Climate-Induced Rice Root-Zone Salinity in the Tidal Swamp Rice Production Zones of the Gambia

Abstract

This research focused on the effects of soil salinity and climate variability on rice cultivation and agricultural productivity in the Lower River Region (LRR) and Central River Region South (CRRS) of The Gambia. The research employed a multidisciplinary approach, utilizing machine learning models, satellite reflectance, soil salinity (ECe) data, and statistical analyses to assess soil salinity dynamics, its impact on rice yield, and potential adaptive strategies for sustainable agriculture. The Random Forest model demonstrates strong predictive capability using Landsat 8 reflectance values. Results indicate significant changes in salinity-affected areas in LRR from 66.34% of the total territory in 2014 to 72.15% in 2021, marking a substantial 9.56% increase. The results indicated a contraction in the combined salinity-affected area in CRRS, dropping from 93.46% of the region’s land in 2014 to 80.29% in 2021, translating to a significant percentage decrease of -13.94 % during the same timeframe. The seasonal variability of root-zone salinity and its impact on rice yield shows that the soils predominantly exhibit acidity, with pH values ranging from 4.0 to 5.8. The GLM offers robust goodness of fit with an impressive R-squared value of 0.98. Tonitaba and Kudang express opposing coefficients on soil ESP, with t estimates (0.84994) and (-1.18268), respectively. The results indicated a statistically significant negative relationship between ESP and sampling time in August, indicating seasonality in salinity conditions. Sodium concentration (Na meq/100g) emerges as the most influential predictor of ESP, soil pH (pHsoil), calcium (Ca+2 meq/100g), potassium (K meq/100), and magnesium (Mg meq/100) manifest a negative association with ESP. The results show a substantial variation in rice yield across these study fields, with Mandina and Tonitaba yielding 30% less than Kudang. Maximum temperature exhibited a robust negative correlation with rainfall in both regions, while minimum temperature showed a medium to strong positive relationship with rain. Linear regression models explained 67% of rice yield variability in LRR and 64% in CRRS, highlighting the substantial influence of year-to-year changes. Random Forest prediction models revealed a 24% change between predicted and actual historical yield in LRR and a 54% change in CRRS. Maximum temperature significantly impacted rice yield, followed by rainfall.