Evaluating Remote Sensing Techniques for Monitoring Grassland Degradation

doi:10.21203/rs.3.rs-4940631/v1

Download PDF

Research Article

Evaluating Remote Sensing Techniques for Monitoring Grassland Degradation

https://doi.org/10.21203/rs.3.rs-4940631/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Ecosystem degradation represents one of the most significant environmental challenges facing the globe. Of particular concern is the impact of grassland degradation on agricultural productivity, species diversity, and soil erosion. This study aimed to compare the applicability of two remote sensing techniques, the Linear Spectral Mixture Model (LSM) and the Grassland Degradation Index (GDI), in assessing and defining the degree of grassland degradation. The results demonstrated that the GDI exhibited superior overall accuracy than LSM, with an accuracy rate of 73.49% as opposed to 63.16% for the LSM. Additionally, the GDI demonstrated a higher F1 score across all evaluated classes, indicating an enhanced capacity to identify true positives and minimise false positives and negatives. Both techniques demonstrated satisfactory performance and can be employed to support restoration and sustainable management studies.

Grassland degradation

Spectral unmixing

Vegetation Indices

Google Earth Engine

Sustainable grazing

One of the most challenging environmental problems currently facing the world is ecosystem degradation (IPCC, 2019). The main impacts of grassland degradation include reduced yield and diversity of high-value forage species, proliferation of toxic and exotic species, and increased soil erosion, severely limiting ecosystem functions and services (SHENG et al., 2022).

The degradation of grasslands has long been associated with anthropogenic factors, including fire management practices and excessive grazing (ZHOU, Y. et al., 2017). In Southern Brazil, there are native grasslands called "Highland grasslands," historically used for grazing mainly cattle and sheep and managed with fire in the winter period to remove dry biomass, aiming to accelerate the vegetation regrowth (BOLDRINI, 1997).

Therefore, assessing the conservation degree of these habitats and their relationship with management strategies is challenging for society. Implementing sustainable agricultural practices, such as improving the management of grasslands for farming and livestock production, can increase productivity while enhancing the adaptability and conservation of these ecosystems (CASTELLANOS et al., 2022).

Studies focused on the degradation of non-forest ecosystems, such as grasslands, consider reintroducing traditional practices of using fire to remove the accumulation of combustible material and stimulate regrowth (OVERBECK et al., 2015).

Remote sensing has become an indispensable tool for regional and global degradation monitoring. Many studies have used spectral responses to define the degradation levels using vegetation indices (GAO et al., 2010; LIU et al., 2019; SUN et al., 2017)

Studies based on the Grassland Degradation Index (GDI), such as the one proposed by GAO et al. (2006), were used to monitor the progress of degradation related to changes in land use and cover (YANG et al., 2019), interaction with climatic factors (AN et al., 2021; ZHOU, W. et al., 2017), and advancing desertification (KUANG et al., 2020).

Another widely used degradation detection and monitoring approach is the spectral mixture analysis, proposed by SHIMABUKURO and SMITH (1991). Studies have used these spectral mixture models to transform spectral information into physical information (fractional values of the components in the pixel), generating, for instance, images of soil fraction, shadow/water, and vegetation (BULLOCK; WOODCOCK; OLOFSSON, 2020; DAWELBAIT; MORARI, 2011; LYU et al., 2020).

Other authors have used the regression-based unmixing approach to recover fractional cover time series of photosynthetic vegetation (PV), non-photosynthetic vegetation (NPV) and soil from Sentinel-2 data (KOWALSKI; OKUJENI; HOSTERT, 2023).

In this context, the aim of this study was to compare the applicability of GDI techniques, and an approach using the linear spectral mixture model (LSM) to assess and define the degree of pasture degradation.

2.1. Study Area

The "Parque Estadual do Tainhas" (PET), located in a region of southern Brazil denominated "Campos de Cima da Serra" (CCS – Highland grasslands), is a State Conservation Area characterised by the grasslands predominance in contact with the Atlantic Forest (SEMA, 2008) (Fig. 01). The region is part of the Serra Geral Mountains, primarily located at elevations above 800 meters. The region's cold climate, high rainfall, and high altitude contribute to a high endemism level in the flora, with plants that have adapted to this specific environment (PILLAR et al., 2009).

This vegetation is subjected to grazing and fire management, with these disturbances being considered vital to maintaining grassland vegetation composed of plants from different groups, thus ensuring diversity and conservation (LUZA et al., 2014). In PET's buffer zone (BZ), some communities depend on agriculture, forestry, and livestock, often using fire to manage rural areas (BOND-BUCKUP, 2008). The PET has some areas acquired through land regularisation, and many of these grasslands need to be assessed for their conservation level and the development of restoration plans.

2.2. Comparative Models

This study employs a comparative methodology to assess the grassland degradation levels. The methodological approach is detailed in (Fig. 02). Images from the Sentinel-2 MSI Multispectral Instrument - Level-2A were used (Table 01). This product provides bottom-of-atmosphere (BOA) images derived from Level 1C products. The images were selected for the years 2020 (before the known fire event), 2021 (right after the known fire event), and 2022 (one year after the known fire event), covering the period from February 1st to April 1st. This interval was chosen because the vegetation was not disturbed by fire and has already recovered from last winter's regrowth.

Table 01

– Select images.
Year	Images
2020	04/02/2020 24/02/2020 29/02/2020 29/02/2020 05/03/2020 10/03/2020 10/03/2020 15/03/2020 15/03/2020 20/03/2020 25/03/2020
2021	08/02/2021 08/02/2021 20/03/2021 20/03/2021 25/03/2021
2022	03/02/2022 03/02/2022 05/03/2022 05/03/2022

2.2.1 Grassland Degradation Index

The initial model was constructed upon the GDI proposed by Gao (2008). The images were processed in Google Earth Engine (GEE), where a calculation of the Normalized Difference Vegetation Index (NDVI) was initially performed (ROUSE; HASS; SCHELL, 1973) as represented by the following equation:

$$\:NDVI=\:\frac{NIR-Red}{NIR+Red}$$

The NDVI is obtained by normalising the reflectance in the near-infrared (NIR) and red (R) bands, with values ranging from − 1 to + 1. However, as far as vegetation is concerned, it is between values ranging from 0 to + 1.

To assess the grassland degradation level, we used the NDVI result for each year and the calculation of Vegetation Cover (VC) using the methodology proposed by GAO et al. (2006), defined by (Eq. 02).

$$\:VC=\frac{NDVI-NDVIs}{NDVIv-NDVIs}\:x\:100$$

where NDVI is Normalized Difference Vegetation Index, NDVIs is the lowest NDVI value found among the pixels representing the area with exposed soil, and NDVIv is the highest NDVI value found among the pixels of the grass area.

The degradation levels were classified into three groups, according to the values described in (Table 02).

Table 02

Degradation Levels
Level	Cover
Non-degraded	GDI > 75%
Moderate	75% ≥ GDI > 50%
Severe	GDI ≤ 50%

2.2.2 Linear Spectral Mixture Model

The second model used the LSM proposed by Shimabukuro (1991). The unmix function available on the GEE platform was applied to generate vegetation, soil, and shadow/water fraction images based on the collection of pixels with the spectral response closest to the theoretical curve expected for pure pixels. LSM assumes that pixel values are linear combinations of the reflectance of different components, called endmembers, and the resulting images are the proportion of each endmember in each pixel. Generating these fraction images is an alternative approach to reducing the dimensionality of an image's data and highlighting the target features desired for digital classification (SHIMABUKURO; SMITH, 1991). The fractions are generated according to (Eq. 3).

$$\:{\text{R}}_{i}=\:{\sum\:}_{j}^{n}{f}_{i}{r}_{i,j}+{\epsilon\:}_{i}$$

where: 𝑅𝑖 represents the spectral reflectance in the 𝑖 spectral band; 𝑟𝑖, 𝑗 is the spectral reflectance of the 𝑗 component in the 𝑖 spectral band (endmember); 𝑓𝑖 is the proportion of the j component within the pixel; and 𝜀𝑖 is the residual of the 𝑖 spectral band.

This research subjected the vegetation and soil fractions to an unsupervised classification to identify degraded areas using the K-means algorithm, starting with K initial centroids, which are randomly chosen. Then, it performs iterative refinement to improve the assignment of points to clusters by minimising the within-cluster sum of squares (WCSS). This process continues until convergence, when the centroids do not change significantly between iterations (AHMED; SERAJ; ISLAM, 2020).

2.3. Grassland Survey Data

We calculated the total sample size for stratified random samples using Eq. (13) from Olofsson et al. (2014), aiming for a 10% error margin. The sample was allocated to strata following the "good practice" guidelines for estimating area (OLOFSSON et al., 2014). The training data selection was utterly independent of the design and reference sample selection (BULLOCK; WOODCOCK; OLOFSSON, 2020). Therefore, with proportions 𝑊𝑖 and assuming hypothetical standard deviations, the required minimum sample size 𝑛 was 163 units.

To collect the samples, we used image mosaics from Planet, with a spatial resolution of 3 meters, geometrically corrected and covered for surface reflectance. After sampling through visual interpretation, some samples were verified in the field, evaluating the characteristics of biomass accumulation, soil cover, and erosion (Fig. 03).

In order to ascertain the occupation of the grassland class, the Mapbiomas data set was employed, which defines land cover classes separately for each Brazilian biome in common cross-cutting land use themes(SOUZA et al., 2020).

2.4. Accuracy Test

To evaluate and compare the generated models' performance, a series of metrics, including confusion matrices, classification reports (precision, recall, and F1-score), and Root Mean Squared Error (RMSE) calculation for both models, were employed.

A confusion matrix is a summary of prediction results on a classification problem. The correct and incorrect predictions are abridged with count values and broken down by each class. This helps to understand the errors made by the classifier and what error types are being made (JAMES et al., 2013).

Classification reports provide detailed metrics about a classification model's performance. These metrics include precision, recall, and F1-score, which offer insights into how well the model correctly identifies and categorises instances (HAN; KAMBER; PEI, 2012).

3.1. Accuracy

The accuracy assessment and area estimation results are shown in (Fig. 04) and (Table 03). The confusion matrix analysis reveals significant performance differences between the GDI and SLM models.

The GDI model demonstrates a superior ability to correctly classify instances of non-grass and non-degraded levels. However, the GDI model still exhibits more classification errors, such as instances where the "Severe" class was incorrectly identified as "non-Grass." In contrast, despite exhibiting a lower overall accuracy, the SLM model performs better in classifying the "Severe" class than the GDI model. Both models encounter comparable challenges in classifying the "Moderate" class, with a notable incidence of misclassifications.

Table 03

Performance Metrics Comparison between GDI and SLM Models.
Class	Metric	GDI	SLM
Non-Grass	Precision	66.67%	63.83%
	Recall	90.91%	68.18%
	F1-Score	76.92%	65.93%
Severe	Precision	77.42%	87.18%
	Recall	57.14%	80.95%
	F1-Score	65.75%	83.95%
Moderate	Precision	82.61%	50.00%
	Recall	51.35%	54.05%
	F1-Score	63.33%	51.95%
Non-degraded	Precision	69.09%	51.16%
	Recall	82.61%	47.83%
	F1-Score	75.25%	49.44%
Overall	Accuracy	71.60%	62.72%
Macro Avg	Precision	73.95%	63.04%
	Recall	70.50%	62.75%
	F1-Score	70.31%	62.82%
Weighted Avg	Precision	73.49%	63.16%
	Recall	71.60%	62.72%
	F1-Score	70.72%	62.86%
RMSE		0.91	1.27

Therefore, the GDI model consistently outperforms the SLM model across all evaluated performance metrics, as seen in (Fig. 05). The GDI model's enhanced accuracy suggests it is more effective at correctly predicting classes. The higher precision and recall indicate that the GDI model is more effective at identifying true positives and has fewer false positives and negatives than the SLM model. Consequently, the GDI model's F1-score, which is a harmonic mean between precision and recall, is also higher, thereby confirming the GDI's overall superior performance compared to the SLM model.

The error distributions from the GDI and SLM models (Fig. 06) are depicted in a histogram plot with density curves. The bars represent the model error frequency, and most are situated around the zero value, which suggests that a considerable proportion of the predictions are closely aligned with the actual values. The number of significant errors at either end of the spectrum (i.e., at -3 and 3) is relatively low, indicating that the model is less prone to making highly inaccurate forecasts.

Nevertheless, the GDI model shows a slightly higher peak frequency and a narrower spread of errors, suggesting more precise predictions overall. In contrast, the SLM model exhibits a broader error distribution, indicating a more significant variability in prediction accuracy. In addition, the density plots provide further confirmation of these observations. Finally, the GDI model displays a more normal distribution with a sharper peak, while the SLM model has a broader distribution with more pronounced tails.

3.2. Comparing the models

A visual representation of the changes in degradation over time is provided in (Fig. 07), allowing for an analysis of degradation and recovery trends in the area under study.

In both models, the areas classified as severely degraded are concentrated in proximity to roads and areas where conversion of use has occurred. The severe class exhibited no significant fluctuations throughout the period under analysis. In addition, the bar graph in (Fig. 08) illustrates the area's evolution with different degradation levels in 2020, 2021, and 2022 for each model.

The "non-degraded" area increased from 2020 to 2022, particularly in the GDI method, demonstrating a notable rise from 55.04% in 2020 to 69.47% in 2022. The SLM method also reflects an increase, albeit with less pronounced fluctuations. In addition, non-grass areas remain relatively constant, with slight variations between the SLM and GDI methods. This suggests that the area without grass cover stays mostly the same yearly.

In contrast, the moderately degraded areas show the most significant variations, reflecting the accuracy assessment results, and are also the class with the most confusion. In 2020, the GDI showed a more substantial amount (12,491 hectares) than the SLM (6,324 hectares). In 2021 and 2022, these areas decrease considerably, especially in the GDI.

Finally, the GDI demonstrates greater variability in the moderate and severe classes, which may be attributed to this method's heightened sensitivity to detecting more nuanced alterations in degradation.

4.1 Degradation and the application of remote sensing techniques

Many grassland monitoring studies that consider grazing have also adopted some vegetation indices and estimated parameters, which is similar to the development of research on grassland degradation monitoring (PEREIRA et al., 2018; SHENG et al., 2022). However, the vegetation indices underestimated the grassland degradation to a certain extent (WANG H. et al., 2021; WANG S. et al., 2022). This may explain the difficulty in defining the Moderate class and the confusion present in the models.

Assessing grassland degradation is challenging, so the first step is to define which characteristics contribute to the degradation in these environments. Elements such as biomass accumulation and productivity (WANG, Z. et al., 2023; ZHANG, Y. et al., 2019) and climate resilience (ZHANG M. et al., 2023) are evaluated as aspects and used as indicative of degradation levels.

Previous studies have employed the GDI with biophysical data on toxic herbs and species richness, yielding an accuracy rate of 98.6% (YANG et al., 2019). In contrast, authors have applied spectral mixing models to time series of Landsat data (BULLOCK; WOODCOCK; OLOFSSON, 2020) and hyperspectral data (LYU et al., 2020) to address the issue of degradation. In contrast, hyperspectral data yielded enhanced spectral separability, thereby facilitating more precise differentiation between cover classes. Nevertheless, the practical deployment of such data on a large scale is impeded by the expense and restricted accessibility of high-spatial-resolution hyperspectral imagery.

In Brazil, standardising the elements for assessing and monitoring these areas is difficult because there is no national classification system for grassland degradation, such as the one used in China (ZHANG Y. et al., 2019). Most remote sensing studies to assess grassland degradation focus on non-natural areas (HOTT et al., 2019; VALLE JÚNIOR et al., 2019).

Given the area's history of fire disturbance, it would be strategic to investigate vegetation recovery as a subsequent step. This would facilitate restoring essential ecosystem services, including forage production, pollination services, soil processes, and life habitat (ALBERTON et al., 2023; VILLARREAL et al., 2016). Also, the various levels of grassland degradation in reverse, as grasslands are recovering, could be defined.

4.2. Limitations and further studies

This study may have limitations as the model demonstrated low ability in distinguishing between grassland and other land uses. While mapping tools such as MapBiomas (SOUZA et al., 2020), which data were used as a base for grassland cover definition, are effective for various applications and provide valuable insights into land cover, they present some classification challenges. These include including wetlands in the classification of grasslands and excluding more degraded areas from the class. This underscores the necessity for establishing parameters to facilitate the identification of degradation. If a more precise grassland cover map had been used, fewer model errors might have happened.

It is recommended that additional studies be conducted using degradation levels established in the field, with the proposed models employed. A further avenue for future research could be the development and implementation of machine learning algorithms that are better able to manage the nuances of land cover classification, including the identification of subtle degradation features.

This comparative study assessed the efficacy of two remote sensing techniques, LSM and GDI, in detecting grassland degradation. The GDI technique demonstrated superior performance in all evaluated metrics, exhibiting greater accuracy and the capacity to identify both non-degraded and severely degraded areas accurately. In contrast, while LSM was less accurate overall, it proved more effective in classifying severely degraded areas.

The analysis indicated an increase in non-degraded areas from 2020 to 2022, highlighting the importance of sustainable management practices in recovering grassland ecosystems. This underscores the pivotal role of continuous monitoring and the implementation of effective restoration strategies to maintain and enhance the health of these vital landscapes.

Despite the effectiveness of these techniques, there are limitations in standardizing elements for assessing grassland degradation in Brazil. Future research should focus on vegetation recovery and restoring essential ecosystem services. Additionally, it is crucial to define degradation parameters specific to the type of grassland and grazing use in the study region.

In conclusion, while both LSM and GDI have their respective strengths, the continuous improvement and application of these remote sensing techniques are essential for effective grassland management and restoration efforts.

Competing interests:

The authors declare no competing interests.

Funding:

No funding was received to assist with the preparation of this manuscript. All costs were covered by the authors.

Author Contribution

P.B.H designed the study, collected the data, performed the analysis, wrote the paper, discussed the results, and contributed to the final manuscript. V.F.N and T.M.K provided critical feedback, supervised the project, verified the analytical methods, discussed the results, and contributed to the final manuscript. C.A.H.O. provide the field data, discussed the results, and contributed to the final manuscript. All authors contributed to the article and approved the submitted version.

Data availability:

The data used in the study are available upon request.

AHMED, Mohiuddin; SERAJ, Raihan; ISLAM, Syed Mohammed Shamsul. The k-means algorithm: A comprehensive survey and performance evaluation. Electronics (Switzerland), [s. l.], vol. 9, no. 8, p. 1–12, 2020. Available at: https://doi.org/10.3390/electronics9081295
ALBERTON, Bruna et al. Monitoring immediate post-fire vegetation dynamics of tropical mountain grasslands using phenocameras. Ecological Informatics, [s. l.], vol. 78, no. June, p. 102341, 2023. Available at: https://doi.org/10.1016/j.ecoinf.2023.102341
AN, Ru et al. Monitoring grassland degradation and restoration using a novel climate use efficiency (NCUE) index in the Tibetan Plateau, China. Ecological Indicators, [s. l.], vol. 131, no. April, 2021. Available at: https://doi.org/10.1016/j.ecolind.2021.108208
BOLDRINI, Ilsi. Campos do Rio Grande do Sul : caracterização fisionômica e problemática ocupacional. Boletim do Instituto de Biociências, [s. l.], vol. 56, no. Universidade Federal do Rio Grande do Sul, Porto Alegre, p. 1–39, 1997.
BOND-BUCKUP, Georgina. Biodiversidade dos Campos de Cima da Serra. Livro de Atividades. [s. l.], vol. xxxx, p. 96, 2008.
BULLOCK, Eric L.; WOODCOCK, Curtis E.; OLOFSSON, Pontus. Monitoring tropical forest degradation using spectral unmixing and Landsat time series analysis. Remote Sensing of Environment, [s. l.], vol. 238, no. April 2018, p. 110968, 2020. Available at: https://doi.org/10.1016/j.rse.2018.11.011
CASTELLANOS, E. et al. in: Climate Change 2022: Impacts, Adaptation, and Vulnerability.Contribution of Work. Chapter 12ed. [S. l.: s. n.], 2022. Available at: https://doi.org/10.1017/9781009325844.014
DAWELBAIT, Mona A A; MORARI, Francesco. Spectral Mixture Analysis and Change Vector Analysis to Monitor Land Cover Degradation in a Savanna Region in Sudan ( 1987-1999-2008 ). International Journal of Water Resources and Arid Environments, [s. l.], vol. 1, no. 5, p. 366–377, 2011.
GAO, Qing zhu et al. Alpine grassland degradation index and its response to recent climate variability in Northern Tibet, China. Quaternary International, [s. l.], vol. 226, no. 1–2, p. 143–150, 2010. Available at: https://doi.org/10.1016/j.quaint.2009.10.035
GAO, Qingzhu et al. Grassland degradation in Northern Tibet based on remote sensing data. Journal of Geographical Sciences, [s. l.], vol. 16, no. 2, p. 165–173, 2006. Available at: https://doi.org/10.1007/s11442-006-0204-1
HAN, Jiawei; KAMBER, Micheline; PEI, Jian. Data Mining. [S. l.]: Elsevier, 2012. Available at: https://doi.org/10.1016/C2009-0-61819-5
HOTT, Marcos C. et al. Analysis of grassland degradation in Zona da Mata, MG, Brazil, based on NDVI time series data with the integration of phenological metrics. Remote Sensing, [s. l.], vol. 11, no. 24, 2019. Available at: https://doi.org/10.3390/rs11242956
JAMES, G., WITTEN, D., HASTIE, T., & TIBSHIRANI, R. An Introduction to Statistical Learning: with Applications in R. Springer. 2013.
KOWALSKI, Katja; OKUJENI, Akpona; HOSTERT, Patrick. A generalized framework for drought monitoring across Central European grassland gradients with Sentinel-2 time series. Remote Sensing of Environment, [s. l.], vol. 286, no. January, p. 113449, 2023. Available at: https://doi.org/10.1016/j.rse.2022.113449
KUANG, Qian et al. A remote sensing monitoring method for alpine grasslands desertification in the eastern Qinghai-Tibetan Plateau. Journal of Mountain Science, [s. l.], vol. 17, no. 6, p. 1423–1437, 2020. Available at: https://doi.org/10.1007/s11629-020-5986-6
LIU, Min et al. The impacts of the eco-environmental policy on grassland degradation and livestock production in Inner Mongolia, China: An empirical analysis based on the simultaneous equation model. Land Use Policy, [s. l.], vol. 88, no. August, p. 104167, 2019. Available at: https://doi.org/10.1016/j.landusepol.2019.104167
LUZA, André Luís et al. Natureza & Conservação. [s. l.], vol. 2, no. 2, p. 166–169, 2014.
LYU, Xin et al. A new method for grassland degradation monitoring by vegetation species composition using hyperspectral remote sensing. Ecological Indicators, [s. l.], vol. 114, no. March, p. 106310, 2020. Available at: https://doi.org/10.1016/j.ecolind.2020.106310
OLOFSSON, Pontus et al. Good practices for estimating area and assessing accuracy of land change. Remote Sensing of Environment, [s. l.], vol. 148, p. 42–57, 2014. Available at: https://doi.org/10.1016/j.rse.2014.02.015
OVERBECK, Gerhard E et al. Conservation in Brazil needs to include non-forest ecosystems. Diversity and Distributions, [s. l.], vol. 21, no. 12, p. 1455–1460, 2015. Available at: https://doi.org/10.1111/ddi.12380
PEREIRA, Osvaldo José Ribeiro et al. Assessing pasture degradation in the Brazilian Cerrado based on the analysis of MODIS NDVI time-series. Remote Sensing, [s. l.], vol. 10, no. 11, 2018. Available at: https://doi.org/10.3390/rs10111761
PILLAR, Valério de Patta; et al. Campos Sulinos - conservação e uso sustentável da biodiversidade. Ministério do Meio Ambiente, Brasília, [s. l.], p. 403, 2009.
ROUSE, J.W; HASS, j. a.;; SCHELL, j. A. Monitoring vegetation systems in the great plains with ERTS. Oxford University, [s. l.], vol. 1, no. NASA-Earth Resources Technology Satellite Symposium, p. 309–317, 1973.
SEMA. PLANO DE MANEJO DO PARQUE ESTADUAL DO TAINHAS. Porto Alegre, [s. l.], 2008.
SHENG, Jun et al. Aboveground productivity and community stability tend to keep stable under long-term fencing and nitrogen fertilization on restoration of degraded grassland. Ecological Indicators, [s. l.], vol. 140, no. May, p. 108971, 2022. Available at: https://doi.org/10.1016/j.ecolind.2022.108971
SHIMABUKURO, Yosio Edemir; SMITH, James A. The Least-Squares Mixing Models to Generate Fraction Images Derived From Remote Sensing Multispectral Data. IEEE Transactions on Geoscience and Remote Sensing, [s. l.], vol. 29, no. 1, p. 16–20, 1991. Available at: https://doi.org/10.1109/36.103288
SOUZA, Carlos M. et al. Reconstructing Three Decades of Land Use and Land Cover Changes in Brazilian Biomes with Landsat Archive and Earth Engine. Remote Sensing, [s. l.], vol. 12, no. 17, p. 2735, 2020. Available at: https://doi.org/10.3390/rs12172735
SUN, Bin et al. Grassland degradation and restoration monitoring and driving forces analysis based on long time-series remote sensing data in Xilin Gol League. Acta Ecologica Sinica, [s. l.], vol. 37, no. 4, p. 219–228, 2017. Available at: https://doi.org/10.1016/j.chnaes.2017.02.009
VALLE JÚNIOR, Renato Farias do et al. Diagnosis of degraded pastures using an improved NDVI-based remote sensing approach: An application to the Environmental Protection Area of Uberaba River Basin (Minas Gerais, Brazil). Remote Sensing Applications: Society and Environment, [s. l.], vol. 14, no. December 2018, p. 20–33, 2019. Available at: https://doi.org/10.1016/j.rsase.2019.02.001
VILLARREAL, M.L. et al. Multi-index time series monitoring of drought and fire effects on desert grasslands. Remote Sensing of Environment, [s. l.], vol. 183, p. 186–197, 2016. Available at: https://doi.org/10.1016/j.rse.2016.05.026
WANG, Hao et al. Satellite-derived NDVI underestimates the advancement of alpine vegetation growth over the past three decades. Ecology, [s. l.], vol. 102, no. 12, p. 1–8, 2021. Available at: https://doi.org/10.1002/ecy.3518
WANG, Shanshan et al. Assessment of Grassland Degradation on the Tibetan Plateau Based on Multi-Source Data. Remote Sensing, [s. l.], vol. 14, no. 23, p. 6011, 2022. Available at: https://doi.org/10.3390/rs14236011
WANG, Zelin et al. Spatiotemporal evolution and attribution analysis of grassland NPP in the Yellow River source region, China. Ecological Informatics, [s. l.], vol. 76, no. May, p. 102135, 2023. Available at: https://doi.org/10.1016/j.ecoinf.2023.102135
YANG, Yanlin et al. Remote-sensing monitoring of grassland degradation based on the GDI in Shangri-La, China. Remote Sensing, [s. l.], vol. 11, no. 24, 2019. Available at: https://doi.org/10.3390/rs11243030
ZHANG, Mengru et al. Contributions of climate change and human activities to grassland degradation and improvement from 2001 to 2020 in Zhaosu County, China. Journal of Environmental Management, [s. l.], vol. 348, no. May, p. 119465, 2023. Available at: https://doi.org/10.1016/j.jenvman.2023.119465
ZHANG, Ying et al. Comprehensive research on remote sensing monitoring of grassland degradation: A case study in the Three-River Source Region, China. Sustainability (Switzerland), [s. l.], vol. 11, no. 7, 2019. Available at: https://doi.org/10.3390/su11071845
ZHOU, Wei et al. Grassland degradation remote sensing monitoring and driving factors quantitative assessment in China from 1982 to 2010. Ecological Indicators, [s. l.], vol. 83, no. December 2016, p. 303–313, 2017. Available at: https://doi.org/10.1016/j.ecolind.2017.08.019
ZHOU, Yuting et al. Examining the short-term impacts of diverse management practices on plant phenology and carbon fluxes of Old World bluestems pasture. Agricultural and Forest Meteorology, [s. l.], vol. 237–238, p. 60–70, 2017. Available at: https://doi.org/10.1016/j.agrformet.2017.01.018
HAN, J., KAMBER, M., & PEI, J. (2012). Data Mining: Concepts and Techniques. Morgan Kaufmann.

No competing interests reported.

Download PDF

Editor assigned by journal
25 Aug, 2024
Submission checks completed at journal
19 Aug, 2024
First submitted to journal
19 Aug, 2024

You are reading this latest preprint version

Evaluating Remote Sensing Techniques for Monitoring Grassland Degradation

Status:

Version 1

Abstract

Figures

1. INTRODUCTION

2. MATERIALS AND METHODS

2.1. Study Area

2.2. Comparative Models

2.2.1 Grassland Degradation Index

2.2.2 Linear Spectral Mixture Model

2.3. Grassland Survey Data

2.4. Accuracy Test

3. RESULTS

3.1. Accuracy

3.2. Comparing the models

4. DISCUSSION

4.1 Degradation and the application of remote sensing techniques

4.2. Limitations and further studies

5. CONCLUSIONS

Declarations

Competing interests:

Funding:

Author Contribution

Data availability:

References

Additional Declarations

Status:

Version 1