Thesis (M.Sc)-Cairo University, 2025.

Bibliography: pages 117-133.

Water scarcity and climate change are two major challenges facing Sudan, which have led
to the migration of many people. The main objective of this study is to evaluate the
potential of single and hybrid machine learning (ML) models in predicting the Green
Water Footprint (GWFP) of sugarcane and cotton under the influence of climate change.
The study will analyze the effects of different input factors, including climate, crop and
remote sensing data, to determine the impact of these factors during the period from 2001
to 2020. Seven models, Random Forest (RF), Extreme Gradient Boosting (XGBoost),
Support Vector Regression (SVR), Hybrid RF-XGB, RF-SVR, XGB-SVR and RF-XGB
SVR were applied with five scenarios. In reed, the highest MBE was obtained under RF
and Sc3 and was 5.14 m3 ton-1 followed by RF-SVR with 5.05 m3 ton-1, while the weakest
MBE was 0.03 under RF-SVR and Sc1. The highest R2 values were achieved using the
SVR model for all scenarios. What is most noticeable is that the R2 values of the dual
hybrid models were higher than those of the triple hybrid models. The highest NSE value
was 0.98 under Sc2 (climatic parameters) and XGB-SVR, while the lowest NSE value
was recorded with SVR and Sc3 (remote sensing parameters) and was 0.09. RMSE did
not have a consistent trend across all ML model combinations and different scenarios but
what is noticeable under all statistical evaluation indices is that Sc3 has the worst
evaluation dealing with remote sensing parameters (EVI, NDVI, SAVI, and NDWI). The
highest significant impact on GWFP came from effective rainfall at 81.67% followed by
relative humidity (RH) at 7.5% and then maximum temperature at 5.24%. The conclusion
from the study is that when predicting GWFP for sugarcane, individual models achieved
equal and in some cases greater results than the dual and triple hybrid models. In the same
context, remote sensing indices had no positive impact on GWFP prediction, with Sc3
reflecting the lowest values for all statistical parameters of all models used, so the study
recommends SVR with Sc1 or Sc4 depending on data availability. While in cotton, the
maximum and minimum RMSE values ranged between 31.35 m3 ton-1 and 166.37 m3 ton
1
, based on the hybrid model RF-XGB-SVR and RF model, respectively, under Sc5 and
(Peeff, Tmax) achieved the highest R2 values using hybrid ML models, whether double
or triple, across all scenarios, reaching values of 1.0 or 0.99. The lowest R2 value, recorded
at 0.0676, was observed under SVR and Sc3, closely followed by XGB and Sc3 with a
value of 0.0767. The study recommends the use of hybrid models to reduce the error term
in predicting GWFP of sugarcane and cotton.

ندرة المياه وتغير المناخ هما تحديان رئيسيان يواجههما السودان، مما أدى إلى هجرة الكثير من الناس. الهدف الرئيسي من هذه الدراسة هو تقييم إمكانات نماذج التعلم الآلي سواء كانت فردية أو هجينة في التنبؤ ببصمة المياه الخضراء (GWFP) لقصب السكر والقطن تحت تأثير تغير المناخ. ستقوم الدراسة بتحليل تأثيرات عوامل المدخلات المختلفة، بما في ذلك البيانات المناخية والمحاصيل والاستشعار عن بعد، لتحديد تأثير هذه العوامل خلال الفترة من 2001 إلى 2020. تم تطبيق سبعة نماذج، الغابة العشوائية (RF)، التدرج الشديد التعزيز (XGBoost)، ودعم الانحدار المتجه (SVR)، وHybrid RF-XGB، وRF-SVR، وXGB-SVR، وRF-XGB-SVR مع خمسة سيناريوهات. في نبات القصب تم الحصول على أعلى MBE تحت RF وSc3 وكان 5.14 م3 طن-1 يليه RF-SVR بمقدار 5.05 م3 طن-1، بينما كان أضعف MBE 0.03 تحت RF-SVR و Sc1 تم تحقيق أعلى قيم R2 باستخدام نموذج SVR لجميع السيناريوهات

Issues also as CD.

Text in English and abstract in Arabic & English.