Using Different Basins in Flood Forecasting; Example of Artvin Floods

Ufuk Yükseler; Ömerul Faruk Dursun

Makale detayları

Using Different Basins in Flood Forecasting; Example of Artvin Floods

Yazarlar:

Ufuk Yükseler | Ömerul Faruk Dursun

Yayıncı:

Fırat University

Ana Başlıklar:

Hidromekanik | İnşaat Mühendisliğinde Sayısal Modelleme | Coğrafi Bilgi Sistemleri ve Mekansal Veri Modelleme

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.36 Issue 2, pp.827-835

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317082

pISSN:

1308-9072

Anahtar kelimeler:

Özet:

 Machine learning methods used recently in determining the sensitivity and risk areas of floods give quite compatible results. As seen in most studies in the literature, a part of the basin where floods occur is estimated with flood trigger parameters based on a part of the basin where data is available and known to have experienced a disaster, or a part where data is missing is estimated. However, the biggest deficiency of these studies in question is the lack of estimation possibilities for basins with no data. The second issue is that machine learning methods create an overlearning problem in estimations within the same basin. In this study, the use of different basins in estimating basins with no data was examined. For this purpose, 4 floods that occurred within the borders of Artvin province in 2009, 2015, 2020 and 2021 were used. The Random Forest method, one of the machine learning methods, was used in the study. The selection of the method was influenced by the fact that there are studies in the literature that have reached very high accuracy values and that it analyzes the event pattern well in complex events such as floods. In this study, training was performed with 1490 point data selected from random points taken from 2009, 2015 and 2020 disasters and 560 test data 2021 disaster were estimated. Validation of the study was carried out using 5 validation methods AUC, ACC, F, P, R and F-Score. Accuracy values in training and test data were determined at the level of 90. When the results were examined, the prediction of the disaster experienced in 2021 with the Random Forest method gave results very close to the real disaster area. The findings show the success of inter-basin predictions in case of data deficiency in the prediction of flood disasters and the creation of risk and susceptibility maps of the basin created for basin planning purposes. It is seen that good results can be obtained in this area with the development of methods and increasing the number of samples and can be used in disaster management issues.

 Machine learning methods used recently in determining the sensitivity and risk areas of floods give quite compatible results. As seen in most studies in the literature, a part of the basin where floods occur is estimated with flood trigger parameters based on a part of the basin where data is available and known to have experienced a disaster, or a part where data is missing is estimated. However, the biggest deficiency of these studies in question is the lack of estimation possibilities for basins with no data. The second issue is that machine learning methods create an overlearning problem in estimations within the same basin. In this study, the use of different basins in estimating basins with no data was examined. For this purpose, 4 floods that occurred within the borders of Artvin province in 2009, 2015, 2020 and 2021 were used. The Random Forest method, one of the machine learning methods, was used in the study. The selection of the method was influenced by the fact that there are studies in the literature that have reached very high accuracy values and that it analyzes the event pattern well in complex events such as floods. In this study, training was performed with 1490 point data selected from random points taken from 2009, 2015 and 2020 disasters and 560 test data 2021 disaster were estimated. Validation of the study was carried out using 5 validation methods AUC, ACC, F, P, R and F-Score. Accuracy values in training and test data were determined at the level of 90. When the results were examined, the prediction of the disaster experienced in 2021 with the Random Forest method gave results very close to the real disaster area. The findings show the success of inter-basin predictions in case of data deficiency in the prediction of flood disasters and the creation of risk and susceptibility maps of the basin created for basin planning purposes. It is seen that good results can be obtained in this area with the development of methods and increasing the number of samples and can be used in disaster management issues.

 Machine learning methods used recently in determining the sensitivity and risk areas of floods give quite compatible results. As seen in most studies in the literature, a part of the basin where floods occur is estimated with flood trigger parameters based on a part of the basin where data is available and known to have experienced a disaster, or a part where data is missing is estimated. However, the biggest deficiency of these studies in question is the lack of estimation possibilities for basins with no data. The second issue is that machine learning methods create an overlearning problem in estimations within the same basin. In this study, the use of different basins in estimating basins with no data was examined. For this purpose, 4 floods that occurred within the borders of Artvin province in 2009, 2015, 2020 and 2021 were used. The Random Forest method, one of the machine learning methods, was used in the study. The selection of the method was influenced by the fact that there are studies in the literature that have reached very high accuracy values and that it analyzes the event pattern well in complex events such as floods. In this study, training was performed with 1490 point data selected from random points taken from 2009, 2015 and 2020 disasters and 560 test data 2021 disaster were estimated. Validation of the study was carried out using 5 validation methods AUC, ACC, F, P, R and F-Score. Accuracy values in training and test data were determined at the level of 90. When the results were examined, the prediction of the disaster experienced in 2021 with the Random Forest method gave results very close to the real disaster area. The findings show the success of inter-basin predictions in case of data deficiency in the prediction of flood disasters and the creation of risk and susceptibility maps of the basin created for basin planning purposes. It is seen that good results can be obtained in this area with the development of methods and increasing the number of samples and can be used in disaster management issues.

 Taşkınların duyarlılık ve risk sahasının tespitinde son dönemlerde kullanılan makine öğrenimi yöntemleri oldukça uyumlu sonuçlar vermektedir. Literatürdeki çoğu çalışmada görüldüğü üzere taşkın yaşanan havzanın veri bulunan ve afet yaşandığı bilinen bir kısmından yola çıkılarak taşkın tetikleyici parametrelerle havzanın tahmin edilmesini istenen veya veri eksikliği bulunan bir kısmı tahmin edilmektedir. Ancak bahse konu bu çalışmaların en büyük eksikliği hiçbir verisi bulunmayan havzaların tahmin olanağının olmamasıdır. İkinci husus ise makine öğrenim yöntemlerinin aynı havza içerisindeki tahminlerde aşırı öğrenme problemi oluşturmasıdır. Bu çalışmada veri bulunmayan havzaların tahmininde farklı havzaların kullanımı incelenmiştir. Bu amaçla Artvin il sınırları içinde 2009, 2015, 2020 ve 2021 yıllarında meydana gelen 4 adet taşkın kullanılmıştır. Çalışmada makine öğrenimi yöntemlerinden Rastgele Orman metodu kullanılmıştır. Yöntemin seçiminde, literatürde oldukça yüksek doğruluk değerlerine ulaşan çalışmalar olması ve taşkın gibi karmaşık olaylarda olay örgüsünü iyi analiz etmesi etkili olmuştur. Bu çalışmada rastgele noktalardan seçilen 1490 noktasal veri ile 2009, 2015 ve 2020 afetlerinden alınan eğitim yapılmış ve 560 test verisi 2021 afeti  tahmin edilmiştir. Çalışmanın doğrulaması 5 adet doğrulama yöntemleri AUC, ACC, F, P, R ve F-Score kullanılarak gerçekleştirilmiştir. Eğitim ve test verilerindeki doğruluk değerleri 90 düzeyinde tespit edilmiştir. Sonuçlar incelendiğinde 2021 yılında yaşanan afetin Rastgele Orman metodu ile tahmini, gerçek afet sahasına oldukça yakın sonuçlar vermiştir. Bulgular, taşkın afetinin önceden tahmin edilmesinde, havza planlamaları amacıyla oluşturulan havzanın risk ve duyarlılık haritalarının oluşturulmasında veri eksikliği bulunması durumunda havzalar arası tahminlerin başarısını göstermektedir. Metotların gelişimi ve örneklem sayısının arttırılması ile bu alanda iyi sonuçlar alınabileceği ve afetlerle mücadele konularında kullanılabileceği görülmektedir.

Makale detayları

Using Different Basins in Flood Forecasting; Example of Artvin Floods

Yazarlar:

Ufuk Yükseler | Ömerul Faruk Dursun

Yayıncı:

Fırat University

Ana Başlıklar:

Hidromekanik | İnşaat Mühendisliğinde Sayısal Modelleme | Coğrafi Bilgi Sistemleri ve Mekansal Veri Modelleme

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.36 Issue 2, pp.827-835

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317082

pISSN:

1308-9072

Flood

random forest method

machine learning

geographic information systems

Artvin provinceTaşkın

rastgele orman metodu

makine öğrenimi

coğrafi bilgi sistemleri

Artvin ili.

Sayı

Makale detayları

Yazarlar: Ufuk Yükseler | Ömerul Faruk Dursun

Yayıncı: Fırat University

Ana Başlıklar: Hidromekanik | İnşaat Mühendisliğinde Sayısal Modelleme | Coğrafi Bilgi Sistemleri ve Mekansal Veri Modelleme

Dergi: Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.36 Issue 2, pp.827-835

Belge Tipi: Makale

Yayın Tarihi: 9/2024

Belge Dili: Türkçe

Document ID (Org No.): 2001317082

pISSN: 1308-9072

Sayı

Yazarlar:

Ufuk Yükseler | Ömerul Faruk Dursun

Yayıncı:

Fırat University

Ana Başlıklar:

Hidromekanik | İnşaat Mühendisliğinde Sayısal Modelleme | Coğrafi Bilgi Sistemleri ve Mekansal Veri Modelleme

Dergi:

Fırat Üniversitesi Mühendislik Bilimleri Dergisi Vol.36 Issue 2, pp.827-835

Belge Tipi:

Makale

Yayın Tarihi:

9/2024

Belge Dili:

Türkçe

Document ID (Org No.):

2001317082

pISSN:

1308-9072