 5 years when using the large grid and small data density (0.02 B/m 2 ) from Landsat 8, and was lowest for PlanetScope 3-m. The exception was May 2022, when LPI was the same for Landsat and PlanetScope 3-m, and the greatest for PlanetScope resampled to 30-m. The patterns of typically larger LPI for 30-m grid cell, and extent of Eroded State appeared in relation to exceptional wet/dry season extremes (Fig. 9 ) for the entire SLUD landscape between May 2022 (dry winter) and September 2022 (wet summer). In May 2022, LPI and Eroded State extent were nearly identical for Landsat 8 and the 30-m PlanetScope, but by September 2022, the LPI estimates from Landsat 8 were significantly greater, and the Eroded State more extensive than those estimated with the 30-m PlanetScope. Additionally, the native 3-m PlanetScope estimates of LPI and Eroded State were the lowest, with a distribution that was primarily confined to larger dirt roads (Fig. 9 ). Discussion Our integration of 3-D UAS and satellite data showed very promising results for representing inter-seasonal and inter-annual changes in the patchiness of bare ground at the landscape scale. These promising results should encourage more evaluations because the integration addresses shortcomings of both the UAS and satellite systems (Villarreal et al. 2025 ). Specifically, the large-scale spatial coverage and regularly scheduled data collections by satellites overcomes the small-scale coverage of UAS systems and the logistical challenge of mounting UAS campaigns on a regular basis. The fine-scale 3-D resolution of UAS data collections overcomes the inability to precisely identify features and their spatial patterns using satellite data. This integration is particularly critical for rangelands, where traditional remote sensing approaches often miss the fine-scale ecological processes that influence state transitions (Steele et al. 2012 ; Karl et al. 2014 ). Given this promise of integrating the data collections, future efforts are likely to define best practices for UAS-based training of satellite data to represent fine-scale features at the landscape-scale. For example, we used Random Forest models to generate the best-fit model of the UAS and satellite data, but other correlation-prediction approaches and specifications may reveal better approaches for the UAS training of satellite data. Our performance-based assessment relied on a natural experiment (Diamond and Robinson 2010 ; Kennedy et al. 2014 ) approach rather than a goodness of fit between a known value and a modeled value across the 5 years and the large landscape. We did not employ the empirically based approach because of the difficulty in obtaining “known” values of LPI for many more polygons across the landscape, but that effort was beyond the time and resources available during this project. However, our natural experiment approach benefitted from 1) the serendipity of a 5-year period of extraordinary inter-seasonal and inter-annual variation in precipitation, 2) a large elevation gradient within the same soil-based ecological site, and 3) our successful deployment of 4 UAS campaigns (2 seasons in 2 years) at the start and end of the 5-year period. Performance Evaluations The response of LPI to precipitation fluctuations was very consistent with the expected negative relationship SPEI (Ponce-Campos et al. 2013 ; Knapp et al. 2017 ; Dash et al. 2018 ). Greater variation in LPI among the three sources of UAS-based satellite estimates in September than May (Table 2 and Figs. 4 and 6 ) is consistent with the greater spatial variation of summer than winter precipitation on the study area (McClaran and Wei 2014 ). It was reassuring that the pattern of lower LPI values in 2023 than 2019 (May and September) was the same for the UAS-trained satellite estimates and the UAS campaigns (Table 2 and Fig. 7 ) (Koh and Wich 2012 ; Klosterman et al. 2018 ). Inconsistency between the modeled and UAS values would have suggested that our 2019 generated Random Forest-based model was not reliable for estimates of LPI in later years. As expected, LPI estimates were consistently lower for the higher elevation Training Area than the larger SLUD landscape that included drier areas at lower elevations (Table 2 and Fig. 9 ) (Williamson et al. 2016 , McClaran et al. 2010 , Breckenfeld and Robinett 2003 ) . This is consistent with our proof-of-concept findings performed in 2019 (Ponce-Campos et al. 2023 ). Because LPI is a measure of spatial patchiness and not spatial extent, we expected and confirmed that our LPI estimates are sensitive to the grid/pixel size in the analysis of the 5-cm pixel in UAS data and the analysis of the 3-m and 30-m pixels from satellites. The role of data density in the UAS-trained satellite estimates of LPI is reasonable (Michaelides et al. 2009 ; Okin et al. 2009 ) given the 50-times greater density for PlanetScope (0.9 B/m 2 ) than Landsat 8 data (0.02 B/m 2 ). Greater data density should provide a more discriminant estimate of bare ground and its connectivity. Implications for Monitoring LPI at the Landscape-scale Our study demonstrates LPI scale sensitivity through a systematic two-step analytical approach that separates the effects of grid cell size from sensor-specific characteristics. The first step involved applying different grid cell sizes (3-m and 30-m) to our high-resolution 5-cm UAS classification data, isolating the pure effect of spatial aggregation on landscape pattern detection. The second step examined scale effects in satellite-derived estimates by comparing LPI values from native 3-m PlanetScope data, resampled 30-m PlanetScope data, and native 30-m Landsat data, effectively decoupling spatial resolution effects from platform-specific characteristics such as data density and spectral band configuration. The difference in the direction of scale-sensitivity between the UAS-based and UAS-trained satellite estimates of LPI have implications for avoiding direct comparisons of those data sets. LPI values increased with smaller grid cells in the UAS-based analysis of training polygons, but the opposite occurred for LPI estimates based on the UAS-trained satellite for the training polygons and the entire SLUD. In the UAS-based 5-cm cover class surface it was obvious that the smaller 3-m grid cell would fit between plants and thus produce a high LPI value. Conversely, the smaller 3-m grid for UAS-trained satellite estimates of LPI were typically smaller than estimates using 30-m grids. We suspect that the greater information in the finer resolution and larger data density improve the discrimination of bare ground in the RF model, but further inquiry is needed. Given these scale dependencies, any LPI monitoring program must reference both grid cell size and data density. For detecting excessive soil erosion or vulnerability at the landscape scale, we suggest a grid size no smaller than 30-m because the energy of water flowing across 30-m of bare ground without obstruction is substantially greater than across 3-m distances (Ludwig et al. 2005 ; Poesen 2018 ). Data density must also be consistently referenced given continuous advances in satellite platforms, with the longest historic representation available at the lowest data density (Landsat 8). These considerations ensure that LPI monitoring programs maintain consistency and comparability across space and time while accurately representing the ecological processes of interest. Implications for Ecological State and Ecological Site Although the adoption of State and Transition Model of vegetation change and response to management has been pursued for nearly 40 years (Westoby et al. 1989 ; Bestelmeyer et al. 2017 ), experts remain the primary sources for distinguishing ecological states, the relative permanence of a state, and the drivers that cause transitions to alternative states. The ecological state (Eroded or non-Eroded) of training polygons was determined by an expert with extensive SLUD landscape experience. Our results suggest that the Eroded state can transition to non-Eroded, and back again (see Supplementary Material, Fig. S1-3). This relative impermanence warrants a state name such as “Erodible” to reflect a potential for accelerated erosion, rather than “Eroded” which implies on-going accelerated erosion that will need major management inputs to stop the erosion processes. This dynamic behavior aligns with emerging understanding of more volatile climate conditions and their effects on ecosystem responses. As demonstrated by Swain et al. ( 2025 ), hydroclimate whiplash events—rapid transitions between extremely wet and dry conditions—are becoming more frequent and intense with climate change. Our observations of apparent state transitions in response to interannual precipitation variability reflect this increasing climate volatility and suggest that rangeland ecosystems may exhibit more dynamic responses than traditionally conceptualized in State and Transition Models. Under these conditions, "Erodible" would serve more as a warning to inspect and monitor ecosystem trajectory rather than an emergency requiring urgent action, acknowledging that state boundaries may be more fluid in response to increasingly variable climate conditions. The system to map ecological sites is based on the soil, climate, and topography that has been used since the 1960s (originally Range Site) to recognize the relative potential to support different vegetations compositions and productivity, and associated land uses such as livestock grazing and wildlife habitat (USDA-NRCS Title190 2022 ). Mapping ecological sites is very dependent on soil maps. In fact, the mapping of ecological sites in our study area was done simultaneously with soil mapping in the 1990s (Breckenfeld and Robinett 2003 ). The SLUD on our study site, includes a range of 12–16 inches of average annual precipitation, and this was based on the precipitation record into the 1980s. Our results show large differences in the inter-seasonal and inter-annual variation in LPI and Eroded State between the upper and lower elevations of the SLUD (Table 2, Fig. 9 ). Given that spatial variation, and the increase of 1 degree C in the annual temperature on the study area since 1996 (McClaran and Wei 2014 ), we advocate for an effort to adjust all ecological site boundaries upslope to account for the increased evaporative demand and greater vapor pressure deficit with higher temperatures (Novick et al. 2016 ; Williams et al. 2020 ) Conclusion We showed promising results from the integration of 3-D UAS and satellite data to map the fine-scale feature of bare ground patchiness over 5 years and 2 inter-annual seasons. The patchiness metric of LPI (largest patch index) has two important features 1) it reflects soil erosion potential because water energy increases without obstructions that slow the water and 2) it is sensitive to the grid/pixel size of the surface data. Therefore, any effort to monitor LPI using a UAS-trained satellite estimate must maintain the same spatial resolution and data density through the time series. We suggest that 30-m is the smallest resolution to represent soil erosion potential. The extraordinarily large inter-seasonal and inter-annual variation in precipitation during our 5-year study provided a natural experiment to evaluate the expected performance of our UAS-trained satellite estimate of LPI across a 100 km 2 landscape. The resulting landscape-scale view of a fine-scale surface feature provided a critical assessment of the 1) commonly applied state and transition model for eroded ecological states and the 2) current mapping boundaries for ecological sites that were based on climate records that are 40 years old and should be adjusted upslope to account for the 1 degree C increase in annual temperature since 1996. While our approach demonstrates promising results for integrating UAS and satellite data, we recognize opportunities for future expansion and refinement. Our analysis focused on a single ecological site with 10 training polygons and four UAS collection periods, providing a robust proof-of-concept that could be extended to additional rangeland systems and longer temporal series to further validate the approach across diverse environmental conditions. The expert-based ecological state classifications and natural experiment validation framework proved effective for our study objectives, though complementary approaches using standardized field measurements and expanded ground truth data could enhance future applications. Declarations The authors declare no conflict of interest. Acknowledgments We gratefully acknowledge the support and contributions of the USDA-ARS Agreement No. 58-2022-0-009 for funding this research. J. Gillan performed the 2019 UAS field campaign and initial data processing. A. Gorlier assisted with the 2019 UAS field campaign and A. Leimroth assisted with the 2023 field campaign. Data sets were provided by the Santa Rita Experimental Range Digital Database. Funding for the digitization of these data was provided by USDA Forest Service Rocky Mountain Research Station and the University of Arizona, Arizona Experiment Station. This research was a contribution from the Long-Term Agroecosystem Research (LTAR) network. LTAR is supported by the United States Department of Agriculture. Author contributions Conceptualization, M.M., G.E.P., P.H.; Formal analysis, G.E.P., M.M., P.H., S.G., C.N. and M.C.; Funding acquisition, P.H. and M.M.; Methodology, G.E.P., M.M. and P.H.; Resources G.E.P., M.M., P.H., S.G., C.N. and M.C., Writing original draft, G.E.P., M.M., P.H. Writing—review & editing G.E.P., M.M., P.H., S.G., C.N. and M.C. All authors have read and agreed to the published version of the manuscript. Github repository: https://github.com/gponce-ars/scaling-uav/ References Anderson K, Gaston KJ (2013) Lightweight unmanned aerial vehicles will revolutionize spatial ecology. Frontiers in Ecology and the Environment 11:138–146. https://doi.org/10.1890/120150 Alvarez-Vanhard E, Corpetti T, Houet T (2021) UAV & satellite synergies for optical remote sensing applications: A literature review. Science of Remote Sensing 3:100019. https://doi.org/10.1016/j.srs.2021.100019 Bestelmeyer BT, Ash A, Brown JR, et al (2017) State and Transition Models: Theory, Applications, and Challenges. In: Briske DD (ed) Rangeland Systems: Processes, Management and Challenges. Springer International Publishing, Cham, pp 303–345 Bestelmeyer BT, Brown JR, Havstad KM, et al (2003) Development and use of state-and-transition models for rangelands. 56. https://doi.org/10.2458/azu_jrm_v56i2_bestelmeyer Bestelmeyer BT, Peters DPC, Archer SR, et al (2018) The Grassland–Shrubland Regime Shift in the Southwestern United States: Misconceptions and Their Implications for Management. BioScience 68:678–690. https://doi.org/10.1093/biosci/biy065 Breckenfeld DJ, Robinett D (2003) Soil and ecological sites of the Santa Rita Experimental Range. In: McClaran, Mitchel P; Ffolliott, Peter F; Edminster, Carleton B, tech coords Santa Rita Experimental Range: 100 years (1903 to 2003) of accomplishments and contributions; conference proceedings; 2003 October 30-November 1; Tucson, AZ Proc RMRS-P-30 Ogden, UT: US Department of Agriculture, Forest Service, Rocky Mountain Research Station p 157-165 30:157–165 Breiman L (2001) Random Forests. Machine Learning 45:5–32. https://doi.org/10.1023/A:1010933404324 Briske DD, Fuhlendorf SD, Smeins FE (2005) State-and-Transition Models, Thresholds, and Rangeland Health: A Synthesis of Ecological Concepts and Perspectives. Rangeland Ecology & Management 58:1–10. https://doi.org/10.2111/1551-5028(2005)582.0.CO;2 Brown DE, Brennan TC, Unmack PJ (2007) A digitized biotic communities map for plotting and comparing North American plant and animal distributions. Canotia 3:1–12 Crawford CJ, Roy DP, Arab S, et al (2023) The 50-year Landsat collection 2 archive. Science of Remote Sensing 8:100103. https://doi.org/10.1016/j.srs.2023.100103 Dash JP, Pearse GD, Watt MS (2018) UAV Multispectral Imagery Can Complement Satellite Data for Monitoring Forest Health. Remote Sensing 10:1216. https://doi.org/10.3390/rs10081216 Diamond J, Robinson JA (2010) Natural Experiments of History. Harvard University Press, Cambridge, MA EDIT, NRCS U-A (2022) The Ecosystem Dynamics Interpretive Tool (EDIT) Gamon JA, Somers B, Malenovský Z, et al (2019) Assessing Vegetation Function with Imaging Spectroscopy. Surv Geophys 40:489–513. https://doi.org/10.1007/s10712-019-09511-5 Gillan JK, Ponce-Campos GE, Swetnam TL, et al (2021) Innovations to expand drone data collection and analysis for rangeland monitoring. Ecosphere 12:e03649. https://doi.org/10.1002/ecs2.3649 Glasbey C (1993) An Analysis of Histogram-Based Thresholding Algorithms. CVGIP Graph Model Image Process. https://doi.org/10.1006/cgip.1993.1040 Google Inc (2015) Google Earth Engine Platform. In: Google Earth Engine. http://earthengine.google.org/. Accessed 30 Jul 2015 Gorelick N, Hancher M, Dixon M, et al (2017) Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment. https://doi.org/10.1016/j.rse.2017.06.031 Karl JW, Gillan JK, Barger NN, et al (2014) Interpretation of high-resolution imagery for detecting vegetation cover composition change after fuels reduction treatments in woodlands. Ecological Indicators 45:570–578. https://doi.org/10.1016/j.ecolind.2014.05.017 Kennedy RE, Andréfouët S, Cohen WB, et al (2014) Bringing an ecological view of change to Landsat-based remote sensing. Frontiers in Ecology and the Environment 12:339–346. https://doi.org/10.1890/130066 Klosterman S, Melaas E, Wang JA, et al (2018) Fine-scale perspectives on landscape phenology from unmanned aerial vehicle (UAV) photography. Agricultural and Forest Meteorology 248:397–407. https://doi.org/10.1016/j.agrformet.2017.10.015 Knapp AK, Ciais P, Smith MD (2017) Reconciling inconsistencies in precipitation–productivity relationships: implications for climate change. New Phytologist 214:41–47. https://doi.org/10.1111/nph.14381 Koh LP, Wich SA (2012) Dawn of Drone Ecology: Low-Cost Autonomous Aerial Vehicles for Conservation. Tropical Conservation Science 5:121–132. https://doi.org/10.1177/194008291200500202 Kuehl RO, McClaran MP, Va J (2001) Detecting fragmentation of cover in desert grasslands using line intercept. 54. https://doi.org/10.2458/azu_jrm_v54i1_kuehl Lloyd CD (2010) Spatial data analysis: an introduction for GIS users. Oxford University Press, Oxford ; New York Ludwig JA, Wilcox BP, Breshears DD, et al (2005) Vegetation Patches and Runoff–Erosion as Interacting Ecohydrological Processes in Semiarid Landscapes. Ecology 86:288–297. https://doi.org/10.1890/03-0569 McClaran MP, Browning DM, Huang C (2010) Temporal dynamics and spatial variability in desert grassland vegetation. In: Webb RH, Boyer DE, Turner RM (eds) Repeat Photography: Methods and Applications in the Natural Sciences. Island Press McClaran MP, Wei H (2014) Recent drought phase in a 73-year record at two spatial scales: Implications for livestock production on rangelands in the Southwestern United States. Agricultural and Forest Meteorology 197:40–51. https://doi.org/10.1016/j.agrformet.2014.06.004 McGarigal K (2014) Landscape Pattern Metrics. In: Wiley StatsRef: Statistics Reference Online. John Wiley & Sons, Ltd Michaelides K, Lister D, Wainwright J, Parsons AJ (2009) Vegetation controls on small-scale runoff and erosion dynamics in a degrading dryland environment. Hydrological Processes 23:1617–1630. https://doi.org/10.1002/hyp.7293 Myers ER, Browning DM, Burkett LM, et al (2024) Novel Use of Image Time Series to Distinguish Dryland Vegetation Responses to Wet and Dry Years. Journal of Remote Sensing 4:0190. https://doi.org/10.34133/remotesensing.0190 Notaro M, Mauss A, Williams JW (2012) Projected vegetation changes for the American Southwest: combined dynamic modeling and bioclimatic-envelope approach. Ecological Applications 22:1365–1388. https://doi.org/10.1890/11-1269.1 Novick KA, Ficklin DL, Stoy PC, et al (2016) The increasing importance of atmospheric demand for ecosystem water and carbon fluxes. Nature Clim Change 6:1023–1027. https://doi.org/10.1038/nclimate3114 Okin GS, Parsons AJ, Wainwright J, et al (2009) Do Changes in Connectivity Explain Desertification? BioScience 59:237–244. https://doi.org/10.1525/bio.2009.59.3.8 Olsoy PJ, Shipley LA, Rachlow JL, et al (2018) Unmanned aerial systems measure structural habitat features for wildlife across multiple scales. Methods in Ecology and Evolution 9:594–604. https://doi.org/10.1111/2041-210X.12919 Otsu N (1979) A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics 9:62–66. https://doi.org/10.1109/TSMC.1979.4310076 Planet Labs P (2018) Planet Application Program Interface: In Space for Life on Earth Poesen J (2018) Soil erosion in the Anthropocene: Research needs. Earth Surface Processes and Landforms 43:64–84. https://doi.org/10.1002/esp.4250 Ponce-Campos GE, McClaran M, Heilman P, Gillan JK (2023) UAV and Satellite-Based Sensing to Map Ecological States at the Landscape Scale. Open Journal of Ecology 13:560–596. https://doi.org/10.4236/oje.2023.138035 Ponce-Campos GE, Moran MS, Huete A, et al (2013) Ecosystem resilience despite large-scale altered hydroclimatic conditions. Nature 494:349–352. https://doi.org/10.1038/nature11836 Scott RL, Johnston MR, Knowles JF, et al (2023) Interannual variability of spring and summer monsoon growing season carbon exchange at a semiarid savanna over nearly two decades. Agricultural and Forest Meteorology 339:109584. https://doi.org/10.1016/j.agrformet.2023.109584 Solazzo D, Sankey JB, Sankey TTs, Munson SM (2018) Mapping and measuring aeolian sand dunes with photogrammetry and LiDAR from unmanned aerial vehicles (UAV) and multispectral satellite imagery on the Paria Plateau, AZ, USA. Geomorphology 319:174–185. https://doi.org/10.1016/j.geomorph.2018.07.023 SPI Tool, University of Arizona (2024) SPI Explorer Tool. https://uaclimateextension.shinyapps.io/SPItool/ Steele CM, Bestelmeyer BT, Burkett LM, et al (2012) Spatially Explicit Representation of State-and-Transition Models. Rangeland Ecology & Management 65:213–222. https://doi.org/10.2111/REM-D-11-00047.1 Swain DL, Prein AF, Abatzoglou JT, et al (2025) Hydroclimate volatility on a warming Earth. Nat Rev Earth Environ 6:35–50. https://doi.org/10.1038/s43017-024-00624-z Tmušić G, Manfreda S, Aasen H, et al (2020) Current Practices in UAS-based Environmental Monitoring. Remote Sensing 12:1001. https://doi.org/10.3390/rs12061001 Turner MG, Gardner RH (2015) Landscape Ecology in Theory and Practice: Pattern and Process. Springer New York Urgeghe AM, Mayor ÁG, Turrión D, et al (2021) Disentangling the independent effects of vegetation cover and pattern on runoff and sediment yield in dryland systems – Uncovering processes through mimicked plant patches. Journal of Arid Environments 193:104585. https://doi.org/10.1016/j.jaridenv.2021.104585 USDA, NRCS (2022) Land Resource Regions and Major Land Resource Areas of the United States, the Caribbean, and the Pacific Basin. U.S. Department of Agriculture USDA-NRCS Title190 (2022) National Range and Pasture Handbook. United States Department of Agriculture van der Leeuw E, van Leeuwen WJD, Marsh SE, Archer SR (2024) Using Fine Resolution Remotely Sensed Data-Derived Land Cover to Inform Dryland State and Transition Models. Rangeland Ecology & Management 96:128–142. https://doi.org/10.1016/j.rama.2024.06.003 Vicente-Serrano SM, Gouveia C, Camarero JJ, et al (2013a) Response of vegetation to drought time-scales across global land biomes. Proceedings of the National Academy of Sciences 110:52–57. https://doi.org/10.1073/pnas.1207068110 Vicente-Serrano SM, Gouveia C, Camarero JJ, et al (2013b) Response of vegetation to drought time-scales across global land biomes. PNAS 110:52–57. https://doi.org/10.1073/pnas.1207068110 Villarreal et al. 2025. Applications of unoccupied aerial systems (UAS) in landscape ecology: a review of recent research, challenges and emerging opportunities Landsc Ecol (2025) 40:43. https://doi.org/10.1007/s10980-024-02040-6 Villarreal ML, Soulard CE, Waller EK (2019) Landsat Time Series Assessment of Invasive Annual Grasses Following Energy Development. Remote Sensing 11:2553. https://doi.org/10.3390/rs11212553 Westoby M, Walker B, Noy-Meir I (1989) Opportunistic management for rangelands not at equilibrium. 42. https://doi.org/10.2307/3899492 Williams AP, Cook ER, Smerdon JE, et al (2020) Large contribution from anthropogenic warming to an emerging North American megadrought. Science 368:314–318. https://doi.org/10.1126/science.aaz9600 Williamson JC, Bestelmeyer BT, McClaran MP, et al (2016) Can ecological land classification increase the utility of vegetation monitoring data? Ecological Indicators 69:657–666. https://doi.org/10.1016/j.ecolind.2016.05.030 Wu S, Wang J, Yan Z, et al (2021) Monitoring tree-crown scale autumn leaf phenology in a temperate forest with an integration of PlanetScope and drone remote sensing observations. ISPRS Journal of Photogrammetry and Remote Sensing 171:36–48. https://doi.org/10.1016/j.isprsjprs.2020.10.017 Table Table 2 is available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformation.docx Table2.png Cite Share Download PDF Status: Published Journal Publication published 18 Nov, 2025 Read the published version in Landscape Ecology → Version 1 posted Editorial decision: Revision requested 06 Aug, 2025 Reviews received at journal 21 Jul, 2025 Reviews received at journal 13 Jul, 2025 Reviews received at journal 25 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers agreed at journal 16 Jun, 2025 Reviewers agreed at journal 15 Jun, 2025 Reviewers invited by journal 15 Jun, 2025 Editor assigned by journal 11 Jun, 2025 Submission checks completed at journal 11 Jun, 2025 First submitted to journal 09 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6857473","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":472102769,"identity":"120a1c22-bfcb-4e68-8bb1-88e76857803f","order_by":0,"name":"Guillermo E. Ponce-Campos","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIiWNgGAWjYBADOQYGxgYoO4E4Lcaka0lsQLAJaJF37z34uKDGJr1/9uHmDx932DHws+cY4NVieOZcsvGMY2m5M84lNhjOPJPMINnzhoCWGTlm0rwNh3MbzjA2JPO2MTMY3CBky/w3IC3/0+WBWg7zttUz2BPSIi/BA9JyIMHgDGNjM2/bYQYDCQJaDHjyko15jiUbbjzD2Mw4s+04j8SZZwX4bWk/e/AxT42dvNwZ9scfPrZVy/G3J2/Ab8sBHlQBHuzqkG1pIKxmFIyCUTAKRjoAACKFRCS4qtbOAAAAAElFTkSuQmCC","orcid":"","institution":"University of Arizona","correspondingAuthor":true,"prefix":"","firstName":"Guillermo","middleName":"E.","lastName":"Ponce-Campos","suffix":""},{"id":472102770,"identity":"d25d933b-ae67-4042-b468-e2451e02751b","order_by":1,"name":"Philip Heilman","email":"","orcid":"","institution":"USDA Agricultural Research Service","correspondingAuthor":false,"prefix":"","firstName":"Philip","middleName":"","lastName":"Heilman","suffix":""},{"id":472102771,"identity":"6a3bc5d1-9ac5-4ec9-b52e-8a52abca8dee","order_by":2,"name":"Cynthia L. Norton","email":"","orcid":"","institution":"University of Arizona","correspondingAuthor":false,"prefix":"","firstName":"Cynthia","middleName":"L.","lastName":"Norton","suffix":""},{"id":472102772,"identity":"ace88b14-5c55-4dfb-bb51-f52ce0066dac","order_by":3,"name":"Shang Gao","email":"","orcid":"","institution":"University of Arizona","correspondingAuthor":false,"prefix":"","firstName":"Shang","middleName":"","lastName":"Gao","suffix":""},{"id":472102773,"identity":"2ab1eb28-17da-4a0a-8922-fe55295e66d3","order_by":4,"name":"Michael Crimmins","email":"","orcid":"","institution":"University of Arizona","correspondingAuthor":false,"prefix":"","firstName":"Michael","middleName":"","lastName":"Crimmins","suffix":""},{"id":472102774,"identity":"1adab322-b1cf-4400-a8a2-552e750e88f2","order_by":5,"name":"Mitchel P. McClaran","email":"","orcid":"","institution":"University of Arizona","correspondingAuthor":false,"prefix":"","firstName":"Mitchel","middleName":"P.","lastName":"McClaran","suffix":""}],"badges":[],"createdAt":"2025-06-09 21:38:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6857473/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6857473/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10980-025-02226-6","type":"published","date":"2025-11-18T15:58:52+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":84888613,"identity":"49910698-5b6d-4fdb-a96b-b57cf526a8b4","added_by":"auto","created_at":"2025-06-18 12:13:29","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":162890,"visible":true,"origin":"","legend":"\u003cp\u003eLocation and boundary (beige) of the Santa Rita Experimental Range in southern Arizona, USA, the SLUD Ecological Site (gray), the 10 training polygons where UAS data was collected, and Training Area boundary (black outline rectangle) that contains the 10 polygons used to create the UAS-trained satellite estimate of bare ground Largest Patch Index.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/d724b3896a2647217d3618da.png"},{"id":84888612,"identity":"2d40d266-c1a1-46ae-aba0-0d8a2585f177","added_by":"auto","created_at":"2025-06-18 12:13:28","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":123470,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow for gathering, processing, modeling, projecting, and interpreting the integration of ground, UAS, and satellite data to represent extent and temporal dynamics of bare ground LPI and the Eroded ecological state.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/55d9043b061e0bfe2d21a9b3.png"},{"id":84888615,"identity":"d3552422-b1bf-4fa4-99bf-92229df9424e","added_by":"auto","created_at":"2025-06-18 12:13:29","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":618185,"visible":true,"origin":"","legend":"\u003cp\u003eHigh-resolution 5 cm land cover classification derived from UAS 1 cm data for a representative 4-ha polygon within the Training Area. Superimposed on the classification is a 30-m grid (red dashed lines) representing Landsat 8 pixels, with white dots at pixel centroids (Figure 2 Boxes 2 and 3). Modified from Ponce-Campos et al. (2023).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/69e042f7e406d949325020df.png"},{"id":84888885,"identity":"03664f1e-5f41-40df-92ab-e6dbd8f0d1ee","added_by":"auto","created_at":"2025-06-18 12:21:29","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":239672,"visible":true,"origin":"","legend":"\u003cp\u003eRelationship between bareground cover (%) and the Largest Patch Index (LPI, %) across two seasonal periods (May and September 2019) and two spatial resolutions (30m Landsat and 3m PlanetScope).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/40e57efa856f88bf9fc2bb84.png"},{"id":84888884,"identity":"82c3a325-9a35-484d-8046-42f7ce0422c8","added_by":"auto","created_at":"2025-06-18 12:21:29","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":65710,"visible":true,"origin":"","legend":"\u003cp\u003eFrequency distributions of LPI values for Eroded and Non-Eroded ecological states during May at 3-m and 30-m spatial resolutions. The red dashed line represents the mean LPI threshold used to distinguish between ecological states.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/e2a3ac5ae57d991672e45bac.png"},{"id":84888620,"identity":"31507ebf-9de2-46fd-8062-24423cb15689","added_by":"auto","created_at":"2025-06-18 12:13:29","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":86731,"visible":true,"origin":"","legend":"\u003cp\u003eInter-annual and inter-seasonal (May and September) timeline of Standardized Precipitation-Evapotransporation Index (SPEI) and Largest Patch Index (LPI) across different satellite grid resolutions (Landsat, PlanetScope 3-m and PlanetScope 30-m). Results are for the Training Area and not the entire SLUD landscape. SPEI for winter (5-month scale ending March) and summer (3-month scale ending September) depicted by the bar plots.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/cc62ec487e60c6fa6587307d.png"},{"id":84888890,"identity":"aa8dcf3f-8f56-4c75-879d-1b05dda6ce5b","added_by":"auto","created_at":"2025-06-18 12:21:29","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":137980,"visible":true,"origin":"","legend":"\u003cp\u003eUAS-based 30 m and 3 m grid cells in May and September of 2019 and 2023. Summary statistics are provided in Table S1.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/7436a947fd8394266f551417.png"},{"id":84888629,"identity":"7d6b5456-3c15-4e97-a8b2-7c3d61663e93","added_by":"auto","created_at":"2025-06-18 12:13:29","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":170580,"visible":true,"origin":"","legend":"\u003cp\u003eNormalized frequency distributions of predicted Largest Patch Index (LPI) values across the Training Area by satellite (Landsat and PlanetScope, including downscale PlanetScope 30m grid) in May and September from 2019 through 2023. The dashed vertical line is the mean LPI for 2019, which provides the reference for inter-annual patterns.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/0c9fb55e64dcffd739a934cd.png"},{"id":84888894,"identity":"938b7435-7686-4392-953e-a95bcf6c6bc5","added_by":"auto","created_at":"2025-06-18 12:21:29","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":510858,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of Largest Patch Index (LPI) values and extent of estimated Eroded ecological state using three combinations of satellite grid resolution and data density for May and September 2022. (a) Landsat (30-m and 0.02 B/m\u003csup\u003e2\u003c/sup\u003e), (b) resampled PlanetScope (30-m and 0.9 B/m\u003csup\u003e2\u003c/sup\u003e), and (c) PlanetScope (3-m and 0.9 B/m\u003csup\u003e2\u003c/sup\u003e). Red pixels for the Eroded state show where the LPI was greater than the LPI threshold defining the Eroded ecological state.\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/039c8604ab43440917dfe048.png"},{"id":96650383,"identity":"7653f0e8-fdef-4aaa-a37d-a503383e4aa5","added_by":"auto","created_at":"2025-11-24 16:11:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2533086,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/b935b35d-c32f-4e08-bb3c-2474014eb950.pdf"},{"id":84889573,"identity":"d5ea7b05-e20f-439c-9d63-e1017e56979f","added_by":"auto","created_at":"2025-06-18 12:29:29","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5300934,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/54e888bf357d6d4583893aec.docx"},{"id":84889572,"identity":"13ec0055-7972-49a4-930f-5f2a877ee70a","added_by":"auto","created_at":"2025-06-18 12:29:29","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":150781,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.png","url":"https://assets-eu.researchsquare.com/files/rs-6857473/v1/9f8407a911bd23fd2da2da20.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at a Landscape Scale","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAs climate change reshapes our world, understanding the subtle shifts in landscape patterns becomes not just a scientific pursuit, but a critical tool for environmental stewardship. The integration of Unoccupied Aerial Systems (UAS) data with satellite-based remote sensing presents a promising frontier in landscape ecology because it offers unprecedented opportunities to bridge the gap between high-resolution, localized UAS observations with satellites\u0026rsquo; broad-scale and dedicated long-term collection from satellites (Steele et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Karl et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Previous studies have highlighted the potential of combining these data sources to improve our understanding of complex ecosystems (Anderson and Gaston \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Solazzo et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Tmušić et al. \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Alvarez-Vanhard et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Wu et al. \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Villarreal et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). While UAS technology provides extremely detailed spatial information at fine scales, its application to landscape-level analysis has been limited by small spatial coverage and infrequent temporal sampling (Koh and Wich \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Klosterman et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Conversely, satellite-based remote sensing offers extensive spatial coverage and high temporal frequency but often lacks the spatial resolution necessary to represent fine-scale ecological processes (Turner and Gardner \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Gamon et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). We performed the assessment across of ~\u0026thinsp;100 km\u003csup\u003e2\u003c/sup\u003e rangeland landscape of mixed-shrub and grass that is representative of a much larger regional landscape in the southwestern US and northern Mexico (Brown et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2007\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eOur research addresses the critical topic of scaling up the UAS-derived information and integrating it with satellite information representing the landscape scale (see review by Villarreal et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; and Dash et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Gillan et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Specifically, we report the results of developing and evaluating a novel landscape-scale predictive model of bare ground patchiness based on the relationship between fine-scale 3-dimensional (3-D) UAS data with the landscape-scale spectral data from satellite sources. Previous efforts to integrate these sources have integrated fine-scale near-surface cameras and aircraft with landscape-scale satellite data to assess ecosystem functioning (Myers et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), identify areas of ecological concern (van der Leeuw et al. \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and the spread of invasive species (Villarreal et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Recently (Ponce-Campos et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), we presented promising results for a UAS-trained landscape-scale predictive model of the patchiness and found that the Largest Patch Index (LPI) was a better predictor of ecological conditions than the percent bare ground or mean fetch (Kuehl et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2001\u003c/span\u003e) distance between plants. ​​LPI is a common metric in landscape ecology to represent spatial arrangement of features by calculating the percentage of a total landscape unit that is covered by the largest contiguous patch of a feature class (McGarigal \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). The present study expands that analysis to a five-year time series and addresses the scale-sensitive behavior of LPI by comparing LPI values calculated at 3-m and 30-m grid cells for the 5-cm UAS data, and by decoupling satellite spatial resolution and data density by resampling 3-m to a coarser 30-m spatial resolution when applying UAS-based training to satellite spectral data. Scale-sensitivity requires deliberate selection of grid size to match the ecological process of interest and the consistent use of a specific spatial scale when comparing LPI values across space or time.\u003c/p\u003e \u003cp\u003eOur attention to scale-sensitivity is based on our interest in detecting changes in the erodibility of hillslopes because of an increase in the patch size of bare ground. The attention to bare ground patchiness is related to the basic hillslope erosion process where the erosion probability increases with the length of bare ground flow path because the resulting absence of vegetation obstructions on the hillslope allows a gain in energy that leads to an increased dislodging and transport of soil (Ludwig et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Poesen \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). In this context, the larger grid size may better represent greater lengths of unobstructed over land flow and increased erosion potential. On the other hand, larger grid size may limit the precision of the spectra-LPI relationship. Finally, the density of data within a grid cell will influence the potential for detecting fine-scale patterning of bare ground patchiness.\u003c/p\u003e \u003cp\u003eThis attention to bare ground patchiness is an element in distinguishing of erosion-prone ecological conditions in the State-and-Transition framework that represents how vegetation composition (States) responds to climate and management practices (Transitions) (Westoby et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Briske et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Steele et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). Specifically, rangeland professionals have identified bare ground patchiness as a distinguishing characteristic of the \u0026ldquo;Eroded\u0026rdquo; state within the landscape unit of similar climate, soil, and geomorphology called the Sandy Loam Upland and Deep ecological site (Bestelmeyer et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2003\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; EDIT, NRCS 2022). Given that preventing and reversing soil erosion is a fundamental goal of rangeland management, we focus our attention on combining fine-scale and large-scale data to detect and monitor the patchiness of bare ground (Olsoy et al. \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eA natural experiment of extraordinary climate variability occurred during the 5-year study (2019-23) and it provided a set of baseline expectations (LPI increases with drier conditions because plant growth is limited) to evaluate the model\u0026rsquo;s performance for representing LPI. There was a \u0026ldquo;whiplash\u0026rdquo; (Scott et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Swain et al. \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) of precipitation from the driest 12 months on record (July 2019 - June 2020) followed by one of the wettest summer monsoon (2020), and later sequences of unusually dry and wet periods.\u003c/p\u003e \u003cp\u003eThe large extent of our study area (~\u0026thinsp;100 km\u003csup\u003e2\u003c/sup\u003e) presented the opportunity to evaluate the performance of UAS-based model for satellite representation of LPI. The ground-truth UAS training polygons used to inform the model were located at a higher and wetter elevation, than other portions of the landscape. Therefore, based on earlier work (Ponce et al. 2023) we expected the model to predict greater LPI values for the lower elevations that are 15\u0026ndash;25% drier than the higher portion of the landscape where the Training Area was located.\u003c/p\u003e \u003cp\u003eOur main objective is to assess the validity of integrating 3-D UAS with satellite data for mapping the patchiness of bare ground at the landscape scale. To this end, we used three performance assessments, and we evaluated a conceptual hypothesis. The performance assessments are 1) the correspondence between UAS-trained satellite estimates of bare ground LPI and the expected responses of LPI to the extreme interannual and inter-seasonal dynamics of precipitation over 5 years, 2) the consistency between the 2019 and the 2023 stand-alone UAS estimates of LPI, and 3) confirming a lower UAS-trained satellite estimate of LPI for the higher elevation Training Area than the portion of the landscape at lower elevations. The conceptual hypothesis is that the scale-sensitive UAS-trained LPI estimates will increase with grid cell size, but the magnitude of that increase will depend on the data density in the grid cell from the satellite sensor.\u003c/p\u003e \u003cp\u003eThis research advances methodologies for integrating UAS and satellite data to map bare ground patchiness at the landscape scale that is used by land managers to assess soil erosion vulnerability. We also evaluate the scale-sensitive behavior of LPI to inform discussions of best practices regarding spatial resolution for measuring LPI.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eStudy Area\u003c/p\u003e \u003cp\u003eThis study area is the ~\u0026thinsp;200 km\u003csup\u003e2\u003c/sup\u003e Santa Rita Experimental Range (SRER, 31.817\u0026deg; N, 110.851\u0026deg; W, ~\u0026thinsp;1,200 m elevation) in southern Arizona (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Established in 1902, the SRER is well-studied, and long-term datasets are easily accessed at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://santarita.arizona.edu/\u003c/span\u003e\u003cspan address=\"https://santarita.arizona.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. We focus on the ~\u0026thinsp;100 km\u003csup\u003e2\u003c/sup\u003e Sandy Loam, Upland, and Deep (SLUD) ecological site because features of soil, elevation, climate, and vegetation are common throughout the arid southwest of the US (Notaro et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Bestelmeyer et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e)\u003c/p\u003e \u003cp\u003eSLUD Ecological Site and Ecological States\u003c/p\u003e \u003cp\u003eOur portion of the SLUD is part of the larger Major Land Resource Area (MLRA) 41\u0026thinsp;\u0026minus;\u0026thinsp;3 (12\u0026ndash;16-inch precipitation zone) (USDA, NRCS \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Baboquivari and Combate soil series define the Sandy Loam Upland (R041XC319AZ) and Deep (R041XC318AZ) sub-types, respectively, and the former has a weakly developed argillic horizon at 30 cm depth (Breckenfeld and Robinett \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2003\u003c/span\u003e). The area experiences a semi-arid climate with a mean annual temperature of 18.4\u0026deg;C and an average annual precipitation of 358 mm. Monsoonal influences concentrate most rainfall during the summer months (June-September), with the remaining precipitation occurring in winter (October-May) (McClaran and Wei \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Vegetation is dominated (10\u0026ndash;30% canopy cover) by the large shrub Mesquite (\u003cem\u003eProsopis velutina\u003c/em\u003e) and various types of grass (1\u0026ndash;5% basal cover) including Lehmann lovegrass (\u003cem\u003eEragrostis lehmanniana\u003c/em\u003e) and Arizona cottontop (\u003cem\u003eDigitaria californica\u003c/em\u003e), the small shrub burroweed (1\u0026ndash;2% canopy cover; \u003cem\u003eIsocoma tenuisecta\u003c/em\u003e) and cacti (1\u0026ndash;2% canopy cover; \u003cem\u003eCylindroopuntia\u003c/em\u003e and \u003cem\u003eOpuntia\u003c/em\u003e species (McClaran et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2010\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThere are five ecological states within the SLUD ecological site, and we focus on the Eroded State and seek to distinguish it from the other four. The Eroded State is characterized by large, contiguous patches of bare ground, and not simply the extent of bare ground. The species composition is not a defining characteristic because large bare ground patches are possible with different compositions of mesquite, shrubs and grasses. The term Eroded is applied because the large patches of bare ground indicate on-going or potential for accelerated soil erosion given the increased opportunity of surface water flow to gain erosive energy in the absence of vegetation that would otherwise disrupt and slow the overland flow (Michaelides et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Okin et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Urgeghe et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eStudy Period Precipitation\u003c/p\u003e \u003cp\u003eThe study period (2019\u0026ndash;2023) included some extreme precipitation variability (SPI Explorer Tool, University of Arizona 2024), that has been referred to as \u0026ldquo;climate whiplash\u0026rdquo; (Scott et al. \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Using the standardized precipitation and evapotranspiration index (SPEI; Vicente-Serrano et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2013b\u003c/span\u003e), July-September 2020 was the driest summer on record (starting 1895); (SPEI \u0026minus;\u0026thinsp;2.47) combined with the next 9 months through June 2021, was the driest 12-months on record (SPEI \u0026minus;\u0026thinsp;2.38). The whiplash occurred when July-September 2021 experienced the 4th wettest summer on record (SPEI\u0026thinsp;+\u0026thinsp;1.95). The \u0026ldquo;whiplash\u0026rdquo; continued through September 2023, where a dry winter (SPEI \u0026lt; -1.0) was followed by a wet summer (SPEI\u0026thinsp;\u0026gt;\u0026thinsp;1.0). The whiplash subsided when a slightly wet winter 2022-23 followed the wet summer of 2022, but the next summer (2023) was very dry (SPEI \u0026minus;\u0026thinsp;2.2).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eUAS polygons and Model Training Area\u003c/p\u003e \u003cp\u003eIn winter 2019, Robinett Rangeland Resources mapped 10 polygons (1\u0026ndash;4 ha) that contained a distinct ecological state (available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bit.ly/srer_es_polys\u003c/span\u003e\u003cspan address=\"https://bit.ly/srer_es_polys\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e).\u003c/span\u003e Half of the polygons were in the Eroded State and half were not Eroded. The 10 polygons are at the higher elevation portion of the SLUD extent, and we designated a best-fit rectangle to serve as a Training Area to build the UAS-trained satellite model of bare ground LPI (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 1). The dispersion of pure Eroded and Non-eroded ecological states illustrates the diversity of conditions across this landscape.\u003c/p\u003e \u003cp\u003eUAS data\u003c/p\u003e \u003cp\u003eWe collected UAS high-resolution RGB imagery across the polygons in May and September of both 2019 and 2023 using a DJI Phantom 4 RTK equipped with a 20-megapixel global shutter camera, flying at 38 meters above ground with 80% image overlap to achieve 1-cm ground resolution, capturing approximately 200 images per hectare using both nadir (single grid) and 30\u0026deg; oblique (double grid) acquisition patterns (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 3; Gillan et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The May surveys captured the time of lowest herbaceous vegetation productivity, and presumably the easiest time to detection bare ground. Conversely, September surveys aligned with peak herbaceous productivity, offering optimal conditions for capturing spectral signatures of herbs at their maximum photosynthetic activity, but possibly a more difficult to detect bare ground.\u003c/p\u003e \u003cp\u003eThe 1-cm resolution imagery was processed to generate 5 cm land cover classifications (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 3) using a supervised Random Forest classification model implemented in Google Earth Engine (GEE). The model combined spectral data (red, green, and blue bands) with three-dimensional structural information derived from point-cloud data using AgiSoft Metashape (Gillan et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). This classification process resulted in four distinct land cover classes: grass, shrub/tree, bare ground, and shadows (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Model inputs included canopy height and spectral data for each pixel, achieving an overall accuracy of 0.92 as documented in the supplemental material of Gillan et al. (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eLargest Patch Index and Scale-sensitivity\u003c/p\u003e \u003cp\u003eLargest Patch Index (LPI), which represents the percentage of a grid cell occupied by the largest contiguous patch of a feature, in this case bare ground, was calculated using the 5-cm resolution UAS-derived 3-D classification data and two grid cell sizes (3-m and 30-m) in the 10 polygons using Google Earth Engine (GEE) (Google 2015) following the methodology described in Ponce-Campos et al. (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Grid cells of 3-m match PlanetScope's native resolution, while 30-m cells correspond to both the Landsat 8 native resolution and the PlanetScope resampled data. LPI values are expressed as percentages (0-100), with higher values indicating larger contiguous patches of bare ground (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 4). The output consists of a point-based shapefile with LPI values for each grid cell. We applied an inner buffer to clip the LPI points where the 5-cm classification image did not fully overlap the satellite grids, thus mitigating edge effects and avoiding incomplete values.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eClearly, bare ground LPI differs but is related to percent bare ground (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e), and LPI is scale-sensitive with greater LPI values at 3-m than 30-m grid cell size.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese plots represent the outcome of overlaying the grid cells on the UAS 5-cm land cover classification in May and September 2019. LPI is regularly less than percent bare ground, but is equal to bare ground when the largest bare ground patch covers 100% of the grid cell. The chances of a 100% LPI value are greater for the 3-m than 30-m grid cells because the smaller grids are more likely to fit between plants than the larger 30-m grids.\u003c/p\u003e \u003cp\u003eSatellite data\u003c/p\u003e \u003cp\u003eSpatial resolution (pixel size) and data density (amount of information per area) are key factors of the satellite imagery for this analysis. Higher spatial resolution (smaller pixel size) typically provides more detailed ground information and results in higher data density. We used data from Landsat 8 and PlanetScope satellite platforms because Landsat provides a longer time-series of collection starting in 1984 at 30-m native resolution and 0.02Bytes/m\u003csup\u003e2\u003c/sup\u003e data density (Crawford et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), whereas PlanetScope, became available in 2019 at a 3-m native resolution and data density of 0.9 Bytes/m2 (Planet Labs \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). offers (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 5). Data density (B/m\u003csup\u003e2\u003c/sup\u003e) was calculated as the product of number of spectral bands multiplied by the Bytes per band divided by the pixel area (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 5). Specifically, the 0.02 B/m\u003csup\u003e2\u003c/sup\u003e data density for Landsat 8 is based on 7 spectral bands (R, G, B, NIR, SWIR1, SWIR2, and Thermal) multiplied by 2 Bytes per band, and the product is divided by the 900 m2 (30 x 30 m) pixel size. The PlanetScope data density (0.9 B/m\u0026sup2;) was derived using the same formula.\u003c/p\u003e \u003cp\u003eTo decouple the spatial resolution and data density effects, we resampled PlanetScope data to 30-m resolution in Google Earth Engine using a resolution reducer during reprojection to match Landsat's coordinate reference system. This approach creates a statistically meaningful aggregation of the original high-resolution data, as the resampling process employs bilinear interpolation that generates a weighted average of the surrounding pixels, effectively incorporating information from all original 3-m pixels within each 30-m cell (approximately 100 original pixels per 30-m cell). This approach preserves spatial patterns while making the data directly comparable to Landsat's 30-m resolution. The resampled PlanetScope data maintained the spectral characteristics of 4 bands with 2 Bytes per band and the original 3-m pixel size.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCharacteristics of the surface reflectance and data density for the satellite platforms.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlatform\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBands\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eData Density\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRes.\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDates\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLandsat 8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eR, G, B, NIR, SWIR1, SWIR2, Thermal.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.02 (B/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMay and September, annually 2019\u0026ndash;2023\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePlanetScope\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eR, G, B, NIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.9 (B/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3 m\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMay and September, annually 2019\u0026ndash;2023\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eGoogle Earth Engine (GEE; Gorelick et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) served as our primary geospatial processing platform for satellite data analysis. Our workflow commenced with the identification and acquisition of relevant Landsat 8 and PlanetScope imagery corresponding to our study dates, clipping them to the SLUD ecological site within the Santa Rita Experimental Range (SRER), encompassing an area of ~\u0026thinsp;100 km\u0026sup2; (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 5). For each satellite platform, we selected two images per year, late May and late September, to match our UAS survey dates. We applied this approach from 2019 through 2023. Image selection criteria focused on minimal cloud coverage and proximity to the UAS survey dates. The full set of predictive variables (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 5) include the original spectral bands and derived indices (Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI), and Enhanced Vegetation Index 2 (EVI2).\u003c/p\u003e \u003cp\u003eUpscaling Largest Patch Index using Random Forest\u003c/p\u003e \u003cp\u003eBuilding upon our previous work (Ponce-Campos et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), we used Random Forest (RF) modeling to build the predictive model for scaling up the UAS-polygon LPI to a landscape-scale metric for each satellite cell in the SLUD across the SRER (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, box 6). Random Forest, an ensemble learning method based on decision trees, offers several advantages for remote sensing applications, including robustness to outliers, ability to handle non-linear relationships, and resistance to overfitting (Breiman \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). The RF model, implemented in GEE was trained using UAS-derived LPI data from 2019 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, box 3). Surface reflectance spectral bands and vegetation indices in both satellite platforms, Landsat 8 and PlanetScope, served as predictors (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, box 5). The RF model training used a K-fold cross validation to assess model performance, which demonstrated that models including all three indices consistently achieved lower RMSE values compared to models with fewer indices. This empirical validation confirmed that the additional information, despite correlation, contributed positively to model performance. The 2019-trained RF model was then applied to predict LPI values across the SLUD for May and September of each year from 2019 to 2023. This created a time series of 5 years of LPI maps across the SLUD at a landscape-scale (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 7).\u003c/p\u003e \u003cp\u003eAssigning Largest Patch Index (LPI) to the Eroded Ecological State\u003c/p\u003e \u003cp\u003eWe assigned the UAS-trained satellite estimates of LPI to a binomial classification of Eroded or Not Eroded ecological state across the entire SLUD ecological site for May and September 2019 through 2023 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, Box 8). Following Ponce-Campos et al. (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), we assigned the pixels as Eroded using the mean value of LPI for all SLUD pixels in May and September 2019, respectively. This approach aligns with commonly used initial thresholding techniques in image analysis (Otsu \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e1979\u003c/span\u003e; Glasbey \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e1993\u003c/span\u003e), particularly when the distribution does not meet specific requirements for more complex thresholding methods. The mean LPI value from 2019 serves as a threshold, with grid cells equal to or exceeding this value were assigned as Eroded state, and those below as non-eroded. The reasoning behind this approach is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, where overlaid histograms demonstrate reasonable separation between ecological states, supporting the utility of this threshold approach for distinguishing between Eroded and non-Eroded conditions across the landscape.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eMapping Eroded State and Largest Patch Index (LPI)\u003c/p\u003e \u003cp\u003eMapping the Eroded state extent started with masking out pixels with LPI values below the established threshold and then converting the remaining contiguous pixels into vector format (polygons). The filtered polygons were rasterized, converting the vector data back into a gridded format consistent with our initial satellite imagery (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, box 9). We applied a 1-ha minimum connectivity criterion (Lloyd \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) to represent the coverage of LPI and Eroded state at a spatial scale that is more representative of features large enough to justify land management attention.\u003c/p\u003e \u003cp\u003ePerformance and Conceptual Hypotheses Assessments\u003c/p\u003e \u003cp\u003eTo judge the performance of the UAS-trained satellite-based LPI we compared the estimates to a set of expected behaviors related to 1) the response to extreme precipitation patterns over the 2019-23 period, 2) the consistency in LPI between 2019 and 2023 stand-alone UAS estimates of LPI collected for the 10 training polygons given inter-annual trends between those years, and 3) greater LPI values at elevations lower than the Training Area used to build the UAS-based satellite estimates of LPI (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 10). We evaluated the conceptual hypothesis that UAS-trained LPI estimates will increase with grid cell size, but that increase will depend on the data density in the cell by comparing the LPI values between 3-m and 30-m grid cell sizes and two data sources that differ by 50 times (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e Box 10).\u003c/p\u003e \u003cp\u003eGiven the strong positive relationship between precipitation and herbaceous ground cover (Ponce-Campos et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Knapp et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), we expect a strong negative relationship between LPI and precipitation from 2019 through 2023. A poor result would suggest a weak sensitivity to temporal changes in bare ground extent, especially given the large inter-annual and inter-seasonal swings in precipitation and growing conditions for herbaceous vegetation. To judge the validity of the UAS-based estimates of LPI in 2023, we compared them to the UAS-based estimates of LPI in 2019. We expected the estimates to be very similar because the UAS-based estimates of LPI in 2023 were very similar to 2019, even though the LPI varied greatly between 2019 and 2023. A poor result would suggest a systemic drift in the RF model parameters over time. Because the Training Area is at a higher elevation (~\u0026thinsp;1200 m) than the majority of the SLUD (~\u0026thinsp;950\u0026ndash;1050 m) it receives more precipitation, therefore, we expect the LPI values in the Training Area to be consistently less than that estimated for a large portion of the entire SLUD landscape (Ponce-Campos et al. scale 2023). A poor performance would suggest that the UAS-based satellite estimate of LPI had a weak sensitivity to the different conditions of vegetation abundance across the diverse landscape.\u003c/p\u003e \u003cp\u003eFor the conceptual hypotheses, we first compare the LPI values calculated at 3-m and 30-m grid cells for the 5-cm UAS data from the training polygons, and second we compare LPI among three sources that vary in grid cell size and data density: 30-m at 0.02 B/m\u003csup\u003e2\u003c/sup\u003e (Landsat) and 3-m and 30-m at 0.9B/m\u003csup\u003e2\u003c/sup\u003e (PlanetScope). We reject the scale sensitivity hypothesis if the LPI values estimated using the 3-m resolution are not different than when using the 30-m resolution, and we will reject the amending role of data density if the LPI is not different for the greater data density source (PlanetScope) than the smaller density source (Landsat 8) at 30-m resolution.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eInter-annual and Inter-seasonal Trends in Largest Patch Index (LPI) and Precipitation\u003c/p\u003e \u003cp\u003eThe September LPI values are positively related to the extraordinarily dry summers in 2020 and 2023 and negatively related to the consecutive wet summers in 2021 and 2022 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). The May LPI values are also positively related to the consecutive dry winters in 2020-21 and 2021-22 and negatively related to the wet winter of 2022-23. The September LPI values had a greater range of values than May LPI over the 5 years. Similarly, the inter-annual range of summer SPEI values were greater across the 5 summers than across the 5 winters. Landsat 8 estimates of September LPI were consistently greater than the native 3-m PlanetScope across the 5 years, but there was not a large difference in May LPI values among the satellites and resolutions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThere is not an obvious lag or trans-seasonal trend in September LPI values (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). For example, the decline in September LPI following wet summers in 2021 and 2022, appears unaffected by the dry winter before those wet summers (2020-21) or the dry winter between those summers (2021-22). However, there was a decline in May LPI in 2023 following a wet 12-month period including the summer of 2022 and the winter of 2022-23. There was no other sequence of consecutive wet summer-wet seasons to judge the consistency of the trans-seasonal trend in the LPI-SPEI relationship.\u003c/p\u003e \u003cp\u003e2019 and 2023 Largest Patch Index (LPI) in Training Area Polygons\u003c/p\u003e \u003cp\u003eIn both May and September, and for both grid sizes (3-m and 30-m), the UAS-based measures of LPI in the 10 Training Area polygons were greater in 2019 than 2023 (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). This pattern of greater LPI values in 2019 than 2023 was repeated in 5 of 6 cases for the UAS-trained satellite estimates of LPI (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). May LPI estimates were always greater in 2019 (~\u0026thinsp;20 LPI) than 2023 (~\u0026thinsp;8 LPI) for the entire Training Area at all three of the satellite spatial resolution-data density combinations (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). The one exception to greater LPI in 2019 than 2023 occurred in September using the UAS-trained Landsat 8 data for the entire Training Area (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). LPI increased from 35 to 48 using the low resolution and low data density Landsat 8 data but declined from 31 to 24 at 30-m resolution and 20 to 17 at 3-m when using the high data density 0.9 B/m\u003csup\u003e2\u003c/sup\u003e PlanetScope data.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eLargest Patch Index (LPI) and Eroded Ecological State in the Training Area and SLUD\u003c/p\u003e \u003cp\u003eAs expected, in both May and September, and across all years the LPI was 30\u0026ndash;60% greater for the entire SLUD landscape than the Training Area that was located at the higher elevations of the SLUD (Table\u0026nbsp;2, Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). The difference between the Training Area and entire SLUD was greater in September than May. However, there was a consistent trend in the LPI response to wet and dry periods between the Training Area and the entire SLUD landscape. The extreme dry and wet events between 2020 and 2023 were associated with the maximum and minimum LPI values for both the Training and SLUD areas. The maximum LPI in May followed dry winters in 2021 and 2022 and the minimum value in 2023 followed a wet winter. The maximum LPI in September followed the dry summer in 2020, and the minimum followed the wet summer in 2022.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs expected, the coverage of the Eroded ecological state followed the same general pattern as the LPI because it is based on a fixed LPI threshold set in 2019 (Table\u0026nbsp;2 and Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). The influence of data density on the Eroded state coverage was greater in September than in May, and this was especially obvious in September 2022 when both the Landsat 8 estimate of the Eroded state was 47% but it was only 3\u0026ndash;7% for the PlanetScope at 3-m and 30-m resolutions. At that time, most of the Eroded state coverage using Landsat 8 data was at the lower elevation (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e), whereas the location of the Eroded state using the PlanetScope sources was limited to larger dirt roads.\u003c/p\u003e \u003cp\u003eLargest Patch Index (LPI), Grid Size, and Data Density\u003c/p\u003e \u003cp\u003eAs expected, LPI was sensitive to grid cell size in the comparison of 3-m and 30-m for the UAS-based LPI estimates in the 10 Training Area polygons in May and September, 2019 and 2023 (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). Sensitivity to grid cell size occurred using the UAS-trained satellite estimates of LPI for the Training Area over 5 years (2019-23) and 2 seasons (May and September) in the Training Area, but there was also a relationship with data density at the 30-m grid (Table\u0026nbsp;2). Lowest mean LPI was estimated using the smallest grid (3-m) and the greatest data density (0.9 B/m\u003csup\u003e2\u003c/sup\u003e) with the native PlanetScope (Table\u0026nbsp;2). In contrast, the May estimated LPI was largest in 4 of 5 years when using the large grid and small data density (0.02 B/m\u003csup\u003e2\u003c/sup\u003e) from Landsat 8, and was lowest for PlanetScope 3-m. The exception was May 2022, when LPI was the same for Landsat and PlanetScope 3-m, and the greatest for PlanetScope resampled to 30-m.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe patterns of typically larger LPI for 30-m grid cell, and extent of Eroded State appeared in relation to exceptional wet/dry season extremes (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e) for the entire SLUD landscape between May 2022 (dry winter) and September 2022 (wet summer). In May 2022, LPI and Eroded State extent were nearly identical for Landsat 8 and the 30-m PlanetScope, but by September 2022, the LPI estimates from Landsat 8 were significantly greater, and the Eroded State more extensive than those estimated with the 30-m PlanetScope. Additionally, the native 3-m PlanetScope estimates of LPI and Eroded State were the lowest, with a distribution that was primarily confined to larger dirt roads (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur integration of 3-D UAS and satellite data showed very promising results for representing inter-seasonal and inter-annual changes in the patchiness of bare ground at the landscape scale. These promising results should encourage more evaluations because the integration addresses shortcomings of both the UAS and satellite systems (Villarreal et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Specifically, the large-scale spatial coverage and regularly scheduled data collections by satellites overcomes the small-scale coverage of UAS systems and the logistical challenge of mounting UAS campaigns on a regular basis. The fine-scale 3-D resolution of UAS data collections overcomes the inability to precisely identify features and their spatial patterns using satellite data. This integration is particularly critical for rangelands, where traditional remote sensing approaches often miss the fine-scale ecological processes that influence state transitions (Steele et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Karl et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eGiven this promise of integrating the data collections, future efforts are likely to define best practices for UAS-based training of satellite data to represent fine-scale features at the landscape-scale. For example, we used Random Forest models to generate the best-fit model of the UAS and satellite data, but other correlation-prediction approaches and specifications may reveal better approaches for the UAS training of satellite data.\u003c/p\u003e \u003cp\u003eOur performance-based assessment relied on a natural experiment (Diamond and Robinson \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2010\u003c/span\u003e; Kennedy et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) approach rather than a goodness of fit between a known value and a modeled value across the 5 years and the large landscape. We did not employ the empirically based approach because of the difficulty in obtaining \u0026ldquo;known\u0026rdquo; values of LPI for many more polygons across the landscape, but that effort was beyond the time and resources available during this project. However, our natural experiment approach benefitted from 1) the serendipity of a 5-year period of extraordinary inter-seasonal and inter-annual variation in precipitation, 2) a large elevation gradient within the same soil-based ecological site, and 3) our successful deployment of 4 UAS campaigns (2 seasons in 2 years) at the start and end of the 5-year period.\u003c/p\u003e \u003cp\u003ePerformance Evaluations\u003c/p\u003e \u003cp\u003eThe response of LPI to precipitation fluctuations was very consistent with the expected negative relationship SPEI (Ponce-Campos et al. \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Knapp et al. \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Dash et al. \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Greater variation in LPI among the three sources of UAS-based satellite estimates in September than May (Table\u0026nbsp;2 and Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) is consistent with the greater spatial variation of summer than winter precipitation on the study area (McClaran and Wei \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). It was reassuring that the pattern of lower LPI values in 2023 than 2019 (May and September) was the same for the UAS-trained satellite estimates and the UAS campaigns (Table\u0026nbsp;2 and Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e) (Koh and Wich \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Klosterman et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Inconsistency between the modeled and UAS values would have suggested that our 2019 generated Random Forest-based model was not reliable for estimates of LPI in later years. As expected, LPI estimates were consistently lower for the higher elevation Training Area than the larger SLUD landscape that included drier areas at lower elevations (Table\u0026nbsp;2 and Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e) (Williamson et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2016\u003c/span\u003e, McClaran et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2010\u003c/span\u003e, Breckenfeld and Robinett \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2003\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e. This is consistent with our proof-of-concept findings performed in 2019 (Ponce-Campos et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBecause LPI is a measure of spatial patchiness and not spatial extent, we expected and confirmed that our LPI estimates are sensitive to the grid/pixel size in the analysis of the 5-cm pixel in UAS data and the analysis of the 3-m and 30-m pixels from satellites. The role of data density in the UAS-trained satellite estimates of LPI is reasonable (Michaelides et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Okin et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2009\u003c/span\u003e) given the 50-times greater density for PlanetScope (0.9 B/m\u003csup\u003e2\u003c/sup\u003e) than Landsat 8 data (0.02 B/m\u003csup\u003e2\u003c/sup\u003e). Greater data density should provide a more discriminant estimate of bare ground and its connectivity.\u003c/p\u003e \u003cp\u003eImplications for Monitoring LPI at the Landscape-scale\u003c/p\u003e \u003cp\u003eOur study demonstrates LPI scale sensitivity through a systematic two-step analytical approach that separates the effects of grid cell size from sensor-specific characteristics. The first step involved applying different grid cell sizes (3-m and 30-m) to our high-resolution 5-cm UAS classification data, isolating the pure effect of spatial aggregation on landscape pattern detection. The second step examined scale effects in satellite-derived estimates by comparing LPI values from native 3-m PlanetScope data, resampled 30-m PlanetScope data, and native 30-m Landsat data, effectively decoupling spatial resolution effects from platform-specific characteristics such as data density and spectral band configuration.\u003c/p\u003e \u003cp\u003eThe difference in the direction of scale-sensitivity between the UAS-based and UAS-trained satellite estimates of LPI have implications for avoiding direct comparisons of those data sets. LPI values increased with smaller grid cells in the UAS-based analysis of training polygons, but the opposite occurred for LPI estimates based on the UAS-trained satellite for the training polygons and the entire SLUD. In the UAS-based 5-cm cover class surface it was obvious that the smaller 3-m grid cell would fit between plants and thus produce a high LPI value. Conversely, the smaller 3-m grid for UAS-trained satellite estimates of LPI were typically smaller than estimates using 30-m grids. We suspect that the greater information in the finer resolution and larger data density improve the discrimination of bare ground in the RF model, but further inquiry is needed.\u003c/p\u003e \u003cp\u003eGiven these scale dependencies, any LPI monitoring program must reference both grid cell size and data density. For detecting excessive soil erosion or vulnerability at the landscape scale, we suggest a grid size no smaller than 30-m because the energy of water flowing across 30-m of bare ground without obstruction is substantially greater than across 3-m distances (Ludwig et al. \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Poesen \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Data density must also be consistently referenced given continuous advances in satellite platforms, with the longest historic representation available at the lowest data density (Landsat 8). These considerations ensure that LPI monitoring programs maintain consistency and comparability across space and time while accurately representing the ecological processes of interest.\u003c/p\u003e \u003cp\u003eImplications for Ecological State and Ecological Site\u003c/p\u003e \u003cp\u003eAlthough the adoption of State and Transition Model of vegetation change and response to management has been pursued for nearly 40 years (Westoby et al. \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e1989\u003c/span\u003e; Bestelmeyer et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), experts remain the primary sources for distinguishing ecological states, the relative permanence of a state, and the drivers that cause transitions to alternative states. The ecological state (Eroded or non-Eroded) of training polygons was determined by an expert with extensive SLUD landscape experience.\u003c/p\u003e \u003cp\u003eOur results suggest that the Eroded state can transition to non-Eroded, and back again (see Supplementary Material, Fig. S1-3). This relative impermanence warrants a state name such as \u0026ldquo;Erodible\u0026rdquo; to reflect a potential for accelerated erosion, rather than \u0026ldquo;Eroded\u0026rdquo; which implies on-going accelerated erosion that will need major management inputs to stop the erosion processes. This dynamic behavior aligns with emerging understanding of more volatile climate conditions and their effects on ecosystem responses. As demonstrated by Swain et al. (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), hydroclimate whiplash events\u0026mdash;rapid transitions between extremely wet and dry conditions\u0026mdash;are becoming more frequent and intense with climate change. Our observations of apparent state transitions in response to interannual precipitation variability reflect this increasing climate volatility and suggest that rangeland ecosystems may exhibit more dynamic responses than traditionally conceptualized in State and Transition Models. Under these conditions, \"Erodible\" would serve more as a warning to inspect and monitor ecosystem trajectory rather than an emergency requiring urgent action, acknowledging that state boundaries may be more fluid in response to increasingly variable climate conditions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe system to map ecological sites is based on the soil, climate, and topography that has been used since the 1960s (originally Range Site) to recognize the relative potential to support different vegetations compositions and productivity, and associated land uses such as livestock grazing and wildlife habitat (USDA-NRCS Title190 \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Mapping ecological sites is very dependent on soil maps. In fact, the mapping of ecological sites in our study area was done simultaneously with soil mapping in the 1990s (Breckenfeld and Robinett \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2003\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe SLUD on our study site, includes a range of 12\u0026ndash;16 inches of average annual precipitation, and this was based on the precipitation record into the 1980s. Our results show large differences in the inter-seasonal and inter-annual variation in LPI and Eroded State between the upper and lower elevations of the SLUD (Table\u0026nbsp;2, Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e9\u003c/span\u003e). Given that spatial variation, and the increase of 1 degree C in the annual temperature on the study area since 1996 (McClaran and Wei \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), we advocate for an effort to adjust all ecological site boundaries upslope to account for the increased evaporative demand and greater vapor pressure deficit with higher temperatures (Novick et al. \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Williams et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2020\u003c/span\u003e)\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eWe showed promising results from the integration of 3-D UAS and satellite data to map the fine-scale feature of bare ground patchiness over 5 years and 2 inter-annual seasons. The patchiness metric of LPI (largest patch index) has two important features 1) it reflects soil erosion potential because water energy increases without obstructions that slow the water and 2) it is sensitive to the grid/pixel size of the surface data. Therefore, any effort to monitor LPI using a UAS-trained satellite estimate must maintain the same spatial resolution and data density through the time series. We suggest that 30-m is the smallest resolution to represent soil erosion potential.\u003c/p\u003e \u003cp\u003eThe extraordinarily large inter-seasonal and inter-annual variation in precipitation during our 5-year study provided a natural experiment to evaluate the expected performance of our UAS-trained satellite estimate of LPI across a 100 km\u003csup\u003e2\u003c/sup\u003e landscape. The resulting landscape-scale view of a fine-scale surface feature provided a critical assessment of the 1) commonly applied state and transition model for eroded ecological states and the 2) current mapping boundaries for ecological sites that were based on climate records that are 40 years old and should be adjusted upslope to account for the 1 degree C increase in annual temperature since 1996.\u003c/p\u003e \u003cp\u003eWhile our approach demonstrates promising results for integrating UAS and satellite data, we recognize opportunities for future expansion and refinement. Our analysis focused on a single ecological site with 10 training polygons and four UAS collection periods, providing a robust proof-of-concept that could be extended to additional rangeland systems and longer temporal series to further validate the approach across diverse environmental conditions. The expert-based ecological state classifications and natural experiment validation framework proved effective for our study objectives, though complementary approaches using standardized field measurements and expanded ground truth data could enhance future applications.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe gratefully acknowledge the support and contributions of the USDA-ARS Agreement No. 58-2022-0-009 for funding this research.\u003c/p\u003e\n\u003cp\u003eJ. Gillan performed the 2019 UAS field campaign and initial data processing. A. Gorlier assisted with the 2019 UAS field campaign and A. Leimroth assisted with the 2023 field campaign.\u003c/p\u003e\n\u003cp\u003eData sets were provided by the Santa Rita Experimental Range Digital Database. Funding for the digitization of these data was provided by USDA Forest Service Rocky Mountain Research Station and the University of Arizona, Arizona Experiment Station. This research was a contribution from the Long-Term Agroecosystem Research (LTAR) network. LTAR is supported by the United States Department of Agriculture.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization, M.M., G.E.P., P.H.; Formal analysis, G.E.P., M.M., P.H., S.G., C.N. and M.C.; Funding acquisition, P.H. and M.M.; Methodology, G.E.P., M.M. and P.H.; Resources G.E.P., M.M., P.H., S.G., C.N. and M.C., Writing original draft, G.E.P., M.M., P.H. Writing\u0026mdash;review \u0026amp; editing G.E.P., M.M., P.H., S.G., C.N. and M.C.\u003c/p\u003e\n\u003cp\u003eAll authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGithub repository:\u0026nbsp;\u003c/strong\u003ehttps://github.com/gponce-ars/scaling-uav/\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAnderson K, Gaston KJ (2013) Lightweight unmanned aerial vehicles will revolutionize spatial ecology. Frontiers in Ecology and the Environment 11:138\u0026ndash;146. https://doi.org/10.1890/120150\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eAlvarez-Vanhard E, Corpetti T, Houet T (2021) UAV \u0026amp; satellite synergies for optical remote sensing applications: A literature review. Science of Remote Sensing 3:100019. https://doi.org/10.1016/j.srs.2021.100019\u003c/li\u003e\n \u003cli\u003eBestelmeyer BT, Ash A, Brown JR, et al (2017) State and Transition Models: Theory, Applications, and Challenges. In: Briske DD (ed) Rangeland Systems: Processes, Management and Challenges. Springer International Publishing, Cham, pp 303\u0026ndash;345\u003c/li\u003e\n \u003cli\u003eBestelmeyer BT, Brown JR, Havstad KM, et al (2003) Development and use of state-and-transition models for rangelands. 56. https://doi.org/10.2458/azu_jrm_v56i2_bestelmeyer\u003c/li\u003e\n \u003cli\u003eBestelmeyer BT, Peters DPC, Archer SR, et al (2018) The Grassland\u0026ndash;Shrubland Regime Shift in the Southwestern United States: Misconceptions and Their Implications for Management. BioScience 68:678\u0026ndash;690. https://doi.org/10.1093/biosci/biy065\u003c/li\u003e\n \u003cli\u003eBreckenfeld DJ, Robinett D (2003) Soil and ecological sites of the Santa Rita Experimental Range. In: McClaran, Mitchel P; Ffolliott, Peter F; Edminster, Carleton B, tech coords Santa Rita Experimental Range: 100 years (1903 to 2003) of accomplishments and contributions; conference proceedings; 2003 October 30-November 1; Tucson, AZ Proc RMRS-P-30 Ogden, UT: US Department of Agriculture, Forest Service, Rocky Mountain Research Station p 157-165 30:157\u0026ndash;165\u003c/li\u003e\n \u003cli\u003eBreiman L (2001) Random Forests. Machine Learning 45:5\u0026ndash;32. https://doi.org/10.1023/A:1010933404324\u003c/li\u003e\n \u003cli\u003eBriske DD, Fuhlendorf SD, Smeins FE (2005) State-and-Transition Models, Thresholds, and Rangeland Health: A Synthesis of Ecological Concepts and Perspectives. Rangeland Ecology \u0026amp; Management 58:1\u0026ndash;10. https://doi.org/10.2111/1551-5028(2005)58\u0026lt;1:SMTARH\u0026gt;2.0.CO;2\u003c/li\u003e\n \u003cli\u003eBrown DE, Brennan TC, Unmack PJ (2007) A digitized biotic communities map for plotting and comparing North American plant and animal distributions. Canotia 3:1\u0026ndash;12\u003c/li\u003e\n \u003cli\u003eCrawford CJ, Roy DP, Arab S, et al (2023) The 50-year Landsat collection 2 archive. Science of Remote Sensing 8:100103. https://doi.org/10.1016/j.srs.2023.100103\u003c/li\u003e\n \u003cli\u003eDash JP, Pearse GD, Watt MS (2018) UAV Multispectral Imagery Can Complement Satellite Data for Monitoring Forest Health. Remote Sensing 10:1216. https://doi.org/10.3390/rs10081216\u003c/li\u003e\n \u003cli\u003eDiamond J, Robinson JA (2010) Natural Experiments of History. Harvard University Press, Cambridge, MA\u003c/li\u003e\n \u003cli\u003eEDIT, NRCS U-A (2022) The Ecosystem Dynamics Interpretive Tool (EDIT)\u003c/li\u003e\n \u003cli\u003eGamon JA, Somers B, Malenovsk\u0026yacute; Z, et al (2019) Assessing Vegetation Function with Imaging Spectroscopy. Surv Geophys 40:489\u0026ndash;513. https://doi.org/10.1007/s10712-019-09511-5\u003c/li\u003e\n \u003cli\u003eGillan JK, Ponce-Campos GE, Swetnam TL, et al (2021) Innovations to expand drone data collection and analysis for rangeland monitoring. Ecosphere 12:e03649. https://doi.org/10.1002/ecs2.3649\u003c/li\u003e\n \u003cli\u003eGlasbey C (1993) An Analysis of Histogram-Based Thresholding Algorithms. CVGIP Graph Model Image Process. https://doi.org/10.1006/cgip.1993.1040\u003c/li\u003e\n \u003cli\u003eGoogle Inc (2015) Google Earth Engine Platform. In: Google Earth Engine. http://earthengine.google.org/. Accessed 30 Jul 2015\u003c/li\u003e\n \u003cli\u003eGorelick N, Hancher M, Dixon M, et al (2017) Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment. https://doi.org/10.1016/j.rse.2017.06.031\u003c/li\u003e\n \u003cli\u003eKarl JW, Gillan JK, Barger NN, et al (2014) Interpretation of high-resolution imagery for detecting vegetation cover composition change after fuels reduction treatments in woodlands. Ecological Indicators 45:570\u0026ndash;578. https://doi.org/10.1016/j.ecolind.2014.05.017\u003c/li\u003e\n \u003cli\u003eKennedy RE, Andr\u0026eacute;fou\u0026euml;t S, Cohen WB, et al (2014) Bringing an ecological view of change to Landsat-based remote sensing. Frontiers in Ecology and the Environment 12:339\u0026ndash;346. https://doi.org/10.1890/130066\u003c/li\u003e\n \u003cli\u003eKlosterman S, Melaas E, Wang JA, et al (2018) Fine-scale perspectives on landscape phenology from unmanned aerial vehicle (UAV) photography. Agricultural and Forest Meteorology 248:397\u0026ndash;407. https://doi.org/10.1016/j.agrformet.2017.10.015\u003c/li\u003e\n \u003cli\u003eKnapp AK, Ciais P, Smith MD (2017) Reconciling inconsistencies in precipitation\u0026ndash;productivity relationships: implications for climate change. New Phytologist 214:41\u0026ndash;47. https://doi.org/10.1111/nph.14381\u003c/li\u003e\n \u003cli\u003eKoh LP, Wich SA (2012) Dawn of Drone Ecology: Low-Cost Autonomous Aerial Vehicles for Conservation. Tropical Conservation Science 5:121\u0026ndash;132. https://doi.org/10.1177/194008291200500202\u003c/li\u003e\n \u003cli\u003eKuehl RO, McClaran MP, Va J (2001) Detecting fragmentation of cover in desert grasslands using line intercept. 54. https://doi.org/10.2458/azu_jrm_v54i1_kuehl\u003c/li\u003e\n \u003cli\u003eLloyd CD (2010) Spatial data analysis: an introduction for GIS users. Oxford University Press, Oxford ; New York\u003c/li\u003e\n \u003cli\u003eLudwig JA, Wilcox BP, Breshears DD, et al (2005) Vegetation Patches and Runoff\u0026ndash;Erosion as Interacting Ecohydrological Processes in Semiarid Landscapes. Ecology 86:288\u0026ndash;297. https://doi.org/10.1890/03-0569\u003c/li\u003e\n \u003cli\u003eMcClaran MP, Browning DM, Huang C (2010) Temporal dynamics and spatial variability in desert grassland vegetation. In: Webb RH, Boyer DE, Turner RM (eds) Repeat Photography: Methods and Applications in the Natural Sciences. Island Press\u003c/li\u003e\n \u003cli\u003eMcClaran MP, Wei H (2014) Recent drought phase in a 73-year record at two spatial scales: Implications for livestock production on rangelands in the Southwestern United States. Agricultural and Forest Meteorology 197:40\u0026ndash;51. https://doi.org/10.1016/j.agrformet.2014.06.004\u003c/li\u003e\n \u003cli\u003eMcGarigal K (2014) Landscape Pattern Metrics. In: Wiley StatsRef: Statistics Reference Online. John Wiley \u0026amp; Sons, Ltd\u003c/li\u003e\n \u003cli\u003eMichaelides K, Lister D, Wainwright J, Parsons AJ (2009) Vegetation controls on small-scale runoff and erosion dynamics in a degrading dryland environment. Hydrological Processes 23:1617\u0026ndash;1630. https://doi.org/10.1002/hyp.7293\u003c/li\u003e\n \u003cli\u003eMyers ER, Browning DM, Burkett LM, et al (2024) Novel Use of Image Time Series to Distinguish Dryland Vegetation Responses to Wet and Dry Years. Journal of Remote Sensing 4:0190. https://doi.org/10.34133/remotesensing.0190\u003c/li\u003e\n \u003cli\u003eNotaro M, Mauss A, Williams JW (2012) Projected vegetation changes for the American Southwest: combined dynamic modeling and bioclimatic-envelope approach. Ecological Applications 22:1365\u0026ndash;1388. https://doi.org/10.1890/11-1269.1\u003c/li\u003e\n \u003cli\u003eNovick KA, Ficklin DL, Stoy PC, et al (2016) The increasing importance of atmospheric demand for ecosystem water and carbon fluxes. Nature Clim Change 6:1023\u0026ndash;1027. https://doi.org/10.1038/nclimate3114\u003c/li\u003e\n \u003cli\u003eOkin GS, Parsons AJ, Wainwright J, et al (2009) Do Changes in Connectivity Explain Desertification? BioScience 59:237\u0026ndash;244. https://doi.org/10.1525/bio.2009.59.3.8\u003c/li\u003e\n \u003cli\u003eOlsoy PJ, Shipley LA, Rachlow JL, et al (2018) Unmanned aerial systems measure structural habitat features for wildlife across multiple scales. Methods in Ecology and Evolution 9:594\u0026ndash;604. https://doi.org/10.1111/2041-210X.12919\u003c/li\u003e\n \u003cli\u003eOtsu N (1979) A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics 9:62\u0026ndash;66. https://doi.org/10.1109/TSMC.1979.4310076\u003c/li\u003e\n \u003cli\u003ePlanet Labs P (2018) Planet Application Program Interface: In Space for Life on Earth\u003c/li\u003e\n \u003cli\u003ePoesen J (2018) Soil erosion in the Anthropocene: Research needs. Earth Surface Processes and Landforms 43:64\u0026ndash;84. https://doi.org/10.1002/esp.4250\u003c/li\u003e\n \u003cli\u003ePonce-Campos GE, McClaran M, Heilman P, Gillan JK (2023) UAV and Satellite-Based Sensing to Map Ecological States at the Landscape Scale. Open Journal of Ecology 13:560\u0026ndash;596. https://doi.org/10.4236/oje.2023.138035\u003c/li\u003e\n \u003cli\u003ePonce-Campos GE, Moran MS, Huete A, et al (2013) Ecosystem resilience despite large-scale altered hydroclimatic conditions. Nature 494:349\u0026ndash;352. https://doi.org/10.1038/nature11836\u003c/li\u003e\n \u003cli\u003eScott RL, Johnston MR, Knowles JF, et al (2023) Interannual variability of spring and summer monsoon growing season carbon exchange at a semiarid savanna over nearly two decades. Agricultural and Forest Meteorology 339:109584. https://doi.org/10.1016/j.agrformet.2023.109584\u003c/li\u003e\n \u003cli\u003eSolazzo D, Sankey JB, Sankey TTs, Munson SM (2018) Mapping and measuring aeolian sand dunes with photogrammetry and LiDAR from unmanned aerial vehicles (UAV) and multispectral satellite imagery on the Paria Plateau, AZ, USA. Geomorphology 319:174\u0026ndash;185. https://doi.org/10.1016/j.geomorph.2018.07.023\u003c/li\u003e\n \u003cli\u003eSPI Tool, University of Arizona (2024) SPI Explorer Tool. https://uaclimateextension.shinyapps.io/SPItool/\u003c/li\u003e\n \u003cli\u003eSteele CM, Bestelmeyer BT, Burkett LM, et al (2012) Spatially Explicit Representation of State-and-Transition Models. Rangeland Ecology \u0026amp; Management 65:213\u0026ndash;222. https://doi.org/10.2111/REM-D-11-00047.1\u003c/li\u003e\n \u003cli\u003eSwain DL, Prein AF, Abatzoglou JT, et al (2025) Hydroclimate volatility on a warming Earth. Nat Rev Earth Environ 6:35\u0026ndash;50. https://doi.org/10.1038/s43017-024-00624-z\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTmu\u0026scaron;ić G, Manfreda S, Aasen H, et al (2020) Current Practices in UAS-based Environmental Monitoring. Remote Sensing 12:1001. https://doi.org/10.3390/rs12061001\u003c/li\u003e\n \u003cli\u003eTurner MG, Gardner RH (2015) Landscape Ecology in Theory and Practice: Pattern and Process. Springer New York\u003c/li\u003e\n \u003cli\u003eUrgeghe AM, Mayor \u0026Aacute;G, Turri\u0026oacute;n D, et al (2021) Disentangling the independent effects of vegetation cover and pattern on runoff and sediment yield in dryland systems \u0026ndash; Uncovering processes through mimicked plant patches. Journal of Arid Environments 193:104585. https://doi.org/10.1016/j.jaridenv.2021.104585\u003c/li\u003e\n \u003cli\u003eUSDA, NRCS (2022) Land Resource Regions and Major Land Resource Areas of the United States, the Caribbean, and the Pacific Basin. U.S. Department of Agriculture\u003c/li\u003e\n \u003cli\u003eUSDA-NRCS Title190 (2022) National Range and Pasture Handbook. United States Department of Agriculture\u003c/li\u003e\n \u003cli\u003evan der Leeuw E, van Leeuwen WJD, Marsh SE, Archer SR (2024) Using Fine Resolution Remotely Sensed Data-Derived Land Cover to Inform Dryland State and Transition Models. Rangeland Ecology \u0026amp; Management 96:128\u0026ndash;142. https://doi.org/10.1016/j.rama.2024.06.003\u003c/li\u003e\n \u003cli\u003eVicente-Serrano SM, Gouveia C, Camarero JJ, et al (2013a) Response of vegetation to drought time-scales across global land biomes. Proceedings of the National Academy of Sciences 110:52\u0026ndash;57. https://doi.org/10.1073/pnas.1207068110\u003c/li\u003e\n \u003cli\u003eVicente-Serrano SM, Gouveia C, Camarero JJ, et al (2013b) Response of vegetation to drought time-scales across global land biomes. PNAS 110:52\u0026ndash;57. https://doi.org/10.1073/pnas.1207068110\u003c/li\u003e\n \u003cli\u003eVillarreal et al. 2025. Applications of unoccupied aerial systems (UAS) in landscape ecology: a review of recent research, challenges and emerging opportunities Landsc Ecol (2025) 40:43. https://doi.org/10.1007/s10980-024-02040-6\u003c/li\u003e\n \u003cli\u003eVillarreal ML, Soulard CE, Waller EK (2019) Landsat Time Series Assessment of Invasive Annual Grasses Following Energy Development. Remote Sensing 11:2553. https://doi.org/10.3390/rs11212553\u003c/li\u003e\n \u003cli\u003eWestoby M, Walker B, Noy-Meir I (1989) Opportunistic management for rangelands not at equilibrium. 42. https://doi.org/10.2307/3899492\u003c/li\u003e\n \u003cli\u003eWilliams AP, Cook ER, Smerdon JE, et al (2020) Large contribution from anthropogenic warming to an emerging North American megadrought. Science 368:314\u0026ndash;318. https://doi.org/10.1126/science.aaz9600\u003c/li\u003e\n \u003cli\u003eWilliamson JC, Bestelmeyer BT, McClaran MP, et al (2016) Can ecological land classification increase the utility of vegetation monitoring data? Ecological Indicators 69:657\u0026ndash;666. https://doi.org/10.1016/j.ecolind.2016.05.030\u003c/li\u003e\n \u003cli\u003eWu S, Wang J, Yan Z, et al (2021) Monitoring tree-crown scale autumn leaf phenology in a temperate forest with an integration of PlanetScope and drone remote sensing observations. ISPRS Journal of Photogrammetry and Remote Sensing 171:36\u0026ndash;48. https://doi.org/10.1016/j.isprsjprs.2020.10.017\u003cstrong\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Table","content":"\u003cp\u003eTable 2 is available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"landscape-ecology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"land","sideBox":"Learn more about [Landscape Ecology](https://www.springer.com/journal/10980)","snPcode":"10980","submissionUrl":"https://submission.nature.com/new-submission/10980/3","title":"Landscape Ecology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Multi-scale analysis, Largest Patch Index, Remote Sensing, Ecological Sites, Ecological States, Rangeland","lastPublishedDoi":"10.21203/rs.3.rs-6857473/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6857473/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIntegrating fine-scale measurements with broad-scale monitoring presents a persistent challenge in rangeland ecology, particularly when scaling detailed Unoccupied Aerial System (UAS) observations to satellite-based landscape assessments. This challenge is especially critical as rangelands face increasing climate variability, requiring reliable methods to detect and monitor ecological changes. We investigated how the Largest Patch Index (LPI) of bare ground patches, derived from 3-dimensional UAS observations, can be scaled to landscape levels for mapping bare ground patchiness across a 100 km\u0026sup2; semi-arid rangeland in southern Arizona. Our findings reveal three key advances in landscape monitoring. First, LPI effectively captured vegetation responses to extreme climate events during 2019\u0026ndash;2023, showing clear sensitivity to both severe drought (SPEI \u0026minus;\u0026thinsp;2.47) and exceptional wet periods (SPEI\u0026thinsp;+\u0026thinsp;1.95). Second, LPI values were consistently 30\u0026ndash;60% higher in lower elevations, validating the ability to detect known ecological gradients. Third, and most notably, that LPI is positively scale dependent between the 3-m and 30-m grid sizes, and that the magnitude of that difference varies with the density of data from the satellite sensors. This previously unrecognized role of data density challenges fundamental assumptions about scale effects in landscape pattern analysis. Our approach demonstrates a practical solution for integrating UAS and satellite observations, providing a new approach for supporting the detection and monitoring of ecological changes across landscapes, a critical need given increasing climate uncertainty.\u003c/p\u003e","manuscriptTitle":"Integrating Unoccupied Aerial Systems and Satellite Data to Map the Patchiness of Bare Ground at a Landscape Scale","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-18 12:13:24","doi":"10.21203/rs.3.rs-6857473/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-08-06T23:39:13+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-21T05:56:17+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-13T06:54:44+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-25T21:32:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"132877248265329100155615524979001021762","date":"2025-06-16T14:37:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"273028804500432345723383515871172706492","date":"2025-06-16T05:22:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"302390986208149309944518277515715355655","date":"2025-06-16T02:22:50+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-16T01:56:52+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-11T08:21:22+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-11T08:17:09+00:00","index":"","fulltext":""},{"type":"submitted","content":"Landscape Ecology","date":"2025-06-09T21:22:01+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"landscape-ecology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"land","sideBox":"Learn more about [Landscape Ecology](https://www.springer.com/journal/10980)","snPcode":"10980","submissionUrl":"https://submission.nature.com/new-submission/10980/3","title":"Landscape Ecology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"897fce28-6527-4227-bb94-13eebc862675","owner":[],"postedDate":"June 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-24T16:07:29+00:00","versionOfRecord":{"articleIdentity":"rs-6857473","link":"https://doi.org/10.1007/s10980-025-02226-6","journal":{"identity":"landscape-ecology","isVorOnly":false,"title":"Landscape Ecology"},"publishedOn":"2025-11-18 15:58:52","publishedOnDateReadable":"November 18th, 2025"},"versionCreatedAt":"2025-06-18 12:13:24","video":"","vorDoi":"10.1007/s10980-025-02226-6","vorDoiUrl":"https://doi.org/10.1007/s10980-025-02226-6","workflowStages":[]},"version":"v1","identity":"rs-6857473","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6857473","identity":"rs-6857473","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}