The influence of floods-recharged soil moisture on tree biochemical and biophysical features in Mbire and Muzarabani semi-arid lands in northern Zimbabwe.

Abstract

In dry lands characterised by scarce precipitation, floods uniquely facilitate soil moisture recharge through deep infiltration. However, the ecological influence of flood-recharged soil moisture (FRSM) on tree growth in semi-arid lands remains poorly understood. To address this knowledge gap, the aim of the study was to assess the influence of floods-recharged soil moisture on tree biochemical and biophysical characteristics in a semi-arid landscape of northern Zimbabwe which was split into the following specific objectives. Firstly, the study sought to assess the utility of remote sensing indices in understanding the nexus between FRSM with tree biochemical and biophysical features in tropical semi‐arid floodplains. Secondly, the study sought to model flood extent mapping in semi-arid floodplains in northern Zimbabwe using Sentinel-1 SAR data. Thirdly, the study sought to assess the utility of deep learning long short-term memory autoencoder algorithm applied on Sentinel-1 SAR data to model FRSM features in semi-arid floodplains in northern Zimbabwe. Fourthly, the study sought to assess the influence of FRSM on short-term biochemical properties of Z. mauritiana tree in semi-arid floodplain in northern Zimbabwe. Finally, the study sought to evaluate the long-term influence of FRSM on Z. mauritiana tree biophysical characteristics in semi-arid lands in northern Zimbabwe. Therefore, to understand the influence of FRSM on trees, this study compared a multipurpose Musawu (Shona) or Jujube (English) (Ziziphus mauritiana) tree species located inside against outside of Mbire and Muzarabani flood-prone semi-arid lands of northern Zimbabwe. The study used an experimental design, flooded constituted the experimental set-up while non-flooded the control set-up for soil moisture, leaf chlorophyll content and tree size variables to enable respective hypotheses of influence of FRSM were tested. Primary and secondary data were collected through fieldwork and downloaded from online repository respectively. Field data measurements included flood extent boundary, tree leaf chlorophyll content measured using chlorophyll meter, tree height measured using a Haglöf Vertex Laser Geo hypsometer, diameter at breast height (DBH) measured using diameter tape and canopy diameter estimated using tape measure respectively. Since floods often occur during cloudy conditions, passive remotely sensed secondary data are commonly inapplicable, hence this study undertook flood extent mapping that used temporal synthetic arperture radar (SAR) data. Firstly, to accurately map flood extent, we innovated an Ensemble of Scenarios Pyramid which is based on change detection and thresholding, utilising a normalized difference flood index (NDFI) framework with Sentinel-1 SAR data. The flood extent map was necessary to spatially discriminate flooded from non-flooded soil moisture, short-term leaf chlorophyll content and longterm tree biophysical characteristics. Secondly, the study innovated a hybrid deep learning Long Short-Term Memory-Autoencoder framework that used temporal Sentinel-1 SAR data to model soil moisture anomaly, lag and memory features compared between flooded and non-flooded locations. Thirdly, short-term post-flood leaf chlorophyll content for flooded tree samples were compared with non-flooded samples using machine learning. Finally, the long-term influence of FRSM on biophysical characteristics was evaluated using canopy diameter, tree height and diameter at breast height (DBH) as proxies of tree growth. The novel Ensemble of Scenarios Pyramid produced a more accurate flood extent map as compared to all the base NDFI scenarios using six metrics (overall accuracy = 93.204%; F1-score = 0.927; Matthews’ correlation coefficient = 0.871; Recall = 0.870; Intersect over Union = 0.865; Kappa = 0.864). The Long Short-Term - AutoEncoder detected positive (wet) FRSM anomaly, with one week lag that occurred starting from the second of February 2017, and initiating a 70-day soil moisture memory inside the flooded as compared to the non-flooded zones. On the shortterm, Random Forest machine learning rejected 43 as unimportant and accepted 16 variables as important to determine leaf chlorophyll content. The flood-related three spatial variables namely distance to river, distance to floodplain and floodplain location were ranked as the most important predictors to determine the short-term FRSM influence on Z. mauritiana leaf chlorophyll content. Thus, on the short-term, the flooded zone (hence FRSM) influenced higher leaf chlorophyll content as compared to the non-flooded zone. On the long-term, flooded trees inside the floodplain were significantly bigger compared to those in non-flooded areas in canopy diameter (p<0.001), tree height (p <0.05) and DBH (p<0.001) using the Levene’s test. Similarly, the Mann-Whitney-Wilcoxn test also showed significant differences in canopy size (p < 0.001), tree height (p < 0.001) and DBH (p < 0.01). Inside flooded locations mapped with the novel Ensemble of Scenarios Pyramid, the innovated Long Short-Term-AutoEncoder modelled that FRSM clearly depicted a minimum of two months longer soil moisture memory than outside. Ultimately, this longer FRSM memory supply soil water for longer to flooded trees than in non-flooded areas, hence the higher leaf chlorophyll content for the former rather than the latter. Consequently, in the long-term, flooded trees inside the floodplain grew bigger than outside as the three selected growth biophysical proxies confirmed using non-parametric tests of differences. In conclusion, the FRSM positively influences soil water recharge in semi-arid lands, which ultimately supports flooded tree growth, thereby confirming the deep infiltration concept noted in the literature review. These results uniquely inform soil moisture and related Z. mauritiana tree management plans in semi-arid lands in view of the current deforestation triggered by its overexploitation by both humans and animals owing to its multiple uses.