Analysing the vegetation dynamics is important in the context of theirsustainability. Vegetation structure changes rapidly due to various reasonsinfluenced by both anthropogenic and natural factors, potentially resulting in thedegradation and destruction. Vegetation phenology is impacted by lifecyclepatterns, season and climatic conditions resulting in changes of their spectralreflectance patterns (Rußwurm and Korner 2017).
Spatio-temporal mapping and monitoring of vegetationhelps in assessing the health of vegetation and imply better management practicesto safeguard them. The availability of the spatio-temporal satellite data aidsin visualizing changing vegetation dynamics in due course of time. Further,development of vegetation indices using remote sensing satellite data providesbetter insights about the growth patterns, seasonal changes and healthconditions of the vegetation (Wang et al 2016). Normalised Difference Vegetation Index (NDVI) is one of the most widelyacknowledged indices for vegetation related studies (Xue and Su 2017). NDVI isa numerical indicator of the greenness of vegetation derived using the visibleand near infrared (NIR) bands (Jeevalakshmi et al 2016, Eslamian and Eslamian2017). Typically, the NDVI values are normalized between +1 and -1, and ahigher NDVI value indicates greener and denser vegetation (Peckham and Jordan2007). The negative NDVI value denotes other non-vegetated classes like water, urbanareas, barren lands, snow, etc (Anonymous 2000). The generalized annual NDVIprofile for vegetation rises with the increase in the plant growth and reachesa peak or plateau (Jose et al.
2002). Later the profile falls off eventuallywith plant death or leaf senescence. Thus, NDVI series provides a means todescribe plant phenology (Viovy et al. 1992,Wang et al 2016). However, the phenological changes in concurrence withtime appear to be very minimal or obscure, as observed in the vegetation oftropical rainforest due to subtle climatic variations (Valtonen et al 2013). The analysis of timeseries is widely used approach in many studies as they work with data of hightemporal resolution and low spatial resolution (Petitjean et al 2014). Theavailability of time series NDVI data enhances information about the vegetationconditions as well as in assessing and monitoring its changes (Ivits et l 2013).Interim changes are better inferred by time series analyses which cannot beaccounted using low temporal satellite data.
Timeseries includes any temporary changes and thus provides better analysis thanany other datasets, as the analysis is based on observations of particular timeframe. Major seasonal changes and inter-annual patterns can be well defined byanalyzing the time series data of vegetation and surface water bodies (Haas et al. 2009; Tulbure and Broich 2013; Cordeiro et al. 2016; Remboldet al. 2015). NDVItime series analysis offers more accurate and efficient results in detectingthe change in vegetation cover (Lyu and Mou2016; Agone and Bhamare 2012), land use and land cover (LULC)classification (Gómez et al. 2016; Anderson1976), estimation and prediction of vegetation, mapping forestdisturbance (Kennedy et al 2010), etc.Time series analysis also enables to estimate and model biomass (Gómez et al.
Best services for writing your paper according to Trustpilot
2014), analyze forestdegradation (Shimabukuro et al. 2014) andassess forest carbon sinks (Gómez et al. 2012). Various types of remotely sensed imagery and processing methods havebeen introduced and used to predict NDVI time series.
Autoregressive integratedmoving average (ARIMA) models are used for forecasting NDVI time series (Stepchenko 2016; Manobavan et al. 2002).These models use the adjoining data to predict the next values in the timeseries, but as these models are parametric and assumption of the data to belinear and stationary, makes them inappropriate for precise prediction of timeseries. Markov chain model simply constructs a probability mass functionincrementally across the possible next states(Stepchenko and Chizhov 2015).
It is memory-less as itconsiders only the present state of the process to predict future. It is asimple and effective method for prediction, but it uses a fixed window (Kriminger and Latchman 2011). Neural networks have become popular in the analysis of remote sensingdata with the increase in availability of satellite data, as they arenon-parametric unlike most of the statistical methods (Foody 2006; Mas andFlores 2008). Many studies have shownthat neural networks work well as they are non-linear models and perform wellwith noisy data as well (Atkinson and Tatnall 1997). Someof the applications of neural networks in remote sensing are land cover mapping (Zhou and Yang 2008), forest change detection (Gopal and Woodcock 1996), and predicting vegetationchanges (Kang et al. 2016). Differentneural networks from the Multiple Layer Perceptron (MLP) (Atkinson and Tatnall 1997), Artificial Neural Network (ANN) (Silva et al.
2014; Miller et al. 1995) toadvanced neural networks like Back-propagation (BP) neural networks (Zhang and Chang 2015) and Convolution NeuralNetworks (Maggiori et al. 2016) are usedfor classification of remote sensing satellite imagery (Atkinson and Tatnall 1997). ANNs can be used for forecasting NDVI index (Kang et al. 2016; Nay et al. 2016), but as ANNs have no memory tostore the information of the past data in the time series, results are less efficient.Recurrent neural network (RNN), which has memory can also be used forpredicting the time series (Stepchenko and Chizhov 2015; Stepchenko 2016), by training the RNN with some past NDVI valuesbut vanishing gradient problem of RNNs makes them less suitable. Further, theyrequire a substantial amount of preceding values as input to the network forprediction of the succeeding values and thus involves lot of computational overheadcompared to Long Short Term Memory (LSTM) network.
The LSTM is a variant of RNN, which has an internalmemory to store the information received till time t, for a long time in the model (Skymind2016; Budama 2015). This property of LSTM makes them very much preferablefor predicting the time series. LSTM is a deep learning neural network which ishighly preferred in predicting time series due to its long-term memory (Gamboa2017). Monitoring vegetationchanges is inevitable in the context of current climatic change conditions andrapid human interventions.
Vegetation, especially island ecosystems arevulnerable to both human induced disturbances and natural forces like tsunamiand volcanic eruptions. Seasonal changes are common in vegetation. However, itis also effected by climate change and may lead to drastic change due tocatastrophes (Verbesselt et al. 2009).
Change detection methodshelp in assessing the change in vegetation (de Beurs and Henebry 2005) and themethods available so far are able to identify changes using low temporalresolution or decadal gap remote sensing data (Coppin et al. 2004). They areable to track the extent of areal changes but are incapable of identifying theminor seasonal or any gradual changes. In view of this drawback,time series data with higher temporal resolution depict better changes invegetation (seasonal, gradual or sudden abrupt changes) and is adopted widelyin assessing minor/shorter phenomenal changes.
Further the use of index derivedfrom time series data (such as NDVI) is beneficial in identifying the changescompared to typical satellite data analysis, where only areal extent iscalculated and compared. Time series data can be analysed using varied ANNmethods and the best one proposed in recent past is the use of LSTM. Though LSTM gained specific significance indifferent themes (like LULC, soil moisture studies, predictions / forecastingan event), its application in vegetation studies is still in nascent stage.
Theresearch of Rußwurm and Korner (2017) can be cited as one example of LSTM study adopted forcrop identification using Sentinel data. However, no specific studies have beencarried out with reference to use of LSTM in forest vegetation and the currentstudy is first of its kind to be cited as an example of LSTM study inprediction of forest dynamics as well as for the study area too. In the view of abovecontext, the main objective of current research is to predict vegetationdynamics using MODIS (Moderate Resolution ImagingSpectroradiometer) NDVI time series data.
Time series is a sequence ofcorrelated scalars or vectors which vary with time. In the present study, thesequence consists of the scalar NDVI values. Time series can be predicted, if oneknows the past values in the series up to time t, and then estimating the valueat time t+s, s = 1, 2..
. The study considered s = 1. LSTM network is trained topredict the NDVI value at t+1.
