Vegetation Change Detection and Recovery Assessment on Post-fire Satellite Imagery using Deep Learning

Vegetation Change Detection and Recovery Assessment on Post-fire Satellite Imagery using Deep Learning | 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 Vegetation Change Detection and Recovery Assessment on Post-fire Satellite Imagery using Deep Learning R. Shanmuga Priya Rajendran, K. Vani K This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3890182/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Wildfire are uncontrolled fires fueled by dry conditions, high winds and flammable materials that tends to have a profound impact on vegetation due to the intense heat generated by it which can cause the destruction of trees, small plants and other vegetation leading to significant consequences including noteworthy changes to ecosystems. Due to the periodic wildfires, vegetation communities in forest systems have changed adaptively to deal with ecological rebuilding. In this study we provide a novel methodology, to understand and evaluate post-fire effects on vegetation. In regions which are affected by wildfire, earth-observation data provided by various satellite sources can be very vital in monitoring vegetation and assessing the effect a wildfire tends to have on it. These effects can be understood by detecting the change of vegetation over years using an unsupervised method termed Deep Embedded Clustering (DEC), which enables us to classify regions on whether there has been a change in vegetation after fire. Appropriate vegetation indices can be used to evaluate evolution of vegetation pattern over the years, for this study we utilized Enhanced Vegetation Index (EVI) based trend analysis. Vegetation recovery maps can be created to assess re-vegetation in regions affected by fire which is performed via a deep learning based unsupervised method, Adaptive Generative Adversarial Neural Network Model (AdaptiGAN) on postfire data collected from various regions affected by wildfire. Through the results obtained from the study we can arrive at a conclusion that our approach tends to have notable merits when compared to pre-existing works. Wildfire change detection Ensemble learning deep learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 Figure 19 Figure 20 Figure 21 Introduction Wildfires, a significant natural disaster, can cause changes to compositions and leads to ecological imbalance along with making major modifications to an ecosystem of a region. The identification of land cover changes and keeping a watch on it is necessary for managing a balanced ecosystem and economical studies at the various levels (Sharma et al., 2019). For continuous environmental monitoring and to thoroughly examine urgent environmental issues including degradation of natural resources, loss in biodiversity, and forest cover loss, forest cover change identification is very important (Negassa et al., 2020 ). For ecological research as well as forest monitoring and management, it’s essential for comprehending how fire affects forests and to identify the patterns of time and space in recovery of forests after fire. (Bonan, 1989; Stueve et al., 2009; Kennedy et al., 2012; Otoda et al., 2013). Traditional approaches including ground surveys are costly and requires a lot of time. This can be attributed to various factors which makes the task of understanding the changes made to the vegetation practically infeasible. To overcome this making use of remote sensing, one can monitor changes across broad areas in an economical and effective manner (Townshend et al., 2012). Change detection can provide crucial data for handling the disaster, policy creation, effective land cover and forest management (Tian et al., 2022 ). Unprecedented fire seasons worldwide prompt a need to understand into how vegetation respond to evolving fire patterns. Usage of Vegetation Indices for assessing and analyzing the regrowth of vegetation in a region after any major disaster is a technique regularly used by researchers for the purpose. We can see that data collected at three different post-fire periods revealed distinctions between vegetation in a control forest and regenerating regions (Úbeda et al., 2006). Due to changes in land use and climate, wildfires have become more frequent and severe and many tend to last long which cause a lot of damage to vegetation resulting in a need for quantifying vegetation recovery in fire-affected regions so that appropriate steps can be taken to speed up this process. Assessing vegetation recovery pose challenges because of extended duration required to capture unique traits of an ecosystem. The use of time-series based remote sensing data allows for assessments made prior to, during and following fire, enhances accuracy. Related works: Accurate and effective methods for collecting quantitative forest change data from remotely sensed photos, developed and tested in a monitoring-based study, are essential for the ongoing monitoring of forest clearing in the MBR (Kristensen et al., 1997). This change detection analysis is a useful method for characterizing the alterations noticed throughout every category of land utilization. The classification of land use would be properly enhanced by high resolution satellite data. For feature extraction, the normalized difference vegetation index technique has been utilized with a range of threshold values. NDVI approach yields more effective outcomes for vegetation with different types as well as dispersed vegetation from a multispectral remote sensing picture (Meera et al., 2015). Satellite data provides unparalleled effectiveness in tracking and measuring extensive changes in landscape across time. But radiometric consistency across multi- temporal imagery must be ensured to get clear detection of changes in terrain (Chen et al., 2005 ). There is a need for Integrating both deep learning techniques and Object-based image analysis techniques for change detection which can make this process to be lot more efficient and accurate (Liu et al., 2021 ).Various approaches have been utilized for changed area detection which includes Patch-Based detection (Joesph et al., 2017), Pixel- Based to Object-Based approaches (Hussain et al., 2013 ), Landsat time series data (Zhu, 2017 ). Accuracy of various object-based detection techniques is dependent upon the quality of image segmentation and many approaches tend to struggle from the under-segmentation error (Wang et al., 2022 ). This can be due to the sparse nature of the features present in images. Enhancing our understanding of how ecosystem imbalance and changes in order of fires, due to climate, can influence post-fire development speed and adaptability is crucial for improving predictions of community responses to fire in the context of climate change (Nolan et al., 2021). When vegetation that is photosynthetic is destroyed by fire, reflectance in the visible to near-infrared range drastically decreases, while reflectance in the short and middle infrared range increases because of increased ash [Lentile et al., 2006; Miller and Thode, 2007]. Due to their ability to absorb residual approximations they can provide a standardized representation of vegetation or something close to it which improves monitoring to a larger extent (Curran, 1981; Goward, 1989; Malingreau, 1989). Vegetation Indices are very useful and are compliant when the region for monitoring is huge (Xiao & Moody, 2005; De Petris et al., 2021). Since satellite data has become accessible daily, it becomes feasible to efficiently track vegetation regrowth in more granular manner using the Normalized Difference Vegetation Index (NDVI) (Lacouture et al., 2020). Studies indicate that due to recurrent fires in a region replacement of a vegetation species by another kind of species takes place and this process can be quite prolonged in nature owing to the nature of both the species (Vasques et al., 2023). This should be taken into consideration while evaluating the possibility of vegetation regrowth, which can be attributed to various factors like soil ph, nitrogen content etc. VIIRS proved to be a good way for estimating spectral indices, over Landsat, given the decommissioning of the MODIS satellites (Jarchow et al., 2018). While previous studies have commonly employed NBR with NDVI for extracting regions of plant regeneration, these approaches are susceptible to atmospheric effects and soil brightness (Kim et al., 2021). Spectral indices, including NDVI, EVI and distinction in various indices spanning the years before and after fire, can be calculated and analyzed (Chen et al., 2011). NDVI proved more effective in tracking vegetation changes (Natalie et al., 2018). To analyze post-fire vegetation recovery, usage of multi wavelength satellite data is an approach which utilized various remotely sensed parameters including visible–infrared vegetation indices (Bousquet et al., 2022). Vegetation indices credibility for effective vegetation monitoring can be further attributed to the fact that it is used for drought monitoring (Benedict et al., 2021). Given the various applications utilizing NDVI derived from MODIS and VIIRS based products we can see that MODIS and VIIRS NDVI data are interchangeable for applications with an uncertainty ranging from 0.02 to 0.05, depending on the scale of spatial aggregation, typically consistent with the individual datasets uncertainty making data provided by the VIIRS- based products especially the NDVI data good enough to perform monitoring (Skakun et al., 2018). Even though the usage of vegetation indices can provide an insight about vegetation recovery in a region, concatenating this along with an unsupervised learning approach is something which is not explored a lot before and can provide more accurate information about that region. In summary, we can come to an understanding that Sparse Autoencoder performs better than simple Autoencoder to extract features which can greatly improve the change detection process when it is used along with Deep Embedded Clustering, typically used for segmentation purposes due to the better performance of the algorithm and its relatively simple implementation. Performing vegetation inference through usage of Sen’s slope based on EVI trend analysis along with ensemble learning can be used to precisely identify whether there is a possibility for vegetation regrowth. GAN’s are of various types and this is dependent on the task they tend to perform (Yumoto et al., 2023), and taking this as inspiration we propose a deep learning based unsupervised adversarial network. AdaptiGAN which is trained on the EVIIRS NDVI composites obtained after several preprocessing steps, can provide a more efficient and accurate description about the level of vegetation recovery in the fire-affected regions. Data and Methodology Data and preprocessing For performing vegetation change detection analysis after fire, a dataset comprising of pre-fire and post-fire satellite images was constructed. This dataset comprised of 3600 pre- fire and post-fire Normalized Difference Vegetation Index (NDVI) images of the EROS Moderate Resolution Imaging Spectroradiometer (eMODIS) collection which is obtained from the from the U.S. Geological Survey (USGS) website ( https://earthexplorer.usgs.gov/ ) for regions which were affected by fire and the search criteria for this was based upon the time periods for pre-fire and post-fire images. Appropriate corrections were made to ensure consistency and accuracy. The QGIS Software is used for the calculation of NDVI index and splitting of region into several classes of vegetation based on index value of each pixel of multi spectral image. For vegetation regrowth assessment remote sensing data is collected for each year starting from 2000 to 2022. The MODIS 250m/pixel 16-day composite vegetation indices dataset from the World Database on Protected Areas (WDPA) dataset, is used for this purpose. From the obtained composites an image-based dataset was constructed. Additionally, for predicting whether vegetation regrowth is possible in a region involves collecting soil data from the soil grid database this data consists of columns including Location Co-ordinates, soil ph, soil nitrogen content, organic carbon content, bulk density and soil groups. All these attributes are then concatenated and represented in the form of a single dataset. The dataset constructed for the purpose of vegetation recovery was downloaded from the USGS website mentioned above. It consisted of 1600 pre-fire images from three main region’s which were more suitable for our study namely: Amazon rainforest of Brazil with location co- ordinates 3.4653° S, 62.2159° W, Knysna region of South Africa with location co-ordinates 34.0351° S, 23.0465° E and Alaska with location co-ordinates 63.5888° N, 154.4931° W. The construction of dataset involved downloading Suomi NPP satellite’s EROS Visible Infrared Imaging Radiometer Suite (eVIIRS) data then loading this in QGIS software followed by creating an output layer which contains the NDVI values obtained by using the raster calculator present in the software. This layer is then exported to .jpeg format used for model training and evaluation purposes. This preprocessing step is performed for all the collected data and the obtained .jpeg file is added to the dataset. Methodology Wildfires cause lots of damage to ecology and economy, vegetation is affected by sudden increase in temperature which can lead to destruction of trees, shrubs and herbs. Loss of habitat is another issue caused by fire, change in vegetation composition is possible as fire-adapted species tend to dominate over others that are less tolerant to fire, there is also a huge question mark over the vegetation regrowth possibilities in regions particularly affected by huge fires and analyzing recovery patterns after fire is key to understanding the effect of fire and what all steps can be taken to improve the recovery if at all the recovery is less. So for this we proposed a integrated framework involving: Vegetation change detection by making use of sparse Autoencoders and Deep Embedded Clustering (DEC) model, an unsupervised learning technique through which change map showing the difference between pre and post fire images is obtained which is then embedded with the pre-fire image to show the change’s caused by fire; Vegetation Regrowth Assessment involves usage of Enhanced Vegetation Index (EVI) to get a time series representation of vegetation over years through browning and greening fraction and additionally performed through the use of AdaptiGAN, an unsupervised deep learning model, which takes preprocessed eVIIRS NDVI image as input and then provides the corresponding recovery map for that image. The architecture diagram for the proposed approach can be seen in Fig. 1 . Change Detection using Deep Embedded Clustering (DEC) Comprehensive assessment of long-term trends in vegetation change at the field scale is required for resource management and ecological assessment. Remote sensing data have been employed widely as greatest asset for change detection. The common approaches deployed for change detection include post-classification comparison, principal component analysis (PCA), and image differencing. New methods are required to effectively use the more complex and diversified remotely sensed data that is anticipated to become so in the near future via satellite and airborne sensors, which is still an active area of research. Vegetation change detection provides an analysis about difference between satellite images which were obtained before and after fire. Dataset comprises 3600 pre & post-fire images obtained from eMODIS NDVI v6. The QGIS Software is used for the calculation of NDVI index and splitting of region into several classes of vegetation based on index value of each pixel of multi spectral image. The change detection process uses Sparse Autoencoder for extracting required features and then this representation is fed as input to the DEC model, an unsupervised learning technique used to iteratively group features, and ensuing assignments are used as supervision to update network weights. The model acquires feature representations through successive iterations by using labeled and unlabeled data points and finding target distributions from prediction alternatively. The distinction between images obtained before and after the fire pertaining to forest region in Jefferson, California is used to create the change map. Figure 2 represents workflow of vegetation change detection. Splitting the Classes The QGIS Software has been used to calculate the NDVI index and produce five different classes ranging from low vegetation to very high vegetation. The Figure 3 below represents the NDVI index value ranging from -1 to +1.The Figure 4 represents the set of classes generated from NDVI. Table 1 shows the Vegetation type and its associated NDVI Index. Sparse Autoencoder : An autoencoder represents a particular kind of Artificial Neural Network(ANN) typically used for unsupervised learning. Autoencoder that attains bottleneck information with an added restriction on sparsity is called as sparse autoencoder. The loss function is designed to push activations inside a layer. L1 regularization or Kullback-Leibler (KL) divergence between appropriate distribution and anticipated mean neuron activation, such that sparsity constraint can be applied, in other words the autoencoder is not only minimizing the difference between the input and the reconstructed output but also encouraging sparsity in the activations of the hidden layer. Table 1 Vegetation and their corresponding NDVI ranges Vegetation NDVI Index No Vegetation Area -1 to 0 Bare Area 0 to 0.1 Low vegetation Area 0.1 to 0.25 Moderate vegetation Area 0.24 to 0.4 High vegetation Area 0.4 to 1 L1 Regularization By scaling the absolute value of the activation vector in layer h for observation i by a tuning parameter λ with its features x and x^ we may add a term to our loss function that penalizes it. $$L\left(x,\widehat{x}\right)+ \lambda . \sum _{i=1}^{n}i.‖a\left(h\right).i‖$$ KL-Divergence: KL-divergence is a measurement of variation in distribution over probability between two samples. A sparse parameter ρ represents average activity of a neuron across a group of samples 𝜌̂. Neurons are induced to fire for a subset of observations by limiting average activity of a neuron and differentiate j it over a group of samples. To contrast expected distribution to actual distributions over all the hidden layer nodes, we can define as a Bernoulli random variable distribution and use the KL divergence. $$L\left(x,\widehat{x}\right)+ \sum _{j=1}^{n}jKL\left(\rho \right||\widehat{\rho }j)$$ Deep Embedded Clustering The model DEC works in two phases: Initialize phase with deep sparse autoencoders Clustering, where they successively repeat calculating a supplementary target distribution and minimize the KL divergence associated with it. Figure 5 represents the block diagram of Deep Embedded Clustering. It briefs workflow of model comprising Autoencoder and Clustering block. Initially a soft assignment is calculated between cluster centroids (classes of vegetation) and embedded points. Then t-distribution computes similarity index between embedded point and centroid t-distribution is used as a non-parametric representation to identify similarity index between embedded points. By employing an auxiliary target distribution and learning from recent high confidence assignments, one can upgrade deep mapping and improve the centroids of cluster. In particular, soft assignment is matched to target distribution to train the model and centroid of clusters. The basic workflow of this change detection process is, images obtained before and after fire are given as inputs simultaneously to our DEC model. Sparse autoencoder used to extract features and pass it to clustering function where each of its similar features is aggregated as a cluster. It gives binary map of pre & post-fire image, by finding difference between the images obtained before and after the fire we get our change map as final output, which in return will be embedded with pre-fire image to produce the changes in vegetation of region. Ensemble Learning Vegetation regrowth assessment following a wildfire is an important aspect of post-fire monitoring and ecosystem recovery. Sen’s slope, also termed as the Sen’s estimator or the Sen’s method, is statistical technique used to assess trends or changes in time series data. While Sen’s slope is typically used for analyzing trends in various fields, such as hydrology and climate science, it can also be applied to assess vegetation regrowth following a wildfire. This involves obtaining time series data representing vegetation indices or other relevant vegetation metrics for the study area. This data should span multiple time periods, covering period before and after the wildfire event. Sens’s slope of EVI index is calculated. Based on the magnitude of slope, browning and greening fraction can be determined. The trend of EVI pattern over the years is analyzed and it is visualized as a graph. MODIS dataset has been acquired for different regions and their NDVI images are converted to RGB frames and aggregated to provide animation representing change in vegetation over the years. The regrowth possibility prediction involves collecting soil data from soil grid database which is then trained using ensemble learning to provide an outcome. Figure 6 represents workflow of vegetation regrowth assessment. The working of Vegetation Regrowth Assessment using Sen’s slope can be described in the following steps: Data Collection Remote sensing data is collected for a specific area over a period of time, such as satellite images from MODIS dataset. Pre-processing Pre-processing the remote sensing data eliminates any noise or artifacts to convert it into a usable format. This may involve tasks such as cloud removal, atmospheric correction and radiometric calibration. Vegetation Index Calculation : Vegetation index, such as Enhanced Vegetation Index (EVI), is calculated from pre-processed remote sensing data. Time-Series Analysis : Vegetation index values for each time step are analyzed using Sen’s slope estimator to calculate the trend or slope of vegetation index over time. The result approximates the pace at which an area’s vegetation is growing again or decreasing. Visualization The results of analysis can be visualized using various techniques, such as a time series plot of vegetation index or a map showing spatial distribution of the vegetation regrowth or decline. Vegetation Regrowth Assessment using Sen’s slope is an excellent technique for providing insights into natural regeneration steps of vegetation as it also allows to identify areas that may need interventions to support recovery in a region, after a disaster like wildfire. Plotting trends based on EVI The following steps describe the process of Plotting trends based on EVI values calculated over a period of time: The World Database on Protected Areas (WDPA) dataset can be used to create MODIS 250m/pixel 16-day composite vegetation indices dataset. Create a collection of images by adding images for every year between 2000 and 2002. Every one of the images has been computed to attain the highest EVI throughout every month of respective years. This is an annual examination of the state of vegetation. In order to prepare for a linear- trend analysis, add the year as a band. Determine each pixel’s Sen’s slope of highest summer EVI over time to estimate a linear trend. Compute and display the regression slope values as histograms. We can determine the browning and greening faction of the vegetation by measuring the slope’s value. Sen’s slope = The dataset for soil properties has been acquired from the soil grid database. Stacking is an ensemble machine learning algorithm to combine results from multiple models. In the proposed work, we consider simple machine learning algorithms including Logistic Regression, SVM, Decision Tree, Random Forest and Naïve Bayes as weak models and their predictions are given to generalizer to provide the final outcome of regrowth possibility. The Fig. 7 below shows the ensemble learning approach using stacking. So, the basic architecture of the utilized Stacking process is: Base Models (Level 0): This involves training diverse base models in this case Logistic Regression, SVM, Decision Tree, Random Forest and Naïve Bayes on the training data. Meta-Model (Level 1): A meta-model in our case the Generalizer is trained on the predictions from the base models. This model learns to combine the base model’s prediction to generate a final prediction. Final Prediction (Output): Based on the soil properties as input the trained model makes predictions about regrowth possibility in a region. Adaptive Generative Adversarial Network (AdaptiGAN) Vegetation Recovery Mapping is a crucial step which can provide valuable information about the ecological, environmental, and management aspects of post-fire landscapes. Traditional approaches which involve field surveys and ground truthing can turn out to be cumbersome and expensive at the same time providing a need for a more efficient approach which can speed-up this process. This involves making use of remote-sensing technologies such as satellite imagery combined with spectral indices like NDVI and deep learning techniques. Now making use of both remote sensing technologies and deep learning together can provide efficient and accurate results that can be used for future planning like resource allocation, habitat restoration in fire affected regions. For this purpose, the proposed approach is collecting satellite data from e-VIIRS products obtained from Suomi NPP satellites, of different fire affected regions which is then preprocessed using QGIS tool via which NDVI is calculated and an image like representation is obtained. This dataset serves as the training data for an unsupervised learning algorithm termed AdaptiGAN which can be adapted based on the kind of data we feed into it. Using this trained model, we can obtain recovery maps for different regions affected by fire for different time periods. Figure 8 represents the framework for the Vegetation Recovery Mapping Approach. The AdaptiGAN is a neural network architecture, a Generative Adversarial Network (GAN), whose generator follows an Encoder-Decoder architecture making use of self-attention mechanisms for recording long-range dependencies which in turn will help in improving feature representation and the discriminator makes use of the PatchGAN architecture which helps in capturing fine-grained details of the input images. With self-attention mechanisms and normalization techniques in place, they are extremely useful in capturing domain- specific features which is very important in our case of generating recovery maps. The key components related to this architecture can be seen in more detail below and the network architecture can be seen in Fig. 9 . Components in the AdaptiGAN architecture Generator: This follows an encoder-decoder architecture. The encoder downsamples the input image to extract features, and the decoder upsamples these features to get the final output image. The downsample layers reduce spatial dimensions and increase the number of channels capturing hierarchical features while upsample layers increase spatial dimensions enabling the generation of a high-resolution image. A self-attention mechanism is introduced after the third downsampling layer to capture long-range dependencies and improve feature representations. Discriminator : The discriminator employs a PatchGAN architecture, where it classifies local patches of the input images as real or fake. This helps in capturing fine-grained details. Convolutional layers with leaky ReLU activations are used for feature extraction and discrimination. Batch normalization is applied to normalize the activations, aiding in the stability and training of the discriminator. Weight Regularization : L2 weight regularization is applied to the convolutional layers of both the generator and discriminator. This helps to fend off overfitting and improves generalization of the architecture. Instance Normalization : Instance normalization is applied in the discriminator after the second and third convolutional layers. It normalizes activations across channels and spatial dimensions independently for each sample. Dropout : Dropout is applied in the generator after the concatenation of feature maps during the decoding process. It helps regularize the network by randomly dropping a fraction of the units during training. Activation Functions : ReLU activation is used in various parts of both the generator and discriminator to introduce non- linearity while Leaky ReLU is used in the discriminator to allow a small, non-zero gradient when the input is negative. Summation: Summation layers are used in the generator to combine feature maps from different stages, aiding in the generation of detailed and realistic images. Output Layer : The generator has a tanh-activated convolutional layer in the output to produce the final generated image. Results and Discussion To verify the viability of our proposed approach, performance of the three modules was tested. There are two areas for the experimental equipment: one for testing and the other for training. In the training phase, deep learning model and machine learning model used for this research were trained in the Google Colaboratory which provided hosted jupyter notebook with python environment implemented on a server with AMD® 7000 Series Ryzen™ 9 7950X CPU @(5.7 GHz) with 16GB memory, Radeon RX 7800 XT (GPU) with 16GB of memory. For testing, performance analysis was performed on the vegetation change detection module by using metrics like precision, recall. This was performed on the same Google Colab platform with python environment. The vegetation regrowth assessment module was then deployed using streamlit web application to display the results for the possibility of vegetation regrowth in a region, which provided an interface where user can enter Location details, soil data to predict regrowth possibility. This can be validated by cross referencing the same location in USGS earth explorer tool to check for vegetation. For the vegetation recovery mapping module metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE) were used .The server used for this testing environment is AMD® 7000 Series Ryzen™ 9 7950X CPU @(5.7 GHz) with 8GB memory, Radeon RX 7800 XT (GPU) with 8GB of memory. Experimental Results and Analysis for Vegetation Change Detection using Deep Embedded Clustering Vegetation change detection to assess long term trends in vegetation change we propose DEC model, change detection is analyzed by finding difference between satellite-based imagery obtained before and after fire. Dataset comprises 3600 pre-fire and post-fire images obtained from eMODIS NDVI v6. Figure 10 . qualitative comparisons result of DEC with other deep convolutional models. Vegetation change detection performance is evaluated based on Precision, Recall, F1 and Accuracy (ACC), in comparison with other trained models. The DEC model achieved an impressive accuracy of 96.17%. Table 2 . comparison results between existing approaches such as STA Net, Bi- attention SFA, and our proposed model DEC. The deep embedded clustering is evaluated based on the loss metrics contributing to minimize the KL divergence. Proposed change detection model shows a loss of around 18.57 which is considerably lower than existing approaches. Table 3 . Loss metrics compared between various existing models such as STA Net, Bi-attention SFA and proposed model DEC. The deep embedded clustering is compared with the various other change detection models and justified with minimal loss value after successful completion of epochs. Table 2 Comparison results of various deep convolutional network models for wildfire prediction. Model Precision Recall F1 Accuracy (%) STA Net 81.25% 82.38 85.68% 87.15% Bi-attention SFA 82.59% 83.79% 87.32% 88.60% DEC 90.07% 91.45% 93.72% 96.17% Table 3 Loss tabulation metrics for vegetation change detection Model Loss STA Net 26.27 Bi-attention SFA 26.77 DEC 18.57 Experimental Results and Analysis for Vegetation Regrowth Assessment using Ensemble Learning After a wildfire, recovery of an ecosystem and post- fire monitoring rely primarily on the regeneration of vegetation. Sen’s slope estimator is used to calculate trend of EVI pattern over the years and it is visualized as a graph. The browning and greening fraction is determined based on the magnitude of slope. Figures 11 & 12 . EVI trend graph of Moore creek, Florida creek, Tosher creek and Tonalite creek along with its browning and greening fraction for ‘ the respective area over the years. Table 4 . Tabulation comparison of browning and greening fraction with its sq.km for moore, florida, tosher and tonalite creek. The dataset for vegetation regrowth assessment is collected from Google Earth Engine fetched from MODIS dataset. Sen’s slope is used for estimating the browning and greening fraction. It is used to analyze the trend of EVI pattern as well. The following sections below show the tabulated results and graphs of various other models and proposed system thereby adding a justification. Ensemble learning can be used to provide the output for regrowth. Figure 13 shows the NDVI images obtained after processing them using QGIS application. Regrowth possibility is predicted by collecting soil properties and trained them using ensemble learning to estimate its results. Performance of the system is tested using Streamlit platform, it is deployed to display results for the possibility of vegetation regrowth. User can enter the location coordinates (latitude and longitude), pH value, nitrogen value and soil group from where data are fetched and possibility of prediction is estimated. Figure 14 shows the results obtained from the streamlit application which predicts possibility and not-possibility of vegetation regrowth. Table 4 Comparison results of browning and greening fraction using Sen’s slope Name Browning fraction Browning sq.km Greening fraction Greening sq.km Moore Creek 0.284 0.529 0.978 0.154 Florida Creek 0.106 0.362 0.8950 3.066 Tosher Creek 0.723 0.945 0.28 0.365 Tonalite Creek 0.45 18.197 0.553 22.368 Experimental Results and Analysis for Vegetation Recovery Mapping using Adaptive Generative Adversarial Network (AdaptiGAN) Vegetation Recovery Mapping is an important step in understanding the ecological, environmental, and managerial elements of post-fire environments. This module involved usage of vegetation indices like NDVI and AdaptiGAN a a deep learning based neural network framework to provide a recovery map absolute error, is a loss function which basically combines the best properties of both MSE and MAE. Its less sensitive to outliers than MSE and provides a smooth transition to MAE at zero error. Here, y is true value, \(\widehat{y}\) is predicted value, and 𝛿 can be a threshold that determines when to switch from quadratic to linear behavior. based on the input image. Figure 15 shows the qualitative comparisons results of AdaptiGAN with other trained models, obtained for the Amazon rainforest. The trained model’s performance was evaluated using evaluation metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Squared Logarithmic Error(MSLE), Root Mean Squared Error (RMSE), Huber loss. The obtained values are tabulated in the Table 5 . Unlike the metrics like MSE, MAE, RMSE, MSLE which are used often for evaluation purposes the Huber loss, also known as the smooth Figure 16 shows the qualitative comparison results of AdaptiGAN with other trained models, obtained for the Knysna Region. Similarly Fig. 17 shows the comparison results obtained for our AdaptiGAN model with other trained models, obtained for Alaska Region. The Fig. 18 below provides a visualization about the loss variation obtained for different models along with our proposed AdaptiGAN model. The Table 5 provides the comparison for various models with our proposed AdaptiGAN model with respect to the chosen performance metrics. Figure 19 provides us the performance-analysis plot for the pre-trained models along with our AdaptiGAN Model. Figure 20 illustrates the “Test for Homoscedasticity” plot for pretrained models compared with our AdaptiGAN model, homoscedasticity can be good for analysis as it provides information about whether the model fully captures the underlying patterns in the data. We can see that AdaptiGAN doesn’t follow a clear trend while other models tend to follow suggesting there are heteroscedastic in nature which can lead to biased estimates. Figure 21 shows the plots for evaluation metrics chosen to quantify the performance of our model. Table 5 Comparison between trained models and our proposed AdaptiGAN model Model MAE MSE MSLE RMSE Huber Loss DNN 0.914 90.716 0.201 9.524 0.999 LSTM 0.723 68.481 0.191 8.275 0.989 CycleGAN 0.593 49.329 0.187 7.023 0.975 AdaptiGAN 0.075 6.863 0.167 2.619 0.926 Conclusions We have proposed and presented an approach for analyzing the effects caused by a disaster like wildfires on vegetation. This approach involves vegetation change detection which is performed using an unsupervised learning algorithm called DEC along with sparse encoders for feature extraction which provides results with an impressive accuracy of 96.17%, vegetation regrowth analysis was performed using Sen’s slope estimator which provided time-series based analysis for regions affected by fire, we considered 4 creeks for this study namely: Moore, Florida, Tosher and Tonalite creek based on which the EVI trend analysis was performed to visualize the greening and browning trend over years by using data from the MODIS dataset to analyze the EVI pattern over years along with this regrowth possibility was predicted using an ensemble learning method called stacking which take soil data as input and provides the regrowth possibility for the region as output on whether there is a possibility for regrowth or not in that region and finally we performed vegetation recovery mapping which required collecting VIIRS data from the USGS website from which NDVI images were extracted, these NDVI images were then used for training the AdaptiGAN model. Using the trained model predictions were made in 3 regions chosen for this research namely Amazon rainforest, Knysna, Alaska regions and the corresponding recovery maps were obtained. The major advantages of our approach include it’s extreme flexibility and can be used for analyzing vegetation for different regions i.e. it’s not region specific, additionally its efficient and can be used by organizations to analyze the effects caused by wildfires without spending a lot of money on ground-based analysis, our approach can meet the needs posed by the real-time disaster response scenarios due to its high accuracy and speedy performance. In the future we are planning to make prediction based on video analysis making use of time-lapse based satellite data and a more comprehensive technique which can effectively make note of all changes and effects caused by a wildfire by additionally incorporating the effect it had on wildlife, what all changes and precautions can be taken to prevent wildfire’s can be developed. Declarations Author Contribution Conceptualization, R.S.P. (R. Shanmuga Priya) and K.V. (K. Vani); methodology, R.S.P. ;.; validation, R.S.P. and K.V.; formal analysis, R.S.P. and K.V.; resources and curation, R.S.P; writing—original draft preparation, R.S.P.; writing—review and editing, R.S.P. and K.V.; visualization, R.S.P.; supervision, K.V.; project administration, R.S.P. and K.V. All authors have read and agreed to the published version of the manuscript. Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on request. Disclosure Statement No potential conflict of interest was reported by the authors References Negassa, M.D., Mallie, D.T., & Gemeda, D.O. (2020). Forest cover change detection using Geographic Information Systems and remote sensing techniques: a spatio-temporal study on Komto Protected forest priority area, East Wollega Zone, Ethiopia. Environmental SystemsResearch,9(1), https://doi.org/10.1186/s40068-020-0163-z. Liu, T., Yang, L., & Lunga, D. (2021). Change detection using deep learning approach with object-based image analysis. Remote Sensing of Environment, 256, 112308. https://doi.org/10.1016/j.rse.2021.112308. Wang, Z., Wei, C., Liu, X., Zhu, L., Yang, Q., Wang, Q., Zhang, Q., & Meng, Y. (2022). Object-based change detection for vegetation disturbance and recovery using Landsat time series. GIScience & Remote Sensing, 59(1), 1706-1721. https://doi.org/10.1080/15481603.2022.21298 70. Joseph Hughes, M., Douglas Kaylor, S., & Hayes, D. J. (2017). Patch-Based Forest Change Detection from Landsat Time Series. Forests, 8(5), 1–22. https://doi.org/10.3390/f8050166. Hussain, M., D. Chen, A. Cheng, H. Wei, and D. Stanley. (2013). “Change Detection from Remotely Sensed Images: From Pixel-Based to Object-Based Approaches.” ISPRS Journal of Photogrammetry and Remote Sensing 80: 91–106. doi:10. 1016/j.isprsjprs.2013.03.006. Zhu, Z. (2017). “Change Detection Using Landsat Time Series: A Review of Frequencies, Preprocessing, Algorithms, and Applications.” ISPRS Journal of Photogrammetry and Remote Sensing130:370–384. doi:10.1016/j.isprsjprs.2017.06.013. Chen, X., Vierling, L., Deering, D. (2005). A simple and effective radiometric correction method to improve landscape change detection across sensors and across time. Remote Sensing of Environment,98(1),63-79. https://doi.org/10.1016/j.rse.2005.05.021. Tian, Y., Wu, Z., Li, M., Wang, B., & Zhang, X. (2022). Forest Fire Spread Monitoring and Vegetation Dynamics Detection Based on Multi-Source Remote Sensing Images. Remote Sensing,14,4431. https://doi.org/10.3390/rs14184431. Meera Gandhi, G., Parthiban, S., Nagaraj Thummalu& Christy, A. (2015) Ndvi: VegetationChange Detection Using Remote Sensing andGis – A Case Study of Vellore District.Procedia Computer Science, 57 (pp 1199-1210). Chen, X., Vogelmann, J. E., Rollins, M., Ohlen, D., Key, C. H., Yang, L., Huang, C., & Shi, H. (2011). Detecting Post-Fire Burn Severity and Vegetation Recovery Using Multitemporal Remote Sensing Spectral Indices and Field- Collected Composite Burn Index Data in a Ponderosa Pine Forest. International Journal of Remote Sensing, 32(23), 7905–7927. https://doi.org/10.1080/01431161.2010.524678. Lacouture, D.L., Broadbent, E.N., Crandall, R.M. (2020). Detecting Vegetation Recovery after Fire in a Fire-Frequented Habitat Using Normalized Difference Vegetation Index (NDVI).Forests,11,749. https://doi.org/10.3390/f11070749. Sharma, R., Rimal, B., Baral, H., Nehren, U., Paudyal, K., Sharma, S., Rijal, S., Ranpal, S., Acharya, R.P., Alenazy, A.A., et al. (2019). Impact of Land Cover Change on Ecosystem Services in a Tropical Forested Landscape. Resources,8,18. https://doi.org/10.3390/resources8010018. Kristensen, P. J., Gould, K., & Thomsen, J. B. (1997). Approaches to field-based monitoring and evaluation implemented by Conservation International. In Proceedings and Papers of the International Workshop on Biodiversity Monitoring, Brazilian Institute for Environment and Renewable Resources, Pirenopolis, Brazil, pp. 129–144. Bousquet, E., Mialon, A., Rodriguez-Fernandez, N., Mermoz, S., & Kerr, Y. (2022). Monitoring post-fire recovery of various vegetation biomes using multi-wavelength satellite remote sensing. Biogeosciences, 19(13), 3317-3336. doi: 10.5194/bg-19-3317-2022. Nolan, R. H., Collins, L., Leigh, A., Ooi, M. K. J., Curran, T. J., Fairman, T. A., et al. (2021). Limits to post-fire vegetation recovery under climate change. Plant, Cell Environ. 44 (11), 3471–3489. doi:10.1111/pce.14176. Jarchow, C.J., Didan, K., Barreto-Muñoz, A., Nagler, P.L., & Glenn, E.P. (2018). Application and Comparison of the MODIS- Derived Enhanced Vegetation Index to VIIRS, Landsat 5 TM and Landsat 8 OLI Platforms: A Case Study in the Arid Colorado River Delta, Mexico.Sensors,18, 1546. https://doi.org/10.3390/s18051546. Kim, Y., Jeong, M.-H., Youm, M., Kim, J., & Kim, J. (2021). Recovery of Forest Vegetation in a Burnt Area in the Republic of Korea: A Perspective Based on Sentinel-2 Data. Applied Sciences,11,2570. https://doi.org/10.3390/app11062570. Úbeda, X., L.R. Outeiro, and M. Sala. (2006). Vegetation regrowth after a differential intensity forest fire in a Mediterranean environment, northeast Spain. Land Degradation and Development 17 (4): 429–440. https://doi.org/10.1002/ldr.748. Vasques, A., Baudena, M., Vallejo, V.R., et al. (2023). Post-fire Regeneration Traits of Understorey Shrub Species Modulate Successional Responses to High Severity Fire in Mediterranean Pine Forests. Ecosystems, 26, 146–160. https://doi.org/10.1007/s10021-022-00750-z. Yumoto, S., Kitsukawa, T., Moro, A., et al. (2023). Anomaly detection from images in pipes using GAN. Robomech Journal, 10, 9. https://doi.org/10.1186/s40648-023-00246-y. Benedict, T.D., Brown, J.F., Boyte, S.P., Howard, D.M., Fuchs, B.A., Wardlow, B.D., Tadesse, T., Evenson, K.A. (2021). Exploring VIIRS Continuity with MODIS in an Expedited Capability for Monitoring Drought-Related Vegetation Conditions. Remote Sensing, 13, 1210. https://doi.org/10.3390/rs13061210. Skakun, S., Justice, C.O., Vermote, E., Roger, J.C. (2018). Transitioning from MODIS to VIIRS: an analysis of inter-consistency of NDVI data sets for agricultural monitoring. International Journal ofRemoteSensing,39(4),971-992. https://doi.org/10.1080/01431161.2017.139597 0. Natalie, R. W., and M. Norman Laura. (2018). ”Analysis of Vegetation Recovery Surrounding a Restored Wetland Using the Normalized Difference Infrared Index (NDII) and Normalized Difference Vegetation Index (NDVI).” International Journal of Remote Sensing 39 (10): 3243–3274. Aksu Prefectural Committee of historical records. Aksu almanac. doi:10.1080/01431161.2018.1437297. Xiao, J., & Moody, A. (2005). A comparison of methods for estimating fractional green vegetation cover within a desert-to-upland transition zone in Central New Mexico, USA. Remote Sensing of Environment, 98(2–3), 237–250. https://doi.org/10.1016/j.rse.2005.07.011 De Petris, S., Sarvia, F., & Borgogno-Mondino, E. (2021). Multi-temporal mapping of flood damage to crops using Sentinel-1 imagery: A case study of the Sesia River (October 2020). Remote Sensing Letters, 12(5), 459–469. https://doi.org/10.1080/2150704X.2021.189026 2 Curran, P. J. (1981). Multispectral remote sensing for esti-mating biomass and productivity. Plants and the Daylight Spectrum. Goward, S. N. (1989). Satellite bioclimatology. JournalofClimate,2(7),710–720. https://doi.org/10.1175/1520- 0442(1989)0022.0.CO;2 Malingreau, J. P. (1989). The vegetation index and the study of vegetation dynamics. In Applications of remote sensing to agrometeorology (pp. 285–303). Springer Netherlands. https://doi.org/10.1007/978-94- 009-2235-8_12 Lentile L.B., Holden Z.A., Smith A.M.S., Falkowski M.J., Hudak A.T., Morgan P., Lewis S.A., Gessler P.E., Benson N.C. (2006) - Remote sensing techniques to assess active fire characteristics and post-fire effects. International Journal of Wildland Fire, 15: 319- 345. doi: https://doi.org/10.1071/WF05097. Miller J.D., Thode A.E. (2007) - Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sensing of Environment, 109: 66-80. doi: https://doi.org/10.1016/j.rse.2006.12.006. Bonan G.B. (1989) - Environmental factors and ecological processes controlling vegetation patterns in boreal forests. Landscape Ecology, 3:111-130.doi: http://dx.doi.org/10.1007/BF00131174. Kennedy R.E., Yang Z.Q., Cohen W.B., Pfaff E., Braaten J., Nelson P. (2012) - Spatial and temporal patterns of forest disturbance and regrowth within the area of the Northwest Forest Plan. Remote Sensing of Environment, 122: 117-133. doi: http://dx.doi.org/10.1016/j.rse.2011.09.024. Stueve K.M., Cerney D.L., Rochefort R.M., Kurth L.L. (2009) - Post-fire tree establishment patterns at the alpine treeline ecotone: Mount Rainier National Park, Washington, USA. Journal of Vegetation Science, 20: 107- 120. doi: http://dx.doi.org/10.1111/j.1654- 1103.2009.05437.x. Otoda T., Doi T., Sakamoto K., Hirobe M., Nachin B., Yoshikawa K. (2013) - Frequent fires may alter the future composition of the boreal forest in northern Mongolia. Journal of Forest Research,18:246-255.doi: http://dx.doi.org/10.1007/s10310-012-0345-2. Townshend J.R., Masek J.G., Huang C.Q., Vermote E.F., Gao F., Channan S., Sexton J.O., Feng M., Narasimhan R., Kim D., Song K., Song D.X., Song X.P., Noojipady P., Tan B., Hansen M.C., Li M.X., Wolfe R.E. (2012) - Global characterization and monitoring of forest cover using Landsat data: opportunities and challenges. International Journal of Digital Earth,5:373-397.doi: http://dx.doi.org/10.1080/17538947.2012.71319 0. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-3890182","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":268950451,"identity":"40cbb4ca-6579-4113-b66a-0ad5aa2d8ee4","order_by":0,"name":"R. Shanmuga Priya Rajendran","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABFklEQVRIiWNgGAWjYDACCSjN2ABhy/GDeAkFJGgxlmwAaTEgQguMnbjhAIiJR4v87OZnj2623ZNnbu89eLui5g7j5vOrEz88MGCQ5xc7gFWLwZ1j5sa5bcWGjT3nki3PHHvGbHbj7WYJoMMMZ85OwK5FIsFMOrctgbFxRo6ZZAPbYTazG2c3gLQkGNzGrkV+Rvo3kBb7xvlvgFr+HeYxnnF28w98Whhu5IBtSWycwWMm2dh2WMKAv3cbXlsMbuSUSeecS0hu7MkxtmzsO2wgcYN3m0WCgQROvwAdtk06pyzBdmP7GcObDd8O1/f3n91880eFjTy/NA6HwYBhA4wlAVYpgUshknVwFv8BwqpHwSgYBaNgRAEAd0JjvRXa0ngAAAAASUVORK5CYII=","orcid":"","institution":"Anna University, Chennai","correspondingAuthor":true,"prefix":"","firstName":"R.","middleName":"Shanmuga Priya","lastName":"Rajendran","suffix":""},{"id":268950452,"identity":"52eae6d4-1cf3-438e-9902-2bb852ca2657","order_by":1,"name":"K. Vani K","email":"","orcid":"","institution":"Anna University, Chennai","correspondingAuthor":false,"prefix":"","firstName":"K.","middleName":"Vani","lastName":"K","suffix":""}],"badges":[],"createdAt":"2024-01-23 06:59:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3890182/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3890182/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":50177122,"identity":"c2a3733a-bc80-4efd-8742-4081035503c8","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":120913,"visible":true,"origin":"","legend":"\u003cp\u003eThe Architecture Diagram for the proposed Approach\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/265927f365cd9d8b48926632.png"},{"id":50177308,"identity":"abbaee10-610b-4a7b-ae1e-fcbed93ef2a9","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":107917,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of Vegetation Change Detection\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/0a5502465368089bf57c5c27.png"},{"id":50177125,"identity":"c6ec2249-59fa-4d4e-b662-ccd4142e5576","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":448174,"visible":true,"origin":"","legend":"\u003cp\u003eNDVI index class range ranging -1 to +1\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/7a1b3639b59052acfc878001.png"},{"id":50177716,"identity":"6898a4b6-7747-496a-a588-9330db05bf69","added_by":"auto","created_at":"2024-01-25 17:07:02","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":285574,"visible":true,"origin":"","legend":"\u003cp\u003eClasses generated from the NDVI index\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/e1b9a43b90905243a2e65033.png"},{"id":50177711,"identity":"3bd6776f-e75e-4abb-a307-6664145a9c24","added_by":"auto","created_at":"2024-01-25 17:07:02","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":26873,"visible":true,"origin":"","legend":"\u003cp\u003eBlock Diagram of DEC\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/7bba72769f9884ec2d1b9cc0.png"},{"id":50177142,"identity":"ba6adf8b-6963-4d37-85ed-0d55df325c94","added_by":"auto","created_at":"2024-01-25 16:51:03","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":166189,"visible":true,"origin":"","legend":"\u003cp\u003eWorkflow of Vegetation Regrowth Assessment\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/d0897df32fa46792f5506fff.png"},{"id":50177310,"identity":"9fbec78a-3c30-4a15-82c1-9760106634c3","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":87512,"visible":true,"origin":"","legend":"\u003cp\u003eThe proposed Ensemble learning approach\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/84acc47961de6161b5ba5868.png"},{"id":50177128,"identity":"14a456ff-f3f7-4aa0-990d-5d9c17ffac01","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":148447,"visible":true,"origin":"","legend":"\u003cp\u003eFramework of Vegetation Recovery Mapping\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/c6aa7e240aa40b513da4a659.png"},{"id":50178425,"identity":"fa0c5d3f-8874-4323-82a6-7b36a6304e12","added_by":"auto","created_at":"2024-01-25 17:23:02","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":265494,"visible":true,"origin":"","legend":"\u003cp\u003eNetwork Architecture of AdaptiGAN\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/d4a466fd4d00130e0d4d751b.png"},{"id":50177134,"identity":"ea4739d4-74c6-4eee-966a-937bfd80b100","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":736857,"visible":true,"origin":"","legend":"\u003cp\u003eVisual comparison of various change detection models. From left to right: original pre- fire, original post-fire, STA Net, Bi-attention SFA, change map of DEC and DEC\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/6a03b4998ddea804431611dd.png"},{"id":50177883,"identity":"9358949e-d516-44ae-ad88-43ff8a95e3f5","added_by":"auto","created_at":"2024-01-25 17:15:02","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":471137,"visible":true,"origin":"","legend":"\u003cp\u003eBrowning and greening fraction of EVI trend for Moore and Florida creek\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/0dc09fc3e48d16c80be935d8.png"},{"id":50177317,"identity":"d541cc68-2a13-47e5-82ef-913b40bb2d72","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":407742,"visible":true,"origin":"","legend":"\u003cp\u003eBrowning and greening fraction of EVI trend for Tosher and Tonalite creek\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/1a11404b259b03bef2dd6f0a.png"},{"id":50177319,"identity":"acd61a93-c426-4b76-ab7f-1b6f3f205694","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":650839,"visible":true,"origin":"","legend":"\u003cp\u003eResults from QGIS application. NDVI images for different regions.\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/991f82d8430619438f60dd76.png"},{"id":50177135,"identity":"5d04388c-bd1d-45e7-803d-8f6a0506bda5","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":441096,"visible":true,"origin":"","legend":"\u003cp\u003eResults from streamlit application. Left to right: Possibility of regrowth and Not-possibility of regrowth\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/6b581657e7aa70df2edea4bd.png"},{"id":50177885,"identity":"4ef50530-2fdc-474a-bf6b-9e0dc981d2df","added_by":"auto","created_at":"2024-01-25 17:15:02","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":384949,"visible":true,"origin":"","legend":"\u003cp\u003eVisual comparison of various Recovery Mapping models for Amazon Region. From left to right: Original post-fire, DNN, LSTM, CycleGAN and AdaptiGAN\u003c/p\u003e","description":"","filename":"15.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/6ba12ae1939db2f2c76fdc2a.png"},{"id":50177717,"identity":"a3cc7efb-7e5c-44ab-a7c4-f3ec3c7abf52","added_by":"auto","created_at":"2024-01-25 17:07:02","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":293924,"visible":true,"origin":"","legend":"\u003cp\u003eVisual comparison of various Recovery Mapping models for Knysna Region. From left to right: Original post-fire, DNN, LSTM, CycleGAN and AdaptiGAN\u003c/p\u003e","description":"","filename":"16.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/67aaa36c40e8959ab3207949.png"},{"id":50177313,"identity":"5e749083-8465-4f06-a85e-041b9e3057b8","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":17,"title":"Figure 17","display":"","copyAsset":false,"role":"figure","size":295423,"visible":true,"origin":"","legend":"\u003cp\u003eVisual comparison of various Recovery Mapping models for Alaska Region. From left to right: Original post-fire, DNN, LSTM, CycleGAN and AdaptiGAN\u003c/p\u003e","description":"","filename":"17.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/554694b27348b3005b7c16fe.png"},{"id":50177321,"identity":"5c3bb52d-748a-471d-9dd3-a85fbbb78d28","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":18,"title":"Figure 18","display":"","copyAsset":false,"role":"figure","size":114451,"visible":true,"origin":"","legend":"\u003cp\u003eLoss Plots for pre-trained models and our proposed AdaptiGAN model\u003c/p\u003e","description":"","filename":"18.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/794ef5c0f728005358ad901c.png"},{"id":50177712,"identity":"65e06863-3ec1-4306-96d2-2b4ecf4ce581","added_by":"auto","created_at":"2024-01-25 17:07:02","extension":"png","order_by":19,"title":"Figure 19","display":"","copyAsset":false,"role":"figure","size":509194,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance Analysis of pre-trained models with proposed AdaptiGAN model.\u003c/p\u003e","description":"","filename":"19.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/590c65616e61a2c9e6946eeb.png"},{"id":50177311,"identity":"5e59a565-4c6e-40bf-ad7e-8adc850f609e","added_by":"auto","created_at":"2024-01-25 16:59:02","extension":"png","order_by":20,"title":"Figure 20","display":"","copyAsset":false,"role":"figure","size":202788,"visible":true,"origin":"","legend":"\u003cp\u003eTest of Homoscedasticity Plot of pre-trained models with proposed AdaptiGAN model.\u003c/p\u003e","description":"","filename":"20.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/2a517dcba16a99cb8329eb56.png"},{"id":50177140,"identity":"0a4a0bac-fdbf-47d5-a497-accc6b8355fc","added_by":"auto","created_at":"2024-01-25 16:51:02","extension":"png","order_by":21,"title":"Figure 21","display":"","copyAsset":false,"role":"figure","size":37308,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation Metrics used for comparing the performance of AdaptiGAN model with other models.\u003c/p\u003e","description":"","filename":"21.png","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/f9fbd05b21ff7823a02a2e8f.png"},{"id":50969606,"identity":"3d51aed4-5925-4e33-89dc-3c07dad3bb15","added_by":"auto","created_at":"2024-02-11 16:22:59","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5943925,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3890182/v1/12d103a0-6202-490c-8a87-5151cac039be.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Vegetation Change Detection and Recovery Assessment on Post-fire Satellite Imagery using Deep Learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWildfires, a significant natural disaster, can cause changes to compositions and leads to ecological imbalance along with making major modifications to an ecosystem of a region. The identification of land cover changes and keeping a watch on it is necessary for managing a balanced ecosystem and economical studies at the various levels (Sharma et al., 2019). For continuous environmental monitoring and to thoroughly examine urgent environmental issues including degradation of natural resources, loss in biodiversity, and forest cover loss, forest cover change identification is very important (Negassa et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). For ecological research as well as forest monitoring and management, it\u0026rsquo;s essential for comprehending how fire affects forests and to identify the patterns of time and space in recovery of forests after fire. (Bonan, 1989; Stueve et al., 2009; Kennedy et al., 2012; Otoda et al., 2013).\u003c/p\u003e \u003cp\u003eTraditional approaches including ground surveys are costly and requires a lot of time. This can be attributed to various factors which makes the task of understanding the changes made to the vegetation practically infeasible. To overcome this making use of remote sensing, one can monitor changes across broad areas in an economical and effective manner (Townshend et al., 2012). Change detection can provide crucial data for handling the disaster, policy creation, effective land cover and forest management (Tian et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Unprecedented fire seasons worldwide prompt a need to understand into how vegetation respond to evolving fire patterns. Usage of Vegetation Indices for assessing and analyzing the regrowth of vegetation in a region after any major disaster is a technique regularly used by researchers for the purpose. We can see that data collected at three different post-fire periods revealed distinctions between vegetation in a control forest and regenerating regions (\u0026Uacute;beda et al., 2006). Due to changes in land use and climate, wildfires have become more frequent and severe and many tend to last long which cause a lot of damage to vegetation resulting in a need for quantifying vegetation recovery in fire-affected regions so that appropriate steps can be taken to speed up this process. Assessing vegetation recovery pose challenges because of extended duration required to capture unique traits of an ecosystem. The use of time-series based remote sensing data allows for assessments made prior to, during and following fire, enhances accuracy.\u003c/p\u003e\n\u003ch3\u003eRelated works:\u003c/h3\u003e\n\u003cp\u003eAccurate and effective methods for collecting quantitative forest change data from remotely sensed photos, developed and tested in a monitoring-based study, are essential for the ongoing monitoring of forest clearing in the MBR (Kristensen et al., 1997). This change detection analysis is a useful method for characterizing the alterations noticed throughout every category of land utilization. The classification of land use would be properly enhanced by high resolution satellite data. For feature extraction, the normalized difference vegetation index technique has been utilized with a range of threshold values. NDVI approach yields more effective outcomes for vegetation with different types as well as dispersed vegetation from a multispectral remote sensing picture (Meera et al., 2015). Satellite data provides unparalleled effectiveness in tracking and measuring extensive changes in landscape across time. But radiometric consistency across multi- temporal imagery must be ensured to get clear detection of changes in terrain (Chen et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2005\u003c/span\u003e). There is a need for Integrating both deep learning techniques and Object-based image analysis techniques for change detection which can make this process to be lot more efficient and accurate (Liu et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).Various approaches have been utilized for changed area detection which includes Patch-Based detection (Joesph et al., 2017), Pixel- Based to Object-Based approaches (Hussain et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), Landsat time series data (Zhu, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Accuracy of various object-based detection techniques is dependent upon the quality of image segmentation and many approaches tend to struggle from the under-segmentation error (Wang et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This can be due to the sparse nature of the features present in images.\u003c/p\u003e \u003cp\u003eEnhancing our understanding of how ecosystem imbalance and changes in order of fires, due to climate, can influence post-fire development speed and adaptability is crucial for improving predictions of community responses to fire in the context of climate change (Nolan et al., 2021). When vegetation that is photosynthetic is destroyed by fire, reflectance in the visible to near-infrared range drastically decreases, while reflectance in the short and middle infrared range increases because of increased ash [Lentile et al., 2006; Miller and Thode, 2007]. Due to their ability to absorb residual approximations they can provide a standardized representation of vegetation or something close to it which improves monitoring to a larger extent (Curran, 1981; Goward, 1989; Malingreau, 1989). Vegetation Indices are very useful and are compliant when the region for monitoring is huge (Xiao \u0026amp; Moody, 2005; De Petris et al., 2021). Since satellite data has become accessible daily, it becomes feasible to efficiently track vegetation regrowth in more granular manner using the Normalized Difference Vegetation Index (NDVI) (Lacouture et al., 2020). Studies indicate that due to recurrent fires in a region replacement of a vegetation species by another kind of species takes place and this process can be quite prolonged in nature owing to the nature of both the species (Vasques et al., 2023). This should be taken into consideration while evaluating the possibility of vegetation regrowth, which can be attributed to various factors like soil ph, nitrogen content etc. VIIRS proved to be a good way for estimating spectral indices, over Landsat, given the decommissioning of the MODIS satellites (Jarchow et al., 2018). While previous studies have commonly employed NBR with NDVI for extracting regions of plant regeneration, these approaches are susceptible to atmospheric effects and soil brightness (Kim et al., 2021).\u003c/p\u003e\u003cp\u003eSpectral indices, including NDVI, EVI and distinction in various indices spanning the years before and after fire, can be calculated and analyzed (Chen et al., 2011). NDVI proved more effective in tracking vegetation changes (Natalie et al., 2018). To analyze post-fire vegetation recovery, usage of multi wavelength satellite data is an approach which utilized various remotely sensed parameters including visible\u0026ndash;infrared vegetation indices (Bousquet et al., 2022). Vegetation indices credibility for effective vegetation monitoring can be further attributed to the fact that it is used for drought monitoring (Benedict et al., 2021). Given the various applications utilizing NDVI derived from MODIS and VIIRS based products we can see that MODIS and VIIRS NDVI data are interchangeable for applications with an uncertainty ranging from 0.02 to 0.05, depending on the scale of spatial aggregation, typically consistent with the individual datasets uncertainty making data provided by the VIIRS- based products especially the NDVI data good enough to perform monitoring (Skakun et al., 2018).\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eEven though the usage of vegetation indices can provide an insight about vegetation recovery in a region, concatenating this along with an unsupervised learning approach is something which is not explored a lot before and can provide more accurate information about that region.\u003c/p\u003e \u003cp\u003eIn summary, we can come to an understanding that Sparse Autoencoder performs better than simple Autoencoder to extract features which can greatly improve the change detection process when it is used along with Deep Embedded Clustering, typically used for segmentation purposes due to the better performance of the algorithm and its relatively simple implementation. Performing vegetation inference through usage of Sen\u0026rsquo;s slope based on EVI trend analysis along with ensemble learning can be used to precisely identify whether there is a possibility for vegetation regrowth. GAN\u0026rsquo;s are of various types and this is dependent on the task they tend to perform (Yumoto et al., 2023), and taking this as inspiration we propose a deep learning based unsupervised adversarial network. AdaptiGAN which is trained on the EVIIRS NDVI composites obtained after several preprocessing steps, can provide a more efficient and accurate description about the level of vegetation recovery in the fire-affected regions.\u003c/p\u003e"},{"header":"Data and Methodology","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eData and preprocessing\u003c/h2\u003e \u003cp\u003eFor performing vegetation change detection analysis after fire, a dataset comprising of pre-fire and post-fire satellite images was constructed. This dataset comprised of 3600 pre- fire and post-fire Normalized Difference Vegetation Index (NDVI) images of the EROS Moderate Resolution Imaging Spectroradiometer (eMODIS) collection which is obtained from the from the U.S. Geological Survey (USGS) website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://earthexplorer.usgs.gov/\u003c/span\u003e\u003cspan address=\"https://earthexplorer.usgs.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) for regions which were affected by fire and the search criteria for this was based upon the time periods for pre-fire and post-fire images. Appropriate corrections were made to ensure consistency and accuracy. The QGIS Software is used for the calculation of NDVI index and splitting of region into several classes of vegetation based on index value of each pixel of multi spectral image.\u003c/p\u003e \u003cp\u003eFor vegetation regrowth assessment remote sensing data is collected for each year starting from 2000 to 2022. The MODIS 250m/pixel 16-day composite vegetation indices dataset from the World Database on Protected Areas (WDPA) dataset, is used for this purpose. From the obtained composites an image-based dataset was constructed. Additionally, for predicting whether vegetation regrowth is possible in a region involves collecting soil data from the soil grid database this data consists of columns including Location Co-ordinates, soil ph, soil nitrogen content, organic carbon content, bulk density and soil groups. All these attributes are then concatenated and represented in the form of a single dataset.\u003c/p\u003e\u003cp\u003eThe dataset constructed for the purpose of vegetation recovery was downloaded from the USGS website mentioned above. It consisted of 1600 pre-fire images from three main region\u0026rsquo;s which were more suitable for our study namely: Amazon rainforest of Brazil with location co- ordinates 3.4653\u0026deg; S, 62.2159\u0026deg; W, Knysna region of South Africa with location co-ordinates 34.0351\u0026deg; S, 23.0465\u0026deg; E and Alaska with location co-ordinates 63.5888\u0026deg; N, 154.4931\u0026deg; W. The construction of dataset involved downloading Suomi NPP satellite\u0026rsquo;s EROS Visible Infrared Imaging Radiometer Suite (eVIIRS) data then loading this in QGIS software followed by creating an output layer which contains the NDVI values obtained by using the raster calculator present in the software. This layer is then exported to .jpeg format used for model training and evaluation purposes. This preprocessing step is performed for all the collected data and the obtained .jpeg file is added to the dataset.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMethodology\u003c/h3\u003e\n\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eWildfires cause lots of damage to ecology and economy, vegetation is affected by sudden increase in temperature which can lead to destruction of trees, shrubs and herbs. Loss of habitat is another issue caused by fire, change in vegetation composition is possible as fire-adapted species tend to dominate over others that are less tolerant to fire, there is also a huge question mark over the vegetation regrowth possibilities in regions particularly affected by huge fires and analyzing recovery patterns after fire is key to understanding the effect of fire and what all steps can be taken to improve the recovery if at all the recovery is less. So for this we proposed a integrated framework involving: Vegetation change detection by making use of sparse Autoencoders and Deep Embedded Clustering (DEC) model, an unsupervised learning technique through which change map showing the difference between pre and post fire images is obtained which is then embedded with the pre-fire image to show the change\u0026rsquo;s caused by fire; Vegetation Regrowth Assessment involves usage of Enhanced Vegetation Index (EVI) to get a time series representation of vegetation over years through browning and greening fraction and additionally performed through the use of AdaptiGAN, an unsupervised deep learning model, which takes preprocessed eVIIRS NDVI image as input and then provides the corresponding recovery map for that image. The architecture diagram for the proposed approach can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eChange Detection using Deep Embedded Clustering (DEC)\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eComprehensive assessment of long-term trends in vegetation change at the field scale is required for resource management and ecological assessment. Remote sensing data have been employed widely as greatest asset for change detection. The common approaches deployed for change detection include post-classification comparison, principal component analysis (PCA), and image differencing. New methods are required to effectively use the more complex and diversified remotely sensed data that is anticipated to become so in the near future via satellite and airborne sensors, which is still an active area of research.\u003c/p\u003e \u003cp\u003eVegetation change detection provides an analysis about difference between satellite images which were obtained before and after fire. Dataset comprises 3600 pre \u0026amp; post-fire images obtained from eMODIS NDVI v6. The QGIS Software is used for the calculation of NDVI index and splitting of region into several classes of vegetation based on index value of each pixel of multi spectral image. The change detection process uses Sparse Autoencoder for extracting required features and then this representation is fed as input to the DEC model, an unsupervised learning technique used to iteratively group features, and ensuing assignments are used as supervision to update network weights. The model acquires feature representations through successive iterations by using labeled and unlabeled data points and finding target distributions from prediction alternatively. The distinction between images obtained before and after the fire pertaining to forest region in Jefferson, California is used to create the change map. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e represents workflow of vegetation change detection.\u003c/p\u003e \u003cp\u003e\u003cem\u003eSplitting the Classes\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe QGIS Software has been used to calculate the NDVI index and produce five different classes ranging from low vegetation to very high vegetation. The Figure 3 below represents the NDVI index value ranging from -1 to +1.The Figure 4 represents the set of classes generated from NDVI. Table 1 shows the Vegetation type and its associated NDVI Index.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSparse\u0026nbsp;Autoencoder\u003c/em\u003e:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAn autoencoder represents a particular kind of Artificial Neural Network(ANN) typically used for unsupervised learning. Autoencoder that attains bottleneck information with an added restriction on sparsity is called as sparse autoencoder. The loss function is designed to push activations inside a layer. L1 regularization or Kullback-Leibler (KL) divergence between appropriate distribution and anticipated mean neuron activation, such that sparsity constraint can be applied, in other words the autoencoder is not only minimizing the difference between the input and the reconstructed output but also encouraging sparsity in the activations of the hidden layer.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u0026nbsp;\u003c/strong\u003eVegetation and their corresponding NDVI ranges\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eVegetation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eNDVI\u0026nbsp;Index\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003eNo\u0026nbsp;Vegetation\u0026nbsp;Area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e-1\u0026nbsp;to 0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003eBare\u0026nbsp;Area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e0 to 0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003eLow\u0026nbsp;vegetation\u0026nbsp;Area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e0.1 to 0.25\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003eModerate\u0026nbsp;vegetation\u0026nbsp;Area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e0.24 to 0.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"59.6875%\" valign=\"top\"\u003e\n \u003cp\u003eHigh\u0026nbsp;vegetation\u0026nbsp;Area\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"40.3125%\" valign=\"top\"\u003e\n \u003cp\u003e0.4 to 1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003eL1 Regularization\u003c/h2\u003e \u003cp\u003eBy scaling the absolute value of the activation vector in layer h for observation i by a tuning parameter λ with its features x and x^ we may add a term to our loss function that penalizes it.\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$L\\left(x,\\widehat{x}\\right)+ \\lambda . \\sum _{i=1}^{n}i.‖a\\left(h\\right).i‖$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003eKL-Divergence:\u003c/h2\u003e \u003cp\u003eKL-divergence is a measurement of variation in distribution over probability between two samples. A sparse parameter ρ represents average activity of a neuron across a group of samples \u0026#120588;̂. Neurons are induced to fire for a subset of observations by limiting average activity of a neuron and differentiate j it over a group of samples. To contrast expected distribution to actual distributions over all the hidden layer nodes, we can define as a Bernoulli random variable distribution and use the KL divergence.\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$L\\left(x,\\widehat{x}\\right)+ \\sum _{j=1}^{n}jKL\\left(\\rho \\right||\\widehat{\\rho }j)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eDeep Embedded Clustering\u003c/h2\u003e \u003cp\u003eThe model DEC works in two phases:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eInitialize phase with deep sparse autoencoders\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eClustering, where they successively repeat calculating a supplementary target distribution and minimize the KL divergence associated with it. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e5\u003c/span\u003e represents the block diagram of Deep Embedded Clustering. It briefs workflow of model comprising Autoencoder and Clustering block. Initially a soft assignment is calculated between cluster centroids (classes of vegetation) and embedded points. Then t-distribution computes similarity index between embedded point and centroid t-distribution is used as a non-parametric representation to identify similarity index between embedded points. By employing an auxiliary target distribution and learning from recent high confidence assignments, one can upgrade deep mapping and improve the centroids of cluster. In particular, soft assignment is matched to target distribution to train the model and centroid of clusters.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe basic workflow of this change detection process is, images obtained before and after fire are given as inputs simultaneously to our DEC model. Sparse autoencoder used to extract features and pass it to clustering function where each of its similar features is aggregated as a cluster. It gives binary map of pre \u0026amp; post-fire image, by finding difference between the images obtained before and after the fire we get our change map as final output, which in return will be embedded with pre-fire image to produce the changes in vegetation of region.\u003c/p\u003e \u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" height=\"282\" width=\"382\"\u003e\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eEnsemble Learning\u003c/h2\u003e \u003cp\u003eVegetation regrowth assessment following a wildfire is an important aspect of post-fire monitoring and ecosystem recovery. Sen\u0026rsquo;s slope, also termed as the Sen\u0026rsquo;s estimator or the Sen\u0026rsquo;s method, is statistical technique used to assess trends or changes in time series data. While Sen\u0026rsquo;s slope is typically used for analyzing trends in various fields, such as hydrology and climate science, it can also be applied to assess vegetation regrowth following a wildfire. This involves obtaining time series data representing vegetation indices or other relevant vegetation metrics for the study area. This data should span multiple time periods, covering period before and after the wildfire event.\u003c/p\u003e \u003cp\u003eSens\u0026rsquo;s slope of EVI index is calculated. Based on the magnitude of slope, browning and greening fraction can be determined. The trend of EVI pattern over the years is analyzed and it is visualized as a graph. MODIS dataset has been acquired for different regions and their NDVI images are converted to RGB frames and aggregated to provide animation representing change in vegetation over the years. The regrowth possibility prediction involves collecting soil data from soil grid database which is then trained using ensemble learning to provide an outcome. Figure\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e6\u003c/span\u003e represents workflow of vegetation regrowth assessment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe working of Vegetation Regrowth Assessment using Sen\u0026rsquo;s slope can be described in the following steps:\u003c/p\u003e \u003cp\u003e \u003cstrong\u003eData Collection\u003c/strong\u003e \u003cp\u003eRemote sensing data is collected for a specific area over a period of time, such as satellite images from MODIS dataset.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003ePre-processing\u003c/strong\u003e \u003cp\u003ePre-processing the remote sensing data eliminates any noise or artifacts to convert it into a usable format. This may involve tasks such as cloud removal, atmospheric correction and radiometric calibration.\u003c/p\u003e \u003cp\u003e \u003cem\u003eVegetation Index Calculation\u003c/em\u003e: Vegetation index, such as Enhanced Vegetation Index (EVI), is calculated from pre-processed remote sensing data. \u003cem\u003eTime-Series Analysis\u003c/em\u003e: Vegetation index values for each time step are analyzed using Sen\u0026rsquo;s slope estimator to calculate the trend or slope of vegetation index over time. The result approximates the pace at which an area\u0026rsquo;s vegetation is growing again or decreasing.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eVisualization\u003c/strong\u003e \u003cp\u003eThe results of analysis can be visualized using various techniques, such as a time series plot of vegetation index or a map showing spatial distribution of the vegetation regrowth or decline.\u003c/p\u003e\u003cp\u003eVegetation Regrowth Assessment using Sen\u0026rsquo;s slope is an excellent technique for providing insights into natural regeneration steps of vegetation as it also allows to identify areas that may need interventions to support recovery in a region, after a disaster like wildfire.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePlotting trends based on EVI\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe following steps describe the process of Plotting trends based on EVI values calculated over a period of time:\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eThe World Database on Protected Areas (WDPA) dataset can be used to create MODIS 250m/pixel 16-day composite vegetation indices dataset.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eCreate a collection of images by adding images for every year between 2000 and 2002.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEvery one of the images has been computed to attain the highest EVI throughout every month of respective years.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eThis is an annual examination of the state of vegetation. In order to prepare for a linear- trend analysis, add the year as a band.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eDetermine each pixel\u0026rsquo;s Sen\u0026rsquo;s slope of highest summer EVI over time to estimate a linear trend.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eCompute and display the regression slope values as histograms.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWe can determine the browning and greening\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003efaction of the vegetation by measuring the slope\u0026rsquo;s value.\u003c/p\u003e \u003cp\u003eSen\u0026rsquo;s slope\u0026thinsp;=\u003cimg src=\"data:image/png;base64,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\"\u003e\u003c/p\u003e \u003cp\u003eThe dataset for soil properties has been acquired from the soil grid database. Stacking is an ensemble machine learning algorithm to combine results from multiple models. In the proposed work, we consider simple machine learning algorithms including Logistic Regression, SVM, Decision Tree, Random Forest and Na\u0026iuml;ve Bayes as weak models and their predictions are given to generalizer to provide the final outcome of regrowth possibility. The Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e7\u003c/span\u003e below shows the ensemble learning approach using stacking. So, the basic architecture of the utilized Stacking process is:\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eBase Models (Level 0):\u003c/h2\u003e \u003cp\u003eThis involves training diverse base models in this case Logistic Regression, SVM, Decision Tree, Random Forest and Na\u0026iuml;ve Bayes on the training data.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eMeta-Model (Level 1):\u003c/h2\u003e \u003cp\u003eA meta-model in our case the Generalizer is trained on the predictions from the base models. This model learns to combine the base model\u0026rsquo;s prediction to generate a final prediction.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eFinal Prediction (Output):\u003c/h2\u003e \u003cp\u003eBased on the soil properties as input the trained model makes predictions about regrowth possibility in a region.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eAdaptive Generative Adversarial Network (AdaptiGAN)\u003c/h2\u003e \u003cp\u003eVegetation Recovery Mapping is a crucial step which can provide valuable information about the ecological, environmental, and management aspects of post-fire landscapes. Traditional approaches which involve field surveys and ground truthing can turn out to be cumbersome and expensive at the same time providing a need for a more efficient approach which can speed-up this process. This involves making use of remote-sensing technologies such as satellite imagery combined with spectral indices like NDVI and deep learning techniques. Now making use of both remote sensing technologies and deep learning together can provide efficient and accurate results that can be used for future planning like resource allocation, habitat restoration in fire affected regions.\u003c/p\u003e\u003cp\u003eFor this purpose, the proposed approach is collecting satellite data from e-VIIRS products obtained from Suomi NPP satellites, of different fire affected regions which is then preprocessed using QGIS tool via which NDVI is calculated and an image like representation is obtained. This dataset serves as the training data for an unsupervised learning algorithm termed AdaptiGAN which can be adapted based on the kind of data we feed into it. Using this trained model, we can obtain recovery maps for different regions affected by fire for different time periods. Figure\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e8\u003c/span\u003e represents the framework for the Vegetation Recovery Mapping Approach.\u003c/p\u003e\u003cp\u003eThe AdaptiGAN is a neural network architecture, a Generative Adversarial Network (GAN), whose generator follows an Encoder-Decoder architecture making use of self-attention mechanisms for recording long-range dependencies which in turn will help in improving feature representation and the discriminator makes use of the PatchGAN architecture which helps in capturing fine-grained details of the input images. With self-attention mechanisms and normalization techniques in place, they are extremely useful in capturing domain- specific features which is very important in our case of generating recovery maps. The key components related to this architecture can be seen in more detail below and the network architecture can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig10\" class=\"InternalRef\"\u003e9\u003c/span\u003e.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eComponents in the AdaptiGAN architecture Generator:\u003c/h2\u003e \u003cp\u003eThis follows an encoder-decoder architecture. The encoder downsamples the input image to extract features, and the decoder upsamples these features to get the final output image. The downsample layers reduce spatial dimensions and increase the number of channels capturing hierarchical features while upsample layers increase spatial dimensions enabling the generation of a high-resolution image. A self-attention mechanism is introduced after the third downsampling layer to capture long-range dependencies and improve feature representations. \u003cem\u003eDiscriminator\u003c/em\u003e:\u003c/p\u003e \u003cp\u003eThe discriminator employs a PatchGAN architecture, where it classifies local patches of the input images as real or fake. This helps in capturing fine-grained details. Convolutional layers with leaky ReLU activations are used for feature extraction and discrimination. Batch normalization is applied to normalize the activations, aiding in the stability and training of the discriminator.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e\u003cem\u003eWeight Regularization\u003c/em\u003e:\u003c/h2\u003e \u003cp\u003eL2 weight regularization is applied to the convolutional layers of both the generator and discriminator. This helps to fend off overfitting and improves generalization of the architecture.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e\u003cem\u003eInstance Normalization\u003c/em\u003e:\u003c/h2\u003e \u003cp\u003eInstance normalization is applied in the discriminator after the second and third convolutional layers. It normalizes activations across channels and spatial dimensions independently for each sample.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003e\u003cem\u003eDropout\u003c/em\u003e:\u003c/h2\u003e \u003cp\u003eDropout is applied in the generator after the concatenation of feature maps during the decoding process. It helps regularize the network by randomly dropping a fraction of the units during training.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003e\u003cem\u003eActivation Functions\u003c/em\u003e:\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eReLU activation is used in various parts of both the generator and discriminator to introduce non- linearity while Leaky ReLU is used in the discriminator to allow a small, non-zero gradient when the input is negative.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003eSummation:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSummation layers are used in the generator to combine feature maps from different stages, aiding in the generation of detailed and realistic images.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003e\u003cem\u003eOutput Layer\u003c/em\u003e:\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eThe generator has a tanh-activated convolutional layer in the output to produce the final generated image.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"373\" height=\"493\"\u003e\u003c/p\u003e"},{"header":"Results and Discussion","content":"\u003cp\u003eTo verify the viability of our proposed approach, performance of the three modules was tested. There are two areas for the experimental equipment: one for testing and the other for training. In the training phase, deep learning model and machine learning model used for this research were trained in the Google Colaboratory which provided hosted jupyter notebook with python environment implemented on a server with AMD\u0026reg; 7000 Series Ryzen\u0026trade; 9 7950X CPU @(5.7 GHz) with 16GB memory, Radeon RX 7800 XT (GPU) with 16GB of memory.\u003c/p\u003e \u003cp\u003eFor testing, performance analysis was performed on the vegetation change detection module by using metrics like precision, recall. This was performed on the same Google Colab platform with python environment. The vegetation regrowth assessment module was then deployed using streamlit web application to display the results for the possibility of vegetation regrowth in a region, which provided an interface where user can enter Location details, soil data to predict regrowth possibility. This can be validated by cross referencing the same location in USGS earth explorer tool to check for vegetation. For the vegetation recovery mapping module metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE) were used .The server used for this testing environment is AMD\u0026reg; 7000 Series Ryzen\u0026trade; 9 7950X CPU @(5.7 GHz) with 8GB memory, Radeon RX 7800 XT (GPU) with 8GB of memory.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003eExperimental Results and Analysis for Vegetation Change Detection using Deep Embedded Clustering\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003eVegetation change detection to assess long term trends in vegetation change we propose DEC model, change detection is analyzed by finding difference between satellite-based imagery obtained before and after fire. Dataset comprises 3600 pre-fire and post-fire images obtained from eMODIS NDVI v6. Figure\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e10\u003c/span\u003e. qualitative comparisons result of DEC with other deep convolutional models.\u003c/p\u003e \u003cp\u003eVegetation change detection performance is evaluated based on Precision, Recall, F1 and Accuracy (ACC), in comparison with other trained models. The DEC model achieved an impressive accuracy of 96.17%. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e. comparison results between existing approaches such as STA Net, Bi- attention SFA, and our proposed model DEC. The deep embedded clustering is evaluated based on the loss metrics contributing to minimize the KL divergence. Proposed change detection model shows a loss of around 18.57 which is considerably lower than existing approaches. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Loss metrics compared between various existing models such as STA Net, Bi-attention SFA and proposed model DEC.\u003c/p\u003e \u003cp\u003eThe deep embedded clustering is compared with the various other change detection models and justified with minimal loss value after successful completion of epochs.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \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 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison results of various deep convolutional network models for wildfire prediction.\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=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSTA Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e81.25%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e82.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e85.68%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e87.15%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBi-attention SFA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e82.59%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.79%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e87.32%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e88.60%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e90.07%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.45%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93.72%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e96.17%\u003c/b\u003e\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\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLoss tabulation metrics for vegetation change detection\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLoss\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSTA Net\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e26.27\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBi-attention SFA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e26.77\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e18.57\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003eExperimental Results and Analysis for Vegetation Regrowth Assessment using Ensemble Learning\u003c/h2\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAfter a wildfire, recovery of an ecosystem and post- fire monitoring rely primarily on the regeneration of vegetation. Sen\u0026rsquo;s slope estimator is used to calculate trend of EVI pattern over the years and it is visualized as a graph. The browning and greening fraction is determined based on the magnitude of slope. Figures\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e11\u003c/span\u003e \u0026amp; \u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e12\u003c/span\u003e. EVI trend graph of Moore creek, Florida creek, Tosher creek and Tonalite creek along with its browning and greening fraction for \u0026lsquo; the respective area over the years. Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e4\u003c/span\u003e. Tabulation comparison of browning and greening fraction with its sq.km for moore, florida, tosher and tonalite creek.\u003c/p\u003e \u003cp\u003eThe dataset for vegetation regrowth assessment is collected from Google Earth Engine fetched from MODIS dataset. Sen\u0026rsquo;s slope is used for estimating the browning and greening fraction. It is used to analyze the trend of EVI pattern as well. The following sections below show the tabulated results and graphs of various other models and proposed system thereby adding a justification. Ensemble learning can be used to provide the output for regrowth. Figure\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e13\u003c/span\u003e shows the NDVI images obtained after processing them using QGIS application.\u003c/p\u003e\u003cp\u003eRegrowth possibility is predicted by collecting soil properties and trained them using ensemble learning to estimate its results. Performance of the system is tested using Streamlit platform, it is deployed to display results for the possibility of vegetation regrowth. User can enter the location coordinates (latitude and longitude), pH value, nitrogen value and soil group from where data are fetched and possibility of prediction is estimated. Figure\u0026nbsp;14 shows the results obtained from the streamlit application which predicts possibility and not-possibility of vegetation regrowth.\u003c/p\u003e\u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison results of browning and greening fraction using Sen\u0026rsquo;s slope\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=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eName\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBrowning fraction\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBrowning sq.km\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGreening fraction\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGreening sq.km\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMoore Creek\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.284\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.529\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.978\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.154\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFlorida Creek\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.106\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.362\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8950\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.066\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTosher Creek\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.723\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.945\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.365\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTonalite Creek\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.45\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e18.197\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.553\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e22.368\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec28\" class=\"Section2\"\u003e \u003ch2\u003eExperimental Results and Analysis for Vegetation Recovery Mapping using Adaptive Generative Adversarial Network (AdaptiGAN)\u003c/h2\u003e \u003cp\u003eVegetation Recovery Mapping is an important step in understanding the ecological, environmental, and managerial elements of post-fire environments. This module involved usage of vegetation indices like NDVI and AdaptiGAN a a deep learning based neural network framework to provide a recovery map absolute error, is a loss function which basically combines the best properties of both MSE and MAE. Its less sensitive to outliers than MSE and provides a smooth transition to MAE at zero error. Here, y is true value, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\widehat{y}\\)\u003c/span\u003e\u003c/span\u003e is predicted value, and \u0026#120575; can be a threshold that determines when to switch from quadratic to linear behavior.\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"344\" height=\"65\"\u003e\u003c/p\u003e \u003cp\u003ebased on the input image. Figure\u0026nbsp;\u003cspan refid=\"Fig15\" class=\"InternalRef\"\u003e15\u003c/span\u003e shows the qualitative comparisons results of AdaptiGAN with other trained models, obtained for the Amazon rainforest.\u003c/p\u003e \u003cp\u003eThe trained model\u0026rsquo;s performance was evaluated using evaluation metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Squared Logarithmic Error(MSLE), Root Mean Squared Error (RMSE), Huber loss. The obtained values are tabulated in the Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e5\u003c/span\u003e. Unlike the metrics like MSE, MAE, RMSE, MSLE which are used often for evaluation purposes the Huber loss, also known as the smooth\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eFigure \u003cspan refid=\"Fig16\" class=\"InternalRef\"\u003e16\u003c/span\u003e shows the qualitative comparison results of AdaptiGAN with other trained models, obtained for the Knysna Region. Similarly Fig.\u0026nbsp;\u003cspan refid=\"Fig17\" class=\"InternalRef\"\u003e17\u003c/span\u003e shows the comparison results obtained for our AdaptiGAN model with other trained models, obtained for Alaska Region. The Fig.\u0026nbsp;18 below provides a visualization about the loss variation obtained for different models along with our proposed AdaptiGAN model.\u003c/p\u003e\u003cp\u003eThe Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e5\u003c/span\u003e provides the comparison for various models with our proposed AdaptiGAN model with respect to the chosen performance metrics. Figure\u0026nbsp;\u003cspan refid=\"Fig18\" class=\"InternalRef\"\u003e19\u003c/span\u003e provides us the performance-analysis plot for the pre-trained models along with our AdaptiGAN Model. Figure\u0026nbsp;\u003cspan refid=\"Fig19\" class=\"InternalRef\"\u003e20\u003c/span\u003e illustrates the \u0026ldquo;Test for Homoscedasticity\u0026rdquo; plot for pretrained models compared with our AdaptiGAN model, homoscedasticity can be good for analysis as it provides information about whether the model fully captures the underlying patterns in the data. We can see that AdaptiGAN doesn\u0026rsquo;t follow a clear trend while other models tend to follow suggesting there are heteroscedastic in nature which can lead to biased estimates. Figure\u0026nbsp;\u003cspan refid=\"Fig20\" class=\"InternalRef\"\u003e21\u003c/span\u003e shows the plots for evaluation metrics chosen to quantify the performance of our model.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison between trained models and our proposed AdaptiGAN model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMAE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMSLE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRMSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHuber Loss\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.914\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e90.716\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.201\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e9.524\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.999\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLSTM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.723\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e68.481\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.191\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8.275\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.989\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCycleGAN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.593\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e49.329\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.187\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e7.023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.975\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaptiGAN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.075\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.863\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.619\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.926\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\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusions","content":"\u003cp\u003eWe have proposed and presented an approach for analyzing the effects caused by a disaster like wildfires on vegetation. This approach involves vegetation change detection which is performed using an unsupervised learning algorithm called DEC along with sparse encoders for feature extraction which provides results with an impressive accuracy of 96.17%, vegetation regrowth analysis was performed using Sen\u0026rsquo;s slope estimator which provided time-series based analysis for regions affected by fire, we considered 4 creeks for this study namely: Moore, Florida, Tosher and Tonalite creek based on which the EVI trend analysis was performed to visualize the greening and browning trend over years by using data from the MODIS dataset to analyze the EVI pattern over years along with this regrowth possibility was predicted using an ensemble learning method called stacking which take soil data as input and provides the regrowth possibility for the region as output on whether there is a possibility for regrowth or not in that region and finally we performed vegetation recovery mapping which required collecting VIIRS data from the USGS website from which NDVI images were extracted, these NDVI images were then used for training the AdaptiGAN model. Using the trained model predictions were made in 3 regions chosen for this research namely Amazon rainforest, Knysna, Alaska regions and the corresponding recovery maps were obtained. The major advantages of our approach include it\u0026rsquo;s extreme flexibility and can be used for analyzing vegetation for different regions i.e. it\u0026rsquo;s not region specific, additionally its efficient and can be used by organizations to analyze the effects caused by wildfires without spending a lot of money on ground-based analysis, our approach can meet the needs posed by the real-time disaster response scenarios due to its high accuracy and speedy performance. In the future we are planning to make prediction based on video analysis making use of time-lapse based satellite data and a more comprehensive technique which can effectively make note of all changes and effects caused by a wildfire by additionally incorporating the effect it had on wildlife, what all changes and precautions can be taken to prevent wildfire\u0026rsquo;s can be developed.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization, R.S.P. (R. Shanmuga Priya) and K.V. (K. Vani); methodology, R.S.P. ;.; validation, R.S.P. and K.V.; formal analysis, R.S.P. and K.V.; resources and curation, R.S.P; writing\u0026mdash;original draft preparation, R.S.P.; writing\u0026mdash;review and editing, R.S.P. and K.V.; visualization, R.S.P.; supervision, K.V.; project administration, R.S.P. and K.V. All authors have read and agreed to the published version of the manuscript.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eData Availability Statement\u0026nbsp;\u003c/strong\u003eThe datasets used and/or analyzed during the current study are available from the corresponding author on request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDisclosure Statement \u0026nbsp;\u003c/strong\u003eNo potential conflict of interest was reported by the authors\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eNegassa, M.D., Mallie, D.T., \u0026amp; Gemeda, D.O. (2020). Forest cover change detection using Geographic Information Systems and remote sensing techniques: a spatio-temporal study on Komto Protected forest priority area, East Wollega Zone, Ethiopia. Environmental SystemsResearch,9(1), https://doi.org/10.1186/s40068-020-0163-z.\u003c/li\u003e\n\u003cli\u003eLiu, T., Yang, L., \u0026amp; Lunga, D. (2021). Change detection using deep learning approach with object-based image analysis. Remote Sensing of Environment, 256, 112308. https://doi.org/10.1016/j.rse.2021.112308.\u003c/li\u003e\n\u003cli\u003eWang, Z., Wei, C., Liu, X., Zhu, L., Yang, Q., Wang, Q., Zhang, Q., \u0026amp; Meng, Y. (2022). Object-based change detection for vegetation disturbance and recovery using Landsat time series. GIScience \u0026amp; Remote Sensing, 59(1), 1706-1721. https://doi.org/10.1080/15481603.2022.21298 70.\u003c/li\u003e\n\u003cli\u003eJoseph Hughes, M., Douglas Kaylor, S., \u0026amp; Hayes, D. J. (2017). Patch-Based Forest Change Detection from Landsat Time Series. Forests, 8(5), 1\u0026ndash;22. https://doi.org/10.3390/f8050166.\u003c/li\u003e\n\u003cli\u003eHussain, M., D. Chen, A. Cheng, H. Wei, and D. Stanley. (2013). \u0026ldquo;Change Detection from Remotely Sensed Images: From Pixel-Based to Object-Based Approaches.\u0026rdquo; ISPRS Journal of Photogrammetry and Remote Sensing 80: 91\u0026ndash;106. doi:10. 1016/j.isprsjprs.2013.03.006.\u003c/li\u003e\n\u003cli\u003eZhu, Z. (2017). \u0026ldquo;Change Detection Using Landsat Time Series: A Review of Frequencies, Preprocessing, Algorithms, and Applications.\u0026rdquo; ISPRS Journal of Photogrammetry and Remote Sensing130:370\u0026ndash;384. doi:10.1016/j.isprsjprs.2017.06.013.\u003c/li\u003e\n\u003cli\u003eChen, X., Vierling, L., Deering, D. (2005). A simple and effective radiometric correction method to improve landscape change detection across sensors and across time. Remote Sensing of Environment,98(1),63-79. https://doi.org/10.1016/j.rse.2005.05.021.\u003c/li\u003e\n\u003cli\u003eTian, Y., Wu, Z., Li, M., Wang, B., \u0026amp; Zhang, X. (2022). Forest Fire Spread Monitoring and Vegetation Dynamics Detection Based on Multi-Source Remote Sensing Images. Remote Sensing,14,4431. https://doi.org/10.3390/rs14184431.\u003c/li\u003e\n\u003cli\u003eMeera Gandhi, G., Parthiban, S., Nagaraj Thummalu\u0026amp; Christy, A. (2015) Ndvi: VegetationChange Detection Using Remote Sensing andGis \u0026ndash; A Case Study of Vellore District.Procedia Computer Science, 57 (pp 1199-1210).\u003c/li\u003e\n\u003cli\u003eChen, X., Vogelmann, J. E., Rollins, M., Ohlen, D., Key, C. H., Yang, L., Huang, C., \u0026amp; Shi, H. (2011). Detecting Post-Fire Burn Severity and Vegetation Recovery Using Multitemporal Remote Sensing Spectral Indices and Field- Collected Composite Burn Index Data in a Ponderosa Pine Forest. International Journal of Remote Sensing, 32(23), 7905\u0026ndash;7927. https://doi.org/10.1080/01431161.2010.524678.\u003c/li\u003e\n\u003cli\u003eLacouture, D.L., Broadbent, E.N., Crandall, R.M. (2020). Detecting Vegetation Recovery after Fire in a Fire-Frequented Habitat Using Normalized Difference Vegetation Index (NDVI).Forests,11,749. https://doi.org/10.3390/f11070749.\u003c/li\u003e\n\u003cli\u003eSharma, R., Rimal, B., Baral, H., Nehren, U., Paudyal, K., Sharma, S., Rijal, S., Ranpal, S., Acharya, R.P., Alenazy, A.A., et al. (2019). Impact of Land Cover Change on Ecosystem Services in a Tropical Forested Landscape. Resources,8,18. https://doi.org/10.3390/resources8010018.\u003c/li\u003e\n\u003cli\u003eKristensen, P. J., Gould, K., \u0026amp; Thomsen, J. B. (1997). Approaches to field-based monitoring and evaluation implemented by Conservation International. In Proceedings and Papers of the International Workshop on Biodiversity Monitoring, Brazilian Institute for Environment and Renewable Resources, Pirenopolis, Brazil, pp. 129\u0026ndash;144.\u003c/li\u003e\n\u003cli\u003eBousquet, E., Mialon, A., Rodriguez-Fernandez, N., Mermoz, S., \u0026amp; Kerr, Y. (2022). Monitoring post-fire recovery of various vegetation biomes using multi-wavelength satellite remote sensing. Biogeosciences, 19(13), 3317-3336. doi: 10.5194/bg-19-3317-2022.\u003c/li\u003e\n\u003cli\u003eNolan, R. H., Collins, L., Leigh, A., Ooi, M. K. J., Curran, T. J., Fairman, T. A., et al. (2021). Limits to post-fire vegetation recovery under climate change. Plant, Cell Environ. 44 (11), 3471\u0026ndash;3489. doi:10.1111/pce.14176.\u003c/li\u003e\n\u003cli\u003eJarchow, C.J., Didan, K., Barreto-Mu\u0026ntilde;oz, A., Nagler, P.L., \u0026amp; Glenn, E.P. (2018). Application and Comparison of the MODIS- Derived Enhanced Vegetation Index to VIIRS, Landsat 5 TM and Landsat 8 OLI Platforms: A Case Study in the Arid Colorado River Delta, Mexico.Sensors,18, 1546. https://doi.org/10.3390/s18051546.\u003c/li\u003e\n\u003cli\u003eKim, Y., Jeong, M.-H., Youm, M., Kim, J., \u0026amp; Kim, J. (2021). Recovery of Forest Vegetation in a Burnt Area in the Republic of Korea: A Perspective Based on Sentinel-2 Data. Applied Sciences,11,2570. https://doi.org/10.3390/app11062570.\u003c/li\u003e\n\u003cli\u003e\u0026Uacute;beda, X., L.R. Outeiro, and M. Sala. (2006). Vegetation regrowth after a differential intensity forest fire in a Mediterranean environment, northeast Spain. Land Degradation and Development 17 (4): 429\u0026ndash;440. https://doi.org/10.1002/ldr.748.\u003c/li\u003e\n\u003cli\u003eVasques, A., Baudena, M., Vallejo, V.R., et al. (2023). Post-fire Regeneration Traits of Understorey Shrub Species Modulate Successional Responses to High Severity Fire in Mediterranean Pine Forests. Ecosystems, 26, 146\u0026ndash;160. https://doi.org/10.1007/s10021-022-00750-z.\u003c/li\u003e\n\u003cli\u003eYumoto, S., Kitsukawa, T., Moro, A., et al. (2023). Anomaly detection from images in pipes using GAN. Robomech Journal, 10, 9. https://doi.org/10.1186/s40648-023-00246-y.\u003c/li\u003e\n\u003cli\u003eBenedict, T.D., Brown, J.F., Boyte, S.P., Howard, D.M., Fuchs, B.A., Wardlow, B.D., Tadesse, T., Evenson, K.A. (2021). Exploring VIIRS Continuity with MODIS in an Expedited Capability for Monitoring Drought-Related Vegetation Conditions. Remote Sensing, 13, 1210. https://doi.org/10.3390/rs13061210.\u003c/li\u003e\n\u003cli\u003eSkakun, S., Justice, C.O., Vermote, E., Roger, J.C. (2018). Transitioning from MODIS to VIIRS: an analysis of inter-consistency of NDVI data sets for agricultural monitoring. International Journal ofRemoteSensing,39(4),971-992. https://doi.org/10.1080/01431161.2017.139597 0.\u003c/li\u003e\n\u003cli\u003eNatalie, R. W., and M. Norman Laura. (2018). \u0026rdquo;Analysis of Vegetation Recovery Surrounding a Restored Wetland Using the Normalized Difference Infrared Index (NDII) and Normalized Difference Vegetation Index (NDVI).\u0026rdquo; International Journal of Remote Sensing 39 (10): 3243\u0026ndash;3274. Aksu Prefectural Committee of historical records. Aksu almanac. doi:10.1080/01431161.2018.1437297.\u003c/li\u003e\n\u003cli\u003eXiao, J., \u0026amp; Moody, A. (2005). A comparison of methods for estimating fractional green vegetation cover within a desert-to-upland transition zone in Central New Mexico, USA. Remote Sensing of Environment, 98(2\u0026ndash;3), 237\u0026ndash;250. https://doi.org/10.1016/j.rse.2005.07.011\u003c/li\u003e\n\u003cli\u003eDe Petris, S., Sarvia, F., \u0026amp; Borgogno-Mondino, E. (2021). Multi-temporal mapping of flood damage to crops using Sentinel-1 imagery: A case study of the Sesia River (October 2020). Remote Sensing Letters, 12(5), 459\u0026ndash;469. https://doi.org/10.1080/2150704X.2021.189026 2\u003c/li\u003e\n\u003cli\u003eCurran, P. J. (1981). Multispectral remote sensing for esti-mating biomass and productivity. Plants and the Daylight Spectrum.\u003c/li\u003e\n\u003cli\u003eGoward, S. N. (1989). Satellite bioclimatology. JournalofClimate,2(7),710\u0026ndash;720. https://doi.org/10.1175/1520- 0442(1989)002\u0026lt;0710:SB\u0026gt;2.0.CO;2\u003c/li\u003e\n\u003cli\u003eMalingreau, J. P. (1989). The vegetation index and the study of vegetation dynamics. In Applications of remote sensing to agrometeorology (pp. 285\u0026ndash;303). Springer Netherlands. https://doi.org/10.1007/978-94- 009-2235-8_12\u003c/li\u003e\n\u003cli\u003eLentile L.B., Holden Z.A., Smith A.M.S., Falkowski M.J., Hudak A.T., Morgan P., Lewis S.A., Gessler P.E., Benson N.C. (2006) - Remote sensing techniques to assess active fire characteristics and post-fire effects. International Journal of Wildland Fire, 15: 319- 345. doi: https://doi.org/10.1071/WF05097.\u003c/li\u003e\n\u003cli\u003eMiller J.D., Thode A.E. (2007) - Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sensing of Environment, 109: 66-80. doi: https://doi.org/10.1016/j.rse.2006.12.006.\u003c/li\u003e\n\u003cli\u003eBonan G.B. (1989) - Environmental factors and ecological processes controlling vegetation patterns in boreal forests. Landscape Ecology, 3:111-130.doi: http://dx.doi.org/10.1007/BF00131174.\u003c/li\u003e\n\u003cli\u003eKennedy R.E., Yang Z.Q., Cohen W.B., Pfaff E., Braaten J., Nelson P. (2012) - Spatial and temporal patterns of forest disturbance and regrowth within the area of the Northwest Forest Plan. Remote Sensing of Environment, 122: 117-133. doi: http://dx.doi.org/10.1016/j.rse.2011.09.024.\u003c/li\u003e\n\u003cli\u003eStueve K.M., Cerney D.L., Rochefort R.M., Kurth L.L. (2009) - Post-fire tree establishment patterns at the alpine treeline ecotone: Mount Rainier National Park, Washington, USA. Journal of Vegetation Science, 20: 107- 120. doi: http://dx.doi.org/10.1111/j.1654- 1103.2009.05437.x.\u003c/li\u003e\n\u003cli\u003eOtoda T., Doi T., Sakamoto K., Hirobe M., Nachin B., Yoshikawa K. (2013) - Frequent fires may alter the future composition of the boreal forest in northern Mongolia. Journal of Forest Research,18:246-255.doi: http://dx.doi.org/10.1007/s10310-012-0345-2.\u003c/li\u003e\n\u003cli\u003eTownshend J.R., Masek J.G., Huang C.Q., Vermote E.F., Gao F., Channan S., Sexton J.O., Feng M., Narasimhan R., Kim D., Song K., Song D.X., Song X.P., Noojipady P., Tan B., Hansen M.C., Li M.X., Wolfe R.E. (2012) - Global characterization and monitoring of forest cover using Landsat data: opportunities and challenges. International Journal of Digital Earth,5:373-397.doi: http://dx.doi.org/10.1080/17538947.2012.71319 0.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Wildfire, change detection, Ensemble learning, deep learning","lastPublishedDoi":"10.21203/rs.3.rs-3890182/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3890182/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWildfire are uncontrolled fires fueled by dry conditions, high winds and flammable materials that tends to have a profound impact on vegetation due to the intense heat generated by it which can cause the destruction of trees, small plants and other vegetation leading to significant consequences including noteworthy changes to ecosystems. Due to the periodic wildfires, vegetation communities in forest systems have changed adaptively to deal with ecological rebuilding. In this study we provide a novel methodology, to understand and evaluate post-fire effects on vegetation. In regions which are affected by wildfire, earth-observation data provided by various satellite sources can be very vital in monitoring vegetation and assessing the effect a wildfire tends to have on it. These effects can be understood by detecting the change of vegetation over years using an unsupervised method termed Deep Embedded Clustering (DEC), which enables us to classify regions on whether there has been a change in vegetation after fire. Appropriate vegetation indices can be used to evaluate evolution of vegetation pattern over the years, for this study we utilized Enhanced Vegetation Index (EVI) based trend analysis. Vegetation recovery maps can be created to assess re-vegetation in regions affected by fire which is performed via a deep learning based unsupervised method, Adaptive Generative Adversarial Neural Network Model (AdaptiGAN) on postfire data collected from various regions affected by wildfire. Through the results obtained from the study we can arrive at a conclusion that our approach tends to have notable merits when compared to pre-existing works.\u003c/p\u003e","manuscriptTitle":"Vegetation Change Detection and Recovery Assessment on Post-fire Satellite Imagery using Deep Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-25 16:50:57","doi":"10.21203/rs.3.rs-3890182/v1","editorialEvents":[{"type":"communityComments","content":1}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"04d53389-aa5a-4795-93c5-5d1af9e2f360","owner":[],"postedDate":"January 25th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-02-11T16:14:50+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-25 16:50:57","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3890182","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3890182","identity":"rs-3890182","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}