Analyzing the Impacts of Climatological Variables on Rainfall using ARIMA and Multi Linear Regressions models: A case study of Algiers

As humanity races for a solution to its making, largely, the crisis is degenerating and becoming vicious- solving one problem creates another. For instance, the call for an energy transition from oil to hydro-power partially tackles the emission issue but offsets marine ecosystem stability. The most significant damage of the humanly induced climate crisis has been observed in the water resource sector. Drier regions are becoming desserts while flood takes over wetter places. The hydrological cycle is most affected as the water crisis surges globally. Precipitation patterns are affected and vary from region to region. Understanding the underlying variables that influence the irregularity is required to tackle this crisis. The current report summarizes several methods used to predict precipitation for the North African city of Algiers, Algeria. The ARIMA and the Multiple Linear Regression models were used to analyze the relationship between rainfall and other parameters such as temperature, runoff, vapor pressure, windspeed, evapotranspiration, soil moisture, etc.). The probability plot was used to determine the best distribution function for precipitation. The results suggest that solar radiation, soil moisture, runoff, and maximum temperature affect precipitation more, recording an R-square value of 35.16% when using the MLR compared to other explanatory variables. For the ARIMA model, The P-value is less than the significance level (p≤0.05). This suggests the coefficient for the autoregressive term is statistically significant; therefore, the term should be maintained in the model. Using the Modified Box-Pierce (Ljung-Box) Chi-Square statistic, it was found that the p-values are all less than the significance level of 0.05 and significant correlations exist for the autocorrelation function of the residuals. Thus, we can conclude that the model meets the assumption that the residuals are dependent. For the probability plot, the 3-Parameter gamma produced the most negligible AD value (11.008). This makes the 3-parameter gamma the best fit for the distribution function.