Abstract: Abstract: Collapsed gully is a kind of soil erosion in South China threating agriculture production and ecological environment due to its considerable sediment yield. Therefore, the prediction and estimation of erosion amount play a significant role in its prevention and control. However, collapsed gullies normally cover a relatively smaller area and eroded more severely than other kinds of soil erosion, which made existing methods and approaches hardly to apply on it. In this research, cellular automation-Markov (CA-Markov) model and Australian National University digital elevation model (ANUDEM) are employed to estimate and predict the erosion amount of collapsed gully. The former is taken to simulate the transition of collapsed gully elevation classes, and the latter is taken to spatial interpolate on the simulation results of collapsed gully elevation classes. Based on the two processes, the continuous elevation data of collapsed gully can be simulated and predicted. Then the erosion amount is computed according to the prediction result and base time data. To test this approach, a collapsed gully located in the Longmen Town, Anxi County, Fujian Province is chosen as the case study. And it contains two parts, which are the confirmation of simulating accuracy and the prediction under 3 types of scenes for annual erosion amount. The results show that the modified CA-Markov model performances well on the simulation of collapsed gully elevation classes, which resulted from a lower area error between simulation results and actual value (1.47%), and a higher Kappa coefficient of 0.958. The goodness of fit (R2) of the interpolation result using ANUDEM is 0.998 7 and its mean error (ME) is 0.27 m, which means ANUDEM is good at the integral restoration of collapsed gullies terrain but not sufficient at details. As a consequence, two types elevation at the base time are chosen to estimate the erosion amount, one is original DEM, and the other one is the elevation data through classes division and interpolated processing. The erosion amount calculated from the second type of base data performances difference rates of 17.55% and 8.66%, which are much lower than the one derived from the first type of base data. Thus, the second type of base data is accessible for computing collapsed gully erosion amount, and the huger the real erosion amount, the higher the accuracy. After the confirmation of modeling accuracy, 3 types of rainfall scenes were set based on the historical records of rainfall, which are normal years, drought years and rainy years. The annual erosion amounts of case collapsed gully under the 3 types of rainfall scenes are 824.69, 731.03 and 924.57 m3, and the difference of the two extreme rainfall scenes is 193.54 m3, which means that the difference of erosion amount between different rainfall scenes should be taken into account when building a sediment storage dam to control the damage of a collapsed gully. The methods proposed in this research can not only provide a novel way to study the erosion process and erosion amount of collapsed gully, but also supply a reasonable reference for the prevention and control of collapsed gully erosion.