Assessment of water productivity of wheat and maize under changing climate in the northwestern Bangladesh using remote sensing data and crop modeling

Abstract

Bangladesh, specifically the northwestern region, is facing twin challenges, namely, food and water security,which are pressing now and likely to increase in the future under the changing climate. Increasing crop water productivity (WP) could be a keystrategy to address both challenges. In this context, this study proposesto delineate wheat and maizecrop fields, to generate spatial WP at present, and to simulate WP of wheat and maize under changing climate along with adaptation strategies in northwestern Bangladesh.Northwestern Bangladesh,including eight districts of Rangpur division,has been selected as the study area. Both primary and secondary data such as the in-situhyperspectral signature of crops using Spectroradiometer, Landsat 8 OLI andSentinel-2 MSI satellite images, weather parameters from BMD stations and GCMs, actual evapotranspiration (ETa) from Lysimeter, crop management data,and GPS survey data from the farmers’ fields over the study area have been collected to accomplish this study. An innovative algorithm has been developed first to delineate wheat fields in a pilot areausing Sentinel-2 satellite data with the help of in-situ hyperspectral data. Here, Sentinel-2 satellite images have been used due to its higher spatial (~10m) and temporal resolution (5 days revisit time). Training datasetshave been prepared based on threshold values of wheat phenology that acquired from the spectral signature of both in-situ and satellite data.A rule-based classification technique has been applied thento map wheat using Sentinel-2 imageries based on the developed training datasetsfor Kaharole Upazila of Dinajpur district. This tested rule-based algorithm has been used further to delineate wheat and maize crop fields over the entire study area. Secondly, WP parameters namely crop yield and ETahave been estimated using Landsat 8 satellite data. Here, Landsat 8 satellite data has been used as it has the Thermal Infra-Red Sensors (TIRS),which are essential for ETa estimation. Based on yield data from the farmer's field and NDVI value generated from the same fields, a regression model has been developed, which has been applied to obtain crop yield maps from the NDVI maps. On the other hand, remote sensing-based well-tested robust algorithm SEBAL has been used to estimate seasonal ETa using Landsat 8 images and ground information. Cloud free single-date images correspond to wheat and maize growth stages such as initial, development, mid-season, and late-season have been used to estimate seasonal ETa. SEBAL-estimated ETa has been validated by newly developed (under this study) Lysimeter-measured ETa. Wheat and maize crop mapshave been then masked to get crop-specific yield and ETa map. Finally, the WP maps of wheat and maize have been produced by the mathematical operation of the yield and ETa maps.On the other hand, AquaCrop, a widely used WP model,has been used to predict future water productivity and yield. However, before any application, the model has been calibrated and validated by comparing with observed data obtained from field experiments. Using this calibrated model, yields and WPhave beensimulated forthe near future (2020-2039), mid future (2040-2059), and far future (2080-2099).Water productivities (WPs) of wheat and maize in the Dinajpur region have been generated based on 11 ensembles mean GCMs climate data under the RCP 8.5 scenario. This high-end scenariohas beenchosen due to a much higher signal to noise ratio for the detection of significant changes in the climate system. Future yield and WPsare then compared to the baseline period (2000-2019) to assess the impact of climate change. Finally, adaptation strategies such as shifting of sowing date and introducing virtual heat-tolerant variety have been selected to reduce the adverse impacts of climate changes on yield and WP of wheat and maize in the future. The first step is to develop an algorithm for wheat and maize classification through a rule-based classification technique. Results show that the NDVI, EVI, SAVI, and RVI-based map accuracy are 83.33 %, 85.19 %, 81.48%, and 72.22%, respectively in the pilot area while compared to ground truth data, indicating the satisfactory (p<0.05) classification results. The Kappa coefficient (~0.80) indicates a strong agreement between classified maps and the reference data. This algorithm-estimated wheat and maize area coverage are110,111 ha and 175,564 ha, respectively,overthe entire study area. The NDVI-based overall accuraciesof the entire study area are 82.30% and 80.15% for wheat and maize, respectively, while compared to ground truth data. The results show that Thakurgaon and Dinajpur are the highest wheat and maize cultivated districts, respectively during the 2017-2018 Rabiseasons. In the case ofthe numerator of WP, i.e. crop yield(Y) estimation, the NDVI-yield modelshave beencalibrated and validated for wheat and maize, which show an acceptable degree of dispersion between observed and estimated values. This model-generated wheat yield map shows the variation in its ranges over the study area. Results show that Thakurgaon and Panchagorh are higher wheat yielding districts, whereas Lalmonirhat and Dinajpur are higher maize yielding districts amongthe study area. The average wheat and maize yields of the study area are3.51, and 9.38 t ha-1, respectively, which are close to the official reported data. In the case of the denominator (ETa) of WP estimation, results show that SEBAL estimated wheat and maize ETa on the image dates are close to ETa estimated by the newly designedLysimeter. The average seasonal ETa for wheat and maize are found as 253 mm and 358 mm, respectively over the study area. In the study area, the estimated highest and lowest ETa demandsare found in Thakurgaon and Panchagorh district, respectively. The estimated WP of wheat and maize are ranged between 1.33-1.46 kg m-3 and 2.52-2.70 kg m-3 with an average 1.39 kg m-3 and 2.62 kg m-3, respectively among the eight study districts. These estimates fall between the ranges of estimates of global wheat and maize WP. Results show that Panchagorh and Thakurgaon districts are the highest water productive for the wheat, whereas Gaibandha and Nilphamari show the minimum level of the WP. However, Panchagorh and Lalmonirhat are the highest water productive for maize, whereas Kurigram, Nilphamari, and Gaibandha show the minimum level of the WP.Among the Y-ETa, WP-ETa, and WP-Y relationships, WP-Y shows the highest correlation r2=0.73 and r2=0.84 for wheat and maize, respectively, which indicates the higher WP observed mainly due to high yielding capacity. The simulated yields using the observed climate data for the baseline period (2000-2019) are found as 3.90 t ha-1 and 9.92 t ha-1 for wheat and maize, respectively, for the Dinajpur region. This study shows that wheat yield could be reduced by 11.59%, 20.92%, and 39.88%, and maize yield could be reduced by 6.38%, 12.22%, and 28.75% in the near future (2020-2039), mid-future (2040-2059), and far-future (2080-2099), respectively. It has been observed that both projected Tmax and Tmin could be increased significantly during the development stage and at the flowering stage of wheat (February) and maize (February-March), which could reduce wheat and maize yield. ETa of wheat and maize during the baseline period (2000-2019) have been found as 271 mm and 387 mm, respectively. This ETa could be reduced by 3.83%, 6.05%, and 3.14% for wheat and reduced by 9.43%, 13.08%, and 20.40% for maize in the three future time slices, respectively. The simulation of WP during the base period (2000-2019) shows that average WP for 1.39 and 2.62 kg m-3 for wheat and maize, respectively. Results show that wheat WP could be reduced gradually by 8.43%, 21.36%, and 33.28% in the three future time slices, respectively. However, maize WP could be increased by 3.77%, 1.92% in the near and mid future whereas it could be decreased by 10.31 % in the far future. There are inconsistent variations in the projected WP due to the variations in both yield and actual evapotranspiration. In the case of wheat, ETa could be decreased similarly proportional to the yield along with the three future time slices, and eventually, WP could be decreased in future three-time slices. In the case of maize, it has been observed that ETa could be decreased more than that of yield during near and mid-future, whereas it could be decreased relatively lessif compared to the yield reduction in the far future, and eventually WP could be decreased. Shifting sowing date as an adaptation option revealed thatwith a 10-day backwards shiftfromthe current optimum sowing date (25 November), wheat WP could be increased by 11.39%, 11.94 %, and 18.42% for three future periods, respectively if compared to the WP without adaptation measure. However, a 20-day backwards shift (5 November) from the current showing date, WPcould increase by 19.56%, 38.74%, and 46.95% for three future periods, respectively. On the other hand, a 10-day forward shift (5 December),WP could be decreased by 27.06%, 17.72 %, and 38.99% for three future periods, respectively if compared with WP without adaptation measures. Similar results have been observed for maize WP, i.e. maize WP could be increased 13.15%, 15.74 %, and 4.23% if sowing dates are shifted10-days backwards and WP could be increased 19.63%, 27.56 %, and 23.39%, respectively if sowing dates are shifted 20-days backwards. WP could be decreased by20.24%, 1.11%, and 23.94%, respectively if the sowing dateisshifted forwardfor three future periodscompared to the WP without adaptation measures. Early sowing of seeds could help to escape the critical stages (flowering) of the higher temperature stress and could increase the yields and WPs significantly in the future time slices. In the case of introducing virtual heat-tolerant variety, results show that both the yield and WP of wheat and maize could be increased significantly along with the future time slices if compared to the benchmark wheat and maize varieties.These results suggest that adjusting sowing dates and introduce heat-tolerant variety might be a powerful tool for mitigating the effect of global warming. Finally, stakeholders, policy, and decision-makers can use the information obtained from this study to define priority areas, set goals for improvement, and justify the type of investment or measures for improving water productive agriculture at present and in the future under changing climate.