Current researches:

---

As climate risk becomes a significant factor affecting banks and financial stability, balance sheet management must take this matter into account. Our research incorporates climate risk into bank balance sheet management.

Using a linear programming duality approach, we determine optimal balance sheets considering climate risk constraints. Our approach allows for the derivation of the shadow price of carbon risk from bank portfolios. The methodology we employ is analytically tractable, and thus we can derive several properties of the balance sheet and its behavior under several different settings, namely, the risk consumption of the assets against the risk limits. Finally, we apply our research to a real-world setting with data, using diversification constraints.

Our research can be used by banks for devising optimal strategies for the balance sheet and computing marginal prices of carbon risk. Similarly, regulators can utilize our framework from a macroprudential point of view to test the impact of carbon risk in balance sheet choices and the channeling of credit. 

Flood forecasting in northern Iran is challenged by strong nonlinear rainfall–runoff dynamics, complex physiographic conditions, and limited lead-time reliability of conventional hydrological models. This study presents a comprehensive evaluation of machine learning–based approaches for short-term flood forecasting in selected river basins of northern Iran. Multiple supervised learning models—including multilayer perceptron artificial neural networks (MLP-ANN), support vector regression (SVR), random forest (RF), extreme gradient boosting (XGBoost), and long short-term memory (LSTM) networks—are implemented to predict river discharge at daily and sub-daily time scales. Input variables consist of antecedent precipitation indices, lagged streamflow, temperature, and soil moisture, derived from ground observations and reanalysis datasets. Model hyperparameters are optimized using k-fold cross-validation and Bayesian optimization. Performance is assessed against benchmark conceptual models using Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE), root mean square error (RMSE), and peak flow error metrics. Results indicate that tree-based ensemble methods and LSTM models consistently outperform shallow learning and conceptual models, particularly in reproducing flood peaks and rising limbs. Feature importance and sensitivity analyses reveal that lagged discharge and cumulative rainfall over short antecedent windows are the dominant predictors of flood occurrence. The study demonstrates the robustness and transferability of machine learning techniques for operational flood forecasting and supports their integration into early warning systems in data-scarce, flood-prone regions.

In a one-price economy, the Fundamental Theorem of Asset Pricing (FTAP) establishes that no-arbitrage is equivalent to the existence of an equivalent martingale measure. Such an equivalent measure can be derived as the normal unit vector of the hyperplane that separates the attainable gain subspace and the convex cone representing arbitrage opportunities. However, in two-price financial models (with a bid–ask price spread), the set of attainable gains is no longer a subspace. We use convex optimization and the conic properties of this region to characterize the “no-arbitrage” principle in financial models with a bid–ask price spread. This characterization will lead us to generate a set of price factor random variables. Under such a set, we can find the lower and upper bounds (super-hedging and sub-hedging bounds) for the price of any future cash flow. We will show that for any cash flow whose price is outside the above range, we can build a trading strategy that provides an arbitrage opportunity. We will generalize this structure to any two-price finite-period financial model.

The Gravity Recovery and Climate Experiment (GRACE) has been successfully used to

monitor variations in terrestrial water storage (GRACETWS) and groundwater storage (GRACEGWS) across the globe, yet such applications are hindered on local scales by the limited spatial resolution of GRACE data. Using the Lower Peninsula of Michigan as a test site, we developed optimum procedures to downscale GRACE Release 06 monthly mascon solutions. A four-fold exercise was conducted. Cluster analysis was performed to identify the optimum number and distribution of clusters (areas) of contiguous pixels of similar geophysical signals (GRACETWS time series); three clusters were identified (cluster 1: 13,700 km2; cluster 2: 59,200 km2; cluster 3: 33,100 km2; Step I). Variables (total precipitation, normalized difference vegetation index (NDVI), snow cover, streamflow, Lake Michigan level, Lake Huron level, land surface temperature, soil moisture, air temperature, and evapotranspiration (ET)), which could potentially contribute to, or correlate with, GRACETWS over the test site were identified, and the dataset was randomly partitioned into training (80%) and testing (20%) datasets (Step II). Multivariate regression, artificial neural network, and extreme gradient boosting techniques were applied on the training dataset for each of the identified clusters to extract relationships between the identified hydro-climatic variables and GRACETWS

solutions on a coarser scale (13,700–33,100 km2), and were used to estimate GRACETWS at a spatial resolution matching that of the fine-scale (0.125◦×0.125◦or 120 km2) inputs. The statistical models were evaluated by comparing the observed and modeled GRACETWS values using the R-squared, the Nash–Sutcliffe model efficiency coefficient (NSE), and the normalized root-mean-square error (NRMSE; Step III). Lastly, temporal variations in GRACEGWS were extracted using outputs of land surface models and those of the optimum downscaling methodology (downscaled GRACETWS) (Step IV). 

Available water resources in the Middle East, as one of the most water-scarce regions of the world, have undergone extra pressure due to climatic change, population growth, and economic development during the past decades. The objective of this study is to detect the trends and quantify the changes in aridity with respect to precipitation and potential evapotranspiration in 20 countries of the Middle East and the adjacent area. A Pixel-wised trend analysis was conducted on precipitation, potential evapotranspiration, and aridity index for 71 years from 1948 to 2018. A nonparametric Mann-Kendall test was used over 14106 points in the study area to detect the trends at monthly and annual time scales. Results showed statistically significant (|Z| >1.96) upward trends in aridity (a downward trend in aridity index) up to 96 percent from December through September in most parts of the region. Aridity in October and November had a downward tendency in most parts of the study area. At the annual time scale, 62.5 percent of the statistically significant trends in aridity were found to be upward (up to 96 percent increase in aridity) due to the combined effects of the decrease in precipitation and the increase in potential evapotranspiration and 37.5 percent of the detected trends were downward (up to 61 percent decrease in aridity). The highest and the lowest trends in aridity were found in the north of Sudan (96 percent increase in aridity) and Eastern Arabia (61 percent decrease in aridity), respectively.

Groundwater salinity in an aquifer system is typically measured through field studies (e.g., groundwater sampling, and direct current resistivity method). The field-based measurements are costly and time-consuming when they are applied over a large domain. In this study, a methodology was developed and evaluated based on available hydrogeology and hydrometeorology data and statistical and machine learning techniques to map the groundwater salinity in the southern coastal aquifer of the Caspian Sea. First, variables affecting groundwater salinity (aquifer transmissivity, distance from the sea, the mean annual precipitation, the mean annual evaporation, elevation, and the depth to the water table) were determined, and the dataset was randomly divided into three subsets of training, testing, and verification. Next, the relationship between groundwater salinity and its controlling factors was established using three methods namely extreme gradient boosting (EGB), deep neural network (DNN), and multiple linear regression (MLR), and the models were evaluated by comparing the measured values and the predicted values using the statistical criteria (R-squared, Nash–Sutcliffe efficiency (NSE), and normalized root-mean-square deviation (NRMSD)). Finally, the optimum model was applied to the set of known input variables to map the spatial variation of the groundwater salinity across the entire southern coastal plain of the Caspian Sea, and the final map was verified using the verification subset. Results showed that consideration should be given to the EGB method, considering its higher performance on the testing subset (R-squared 0.89, NSE 0.87, NRMSD 0.45). In-depth analysis of the variables showed that the aquifer transmissivity is the most crucial parameter affecting groundwater salinity in the region. The adopted approach could potentially be used for groundwater management purposes in the study area and similar settings elsewhere.

Assessment of water-induced soil erosion as a crucial part of soil conservation plans is costly and time-consuming when applied to an extensive area. In this study, we propose a methodology based on recording the annual soil erosion in a portion of the study area using erosion pins and assessing the spatial distribution of soil erosion for the entire area using machine learning techniques. First, soil erosion pins were installed, and the amount of soil loss in each pin was recorded. The controlling factors of soil erosion (percentage of vegetation canopy, curvature, slope degree, slope length, percentage of sand, percentage of silt, and percentage of clay) were determined, and the dataset was divided into training (75% of the data) and testing (25% of the data) subsets. Three machine learning algorithms, namely boosted regression trees (BRT), deep learning (DL), and multiple linear regression (MLR), were employed to identify the relationship between soil erosion and its controlling factors. Then, the methods were evaluated by comparison between the predicted and observed values on the testing subset using statistical coefficients including coefficient of determination (R-squared), normalized root mean squared error (NRMSE), and Nash-Sutcliffe efficiency (NSE). Results show that the BRT outperformed the other algorithms in the assessment of the annual soil erosion (R-squared: 0.92, NSE: 0.9, and NRMSE: 0.32). Finally, the optimal algorithm (BRT) was selected to estimate the spatial distribution of soil erosion across the entire study area, and the final erosion map was verified using additional verification pins.

Monitoring temporal variation of streamflow is necessary for many water resources management plans, yet, such practices are constrained by the absence or paucity of data in many rivers around the world. Using a permanent river in the north of Iran as a test site, a machine learning framework was proposed to model the streamflow data in the three periods of growing seasons based on tree-rings and vessel features of the Zelkova carpinifolia species. First, full-disc samples were taken from 30 trees near the river, and the samples went through preprocessing, cross-dating, standardization, and time series analysis. Two machine learning algorithms, namely random forest (RF) and extreme gradient boosting (XGB), were used to model the relationships between dendrochronology variables (tree-rings and vessel features in the three periods of growing seasons) and the corresponding streamflow rates. The performance of each model was evaluated using statistical coefficients [coefficient of determination (R-squared), Nash–Sutcliffe efficiency (NSE), and root-mean-square error (NRMSE)]. Findings demonstrate that consideration should be given to the XGB model in streamflow modeling given its apparent enhanced performance (R-squared: 0.87; NSE: 0.81; and NRMSE: 0.43) over the RF model (R-squared: 0.82; NSE: 0.71; and NRMSE: 0.52). Furthermore, the results showed that the models perform better in modeling the normal and low flows compared to extremely high flows. Finally, the tested models were used to reconstruct the temporal streamflow during the past decades (1970–1981).

The focus of this study is to locate shallow groundwater (SGW) occurrences in arid lands using the Western Desert (WD; area: ∼680,000 km2) of Egypt as a test site. The SGW in the study area originated from paleo-precipitation during previous wet climatic periods. In wet periods, fossil groundwater was at higher levels, ascended along high-angle faults, and discharged at the surface. In contrast, at present, the water levels are lower, and the discharge occurs at near-surface elevations. Spring locations were identified as the dependent variable, while the independent variables included remote sensing–based variables and geomorphological features indicative of current or paleo discharge locations, including elevation, slope, curvature, distance to sapping features, soil moisture, NDVI, radar backscatter coefficient, and brightness temperature. Relationships between SGW occurrences (target) and their controlling factors (independent variables) were established using extreme gradient boosting (XGB), support vector machine (SVM), and logistic regression (LR) methods. The trained models were used to map SGW locations across the entire WD. Findings include the following: (1) the XGB yielded the most favorable result in identifying SGW locations (overall accuracy: 0.93) compared to SVM (overall accuracy: 0.88) and LR (overall accuracy: 0.87); (2) areas with a very high probability of SGW occurrences were found in lowlands and proximal to sapping features; (3) the overwhelming majority of the cultivated lands within the southern and central sections of the WD lie within areas identified as high and very high probability SGW locations; (4) our models identify-two previously unrecognized major SGW occurrences, an eastern zone (EZ; length: 800 km; width: 9 to 80 km; area: 43,000 km2) and an east–west trending northern zone (NZ) centered over the Qattara depression (length: 500 km; width: 200 km; area: 62,150 km2); (5) additional criteria were used to refine the modeled (XGB) SGW distribution (southern and central WD: 43,200 km2 to 23,400 km2; EZ: from 21,300 km2 to 17,400, and NZ: from 62,150 to 30,700 km2) including presence of shallow aquifers to accommodate rising Nubian waters, Nubian water salinity (fresh to brackish), and low to moderate thickness (<1 km) of post-Nubian successions that rising waters interact with. The techniques are cost-effective and efficient and could be readily applied to large sectors of Saharan Africa and Arabia, whose landscape and fossil aquifers bear many resemblances in their geologic, climatic, and geomorphic characteristics to the Nubian Sandstone Aquifer System.