Analysis of vegetation index with artificial neural network and wavelet transform

In this study, the LAI (Leaf Area Index) data for 2011-2016 of the Marmara region were taken from MODIS TERRA satellite belonging to NASA to examine spatio-temporal features of vegetation index in terms of urbanization, migration etc. The ever-increasing greenhouse gases in the atmosphere cause the most important problem of our era, the global warming. Global warming is causing rise in sea level due to melting glaciers and while in some regions hurricanes and floods are seen, in some others, drought occurs due to the reduction of rainfall. Fossil fuels, rapid population growth, insufficient afforestation, and destruction of forests increase the impact of global warming. The effects of climate change have also been felt in Marmara Region, where rapid population growth has been observed in Turkey. Decrease in the amount of rainfall, increase in fires, decrease in plant species due to increased surface temperature are the main causes of these effects. Drought causes decline in agricultural yield. Tourism is also one of the sectors affected by these adversities. In this study, the effects of climate change were observed in Marmara Region by analyzing vegetation cover based on mathematical modelling. A hybrid system was created with Wavelet Transform and Artificial Neural Networks. Thus, predictions had lower error rate. In addition, rainfall data from 10 stations (Edirne, Tekirdağ, Çanakkale, Balıkesir, Bursa, İstanbul, Sakarya, Kocaeli, Bilecik, Kırklareli) in the Marmara Region have been interpreted statistically. Hereby, the change in rainfall amount due to global warming has been analyzed. Generally, while the highest LAI values are recorded in April, May and June, the lowest LAI values are recoreded in November, December and January. When it comes to precipitation, the highest values are recorded in fall season and the lowest values are recorded in July and August. As a result, a negative correlation between LAI and precipitation has been observed. Accordingly, depending on cities, 5-6 months time lag between LAI and precipitation attracts attention. It has been determined that the delay experienced in soil while transferring precipitation to the plant members causes this time lag. Entropy graphs obtained by continuous wavelet transform show an increase in the frequency of heavy precipitation. Precipitation was divided into wavelets by Daubechies algorithm and sent as an input to the Levenberg-Marquardt backpropagation artificial neural network algorithm. According to YSA training results, the lowest MSE and RMSE values were recorded in Çanakkale. It can be sait that land-sea interaction in Çanakkale may effect this outcome. MB and RMB values are the lowest in Bursa. NRMS value is the lowest in Balikesir province. As a result of this study, it was observed that Wavelet-ANN model generates promising rain depended LAI predictions. In other words, this model is quite feasible to use in studies about vegetation data.